<%BANNER%>

Hydro-Climatic Influences of El-Nino/Southern Oscillation on Nutrient Loads in the Southeast United States

Permanent Link: http://ufdc.ufl.edu/UFE0041406/00001

Material Information

Title: Hydro-Climatic Influences of El-Nino/Southern Oscillation on Nutrient Loads in the Southeast United States
Physical Description: 1 online resource (266 p.)
Language: english
Creator: Keener, Victoria
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2010

Subjects

Subjects / Keywords: climatology, elnino, enso, hydroclimatology, hydrology, nutrients, simulation, spectral
Agricultural and Biological Engineering -- Dissertations, Academic -- UF
Genre: Agricultural and Biological Engineering thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: As global climate change becomes more of a problem, it is crucial to understand regional variability. In the southeast U.S.A., natural resource managers wish to reduce climate associated risks. This research explores the relationship between the El-Nino/Southern Oscillation (ENSO), a major driver of global climate variability, with hydrology and water quality in two watersheds. In basin S-191 of Lake Okeechobee, we quantified relationships between 36 years of WAM simulated phosphorus (P) loading and categorical measures of ENSO. Results showed ENSO strongly affects seasonal and monthly P runoff, with significant loading in spring of El Nino and summer of La Nina years. Greater P load in certain months was consistent with greater precipitation, with large flow and nutrient flushes following drought. Observed data from the Little River Watershed (LRW) was used with wavelet analysis to quantify the significance of a teleconnection between the NINO 3.4 index and precipitation, stream flow, nitrate concentration and load. Areas of high power and inter-annual variability manifested in the 3-7 year ENSO periodic signal. To explain the powerful relationship between ENSO and stream flow, a calibrated SWAT model of the LRW mechanistically identified how the power of a climate signal an be increased through natural processes. Using SWAT and wavelet analysis, we found that the presence of a confining layer in the LRW increased the groundwater/baseflow signal such that the ENSO signal in flow was more significant than the chaotic precipitation signal. Finally, significant 3-7 year reconstructed components from the wavelet analyses were extracted and used to create a monthly vector time series model that more accurately forecasts 1-3 month NO3 loads than time-domain signals only. IPCC reports have concluded that climate variability and extreme events will be more common in the future. Research that focuses on understanding and predicting the effects will be increasingly helpful for making robust management decisions in an uncertain world. Models of water quality and ENSO can help stakeholders effectively manage their risk in the near future, and are a step towards faster integration of climate information into daily decision making.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility: by Victoria Keener.
Thesis: Thesis (Ph.D.)--University of Florida, 2010.
Local: Adviser: Jones, James W.

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2010
System ID: UFE0041406:00001

Permanent Link: http://ufdc.ufl.edu/UFE0041406/00001

Material Information

Title: Hydro-Climatic Influences of El-Nino/Southern Oscillation on Nutrient Loads in the Southeast United States
Physical Description: 1 online resource (266 p.)
Language: english
Creator: Keener, Victoria
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2010

Subjects

Subjects / Keywords: climatology, elnino, enso, hydroclimatology, hydrology, nutrients, simulation, spectral
Agricultural and Biological Engineering -- Dissertations, Academic -- UF
Genre: Agricultural and Biological Engineering thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: As global climate change becomes more of a problem, it is crucial to understand regional variability. In the southeast U.S.A., natural resource managers wish to reduce climate associated risks. This research explores the relationship between the El-Nino/Southern Oscillation (ENSO), a major driver of global climate variability, with hydrology and water quality in two watersheds. In basin S-191 of Lake Okeechobee, we quantified relationships between 36 years of WAM simulated phosphorus (P) loading and categorical measures of ENSO. Results showed ENSO strongly affects seasonal and monthly P runoff, with significant loading in spring of El Nino and summer of La Nina years. Greater P load in certain months was consistent with greater precipitation, with large flow and nutrient flushes following drought. Observed data from the Little River Watershed (LRW) was used with wavelet analysis to quantify the significance of a teleconnection between the NINO 3.4 index and precipitation, stream flow, nitrate concentration and load. Areas of high power and inter-annual variability manifested in the 3-7 year ENSO periodic signal. To explain the powerful relationship between ENSO and stream flow, a calibrated SWAT model of the LRW mechanistically identified how the power of a climate signal an be increased through natural processes. Using SWAT and wavelet analysis, we found that the presence of a confining layer in the LRW increased the groundwater/baseflow signal such that the ENSO signal in flow was more significant than the chaotic precipitation signal. Finally, significant 3-7 year reconstructed components from the wavelet analyses were extracted and used to create a monthly vector time series model that more accurately forecasts 1-3 month NO3 loads than time-domain signals only. IPCC reports have concluded that climate variability and extreme events will be more common in the future. Research that focuses on understanding and predicting the effects will be increasingly helpful for making robust management decisions in an uncertain world. Models of water quality and ENSO can help stakeholders effectively manage their risk in the near future, and are a step towards faster integration of climate information into daily decision making.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility: by Victoria Keener.
Thesis: Thesis (Ph.D.)--University of Florida, 2010.
Local: Adviser: Jones, James W.

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2010
System ID: UFE0041406:00001


This item has the following downloads:


Full Text

PAGE 1

HYDRO-CLIMATIC INFLUENCES OF EL -NINO/SOUTHERN OSCILLATION ON NUTRIENT LOADS IN THE SOUTHEAST UNITED STATES By VICTORIA WHITWORTH KEENER A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2010 1

PAGE 2

2010 Victoria Whitworth Keener 2

PAGE 3

To my wonderful grandmother, Yue Mei-Fong, who continues to be a woman ahead of her time 3

PAGE 4

ACKNOWLEDGMENTS Completing a Doctoral dissertation is as much an exercise in perseverance as one of learning and research. Of the multitude of times I was discouraged, depressed, or despairing of ever finishing, there have been people that empat hized with me, then promptly reminded me that not only could I do it, in fact, I had to do it, and should probably get back to work and stop complaining. I would like to thank my advisor, Dr. James Jones, for his encouragement, intelligence, generosity, and kindness. I can not imagine a more positive mentoring experience with better opportunities than the ones he has provided fo r me. I also thank my Committee members, Drs Wendy Graham, Greg Kiker, Ken Campbell, and Mike Annable for their thoughtful input and help in completing this research. For their graciousness in both collecti ng and sharing massive amount s of data with me, I thank the members of the USDA-A RS Southeast Watershed Resear ch Laboratory in Tifton, Georgia, without whom most of this research would not have been possible. The friends I made in graduate school ar e for-a-lifetime, and I would have not been able to make it through without their understanding and commiseration. Special thanks go out to my Tanks In Series band -mates David Kaplan and Dan Tankersley, with whom I made beautiful music, and released a lot of stress by learning how to play the drums. Other creatures and molecules that made graduate school endurable include my cats, Odette (R.I.P), B un-Bun, and Frey, and caffeine. A significant part of maintaining my happi ness and sanity is attributable to my husband, Keith. Although not a day passed near the very end that I did not envy his already being done with his Ph.D., his love, c onfidence, and laid-back attitude was a large part of what allowed me to get th rough Graduate School in (mostly) one piece. 4

PAGE 5

Most importantly, I thank my par ents, Tom and Nanda Keener, for their unwavering love, support, inspiration, happine ss, and belief in me. This dissertation would not have been possible without the traits I inherited and learned from each of them. From my father: logic, scientific curiosity, a love of nature and the outdoors, the desire to question the dominant paradigm, and impatience. From my mother: patience (to balance the impatience inherited from my father), perseverance, assertiveness, the ability and inclination to understand that with which you disagree, and the power to stay awake and alert late into the night. I have a bad habit of forgetting to thank them, which is because they are such wonder ful parents that it is all too easy to take them for granted. Much like breathing or having a hear tbeat, they are always there, keeping me going, and for that I love and thank them with all my heart. 5

PAGE 6

TABLE OF CONTENTS page ACKNOWLEDG MENTS..................................................................................................4 LIST OF TABLES............................................................................................................9 LIST OF FI GURES ........................................................................................................10 ABSTRACT ...................................................................................................................13 CHAPTER 1 INTRODUCTION AND LI TERATURE REVIEW.....................................................15 Motivating Issues....................................................................................................15 Global Climate Change and Water S upply .....................................................15 Climate Indices and the El-Ni o/Southern Osc illatio n....................................17 Global ENSO effects...........................................................................20 Paleo-climate and fu ture of ENSO .......................................................22 Climate and Agriculture in t he Southeast United Stat es.................................25 Objectiv es......................................................................................................27 Literature Review....................................................................................................28 Agricultural Polluti on: Nutri ents......................................................................28 Water Management Policie s in the S outheast................................................31 The Watershed Restoration Act (WRA) and TMDL standards............31 Best Management Prac tices (B MP).....................................................33 Simulation of Natu ral Syst ems.......................................................................36 Global circulation models and ENSO..................................................39 Agro-hydrologi cal mode ls....................................................................40 Time series model s.............................................................................42 Spectral analysi s and models..............................................................44 2 EFFECTS OF CATEGORICAL EL NIO/SOUTHERN OSCILLATION ON SIMULATED PHOSPHORUS LOAD ING IN SOUTH FLORIDA.............................49 Introducti on.............................................................................................................49 Data and Me thods..................................................................................................53 The Lake Okeechobe e Watershed.................................................................53 History and hy drology ..........................................................................54 Geology an d soils ................................................................................56 Ecology and wildlife .............................................................................57 Management and rest oration...............................................................58 Data ............................................................................................................59 Watershed Assessment Model (W AM)...........................................................61 Analysis and Statisti cs....................................................................................64 Precipitation and stream flow...............................................................64 6

PAGE 7

Phosphorus concentration and loading: JMA ENSO classification......64 Phosphorus concentration and loadi ng: M-ENSO clas sification..........69 Result s....................................................................................................................70 Monthly Precipitatio n and ENSO Phase.........................................................70 Monthly Simulated Stream Flow and ENSO Phase.......................................72 Simulated Monthly P Concent ration and ENSO Phase..................................73 Simulated Monthly P Load and ENSO Phase................................................74 Simulated Seasonal P Load, Concentra tion, Stream flow, and JMA ENSO Phase..................................................................................................76 February-March-Apri l...........................................................................78 May-JuneJuly.....................................................................................79 August-Septem ber-Oct ober.................................................................80 November-Decem ber-Januar y............................................................81 M-ENSO Statistics (ANOVA) of M onthly Precipitation, Flow, P Concentration and Load ......................................................................82 Summary and Di scussion.......................................................................................83 3 EL-NIO/SOUTHERN OS CILLATION (ENSO) INFLUENCES ON MONTHLY NO3 LOAD AND CONCENTRATION, ST REAM FLOW AND PRECIPITATION IN THE LITTLE RIVER WATERSHE D, TIFTON, GE ORGIA................................102 Introducti on...........................................................................................................102 Data and Me thods................................................................................................106 Field Site: Little Riv er Waters hed.................................................................106 History and moni toring .......................................................................107 Soils and geology ..............................................................................109 Climate and hydr ology.......................................................................110 Land use, management and policie s.................................................111 Little River Wate rshed Data .........................................................................113 Wavelet Anal ysis..........................................................................................115 Cross Wavelet and Coher ence Transfo rms.................................................118 Cross-Correlation Time Series Anal ysis.......................................................119 Results ..................................................................................................................120 Wavelet Anal ysis..........................................................................................120 Cross-Wavelet Analysis...............................................................................124 Wavelet Coherence Analysis .......................................................................125 Cross Correlation Analysis...........................................................................127 Summary and Di scussion.....................................................................................128 4 INTEGRATION OF ENSO SIGNAL POWER THROUGH HYDROLOGICAL PROCESSES IN THE LITTLE RIVER WATERS HED..........................................146 Introducti on...........................................................................................................146 Data and Me thods................................................................................................150 Field Site and Data: The Li ttle River Wa tershed..........................................150 SWAT Model Hydrologic Cali bration and Vali dation....................................151 Water Budget and Exc eedance Curves.......................................................156 7

PAGE 8

Wavelet Anal ysis..........................................................................................157 Results ..................................................................................................................158 Observed Precipitation and Stream Flow..................................................... 158 SWAT Water Budget and Exceedance Curves by ENSO Phase.................159 Univariate Wavelet Analysi s.........................................................................162 Cross Wavelet and Wavelet Coherence An alysis........................................165 Summary and Di scussion.....................................................................................167 5 AN ENSO BASED MULTIVARIATE TIME-SERIES MODEL FOR FLOWS AND NITROGEN LOADS IN THE LITTL E RIVER WATERSHED.................................182 Introducti on...........................................................................................................182 Data and Me thods................................................................................................187 Field Site and HydroClimate Da ta...............................................................187 SWAT Nutrient Calibration and CP Implement ation.....................................187 Wavelet Recons truction ...............................................................................190 Wavelet Time Series Model: VARX(p ,s).......................................................191 Forecasting and Un certainty........................................................................194 Tercile Anal ysis............................................................................................195 Results ..................................................................................................................195 VAR, VARX, and W-VAR Time Series M odels.............................................196 Bi-monthly updated W-VAR forecasts and SWAT simu lations.....................201 Summary and Di scussion.....................................................................................203 6 CONCLUSION S...................................................................................................224 APPENDIX A WAVELET ANALYSIS OF LAKE OKEECHOBEE BASIN S-191 MONTHLY OBSERVED PRECIPITATION AND WA M SIMULATED STREAM FLOW, TOTAL P LOAD AND CONCENTRATION...........................................................236 Introducti on...........................................................................................................236 Data and Me thods................................................................................................238 Data: WAM, Basin S-191 and t he NINO 3.4 Index.......................................238 Wavelet Anal ysis..........................................................................................238 Cross-Correlation Analysis...........................................................................239 Results ..................................................................................................................239 Univariate Wavelet Analysi s.........................................................................239 Cross-Wavelet and Wavelet-Coherence An alysis........................................241 Cross Correlation Analysis...........................................................................243 Summary and Di scussion.....................................................................................244 LIST OF REFE RENCES.............................................................................................251 BIOGRAPHICAL SKETCH ..........................................................................................266 8

PAGE 9

LIST OF TABLES Table page 2-1 Example of a 3 2 contingen cy table.................................................................87 2-2 Total P load (kg) anomaly summar y statistics separa ted by month....................87 2-3 Total P load (kg) anomaly summar y statistics separated by JMA and MENSO pha se.......................................................................................................87 2-4 Seasonal contingen cy table for P loads ..............................................................88 2-5 Seasonal contingency t able for P conc entrati ons...............................................88 2-6 Single factor ANOVA results between M-E NSO phas es....................................88 3-1 LRW time series dat a and sources used..........................................................133 4-1 Reference probabilities from all M-ENSO separated exceedance curves........172 5-1 Summary of SWAT performance m easures for calibra tion period....................208 5-2 Time series descriptions and attr ibutes for W-VAR and VARX mode ls............208 5-3 Tercile predictions for 2003 W-VAR, VARX forecasts and SWAT simulation...209 5-4 Summary of time series m odel forecast per formance......................................210 9

PAGE 10

LIST OF FIGURES Figure page 1-1 Global air circul ation patte rns.............................................................................47 1-2 ENSO anomaly mechani sms..............................................................................47 1-3 Physical locations of di fferent ENSO indices......................................................48 2-1 Location of Lake Okeec hobee and basin S-191.................................................89 2-2 Monthly precipitation and simulated P load anomaliesin each ENSO phase.....90 2-3 Box and whisker plots of observe d JMA precipitati on anoma lies.......................91 2-4 Box and whisker plots of observe d M-ENSO precipitat ion anomalies................92 2-5 Box and whisker plots of simula ted JMA stream fl ow anoma lies........................93 2-6 Box and whisker plots of simulat ed M-ENSO stream fl ow anoma lies.................94 2-7 Box and whisker plots of simulat ed JMA TP concentra tion anomal ies...............95 2-8 Box and whisker plots of simulated M-ENSO TP concentration anomalies........96 2-9 Box and whisker plots of monthly JMA P l oad anomalie s...................................97 2-10 Box and whisker plots of m onthly M-ENSO P l oad anomalie s............................98 2-11 Seasonal JMA P l oad anomaly box plots...........................................................99 2-12 Seasonal JMA stream fl ow anomaly box plots.................................................100 2-13 Seasonal JMA TP concent ration anomaly bo x plots.........................................101 3-1 The Little River Watershed (LRW) and sub-basins, Tifton, Georgia.................134 3-2 LRW univariate wavelet power spec tra............................................................135 3-3 LRW cross wave let spectr a..............................................................................137 3-4 LRW wavelet c oherence analysi s.....................................................................138 3-5 Cross-correlat ion analyse s...............................................................................139 3-6 Box and whisker plots of observed M-ENSO NINO 3.4 SST anomalies...........140 3-7 Box and whisker plots of observed M-ENSO precipitatio n anomalies..............141 10

PAGE 11

3-8 Box and whisker plots of observed M-ENSO stream flow anomalies...............142 3-9 Box and whisker plots of observed M-ENSO NO3 concentration anomalies....143 3-10 Box and whisker plots of observed M-ENSO NO3 load anoma lies...................144 3-11 Raw monthly LRW time seri es..........................................................................145 4-1 Observed land-use from 19752003 in the LRW basin K. ................................173 4-2 Cartoon of relevant SWAT simu lated hydrological processes..........................173 4-3 Univariate observed LRW wavelet power spectra............................................174 4-4 Average M-ENSO annual water yield of selected SWAT variabl es..................175 4-5 Average M-ENSO annual water yield as percentage of observed annual average precipitation of selected SWAT va riables ...........................................175 4-6 Probability of exceedance curves in different M-E NSO phases ........................176 4-7 Univariate SWAT wave let power s pectra.........................................................178 4-8 SWAT cross wave let spectr a............................................................................180 4-9 SWAT wavelet c oherence anal ysis..................................................................181 5-1 SWAT simulated total monthly st ream flow in LR W basin K.............................211 5-2 SWAT simulated total monthly tota l nitrogen load in LRW basin K..................212 5-3 Original observed time series and si gnificant reconstructed components........213 5-4 VAR(1) time series model for streamflow and ni trate l oad................................214 5-5 VARX(1) + Observed NINO 3.4 time se ries model for str eamflow and nitrate load................................................................................................................... 215 5-6 VARX(1) + Predicted NINO 3.4 time series model for streamflow and nitrate load................................................................................................................... 216 5-7 W-VAR(5) time series model fo r streamflow and ni trate l oad...........................217 5-8 VAR(1) time series model for str eamflow and total ni trogen load.....................218 5-9 VARX(1) + Observed NINO 3.4 SST ti me series model for streamflow and total nitrogen load.............................................................................................219 11

PAGE 12

5-10 VARX(1) + Predict ed NINO 3.4 SST time series model for streamflow and total nitrogen load.............................................................................................220 5-11 W-VAR(7) time series model for streamflow and total nitrogen l oad................221 5-12 Bi-monthly updated W-VAR(5) time seri es model for str eamflow and nitrate load................................................................................................................... 222 5-13 Bi-monthly updated W-VAR(7) time seri es model for streamflow and total nitrogen lo ad.....................................................................................................223 A-1 WAM simulated S-191 univariat e wavelet power spectra.................................246 A-2 WAM simulated S-191 cross wavelet s pectra...................................................248 A-3 WAM simulated S-191 wavelet coherence sp ectra..........................................249 A-4 WAM simulated S-191 cross-correlation an alysis.............................................250 12

PAGE 13

Abstract of Dissertation Pr esented to the Graduate School of the University of Florida in Partial Fulf illment of the Requirements for t he Degree of Doctor of Philosophy HYDRO-CLIMATIC INFLUENCES OF EL -NINO/SOUTHERN OSCILLATION ON NUTRIENT LOADS IN THE SOUTHEAST UNITED STATES By Victoria Whitworth Keener May 2010 Chair: James W. Jones Major: Agricultural and Biological Engineering As global climate change becomes more of a problem, it is crucial to understand regional variability. In the s outheast U.S.A., natural res ource managers wish to reduce climate associated risks. This research explores the relationship between the ElNio/Southern Oscillation (ENSO), a major dr iver of global climate variability, with hydrology and water quality in two watersheds. In basin S-191 of Lake Okeechobee, we quantified relationships between 36 years of WAM simulated phosphorus (P) loading and categorical measures of ENSO. Results showed ENSO strongly affects seasonal and m onthly P runoff, with significant loading in spring of El Nio and summer of La Nia years. Greater P load in certain months was consistent with greater prec ipitation, with large flow and nutrient flushes following drought. Observed data from the Little River Watershed (LRW) was used with wavelet analysis to quantify the significance of a teleconnection between t he NINO 3.4 index and precipitation, stream flow, nitrate c oncentration and load. Areas of high power and inter-annual variability manifested in the 3-7 year ENSO periodic signal. To explain the powerful relationship betwe en ENSO and stream flow, a calibrated SWAT model of the LRW mechanistically i dentified how the power of a climate signal 13

PAGE 14

can be increased through natural processes. Using SWAT and wavelet analysis, we found that the presence of a confin ing layer in the LRW increased the groundwater/baseflow signal such that the ENSO signal in flow was more significant than the chaotic precipitation signal. Finally, significant 3-7 year reconstructed components from the wavelet analyses were extracted and used to create a monthly vector time series model that more accurately forecasts 1-3 month NO3 loads than timedomain signals only. IPCC reports have concluded that climat e variability and extreme events will be more common in the future. Research t hat focuses on understanding and predicting the effects will be increasingly helpful for making robust management decisions in an uncertain world. Models of water quality and ENSO can help stakeholders effectively manage their risk in the near future, and are a st ep towards faster integration of climate information into daily decision making. 14

PAGE 15

CHAPTER 1 INTRODUCTION AND LI TERATURE REVIEW Motivating Issues Global Climate Change and Water Supply Global climate change has been called the socio-economic, environmental, and energy challenge of our lifetime. The possible consequences of a changing climate are both far-reaching and severe, making resear ch that can help us better understand, adapt to, mitigate, or stop and reverse it s consequences a crucial part of modern environmental science. Warming of the aver age global climate ov er the past century has been proven unequivocally [ IPCC et al. 2007], and because the hydrological cycle is linked closely to variations in climate, citizens and water resource managers both will have to deal with new challenges associated with both water quantity and quality. Over the past several decades, research into global warming has revealed significant changes in precipitation patte rns, weather extremes su ch as floods, droughts and storms, snow-pack duration and amount, increas ed evaporation, increased wildfire risk, and changes in soil moisture and runoff [ Mote et al. 2005; Stewart et al. 2005; Westerling et al. 2006]. These relatively sudden changes make it increasingly difficult to efficiently adapt current anthropological practi ces such as large-scale agriculture and maintaining a municipal water supply, while continuing to provide clean, safe drinking water and adequate food to an ev er increasing population. These systemic changes in the hydrological cycle alter global agricultural practices and economies, affecting the health and welfare of the environment and the people. With the exception of a few industrialized nat ions, global municipal water use over the last few decades has increased due to populati on growth and a higher standard of living 15

PAGE 16

[ Bates et al., 2008]. Global irrigation water withdrawals have seen the largest increase, accounting for approximately 70% of tota l withdrawals, and more than 90% of consumptive use [ Cai and Rosegrant 2002]. This drastic increase in agricultural water usage has contributed to the gener al decline of surface water quality in recent decades [ UN, 2006], and it is very likely that due to incr easing availability and price of fertilizers and pesticides in developing countries and in creasing population a nd demand for meat, declining water quality will continue to be a problem into the future [ IPCC et al. 2007]. The mass use of fertilizers and pesticides in modern agricultural practices is commonly used to harvest maximum yields fr om the available land and possibly less than desirable environmental growing co nditions. Nitrogen (N) and phosphorus (P) are two nutrients of limited resource found in fertilizers and from animal production, which have a large impact on ecosystem function due to their leaching into surface and groundwaters. During the last 50 years of ag ricultural improvement s, fertilization has reached a point where humans annually rel ease as much N and P to terrestrial ecosystems as all natural sources combined [Tilman et al. 2001]. The abundance of N and P can eutrophy surface waters, causing blooms of algae in lakes and streams, estuaries, and even has created a large dead zone in the Gulf of Mexico [ Tilman et al. 2001; Carpenter et al. 1998; Rabalais et al., 2002]. Combined with future climate uncertainties, it is likely that non-point source agricultural nutrient pollution will continue to have a major negative impact on global ec osystems unless it is addressed through research and management. Specifically, it is difficult to say exac tly what climate change effects will be, as climate signals are fairly chaotic and noisy, encompassing annual, inter-annual, 16

PAGE 17

decadal, or much longer periods of variability. This climate variability combined with the effect of exogenous variables and the la ck of extensive monitoring systems, both spatially and temporally, can make extracti ng short or long term climatic patterns an uncertain process. For both short and mi ddle term risk managem ent planning, interannual modes of climate variabi lity and their seasonal expression are of interest. There is a need to identify climate non-stationarities and their links to watershed water quality outcomes. In this dissertati on, the focus is on isolating, modeling and forecasting the effects of climate variability on hydrology, specifically water quality, in the southeast United States, rather than making statem ents about global warming or any possible anthropogenic causes thereof. Vari ability for our purposes is defined as fluctuations in climate from the monthly to seasonal and multi-annual sca le, and will be quantified via standard climate indices, while climate change refers to re corded multivariate trends over decades to centuries. Climate Indices and the El-Ni o/Southern Oscillation Climate indices are researcher-created diagnostic monitoring tools that describe an important or significant pattern or st ate of a climate system Generally, climate indices are represented as time series, wit h one index value repres enting a particular point in time. There are dozens of indices, which can describe any atmospheric event including monsoon precipitation, air pressu re differences, sea surface temperatures (SSTs), hurricane activity, or solar radiati on. Spatially averaged areas of sea surface temperatures in various parts of the world are particularly relevant to describing climate phenomena in specific locations, and the El -Nio/Southern Oscillation (ENSO) has proven to be one of the most consistent in describing low-frequency climate variability on both a regional and local scale [ Ropelewski and Halpert 1986]. The ENSO 17

PAGE 18

maintains an irregular 2-7 year periodicity that gives it a level of predictability, yet retains some variability in its occurrence, magnitude and climate consequences around the world [ Cane 2005]. While the definition of ENSO has changed s lightly since it was coined, it has always remained centered on SSTs in a narrow band of the equatorial Pacific Ocean. The original term, El Nio was from South American fisherman and sailors who noticed an unusually warm band of water off the coast of Peru and Ecuador in the Pacific during the end of the year, near Christmastime, that disrupted normal fish and bird populations. The term El Nio therefore refers to The Christ Child. It was not until 1969 that Dr. Jacob Bjerknes of UCLA described the full exte nt of the physical mechanism of El Nio, including the atmos pheric pressure component referred to as the Southern Oscillation [ Bjerknes 1969]. The entire ENSO phenomenon is coupled through the ocean and the atmosphere, centered over the tropical Pacific Ocean. T he mechanism of ENSO can be divided into the cold (La Nia) and warm (El Nio) phases each representing a deviation from the average or neutral condition. The main oceandrivers of ENSO are the unusual normal conditions of the equatorial Pacific Ocean, where the east may be 4-10 C colder than the west, due to the raising of the thermoc line from equatorial upwelling in the east and the transport of cold water from the South Pacific [ Cane 1986]. These ocean dynamics are driven by easterly trade winds (Figure1-1) which in turn are partly driven by SST differences causing higher pressures in the east, forcing surface air to flow along this gradient [ Bjerknes 1969]. The 3-7 year oscillation associated with ENSO was not more fully explained until two decades after Bjerk nes groundbreaking research, and has to 18

PAGE 19

do with the depth of the warm water above the thermocline. The changes involving the trade winds and SST are tightly correlated, yet the thermocline depth lags behind. Although physically very complex, it is the lagged change in the mean depth of the tropical thermocline which creates an oscilla ting system, as every oscillation must by definition contain an out-of-phase variable-technically, the delayed oscillator [ C. Wang, 2001]. Taken all together, this forms the coupled ocean-atmosphere positive feedback system that results in what we call the El-Nio/Southern Oscillation. During El Nio events, low air pressure in the eastern Pacific weakens the atmospheric pressure gradient heading west ward. This causes unusually high SSTs and increased convection in the central and eas tern equatorial Pacific. In a La Nia, trade winds strengthen, amplifyi ng the SST gradient so that lower than average SSTs are recorded instead (Figure 1-2). The relative strength and precise o scillation of these warm or cool phase events depends on the str ength of the mean winds, how much heat is generated by SST gradients from specif ic temperatures and humidity, how deep the thermocline is, and chaotic system dynamics. T he details of these mechanics were fairly recently published by Federov and Philander, although there is not yet a comprehensive theory as to why the average oscillation s tend to remain between 2-7 years [ Fedorov and Philander 2001]. Even as a climate index, the definition of ENSO is not standard, however. There are several accepted indices in use, including the NINO1 & 2, NINO 3, NINO3.4, NINO4, Multivariate ENSO Index (MEI), and Japan Me teorological Agency (JMA) index. Each uses slightly different defin itions of ENSO coordinates and phases (Figure1-3), and is the most relevant to slightly different r egions around the world. The MEI is a composite 19

PAGE 20

index using not only SST, but also surface ai r temperature, sea-level pressure, zonal and meridional surface wind, and cloudiness [ Wolter and Timlin 1993]. For having the most relevance to the southeast United Stat es, the JMA index and the NINO 3.4 index were used in this research. Global ENSO effects Due to global ocean and atmospheric fl ow patterns, ENSO effects can have a significant effect or little effe ct at all in different locations throughout the world, which are presented in detail by Molnar and Cane [ Molnar and Cane, 2007]. One of the clearest ENSO signals is seen where the phenomenon was first named -off the coast of Peru, Venezuela, eastern Colombia, and northeastern Brazil and the Amazon Basin. In El Nio years, these areas commonly receive lowe r than normal precipitation especially in boreal (November to April) winter and the preceding late summer and autumn [ Aceituno 1988; Kiladis and Diaz 1989; Mason and Goddard 2001; Ropelewski and Halpert 1987]. La Nia events in the same loca tions are associated with greater than normal precipitation, a nd lower temperatures [Aceituno 1988; Ropelewski and Halpert 1987]. Another well known and clear signal that has been record ed for over a century is the negative relationship between precipitat ion from the Indian Monsoon and El Nio events [ Ropelewski and Halpert 1987; Charles et al. 1997]. As a critical component of Indian sub-continent water supply and agriculture, historic monsoon failures have often coincided with strong El Nio events [ Kiladis and Diaz 1989; Ropelewski and Halpert 1987; Charles et al., 1997]. Australia also experienc es a strong ENSO signal, with drought commonly occurring over much of the continent during El Nio phases [ Kiladis and Diaz 1989; Ropelewski and Halpert 1987]. During an El Nio event, central 20

PAGE 21

Australia will remain dry for most of that year, while the north and northeastern areas will experience drought localiz ed in the winter, and the southern areas during the summer [Ropelewski and Halpert 1987]. During La Nia events in Australia, the same areas and seasons are associated with flooding. Much of Asia is affected by the Pacific Decadal Oscillation (PDO), but China also correlates with ENSO. During El Nio summers and autumns, China north of the Yangtze River and east towards Beijing tend s to have less precipitation than normal [ Mason and Goddard 2001; Ropelewski and Halpert 1987; Lau and Sheu 1991]. China south of the Yangtze River, however, has the opposite correlation with ENSO, with flooding and heavy rainfall du ring strong El Nio winter seasons, but with little to no connection to La Nia years [ Mason and Goddard 2001; Lau and Sheu 1991]. Many different climate indices affect r egions of Europe, including ENSO. During strong El Nio winters, west and cent ral Europe may experience increased temperatures and precipitation, while nor th Europe receives less. During La Nia events, higher pressures are observed over central Eu rope, and increased temperatures and precipitation are seen in northern Europe [ Fraedrich and Muller 1992]. During El Nio years, the southeastern area of Africa and t he eastern equatorial regions have also been shown to have gr eater temperatures and precipitation than normal [ Ropelewski and Halpert 1987; Halpert and Ropelewski 1992]. Finally, in the North and Central Amer ican continent, there are a variety of significant ENSO correlations. West and s outh Canada and the nort hern United States experience warmer winters and less precipitation during El Nio events [ Ropelewski and Halpert 1986; Rasmusson and Wallace 1983; Kiladis and Diaz 1989; Halpert and 21

PAGE 22

Ropelewski 1992]. The southwestern United States has greater than average precipitation during El Nio summers, while the north central and northwestern receives less. Again, this pattern is reversed during strong La Nia years [ Ropelewski and Halpert 1986; Mason and Goddard 2001; Rasmusson and Wallace 1983]. While most of the world tends to warm up during El Nio, the Gulf of Mexico area encompassing both Mexico and the southeaster n United States cools and receives more precipitation that normal in the winter season, while La Nia winters are drier than average [ Hoskins and Karoly 1981; Ropelewski and Halpert 1986; Kiladis and Diaz, 1989; Mason and Goddard 2001; Schmidt et al. 2001]. This distinctive winter El Nio pattern in the southeast United States is now understood as a deflection of the subtropical jet due to stronger Hadley Circu lation over the eastern Pacific Ocean [ Cane 2005]. It is here, in Florida and Georgia, t hat we are analyzing data from two different watersheds for their ENSO responses. Paleo-climate and future of ENSO A subject of interest for c limatologists is comparing ancient Earths climate, or, paleo-climate, to modern c onditions. In the mid-Plio cene (pre-Ice Age) epoch, approximately 3-5 million year s ago, Earths climate was much warmer than current conditions, with extremely high carbon dioxide (CO2) concentrations and a coupled ocean-atmosphere system resembling that of modern El Nio conditions [ Molnar and Cane 2007]. By learning about the climate f eedbacks and causes of warming or cooling in Earths past when anthropogenic change wa s not a variable, researchers hope to understand Earths natural periodic variability. The mid-Pliocene climate closely resemble s that associated with modern El Nio teleconnections and conditions [ Molnar and Cane, 2007], suggesting that a permanent 22

PAGE 23

El Nio-like state most similar in prec ipitation and warming patterns to the 1997-98 Nio of the Century was t he normal condition. These recent climate reconstructions are based on data collected from oxygen isotope analysis from Foraminifera specimens from marine cores [ Koutavas et al. 2006], coral bands that show the current incarnation of ENSO existing for the last 130,000 years [ Tudhope et al. 2001], glacial varve spectral analysis [Huber and Caballero 2003] and proxy records such as mollusk fossils, pollen taxa, lake sediment, and general animal and plant fossil types and locations. This permanent El Nio condition, or El Padre as it has been coined [ Bonham et al. 2009], was characterized by less of an east-west temperature gradient in the Pacific Ocean, with no o scillations in ENSO phase t hat we recognize today. As CO2 levels in the mid-Pliocene were 30% higher than pre-industrial conditions and global average temperature was 3 higher than the current [ Haywood and Valdes 2004], the geographic and greenhouse gas condition s make it the best analogue to those occurring currently. The drastic changes in the past four million years of SST and thermocline temperature and depth in the equatorial Pacific are unusually active, relative to the age of the Earth [ Molnar and Cane 2007]. It is still under question why despite having equal or arguably higher current atmospheric CO2 levels, Earths current average surfac e temperature is still significantly cooler than average temperatures in t he mid-Pliocene. Although there are many possible variables under study, it is apparent that greenhouse gases are not the entire story behind either global warming, and glob al phenomena such as the ENSO could be an important part of being able to more accu rately model both past and future climates. Interestingly, although most of the world was warmer and drier in the early to mid23

PAGE 24

Pliocene than today, many studies have shown that the Gulf of Mexico region, including the southeast seaboard of t he United States, was actually cooler and wetter [ Graham 1989]. As is discussed in the section in Ch apter 1, Southeast United States Climate, current conditions in the southeast are also on average cooler, despite global temperatures risinganother analog to the present. Also of interest is how ENSO may change in the future, so current models are not rendered obsolete immediately if they are predicting conditions predicated on an ENSO that no longer exists. As it is difficult to accurately model current global ENSO dynamics and the other low frequency climate patterns it is affected by, it is even more difficult to predict future ENSO patterns. Much of t he work in the past two decades involves working out the details of ENSO specific teleconnections, or pat terns between regional precipitation or temperatur es and larger modes of climate variability, which helps modelers create a more accurate model of ENSO. Accurate prediction of ENSO phase up to 12 months in advance has been claimed [ Cane 1986; Latif et al. 1998], however, unbiased skill estimates suggest that many EN SO events can only be predicted several months in advance, and sometimes only after the start of an event [ Goddard et al. 2001]. In a review of 20 Global Circulation Models (GCM) that modeled future ENSO conditions with a 1% increase in greenhouse gas concentrations per year of simulation, there was no significant trend in seeing eit her a more Nia or more Nio like state [ Collins 2005]. Although some studies have found modest trends in both directions, there is little conclusive evidence to sa y exactly how ENSO dynamics will change in either the near or distant future. 24

PAGE 25

Climate and Agriculture in the Southeast United States The humid sub-tropical southeast sunbelt re gion is important for many reasons, including its agricultural importance both lo cally and for the entire country. Although the number of farms has decreased by 80% in the last 50 years in the southeast [ USDA 1999], 25% of all national crops, including softw ood forests such as Loblolly pines, and 50% of all fresh winter vegetables are still grown there [ Hansen et al. 1999]. Both weather and climate are very important determinants of the population growth, and hence the economic development of the regi on, agricultural production, and how well the unique ecosystems in the area will fare in the near and far future. Because of the sunbelts rapidly increasing popu lationespecially in the coastal regions, which are projected to increase by another 40% bet ween 2000 and 2025It is especially crucial to understand how the changing climate will affect flora, fauna, and the human population. Prior to European settlement, the S outheast was mainly upland forests, grasslands, and up to 33% wetlands [ Dahl, 1990]. By 1990, wetland area had been reduced to about 16% of the Southeastern landscape, and is still decreasing in area, although Florida retains the most wetland area of all US states, with 9.3 million acres [ Dahl, 1990; Hefner et al. 1994]. Research has also sh own that C3 and C4 plant abundance in central Florida has changed drastically on a thousands-of-year scale. In an analysis of the past 62,000 years in Flor ida, it was found that during times of precipitation increase and low atmospheric CO2, C4 plant abundance decreased dramatically, while C3 plants and pine forests blossomed [ Huang et al. 2006]. In a future with higher CO2 concentrations, these relationships may be very telling in what flora and fauna populations will flourish. 25

PAGE 26

The southeast in general receives gener ous amounts of summer rainfall, with annual amounts in Florida alone totaling 53 inches [ Purdum 2002]. Even with this relative abundance of water, however, there are still both heated water quantity and quality debates in the southeast, as the timing and amount can be quite variable. During the last century, temper ature trends have varied inte r-decadally, with a warm period during 1920-40, and a cooling trend through the 1960s. The 1970s were the beginning of another warming trend, and 1990s temperatures reached peaks as high as those of the 1920s [Burkett et al. 2001]. These cooling and warming trends have made the Southeast the only region in the US to show an overall cooling of 1-2 C, although north Louisiana and peninsular Florida have shown a slight (1 C) warming [ Burkett et al. 2001]. In terms of temperature extremes over the last cent ury, the Southeast has shown an average annual decrease of 5 days exceeding 90 F, and an increase of six days in the annual number of da ys below freezing [ Burkett et al. 2001]. These extreme temperature trends can again be explained by the mid 20th century cooling trend balancing out the early and late warming trends in the regional average. Rainfall trends in the last century, on the other hand, show overall strong increases of 20-30% or more in Mississippi, Alabama, and Louisiana, with mixed results in Georgia and Florida. Although increased pr ecipitation is generally thought of as beneficial for agricultural production, especia lly in the southeast, where rain-fed crops are more common than irrigated ones, the precipitation increas e can mainly be attributed to intense events, which are generally damaging to agriculture and do not provide even moisture, speed erosion, and cause floods [ Rosensweig and Hillel, 1995]. Stream flow analysis is usually correlated to precipitation trends, ho wever, stream flow 26

PAGE 27

trends in the latter half of the 20th century show little change in maximum daily discharge, but increases in both annual m edian and minimum flows in Mississippi and Georgia [ Lins and Slack, 1999]. In the southeast United States, farme rs and ranchers use weather forecasts on a daily basis to manage their crops. Despite th is, a case study in Florida found that only a few farmers had integrated climate forecasts into their decision making [ Jagtap et al. 2002]. It was also found that there were varying degrees of flexibility to adjust management using climate forecasts, dependi ng on the size and type of farm and the specific situation. Using modeling studies and availabl e data on crop production in Florida, results suggest that there is a lar ge potential benefit to farmers who use ENSO based climate forecasts in planning their growing season [ Hansen et al. 1999, 1997; Jagtap et al. 2002; Messina et al. 2001]. Several of the same studies have also found that there are associated risks with usi ng ENSO climate forecasts, such as overproduction. These reasons prevent ENSO forecasts in agriculture from being quickly and widely adopted in the southeast. Ho wever, the voluntary incorporation of ENSO forecasts into managing agricultura l water quality outputs may be a less risky proposition, while also potentially saving farmers money on fertilizer and improving ecosystem function in the southeast United States. Objectives Water management agencies in the southeast United States wish to reduce climate associated risks in managing water resources and agricultural systems. The main goal of this research is targeted at exploring the relationship between and modeling the climate variability at the seasonal and inter-annual level via the El Nio/Southern Oscillation and hydrology and wa ter quality at the watershed scale in two 27

PAGE 28

basins in the southeast United States. If ni trogen and phosphorus loads in the southeast U.S. can be correlated with climate variabilit y via ENSO phase or climate indices, then we can make seasonal or monthly predi ctions of loads based on sea surface temperatures. This model could aid decisio n makers in making seasonal choices in water and agricultural management that woul d reduce the chance of exceeding stated TMDL goals for problematic nutrient loads. In greater detail, this dissertation will: 1. Investigate and quantify the significanc e of the relationship between discrete annual ENSO phase based water quality via nut rient loading in surface waters of basin S-191 of the Lake Okeechobee wate rshed, Florida, using hydrological simulation with Watershed Assessment Model (WAM) so as to remove confounding land-use change an d management variables. 2. Show a climatic teleconnection by extr acting significant ENSO spectra from observed hydrology data in the Little River Watershed (LRW) in Georgia, and compare it to the continuous classification of ENSO through monthly NINO 3.4 SST. 3. Show, through simulation using the So il and Water Assessment Tool (SWAT), the mechanism of hydrological increase of ENSO signal pow er in the stream flow and nutrient loads in the Little River Watershed. 4. Extract dominant modes of variability in each LRW and NINO 3.4 time series to rebuild and create a multivariate wavele t based time series model, able to make seasonal predictions of nutrient loads in the LRW based on NINO 3.4 SST forecasts. Literature Review Agricultural Pollution: Nutrients Clean and abundant water is necessary for human and animal life, providing drinking water, recreation, fishing habitat, i rrigation, hunting, trans portation, industry, and support of biodiversity. Pollution of ground and surface waters negatively impacts all these necessities, making it a major c oncern in the southeast. Water pollution inputs into the environment can be divided into tw o main types: point and non-point source. A point source discharge tends to be contin uous and relatively localized, with little 28

PAGE 29

temporal variation, such as effluent from an industrial plant or a municipal sewage treatment center. As a result, point source pollution can be easier to locate, regulate and fix by treating it at the source. Non-point sources of pollution are often intermittent or irregularly discharged, and ar e many times linked to activi ty from extensive areas of agricultural land. These sources may travel underground, in the at mosphere, or over land to a receiving water body, making them very difficult to monitor, regulate and fix. In the well known examples of the impact ed Chesapeake Bay watershed and the hypoxic zone in the Gulf of Mexico, 35 states are kn own to contribute nutrient loadings, making regional regulation very difficult [ U.S. EPA 1996]. As a result, control of non-point sources is usually centered on continued good land management practices, which makes a climate forecast system that could reduce variability in best management an attractive prospect. The main non-sediment non-point water po llutants from agricul ture stem from use and overuse of chemical fertilizer, specifically those containing nitrogen (N) and phosphorus (P). Although beneficial and necessary for plant growth, N and P in overabundance can result in negative ecosyst em impacts such as eutrophication and toxic fish kills. Nitrogen is a limiting nutri ent in the productivity of many aquatic and terrestrial ecosystems, for it is a necessa ry component of all amino acids and proteins yet can only be utilized by many plants and animals in its non-gaseous form. The Nitrogen Cycle is how life on earth fixes atmospheric N2 (gas) into widely useable forms of N such as nitrate (NO3), nitrogen dioxide (NO2), or ammonia (NH3) [Postgate, 1998]. Nitrogen is fixed naturally through a number of bacteria, prokaryotes, fungi and microorganisms via nitrogenase enzymes. Some plants and the majority of legumes can 29

PAGE 30

fix N through a symbiotic relationship with Rhizobia bacteria in their root systems. However, nitrate is water soluble, and can enter ground and surface waters easily, contributing to eutrophication and declines in water quality. Phosphorus, another limiting nutrient, is essential for life in DNA and RNA molecules, ATP, and cell membranes. The majori ty of natural P is locked up in inorganic mineral deposits such as phosphates (PO4), which slowly weather and dissolve into soil water to be taken up by plants and converted to organic orthophosphate (PO4 3-), and in turn consumed by animals. A limiting factor in plant growth, P is a major component of fertilizers and is partly responsible for the major increases in crop yields seen throughout the last century. The overuse of P fertilizer has affected water quality and is a main pollutant in eutrophic lakes today [ Carpenter et al. 1998; Gakstatter et al. 1978]. Non-point sources are a major part of wate r pollution in the Unit ed States. Even if point source inputs were eliminated entir ely, 72-85% of eutr ophic lakes would still require non-point P control to meet water quality standards [ Gakstatter et al. 1978]. Eutrophication occurs when large N and P i nputs cause excessive blooms of aquatic weeds or algae, and is a problem in fresh, brackish and salt water. The overgrowth and decomposition of the plants and algae can cr eate hypoxic zones, destroy fish and aquatic insect habitat, kill entire fish populat ions, destroy coral reefs and aquaculture, and create toxic byproducts that make the water undrinkable [ Carpenter et al. 1998; Gakstatter et al. 1978]. Virtually every state is adv ersely affected by nutrient water quality degradation, and 49 states have listed Clean Water Act Section 303(d) nutrient-related water 30

PAGE 31

impairments as defined by the U.S. Enviro nmental Protection Agency (U.S. EPA) [ U.S. EPA, 1998]. The EPAs national strategy for nut rient criteria in 1998 was followed in 2001 by an action plan for the es tablishment of specific regional numeric nutrient criteria for streams, river, lakes, and reservoirs, of which there has been progress. Dictating quantitative nutrient loads is very tricky, as natural regional le vels vary widely both spatially and temporally; however the establis hment of these limits is very important to create measurable and objective water qualit y baselines to measure progress against and create appropriate laws and regulati ons. In 2008, the EPA released a report detailing the progress of a ll states in defining r egional nutrient loads [ U.S. EPA 2008]. The southeast United States is doing well over all, with Alabama, Ge orgia, and Florida already using some specific nutrient st andards for chlorophyll-a, total N and P, and Louisiana and Mississippi planning to start im plementation of their standards in the next few years [ U.S. EPA 2008]. The next section will detail re levant regional nutrient related policies and Best Management Practices (BMPs). Water Management Policies in the Southeast Although there are state r egulations and policies too numerous to describe in detail, this section will focus on the implem entation of federal nutrient related water quality standards and management from the US EPA in Flori da and Georgia, where the two watersheds under study in this dissertation are located. The Watershed Restoration Act (WRA) and TMDL standards In 1999, the Florida legislature enact ed the Florida Watershed Restoration Act (FWRA) with the intent of preserving and improving water quality. This was accomplished through the development of EPA encouraged Total Maximum Daily Loads (TMDL) for ground and surface waters. In turn, TMDLs are required by the 31

PAGE 32

federal Clean Water Act (CWA) as an extensi on of the nutrient control program. The FWRA created a process to collect data on and identify impaired water through Florida, and is enforced by the Flori da Department of Environment al Protection (DEP) and the Department of Agriculture and Consumer Services (DACS). One of the main innovations in the FWRA program was the regulation of water bodies via watershed delineation instead of political boundaries, which has been t oo frequently used in the past. The federal TMDL program was created to systematically establish the maximum pollution specific bodies of water could handle while still meet ing water quality standards and controlling both point and nonpoint source pollution. By providing measurable nutrient water qua lity baselines, the TMDL program makes measuring progress easier and facilitates the writing of National Pollution Discharge Elimination System (NPDES) permits, which allow polluters a certain amount of loading. Given that Florida has 52,000 miles of rivers and stre ams, 800 lakes, 4,500 square miles of estuaries and more than 700 springs, gatheri ng enough data to make informed TMDL decisions for each is a monumental undertaking [Florida DEP, 2005]. The DEP addressed this by splitting up the state into five groups of basins, each of which would be researched one at a time over a five year cycle and put into practice. The Lake Okeechobee basin was in the fi rst group of watersheds, meaning that TMDLs have been established for basin S-191, a basin used as a field site in this dissertation. TMDLs are decided upon by a process of data collection and monitoring, historical data, public meetings, and the env ironmental goals decided upon. These limits can later be altered by public hearings, new information, or updated Basin Management 32

PAGE 33

Action Plans (BMAP). By addressing all wate r users and stakeholders in the process, there is environmental educat ion of both the public and agr icultural partnerships to employ appropriate BMPs to achieve t he mandated TMDL. The specifics of these loadings for S-191 are discussed in greater detail in Chapter 2. The state of Georgia has a grant system under the federal Clean Water Act section 319(h) to specifically address cl eaning up non-point pollution sources. The Environmental Protection Division of the Georgia Department of Natural Resources oversees the grants and implement ation, and the basin of interest in this dissertation, the Little River Watershed, has a four year TMDL development project grant under this section, that will be discussed in Chapter 3. The Soil and Water Commission works with landowners and agricultural users to devel op nutrient management plans and install appropriate BMPs under one year contracts. After a BMP is installed and inspected, landowners are reimbursed fo r 60% or more of the cost, depending on its size. BMP implementation in both Florida and G eorgia programs are voluntary, although stakeholders must meet NP DES guidelines or TMDLs. If landowners successfully implement a BMP and are certif ied by the state, actions cannot be brought against them if they are still not m eeting the TMDL limits. The volunt ary nature of BMPs makes them a contentious issue for those concerned with ec ological restoration projects, as their validity and effectiveness is still under discussion. Best Management Practices (BMP) The history of the environment al BMP stems from the mo re managerial aspect of Best Practices. In a corporate environmen t, Best Practice is a buzzword used to describe the process of developing and followin g a standard way of doing something. It is not necessarily the best way of doing anything, but is instead a standard operating 33

PAGE 34

procedure that can be followed uniformly bet ween companies and industries. Often, the benefit of Best Practice implementation in th is context is that or ganizations that have poorly designed (or evolved) processes are given a choice between a typically expensive modification to thei r system, or choosing to follow a Best Practice. The rate of enormous technological change over the pas t century forces rapid adaptation and versatile Best Practices. Ideally, Best Practi ce in these cases is meant to get things done, with the added benefit s of quality result s and consistency. In the United States, BMPs are practical control measures that have been shown to minimize environmental damage via nonpoint pollution in water bodies. The evolution of BMPs, however, is not easily traceable, and the implementation and enforcers of BMPs are not obvious. BMPs originated duri ng the Dust Bowl era in the United States [ Ice 2004]. As drought and soil erosion continued to negatively affect a largely farmbased citizenship, the Soil Conservation Act of 1935 started advising farmers on methods to manage their lands to protect es sential watershed functions. These land management practices were the predecessors of modern day BMPs. In 1972, the Federal Water Pollution C ontrol Act Amendments separated point from nonpoint pollution control measures, and defined BMPs to control nonpoint sources. In these amendment s, BMPs were defined as vo luntary, incentive, or regulatory control programs, gi ving states a large measure of control as to how they would be enforced and defined. Point source po llution, on the other hand, is subject to federal laws and permits. BMPs are inherently a compromise between the environmental ideal zero po llutant state, and practi cal management options, and as such are continually adjusting to adapt to new ideas in the scope of both science and 34

PAGE 35

society. Best Management Practices can be t hought of as a narrow and specific subset of Adaptive Management, in which an optimal decision is reached in the face of uncertainty through an iterative process, with an aim of reducing uncertainty over time via system monitoring. BMPs become a s pecific subset of Adaptive Management because of their narrow aim, and because the optim al decision that is sought is also in flux. Different states use different BMP in centives and implementation, such as performance bonds, permits, required writt en management plans, on-site project reviews, random inspections, cost share pr ograms, and education. In some studies, BMP effectiveness has been demonstrated to reduce water quality impacts from preBMP era forestry practices by 90% or better [ Ice 2004], showing that these practices do have noticeable and significant effects, as co ntinually evolving and innovating measures that meet demands for tighter control of water quality impacts while remaining economically viable for stakeholders [ Veith et al. 2003; Malik et al. 1994]. Compared to other environmental wa ter quality policies such as wetland mitigation, BMPs also have been successful Part of their success is certainly dependent on the fact that many stakeholders are indivi duals with a connection to the land and environment, or small businesses, rather than sprawling corporations with a financial bottom line. However, it seems t hat BMPs also represent a compromise between top down and bottom up management. The lack of inflexible regulations enforced by a federal bureaucra tic institution are instead carried out by researchers and the Water Management Districts (WMD) in Flori da, which gives stakeholders a feeling of being more personally involved in t he adoption and success of the process. 35

PAGE 36

In a study of 400 private forest owners in Belgium, it was found that management suggestions were most accepted when the owners were well educated and informed, and when implementation did not decrease their profits. Additionally, those who were purely motivated by profit had no interest in the ecological aspects of forest management, implying that owners of mid-size farms in the Ever glades region may indeed have a deeper connection to their land that makes them more open to environmentally sound practices. The Belgian study also found that subsidies were not effective in persuading forest owners to c hange their management prac tices. Again, the differences between Belgian forest owner s and landowners and stakeholders in the southeast United States may lay in their fundamental environmental concern, which makes them more amenable to changing their operations by working with researchers and receiving subsidies [ Serbruyns and Luyssaert 2006]. Another factor in BMPs success may be that they inherently take a systems view, by necessity of dealing with nonpoint pollution, rather than trying to optimize specific parts unless it is absolutely nece ssary. For example, water quality is normally monitored at the basin level, rather than the fa rm level. Only if the water quality has not improved do researchers start m onitoring at the farm level. At the watershed level in the end, some farms may in fact still be releasin g quite a bit of P, while others may have drastically reduced their loads. The average decrease, however, is enough to make a difference in the ecosystem being restored. Simulation of Natural Systems The simulacrum is never what hides the tr uth-it is truth that hides the fact that there is none. The simulacrum is true. Jean Baudrillard (Simulation and Simulacra) [ Baudrillard, 1985] 36

PAGE 37

In the last 50 years of continually im proving computer technology, the use of mathematical models to simulate and help us better understand complex or chaotic natural systems has made incredi ble progress. A continuing deb ate in the mathematical modeling community is the relative advantage of using empirical models versus physical models. An empirically based relationship is one based upon directly observed values or experimental data. These models are ofte n based off of statistical or probability theory, and accurately predict observed results from other variables, yet have little to no theoretical reasoning to explain why the relationship works. Examples of empirical relationships commonly used in hydrological process modeling include the SCS Curve Number met hod for estimating runoff in an area from precipitation, the original Darcys Law describing flow through a porous medium (now given a theoretical basis), and Mannings Eq uation for open channel flow (also recently given a theoretical basis). Often, accurate empirical relationships derived from careful experimentation are later proven to have a theoretical basis as well. This fact can be used to make a strong case for the use of empirical re lationships in models, as relationships derived from careful and reproducib le experimentation often lead to future understanding of a process. In fact, the Manning equation, derived in 1885, was not formally physically explained until 2002 [ Gioia and Bombardelli 2002], using phenomenological turbulence theor y. Even the specifics of gravity are not physically explained, yet without using the concept, many of our physically based models would be useless. One of the main disadvantages of using em pirical relationships in models is that the relationship is only as good as the data, which may be limited in spatial or temporal 37

PAGE 38

scope. Additionally, noise and measurement errors can influence the accuracy of relationships. To address these critiques, t he model developer must be very aware of potential issues in collected data and predict ion uncertainties in the model itself. Accounting for uncertainties in data, model calibration, and model structure is becoming a very sophisticated process, and recent st ochastic techniques add to the power of empirical models. A theoretical or physically based model is one in which specific physical processes are accounted for to make pr edictions based on acc epted laws and theory. Theoretical models simplify the physical system to some extent, and often include empirical components, just as empirical models often use relationships based on theory. A main benefit to using a theoretical model is the increased understanding of a process that is being simulated. Another advant age of theoretical models is their ability to be more easily applied to different situat ions and case studies than empirical models. A theoretically based model should be able to account for the different variables and factors in a process no matter where the location is, as long as certain underlying assumptions are met. An effective empiri cal methodology may be employed in different situations, although a specific model may be irrelevant. There are advantages and disadvantages inherent in using each type of modeling system on their own, which is why this dissertation attempts to use both individually and in combination to achieve a mo re accurate representation of reality. As the post-modern philosopher Jean Baudrillard s uggests in his quote, truth or reality in the world of simulation is a relativistic te rm, and using a wide variety of methods to explore reality as we perceive it through ti me may be the best approach we have. As a 38

PAGE 39

modern scientist, it is essentia l to understand how your research relates to the greater knowledge base of the world, and how the ultima te goal of scientific thought is not to provide definitive answers, but instead to discover deeper problems using new methods and subject the best answers to ever more ri gorous examination. The next sections will detail some relevant models in this dissertation. Global circulation models and ENSO Historically, the desire to accurately simulate and predict Earths climate probably evolved from our desire to forecast weather Although there is no specific distinction, operational weather forecasting generally extends from hours to days, and climate forecasts pertain to weeks, months, and y ears. The feasibility of attempting to numerically simulate long-term global climat e using computer models was first proved by Norman Phillips in 1956 [ Phillips 1956], and the first Gl obal Circulation Models (GCM) used first principles of flui d dynamics rather than energy balances. Parallel GCMs were developed ar ound the world by different groups incorporating different models, and today t here are still many GCMs that are commonly used in conjunction. The presence of multiple models is advantageous, as the current understanding of the extremely complex physics and chaotic dynamics in the ocean and atmosphere is still limited, and having severa l model outputs allows decisions to be made based on ensemble means or forecasts in which the uncertainty of all models is incorporated. A key strength of modern GCMs la ys in their coupled nature: atmosphere models are coupled with ocean, sea-ice, and land-surface process models incorporating greenhouse gases and carbon sequestrat ion. In turn, these m odels can be statistical, physical, probabilistic, or a combination of methods. Some more well known models 39

PAGE 40

include the NASAs GISS Model II, the H adley Centers HadCM3, and NCARs CCM3. While beneficial for investigating long-te rm trends and scenarios of climate change, the GCM is not fundamentally well equipped for predicting region al climate impacts, as the area extent of the map grids t ends to be of a low-resolution. There is a large body of developing research into GCM downscaling techniques to create more accurate Regional Simulation Models (RCMs) from av ailable GCM forecasts. These downscaling methods may also be statistical or physically based, and are continually evolving to provide better regional climate predictions. As pertaining to this dissertation, GCMs are of interest because of their relative inaccuracies and disparate predictions in fo recasting and simulati ng ENSO events. Due to the many interacting ocean and atmospher ic processes and relative sensitivity to boundary conditions, modeling ENSO accurately in a GCM is a considerable challenge [ Bonham et al. 2009]. Only very recently have some GCMs created emergent ENSOlike precipitation and circulation patterns within their simulations, although the future severity and occurrence of ENSO events remains unclear [ Bonham et al., 2009; AchutaRao and Sperber 2002]. The potential socio-econom ic impacts of changes in ENSO are extensive, making a method that can reduce the uncertainty of ENSO impacts beneficial. Agro-hydrological models Compared to a GCM, agro-hy drological models are relatively smaller in scale and scope of output; however, they can simu late details that a GCM has no hope of being able to do. There are a large variet y of models to choose from depending on the required local ecology, necessary model complexity, data availability, spatial and temporal simulation extent, and final model performance. Generally, these models can 40

PAGE 41

be grouped in two main categories: those able to operate at a variety of spatial scales that can handle a wide variety of conditions, or those more specifically suited to detailed field scale simulation. It should be noted that a hydrological model does not exist that is able to simultaneously operate at multiple spatial scales, due to the complexity of the land/water processes and interactions. Academic models able to handle a variety of scales include AnnAGNPS [ Bingner and Theurer 2007], BASINS [ Duda et al. 2006], SWAT [ Neitsch et al. 2005], and WAM [ Soil and Water Engineering Technology, Inc. 2004], and MIKE-SHE [ DHI 2004] to name just a fracti on. Field scale models include GLEAMS [ Leonard et al. 1987], DRAINMOD [ Skaggs 1981], FHANTM [ Fraisse and Campbell 1997], and EAAMOD [ Soil and Water Engineering Technology, Inc. 2000]. A commonality of all these models is that they are driven by precipitation and irrigation inputs. Main tasks that agro-hydrological m odels should be able to simulate are hydrological processes such as water and so lute transport through soil in one or more dimensions, evaporation and transpiration, both ground and surface water flow, vegetation effects which could include agricul tural input, pollutant transport such as nutrients, and BMP evaluation. In most models, transport is through a series of cells or basin units whose number and size are decided on by the user, as are many parameters and inputs if not otherwise available as observed data points. The strength of these large hydrologi cal model packages lies in their broad applicability and the relative ease of generat ing multiple output scenarios once the model is calibrated. Their weaknesses, howev er, are the large amount of data required 41

PAGE 42

to accurately calibrate them, and the user s ubjectivity in many of the parameter input selections and large possi bility of human error. Time series models A time series is an ordered sequence of observed data, usually through equally spaced time intervals. A time se ries may be discrete, i.e. tak en at specific time intervals such as a daily stock price, or continuous, su ch as an electric voltage, meaning that it can be recorded continuously in any time inte rval. In studying a given time series, one can understand the generating mechanism, forecast future values and show the control parameters of a system. Statistics are essent ial in describing relationships in a system of ordered, correlated data, and the methods that can be used to analyze a time series are generally referred to as time series analysis. Unlike the models previously discussed, time series simulations are purely statistically based, physical interpretations only being assigned subjectively afterwards. In most areas of research an observed time series ( x,t ) is a realization or sample function from a given stochastic process, Z( t) which in turn is just all realizations { Zt1, Zt2Ztn}, of a set of time ( t ) indexed variables. Processes are characterized as being strictly stationary if it is nth order distribution functions are each time invariant, but may be characterized as having a lesser order of st ationarity. If the dist ribution function is not time invariant, (for example if the mean function or variance c hanges over time), then it is called non-stationary. Special measures must be taken to make non-stationary time series stationary before putting them into a model. The crux of time series modeling theory is based on autocorrelation ( k) and autocovariance ( k) functions. If the covariance (E quation 1-1) describes how two variables vary together through time via their expected value, the autocovariance 42

PAGE 43

simply describes how one variable varies through time against a time shifted ( k) version of itself (Equation 1-2). Cov(X, Y) = E((X x)(Y y)) (1-1) k = Cov(Zt, Zt+k) = E(Zt )(Zt+k ) (1-2) The autocorrelation function, or ACF, is simply the autocovariance normalized by the autocovariance at ti me zero (Equation 1-3). k = [Cov(Zt, Zt+k)] / [ var(Zt) var(Zt+k)] = k/ 0 (1-3) The other most used function is the partial autocorrelation function ( kk), or PACF. The PACF is a measure of the conditional correlation between Zt and Zt+k after their linear dependency upon the variables in-bet ween them has been removed (Equation 14). kk = Corr(Zt, Zt+k | Zt+1,.,Zt+k-1) (1-4) The most basic models can be classi fied singly or in combination as: autoregressive processes, AR(p ), moving average processes, MA( q ), and as integrated I(d ), or seasonal (S). A model with all the parameters would be SARIMA( p,d,q ) x ( P,D,Q )s, with capital letters referring to seasonal components and sub-index s to the seasonal period. Model identification is usually done through quantitative goodness-offit measures such as the AIC, Akaike's information criterion, or BIC, Bayesian information criterion, and looking at specif ic patterns in the graphics of the ACF and PACF at different lag times. However, time series model selection has also been called as much of an art as a science, as there is user subjectivity in sacrificing accuracy for the number of model parameters or vise ve rsa, and recognizing the patterns seen in the ACF and PACF graphs. 43

PAGE 44

Time series models may also be creat ed to describe the relationships in multivariate time seri es, called vector ARMA( p,q ), or VARMA processes. This type of model will be described in det ail in Chapter 5. There have been relatively recent advances in time series model techniques for considering the variance of the current error term as a function of the variances of the previous time periods' error terms. This family of models is called an autoregressi ve conditional heteroscedasticity (ARCH( q )) or generalized ARCH (GARCH( p,q )) model. These methods are used mainly in finance and economics when dealing with exceptionally volatile time series variables [ Bollerslev 1986]. Spectral analysis and models The last type of model to be dealt with in this dissertation is spectral models. Though most time series analysis studies re lationships in the time domain, an alternative view in the frequency domain descri bes a variable in terms of its sinusoidal behavior in a range of applicable frequencies, i.e. how often an event happens. The study of Fourier analysis is integral to the study of the frequency domain, as the 18th French mathematician Joseph Fo urier lent his name to the concept that any periodic function could be represented as harmonically related sinusoidsalthough the field has vastly expanded since that time An advantage of studying time series in the spectral domain, especially in geophysica l data, is that hidden periodicities may emerge from noisy or chaotic systems that have a clear physi cal basis. This allows what started as an exercise in empirical modeling to move into becoming a physical model. In a simple real and finite continuous sequence defined over ( -P, P), a Fourier series representation, f(t) is accomplished using an orthogonal combination of sines and cosines (Equation 1-5). 44

PAGE 45

P P P P P P j j j adt P jt tf P b dt P jt tf P a dttf P a P jt b P jt a tfj j sin)( 1 cos)( 1 )( 1 sin cos 2 )(0 01 (1-5) A Fourier transform (Equation 1-6) comes fr om the study of the Fourier series, involving the use of Eulers formula and complex exponentials to decompose a time series into oscillatory functions. The Fourier transform of a function f(t) is the integral over all times t multiplied by Eulers formula representing a basic waveform. .number realevery for ,)()(2 dtetfFit (1-6) However, when dealing with lo ng-term environmental time series data, there are a number of disadvantages in using Fourier transforms. When doing a Fourier transform using this type of data, one chooses a s liding window to extract localized frequency response. The window thereby imposes a scal ed response interval into the analysis, which effectively blinds the researcher fr om gaining the whole picture of the power frequency response at all scales thr ough the whole length of record [ Kaiser 1994]. Additionally, Fourier analyses are not pr epared to handle non-stationary time series, which makes up much environmental data. Wavelet analysis is a spectral method which can explore non-stationary environmental data over all frequencie s of interest simultaneously [Daubechies 1990]. The basic concept is that a different wavelet function, o( ), is a localized, zero-mean 45

PAGE 46

function with a non-dimens ional time parameter, exists for every time series. Although for our purposes in this dissertat ion only continuous analysis is used, a wavelet function may be discrete or continuous, orthogonal or non-orthogonal. The wavelet function serves the same purpose as Eulers formulamoving the time series of interest into the frequency domain. The continuous wavelet transform, Wn( s), is the convolution of the time series, xn with a scaled version of o( ) (Equation 1-7), where s is the wavelet scale, n is the localized time index, (*) indicates the complex conjugate. s tnn xsWN n n n)'( )(* 1 0' (1-7) By varying s and translating along the time index, n (similar to the sliding window of the Fourier transform), a picture is created against appropriate background spectra showing the amplitude versus the scale, and how this varies through the time of interest or record. More detail on the technical aspe cts of wavelet transform will be given in the introduction of Chapter 3. Wavelet analyse s have been used for data compression and signal analysis in electrical and computer engineering, molecular dynamics and DNA analysis, astrophysics, and turbulence theory [ Chan and Shen 2005; Askar et al. 1996; Mouri et al. 1999; Zhong and Yang 2007], and has more recently been applied to geophysical data such as rainfall, atmospheric pressure, infrared radiance, chlorophyll concentration, river stream flow and sea surface temperatures [ Serrano et al. 1992; Rajagopalan and Lall 1998; Nezlin and Li 2003; Labat 2008; Y. Wang 1996]. The application of wavelet transform is growing in popularity, as more scientific fields recognize the powerful information it can provide, coupled with the advantages it has over traditional Fourier types of analysis. 46

PAGE 47

Figure 1-1. Global air circul ation patterns are dominated by the trade winds, tropical easterly winds that are found near the equator. Trade winds are responsible for steering the flow of tr opical storms that affect North and South America, Asia, and India, and African dust. (from NASA Jet Propulsion Laboratory http://sealevel.jpl.nasa.gov/overview/climate-climatic.html) Figure 1-2. As opposed to neutral conditions (left), during an El Nio event (right) sea surface temperatures (SST) in the Pa cific equatorial region are warmer than usual, increasing convection of moist air into global circulation. During a La Nia event, cooler water decreases convection. 47

PAGE 48

Figure 1-3. The Nio 3 Region is bounded by 90 150W and 5S 5N. The Nio 3.4 Region is bounded by 120 170W and 5S 5N. Nio 4 is bounded by 160E 150W, and from 5S 5N. Nio 1 is defined by 8090W and 510S, Nio 2 by 8090W and 05S. The JMA (not pictur ed) extends from 4S 4N, 150 90W. 48

PAGE 49

CHAPTER 2 EFFECTS OF CATEGORICAL EL NIO/SOUTHERN OSCI LLATION ON SIMULATED PHOSPHORUS LOADING IN SOUTH FLORIDA Introduction Indices of climate variability have been moni tored regularly for the greater part of the twentieth century. The El Nio/S outhern Oscillation (ENSO) phenomenon has emerged as one of the most consistent indices for describing low-frequency climate variability on global an d regional scales [ Ropelewski and Halpert 1986, 1987]. As the Florida water management agencies are intere sted in reducing climate-associated risks in managing water resources and agricultura l systems, ENSO could provide a basis for correlating water quality, especially nutri ent loading, with climate variability. Different ENSO phases are based on an index of warming or cooling of surface sea temperatures in the equatorial Paci fic Ocean and associated pressure and wind pattern changes. ENSO phase classifications us ed in this research are from the Japan Meteorological Agency (JMA) index [Japan Meteorological Agency 1991], which are based on observed data from 1949 to the present, and a new alternative monthly ENSO classification, which has been initially refe rred to as the M-ENSO Index. The JMA index is a six-month running average of spatially averaged surface sea temperature anomalies over the tropical Pacific Ocean (4 S to 4N, 150W to 90W). The ENSO year (defined as October through the following September) is classified as El Nio if the running six month surface sea temperatures average, that must include the reference season October to December, is at least +0.5 C higher than average, La Nia if it is 0.5C lower than average, or neutral (all other values). A main disadvantage of the annual JMA ENSO classificati on is that the summer months are determined using the SST conditi ons from the previous October, as 49

PAGE 50

opposed to more current conditions. The MENSO Index still uses the JMA defined sixmonth duration during a refer ence season, however, that reference season is from November to January, and the El Nio or La Nia event stops for a given month when the SST anomalies change to a different phase [ Grard-Marchant et al. 2008]. This allows months within an ENSO year to be cl assified as different phases. Using the MENSO classification on recorded SST data from 1900-2007, the total number of La Nia months from April to Sept ember is 25-35% less than with the annual classification. The number of El Nio months is about 10% less using M-ENSO [ Grard-Marchant et al. 2008]. Much of the southeastern U.S., especially Florida, is strongly affected by ENSO. El Nio winters in Florida tend to be cooler and wetter, whereas La Nia winters tend to be warmer and drier [Hanson and Maul 1991; Kiladis and Diaz 1989; Schmidt et al. 2001]. Previous studies have correlated ENSO phase with stream flow [ Chiew et al. 1998], rainfall in the western U.S. [ Rajagopalan and Lall 1998], crop yield in Florida [ Handler 1990; Hansen et al. 1997; Hanson and Maul 1991], and even cholera dynamics worldwide [ Pascual et al. 2000]. In Florida, water quality and quantity are important concerns for environmental quality and standard of living. T he importance placed on water pollutants is reflected in Florida's policies with respect to ecosyst em and Everglades rest oration, and in the prevalence of its water management agencie s. Although ENSO effects on many environmental data have been explored, ther e has been no research exploring ENSO effects on nitrogen (N) and phosphorus (P) runoff pollution in Florida. In agrohydrological watersheds, P and N are important indicators of water quality and best 50

PAGE 51

management practices (BMPs) and are a focus of efforts to reduce nutrient loading in water bodies. In this study, we are focusing on both JMA and M-ENSO effects on simulated nutrient loads rat her than historical data because major changes in land use and land management in Florida over the last 50 years make the use of historical data impractical. However, measured P loads were determined to be reasonably represented by model predictions. A major advantage of using simulated nutrient loads is our ability to look for the presence of direct associ ations between simulated nutrient loads and ENSO phase without confounding l and use and management variables. The agro-hydrological model WAM (Wat ershed Assessment Model) is a watershed-scale, continuous, spatially distri buted simulation model developed by Soil and Water Engineering Technologies, Inc. (SWE T) of Gainesville, Fl orida. Using WAM, scientists and engineers have successfully modeled more than 25% of the watershed areas of Florida [ Jacobson 2002; Ouyang, 2003]. Among the most well calibrated and validated processes that WAM simulates ar e daily stream flow and N and P loads in systems with high water tables [ Soil and Water Engineering Technology, Inc. 2004]. WAM was specifically calibrated to south Fl orida's hydrology, as models of specific regions are useful for ex ploring effects of management on runoff and pollution. Additionally, by assuming that land management has remained constant over time, the effects of climate variability on pollution can be isolated and explored. For these reasons, it is reasonable to use a watershed model, such as WAM, to study the effects of climate variability manife sted through ENSO phases on simulated water quality, and to assume future potential in tailoring land management prac tices to climate prediction. 51

PAGE 52

Nitrogen and P loads into Lake Okeechob ee were simulated for 36 years (19652001) in basin S-191, a 486.9 km2 sub-basin directly northeast of Lake Okeechobee in south Florida (Figure 2-1a). Then, we exami ned potential P load differences at different temporal scales based on historical annual JMA and monthly M-ENSO classifications. Our hypothesis was that there would be statistically significant and quantifiable differences in simulated monthly and/or s easonal P load for the three ENSO phases, and that the M-ENSO index woul d capture summer climate trends more accurately than the JMA because of the finer temporal resolu tion. Specifically, we expected to predict greater nutrient loads during El Nio winters and La Nia summers because of the greater precipitation during these phases in south Florida. We also expected nutrient loads in neutral years to be closer to aver age values. By investigating the trends in nutrient load during the diffe rent ENSO phases, the objective was to answer the question of whether or not P loads have climate-based predictability. In the southeastern U.S., and more specifically the Lake Okeechobee watershed, P loading is more of an issue than N because pl ant and microbial growth in the lake is most limited by P availabilit y. Traditionally, it had been thought that P applied to soil for crop production would remain in place despite the action of water and independent of soil type. As a result, inorganic phosphate fertilizers often were applied in excess of need because cost was not prohibitive. This was called "banking" phosphorus to insure that it would not be a limiting factor in cr op growth. While both N and P are integral lifesustaining components of a biological system an excess or dearth of either one can create an ecological mess. 52

PAGE 53

Between the early 1970s and mid-1980s, tota l P loads to the open-water region of the lake nearly tripled [ Flaig and Reddy 1995], causing massive hyper-eutrophication and fish kills. Monitoring and managing P loads in Lake Okeechobee are also part of the Lake Okeechobee Water Management Projec t (LOWP) and the Comprehensive Everglades Restoration Proj ect (CERP), neither of whic h has yet looked at using climate forecasts to manage nutrient loads Although WAM simulated both N and P loads, and the consequent analysis was done with both nutrients, this research presents only the P-related results, as N loading is not a problem in Lake Okeechobee. In general, simulated N loads were of a lesse r magnitude than P load s, but followed the same trends and patterns over the time series. Data and Methods The Lake Okeechobee Watershed Lake Okeechobee (LO) and its watershed are central components of south Florida's Kissimmee-Okeechobee-Everglades ecosystem, which extends from the headwaters of the Kissimm ee River in the north to Florida Bay in the south. The shallow lake is the second largest freshwater body loca ted in the continental United States. It is home to one of the nation's prized bass and speckled perch fisheries, as well as an economically important commercial fishery [ Flaig and Reddy 1995; Rosen et al. 1996]. At the same time, it provides habitat fo r a wide variety of wading birds, migratory waterfowl, and the federally endangered Ever glades Snail Kite. Lake Okeechobee can also be a backup water supply for the communiti es of the lower east coast of Florida. The lake supplies drinking water, irrigat ion water for the expansive Everglades Agricultural Area (EAA) to the South, and is a critical supplemental water supply for the 53

PAGE 54

Everglades National Park. Given these co mpeting demands, management of the water resource is a major challenge. History and hydrology Geological surveys indicate that La ke Okeechobee formed about 6,000 years ago, when ocean waters retreated from covering what is now Florida, leaving a large shallow lake with a maximum depth of about 19-21 feet [ Florida DEP, 2005]. Much of south Florida remained under sheet flow conditions from the lakes outflow until the late 19th century, when real estate developers c onstructed a canal connecting Okeechobee with Lake Hicpochee, providing another outlet. Additional canals were built in the early 1900s to provide slow continuous drainage for agricultural and flood prevention purposes, with the ultimate goal of draining the entire northern hal f of the Everglades. With the advent of small farming communities around the lake, a muck levee was constructed around the southern shore. However, two large hurricanes in the 1920s caused major flooding and human casualties, at which point federal help was enlisted through the US Army Corps of Engineers (U SACE) to prevent further losses of this nature. The USACE built the Herbert Hoover Dike around Lake Okeechobee, and created a system of canals and le vees that artificially cont rolled all surface inputs and outputs to the lake, except for Fisheat ing Creek, which remained natural. This system allowed the lake levels to be monitored and controlled to prevent flooding and measure irrigation allowances, a nd as a result, much of the land directly surrounding the lake was converted to agricul tural uses, with dairy and beef farms in the north and sugar cane and vegetables in the south [ U.S. EPA 1996]. These land use changes dramatically increased the loads of N and P inputs into the lake, which resulted in much of the hyper-eut rophication problems being deal t with today. An abundance of 54

PAGE 55

P has caused excessive algal growth in Lake Okeechobee, and hyper-eutrophic P levels cause algal blooms, or excesses of growth, in which dissolved oxygen is depleted, in turn killing other aquatic biot a and increasing the am ount of treatment needed to make the water safe for drinking. Many algal blooms have been shown to be in proximity to areas with high soil P, such as agriculture or poultry and dairy farms, leading researchers concerned with water quality to examine the relationships between soil, hydrology, and land use. In total, the LO watershed covers a huge area of 12,000 km2 that extends from Orlando in central Florida, to the Everglades in the southern tip. Except for the natural input of Fisheating Creek, by this point all surface runoff to the la ke is directed through man-made canals and gated streams. Hydrological inputs to t he lake are mostly from precipitation (39%), the Kissimmee River (31%), S-191, the Taylor Creek/Nubbin Slough (5%) and multiple small inflows such as Fisheating Creek [ Florida DEP, 2001]. Major outflows are evapotranspiration (66%), the Caloosahatchee River (12%), canals that discharge to the Everglades ( 18%), and the St. Lucie Canal (4%) [ Florida DEP, 2001]. Precipitation in Lake Okeechobee is very seasonal; about 75% of the 53 inches/year occur during the summer storm season from May to October [ Purdum et al. 1998]. North of LO, elevations in the watershed range from 4 to 23 meters, but south of the lake, the landscape has an even lower gr adient that is poorl y drained with many marshes and wetlands. Field runoff and nutri ent transport in the LO watershed are controlled by water table fluctuation, mostly within one to two mete rs from the surface [ Knisel et al. 1985]. Surface runoff occurs when soil pore spaces are filled with water 55

PAGE 56

and the water table is at the gr ound surface. This suggests a slow lateral or vertical groundwater infiltration may be an important fa ctor in nutrient transport in areas with more slope [ Flaig and Reddy 1995]. Additional well-field and tracer studies done in basin S-191 indicate that the capacity of a region for sub-surface storage is a better indicator of surface runoff potential than slope [ Flaig and Reddy 1995], however, subsurface storage capacity is in turn determi ned by a combination of topography and local boundary conditions. The different P flow-paths and topographical conditions throughout the watershed indicated that non-point source P management and reduction must be done on a case by case basis, rather than an overarchi ng management plan. Geology and soils Much of Floridas underlying geology is marine sediment made up of sand, gravel, and clay lenses, with a network of artesi an aquifers at depths below 40 meters. The local soils north of Lake Okeechobee are pr imarily poorly-drained, sandy and naturally infertile Spodosols, or flatwoods, (45%), with some deposits of Entisols and Histosols near the Everglades Agri cultural Area (EAA) [ USDA-ARS 2005]. Spodosols in Florida are of the Aquod, or wet type, from 0.5 to 3 m depth, char acterized by a shallow and fluctuating water table. Natural vegetation in Aquod soils is water loving, and in the LO watershed includes palms, pines, wet prairie grasslands and wetland species. Most of the soils in the watershed are composed of greater than 90% sand, with high infiltration rates and poor internal drainage due to low permeability of the spodic horizon [ Flaig and Reddy 1995]. The sandy soils in the LO watershed generally do not retain P in their surface horizons, making leaching into gr ound and surface waters a problem, however, P can be retained in Spodosols given the si multaneous presence of Alor Ferich spodic horizons [ Yuan 1965]. 56

PAGE 57

In the 1980s, researchers found that the Taylor Creek/Nubbin slough (S-191) basin of the LO watershed provided only 5% of the water input to the lake, yet 28% of the total P load [ Federico et al. 1981]. Specifically, the Taylor Creek/Nubbin Slough basin is mainly comprised of the fine sandy soils Myakka/Immokalee/Waveland and Wabasso/Felda/Pompano, characterized by hi gh hydraulic conductivities of over 16 cm/hr [Campbell et al. 1995]. This and several other hi gh-impact basins were identified as having multiple dairy operations, which we re putting high levels of P into the fine sandy soils that have high infiltration rates and do not retain soluble-P well [ Knisel et al. 1985]. The lack of deep percolation into spodic hor izon soils that could potentially retain the excessive P only compounded the problem [ Yuan 1965], resulting in high concentrations of soluble P being transported via lateral seepage above the spodic layer or overland flow from the dairies to adjacent streams and wetlands that ultimately discharge into the lake [ Campbell et al., 1995]. Ecology and wildlife Lake Okeechobee is a naturally eutrophic lake meaning that its default state is a murky appearance with high levels of nutrients. This serves to make it an extremely productive ecosystem, given that it does not reach the hyper-eut rophic state of the 1970s and 1980s. Additionally, the Okeechobee Ba sin contains approximately 4,000 km2 of wetland area, mostly in the Lower Kissi mmee River Basin, that are connected via canals and streams to Lake Okeechobee [ Rosen et al. 1996]. Both these wetlands and the surrounding watershed areas provide cruc ial wildlife habitat for birds, fish, and plants. Lake Okeechobee itself supports a signifi cant largemouth bass and black crappie sport fishery, as well as commercial fisheries for catfish and bream that in 1993 were 57

PAGE 58

generating over $30 million per year in the local economy [ Fox et al. 1993]. There are over 40 recorded species of fish in the lake that depend on macro-invertebrate and zooplankton populations, which in turn pr ovide food for numerous migratory wading birds, raptors, and alligators, including the federally endangered Snail Kite [ Havens and Gawlik 2005]. The lakes natural littoral zone wa s much larger than what it is today under current levee conditions, and has cons equently altered the vegetation, reducing populations of native willow, spikerush, and sawgrass, and encouraging the growth of invasive species such as cattail, torpedograss and hydrilla [ Havens and Gawlik 2005]. Submerged aquatic vegetation p opulations have also declined causing increases in algal growth and hyper-eutrophication, as th ere are more available nutrients for algae growth. Management and restoration The Lake Okeechobee Protection Bill was passed in the Florida legislature in 2000, which requires several different governing bodies to work together to implement P TMDL regulations. The Bill operates in c onjunction with the Comprehensive Everglades Restoration Act (CERP), as outflow from t he lake is an input to ecosystems in the Everglades. In 2001, TMDL limits were dec ided upon for the lake with an overarching adaptive management approach: if newer resear ch and monitoring revealed better load limitations, the TMDL regulations woul d be changed to reflect that research. Current TMDLs are based on various combinations of computer model simulations of the lake with an optimal bal ance between its designated Class 1 uses of urban water supply, agricultural irrigation, Everglades flow, habitat, aquifer recharge, flood control, recreation, and navigation [ Florida DEP, 2001], and historical record. Annually, the total allocation for all non-po int source inputs to Lake Okeechobee is 140 58

PAGE 59

metric tons of P, with an in-lake P concentra tion target of 40 ppb in the pelagic zone. Inlake concentrations of P in the water column from 1995-2000 average about 100 ppb or 641 metric tons/year, with recorded concentrati ons as high as 400 ppb during intense storms or hurricanes that stir up sediments from the lake bottom [ Florida DEP 2001]. To achieve the ultimate desired concentra tion conditions in the lake, researchers calculated that the total annual P load must be less t han 423 metric tons/year [South Florida Water Management District (SFWMD) 2001], which has never been close to being met since implementation in 1995. Basi n S-191 is a main offender in failing to meet its TMDL targets, with specified basin management action plan (BMAP) goals of 0.18 ppm and 25.1 metric tons/year TP, and recorded rolling averages from 1995-1999 of 0.65 ppm and 91.3 metric tons/year TP [ Florida DEP, 2001]. There are various committees of researchers, advisors, and managers that monitor and decide on how the LO watershed should be managed and what kinds of BMPs are most effective. To date, reduction strategies have not been effectiv e in meeting the decided upon phosphorus TMDLs, and the results of this research only reinforce that the current limits may not be feasible. Whether this is a problem of BMP implementation and load reduction efforts or realistic P load expectations remains to be seen. Data Basin S-191 (Taylor Creek/N ubbin Slough) is a 486.9 km2 watershed with 85 defined stream reaches in WAM (Figure 2-1b) Stream flow is generally slow-moving and characterized by broad flood-plains with poorly defined stream channels, typical of the Coastal Plain region. Land use has varied hi storically, but for the purposes of this study, static land use from the year 2000 wa s used for all simulation years. In 2000, 56% of the watershed area was designated as improved pasture, 6% as scrub/brush 59

PAGE 60

land, 7% as freshwater marsh, 4% as fiel d crops, and all other land uses less than 4% each. Required WAM inputs such as daily pr ecipitation, daily and mean monthly maximum and minimum temperatures, sola r radiation, and wind speed data were obtained online at the SF WMD DBHYDRO browser (www.sfwmd.gov/org/ema/dbhydr o). Precipitation and water quality stations in S-191 are in 14 locations throughout the basin (Fi gure 2-1b), which are grouped into spatially distinct "rain zones" within the waters hed by WAM. "Rain zones" are user-defined Thiessen polygons within the GIS interface that allow the user to assign different weather stations to represent different areas of a particu lar watershed. The cumulative annual rainfall averaged yearly over all weather stations from 1967 to 2001 was 115.6 cm, with a minimum of 73.1 cm in 1981, and a maximum of 156.9 cm in 1982. GIS coverages including land use for the y ear 2000, soils, hydrography, topography, and relevant hydrological structures were pr ovided by the Lake Okeechobee CERP Project Manager. WAM was run using a uniform cell si ze of 1 ha. Calibration and validation of the model over all 20 basins surroundi ng Lake Okeechobee (including S-191) was performed by SWET before WAM was provi ded to the SFWMD, using data from 19912000 [ Jacobson 2002]. Historical annual ENSO climate data used in this analysis include 9 JMA El Nio years starting with October of the current year unt il the following Septem ber, (1965, '69, '72, '76, '82, '86, '87, '91, and '97), 10 La Nia y ears (1967, '70, '71, '73, '74, '75, '81, '88, '98, and '99), and 17 neutral years (remaining years from 1967 to 2001). These data are classified by the Center for Ocean-Atmos pheric Prediction Studies (COAPS) based on 60

PAGE 61

the JMA's recording of surface sea te mperatures anomalies. Using the M-ENSO classification [ Grard-Marchant et al. 2008], there are 109 El Nio months, 96 La Nia months, and 236 neutral months. The M-ENS O re-classification of this data as compared to the annual method results in about the same number of total El Nio months, a 20% decrease in the number of La Nia months, and a 15% increase in neutral months. Watershed Assessment Model (WAM) WAM's function is to serve as a tool for watershed assessment, using the appropriate model components and available data sources [ Soil and Water Engineering Technology, Inc. 2004]. WAM includes four nutrient sub-models for different land uses: the Groundwater Loading Effect s of Agricultural Manage ment Systems (GLEAMS) model [ Leonard et al. 1987], the Everglades Agricu ltural Area Model (EAAMOD) [ Bottcher et al. 1998; Soil and Water Engineering Technology, Inc. 1996], and two submodels developed by SWET specifical ly for wetland and urban landscapes [ Soil and Water Engineering Technology, Inc. 2004]. For basin S-191, both GLEAMS and EAAMOD were used to simulate daily nutri ent loads based on recorded land use, precipitation, and simulated stream flow time series. Stream reaches in the model are routed to the outlet by so lving the continuity equat ion and Manning's equation for uniform channel flow with a variable time step of approximately 15 min, based on the simulated stream velocity (see [Jacobson and Bottcher 1998], for details). There are no inter-cell interactions in WAM simulations. WAM simulated nutrient loads in 85 userdefined stream reaches of basin S-191, which ultimately merge at a single reach (reach 2), which enters Lake Ok eechobee (Figure 2-1b). 61

PAGE 62

For the S-191 watershed, six years (1995-2000) were used as a calibration period for flow and P [Jacobson 2002], while validation was pe rformed using the years 19911994. WAM is a physically based model that ut ilizes the characteristics of the watershed in the generation of flows and pollutant loadings. Thus, calibration in the classic sense, whereby various parameters ar e modified until a goodness-of-f it is achieved, is not done. Rather, known data from the watershed are used to verify that the model is appropriately simulating the physical dynam ics of the watershed. Measured flow and P concentration for the period 1991-2000 were pr ovided by SFWMD. While daily flow measurements were provided, P measurements were more sporadic, ranging from daily to bi-weekly grab samples. Daily concentrations were calculated by SWET by interpolating between grab samples, which caus es errors in actual concentrations and therefore P loads as well. In fact, the flow, concentration, and load data contain measurement uncertainty that can be esti mated and incorporated into the performance evaluation of WAM. Using a goo dness-of-fit indicator, such as a Nash-Sutcliffe (NS) coefficient [Nash and Sutcliffe 1970], which can be modified to include measurement uncertainty, allows model evaluation to cons ider this inherent data uncertainty (Harmel and Smith, 2007). In this analysis, the NS coefficient was used to quantify WAM goodness-of-fit both with (modifi cation 1) and without consi deration of measurement uncertainty. The probable error range (PER) va lues used to compute the modified NS coefficients were based on previous uncertainty estimates [ Harmel et al. 2006]. In the initial calibration and validation simulations, the predicted flows were consistently below measured values. Examini ng the crop coefficients revealed that the leaf area index (LAI) estimates were bei ng reduced too much in the winter for the 62

PAGE 63

regional grasses. Increasing these LAI coefficients improved the match between simulated and measured values [ Jacobson 2002]. The NS coefficients for simulated and measured monthly flows were 0.79 for the calibration period and 0.66 for the validation period, which represent good" model performance according to [ Moriasi et al. 2007]. When measurement uncert ainty in the flow data was taken into account using a PER of 6%, the modified NS coefficients showed an increase to 0.99 for the calibration period and 0.99 for the validatio n period. When large differences between simulated and measured P loads occurred, the correctness of land uses of contributing areas was examined. For example, some areas were identified as "improved pasture" when "intensive pasture" was more appropr iate. The management practices for the basins were also evaluated to ensure appropriate representation of current practices [ Jacobson 2002]. The NS coefficients for si mulated and measured monthly P concentrations were 0.53 for the calibration period and 0.39 for t he validation period, which were less accurate than the flows, as expected, and indicate "unsatisfactory" model performance. When uncertainty of c oncentration data was taken into account using a PER of 27.1%, the modified NS coeffici ents increased to 0.81 for the calibration period and 0.75 for the validation period, indicating that WAM simulated P concentration values that are as good as the uncertain data. The NS coefficients for simulated and measured monthly P loads were 0.73 for the calibration period and 0.63 for the validation period. With a load PER of 27.8%, the modified NS coefficients increased to 0.94 for the calibration peri od and 0.93 for the validation period. 63

PAGE 64

Analysis and Statistics Precipitation and stream flow Observed daily precipitation data from a ll 14 weather stations were averaged and summed into monthly cumulative values, normalized by monthly average, and averaged over all years and by JMA or M-ENSO phase for each month and over all months by year. These anomalies are visualized in both monthly box plots and scatter plots over all years (Figures 2-2, 2-3, and 2-4) in the Results and Discussion section. Simulated daily stream flow data from WAM were averaged into a monthly rate, normalized by monthly average, and averaged over all years and by JMA or M-ENSO phase for each month and over al l months by year. The resulting stream flow anomalies can be seen in monthly and seasonal box plot s for each ENSO categorization (Figures 2-5, 2-6 and 2-12) in the Re sults and Discussion section. Phosphorus concentration and lo ading: JMA ENSO classification Simulated daily sediment and solubl e P concentrations were summed and averaged into total daily P, and were t hen averaged basin-wide into an average value per month and normalized by the monthly av erage over all years. Simulated average daily total P loads were computed by multiply ing daily flow by P concentrations for both sediment and soluble P, and averaged into basin-wide monthly and seasonal values across all sections of stream reaches. Seasons were defined by known southern Florida precipitation and climate regi mes and were designated into four categories within their ENSO year: February-April (FMA), MayJuly (MJJ), August-Oc tober (ASO), and November-January (NDJ). Load and concentr ation anomalies were computed by normalizing loads against the average climatol ogy-based monthly load, which included all ENSO phases (load anomal y = summed monthly load average monthly load), and 64

PAGE 65

by separation into JMA-defined ENSO phases. Results were visualized in both box plots and scatter plots. Temporal differences in P loads and concentrations were analyzed as "anomalies" rather than actual values, as is commonly done in climate science when the year-to-year variability is high. This proc edure normalizes the data to remove bias and effectively see anomalies in the data [Mason and Goddard 2001; Rajagopalan and Lall 1998; Ropelewski and Halpert 1987]. Especially when comparing data directly with the JMA index, which is defined by temperat ure anomalies, it makes sense to analyze precipitation and nutrient vari ables in the same manner. Measures of the modeled dat a's prediction variability at the monthly, seasonal, and JMA phase-based level were calculated and compared via maximum and minimum values, range, and standard deviation. Data ar e graphically presented in box plots for P load (Figures 2-9 and 2-11) and P concentration (Figures 2-7 and 2-13) in the Results and Discussion section. Differences between nutrient loads and concentrations in JMA ENSO phases and months can be seen, and variability evaluated from the statistics calculated. The pooled variance between months or di fferent JMA Index ENSO phases was too large, and the sample size (number of ENSO phase events) was too small to identify high probability of forma l significance. Thus, statistica l tests of significance, such as Student's t-test or classic ANOVA, were not used for assessing variance and significance. Instead, contingency tables were used. This simple and proven alternative to parametric statistics is appropriate fo r estimating the probabi lities of climate anomalies based on ENSO phase [ Mason and Goddard 2001; Ropelewski and Halpert 1987; Wilks 1995]. This methodology was used to determine whether the events of 65

PAGE 66

interest (high or low nutrient loads) were more or less likely to occur given different values of the independent variable (ENS O phase). Contingency table theory and the hypergeometric probability distribution were used to determine the significance of the number of times that the seasonal simu lated nutrient load during each ENSO phase was above or below a pr e-defined threshold. Using the methodology outlined by Mas on and Goddard, three-month seasonal normalized predicted mean nutrient load thresholds were calculated for each ENSO year (1965-2001) and then sorted and separated into three terciles: low (T1), mid (T2), and high (T3) for general comparison. Then a TMDL-based categorization of above or below a TP concentration of 0.22 mg/L, as there was no significant probability of any season or phase having a concentration below the current BMAP policy limit of 0.18 mg/L. The low tercile indicates load anomalie s dominated by negative values, while the high tercile is dominated by positive values The mid tercile indicates both positive and negative anomalies of smaller magnitude, a nd the policy-based table indicates which months and JMA ENSO phases are below an investigatory concentration based on the actual basin S-191 TP target of 0.18 mg/L. The number of times that the simulated nutrient load(concentration) during the separate ENSO phases was in each tercile(concentration level) was tabulated for each season. These frequencies indicate the likelihood of observing a certain tercile(conc entration level) of load(concentration) in each ENSO phase and season over all simulated years. There were a total of 12 contingency t ables created for each of the three ENSO phases in each of the four tri-month seasons. Using Fishers exact test [ Fisher 1935], the probability that a certain ta lly of tercile of load within a given season and ENSO 66

PAGE 67

phase is statistically greater t han the other terciles is calcul ated. This probability is equal to the right tail of the hyper geometric distribution (Equation 21), and the information is represented in the conti ngency table (Table 2-1). r n xr bn x b nbrxHxXPbr xx),min(),,;()( (2-1) In Equation 2-1 and Table 2-1, for all calculations there are a total of n = 36 years, of which r are either El Nio (9), La Nia (10), or neutral (17). The total number of T1 or T3 events counted within a certain ENSO phase across all four seasons is represented by b and a respectively. The number of ta llied T1 or T3 events within the particular season and ENSO phase of interest is represented by x or y, respectively. For example, in the November to January season during El Nio, n = 36 total years, r = 9 El Nio years, b = 10 total T1 events across all seasons in El Nio, a = 14 total T3 events across all seasons in El Nio, while t he tallied T1 in the NDJ season is x = 0 and the tallied T3 in the NDJ season is y = 3, leaving the remainder tallied in T2 as r-y-x = 6 (note that the total T1+T2+T3 events equals r the total number of El Nio years). According to Fishers exact test in this example, for the tallied number of events in a tercile to be significantly greater than the ot her terciles with 95% c onfidence, there must be at least 5 events. To achieve 99% confid ence, there must be at least 6 events. As the test used is right-tailed, we are only i dentifying those loads which are significantly greater than the others, as high nutrient loads are of more risk and management significance. Therefore, during the NDJ season in El Nio years, the mid-tercile, T2 = 6, is the only one in which the number of event s is greater than w hat would be expected 67

PAGE 68

from random chance, wit h 99% confidence. These probabilities indicate the thresholds, pre-defined using the hypergeom etric distribution and Fisher's exact test, that must be met or exceeded by the tallied load data to a ccept a given tercile's (or concentration levels) significance at the 95th or 99th percentile within each ENSO phase, between seasons. The relevant assumptions for Fisher's exact test and the hypergeometric distribution are similar to t hose of a Monte Carlo signific ance test, for which it is assumed that the observation in one year is independent of the obser vations in all other years. Independence can be confirmed using autoregressive methods [ Mason and Goddard, 2001; Wilks 1995]. For our purposes, if t he selected ENSO years (the r sample) are evenly distribut ed through the entire period (n ), then autocorrelation and trend effects are not important, and errors will be conservative [ Mason and Goddard 2001]. In our relatively short 37-year per iod, during which EN SO patterns can be considered climatologically and geophysically stable, each ENSO phase was evenly distributed through the tested period with very little clustering. To be certain that the underlying relative frequencies of a particula r ENSO phase differ from one-third, the expected probability of any of the three phas es occurring due to random chance, for a period of about 40 years within which there are approximately ten La Nia or El Nio years of interest, relative frequencies of 50% or more are needed (for details, [ Mason and Goddard, 2001]). The contingency tables were derived fr om historical JMA and M-ENSO events, and the resulting probabilities ar e not meant to serve as a fo recast. If the probabilities calculated using contingency tables were used to forecast, the table should be 68

PAGE 69

constructed so that the probabi lities are contingent on foreca sts of ENSO events, rather than observed events. Rather, these contin gency table probabilitie s are only meant to provide a quantitative m easure of the significance of the relative seasonal nutrient loads in each ENSO phase. Phosphorus concentration and lo ading: M-ENSO classification The recent Zierden M-ENSO Index [ Grard-Marchant et al. 2008] was used to try and quantify statistically significant differences between months in different ENSO phases. As was previously noted, no stat istical difference was found between ENSO phases when a variety of statistical te sts were done using yearly ENSO JMA classifications to look at differences in obs erved precipitation and simulated stream flow and phosphorus concentrations and loads. With the finer temporal resolution of the Zierden Index to better capture the changes in SST, more standard statistical tests could be used, as it has been initially noted that the M-ENSO index is better able to quantify different ENSO events in the s outheast United States than the annual JMA index, especially in the summer [ Grard-Marchant et al., 2008]. The finer temporal scale allowed t he use of a more standard single-factor ANOVA to determine if there was a statis tically significant difference between the means of the three M-ENSO phases, and mo re specifically, which two ENSO phases were different. ANOVA analysis was performed using a p-level of 0.05 on the observed precipitation data, and the simulated stream flow, P concentration, and P load anomalies. 69

PAGE 70

Results Monthly Precipitation and ENSO Phase Because certain months in El Nio or La Nia years in Florida can exhibit more rainfall than others, it logically follows t hat increased precipitat ion would result in increased runoff and, consequently, incr eased nutrient loading. The south Florida/Everglades area is semi-tropical in terms of climate patterns, with a general wet season from May to October an d a mild dry season from November to April. El Nio winters in Florida are usually wetter than La Ni a or neutral years, with southern Florida showing the strongest response [ Livezey and Smith 1999; Schmidt et al. 2001]. Our results were consistent with these trends. Su mmer and fall precipitatio n trends during El Nio were clearly demarcated, although anoma lies tended to be small, and spring in southern Florida can actually be drier than during other phases [ Sittel 1994]. La Nia winters and summers were shown to be drier overall across Florida as a whole, while La Nia springs are drier in southern Florida [ Livezey and Smith 1999; Sittel 1994]. Figure 2-2a shows spatially averaged and normalized monthly tota l precipitation, averaged across all stream reaches for all years and separated by ENSO phase. Error bars are not shown on this gr aph, as the small ENSO phase sample sizes make error bars large enough to obscure the visible pa tterns. Straight line segments connect adjacent data points and are not r epresentative of a mathematic al function, but they are included to improve visualization of temporal patterns. As expected from previous observed ENSO patterns in south Florida [ Livezey and Smith 1999; Schmidt et al. 2001], above-average precipitation was obser ved in November, December, January, February, and August during El Nio years, and from April to June during La Nia years. Neutral year precipitation was clos er to average and had less overall variability. 70

PAGE 71

Figure 2-3 shows total precipitation (cm) anomalies averaged for all precipitation stations, and presents spatiall y averaged precipitation anom aly means by month and by JMA ENSO phase. In neutral ye ars, monthly precipitatio n anomalies tended to stay closer to the average (Figure 2-3a), with greater general variability than the nutrient loads, seen in Figure 2-4a with some sma ll ranges, especially in the winter months. However, La Nia and El Ni o precipitation trends (Fi gures 2-3b and 2-3c) did not closely match their corresponding nutrient load trends (Figures 2-4b and 2-4c). Some of this trend is likely due to lags between in filtration and runoff responses, in which extended dry months fo llowed by high-intensit y precipitation causes greater P runoff. In contrast, observed monthly precipit ation anomalies classified using the MENSO index are shown in Figure 4. While the neutral and El Nio months precipitation (Figures 2-4a and 2-4c) retains the same general pattern, the La Nia months show major differences (Figure 2-4b) Specifically, the largest di fferences are seen during the summer months during M-ENSO La Nia mont hs that commonly experience intense convective thunderstorms in Florida: Ap ril, May, and June (Figure 2-4b). This information is consistent with previous res earch on both ENSO in south Florida and the increased accuracy of M-ENSO in the summer m onths, as it was found that significantly lower La Nia summer precipitation levels were found when comparing the M-ENSO with the JMA [ Grard-Marchant et al. 2008]. The M-ENSO classification both reduces the range and average of the summer months La Nia pr ecipitation, which suggests that perhaps the increase in the precipitation in the re st of an annual JMA La Nia is responsible for the signific antly higher precipitations that are commonly expected. 71

PAGE 72

Monthly Simulated Stream Flow and ENSO Phase Florida precipitation patterns are altered by ENSO phase, as we have just seen. In the Lake Okeechobee area of south Florida, stream flow has a more complex relationship with ENSO. Generally stream flow increases during El Nio fall (OctoberDecember) and winter (January-March) seas ons, while strong La Nia events cause decreases in stream flow for fall through spring, and may even extend into the summer months [Schmidt et al. 2001; Zorn and Waylen 1997]. Stream flow patterns are a complex combination of precip itation, local basin characteristics, human construction, and climate, and while flow patterns smooth out the noise of a rainfall time series, a strong ENSO signal may be lagged or harder to identify. These seasonal ENSO flow patterns are br oadly replicated in the JMA (Figure 2-5) and M-ENSO (Figure 2-6) results seen for basin-w ide stream flow in S-191 of the Lake Okeechobee watershed, however, taken mo nth-by-month, the analysis is more complex. In both ENSO designations, neutra l months and years look similar to each other (Figures 2-5a, 2-6a) in that they s hare a pattern marked by more outlier points. Compared to neutral years, El Nio years in the JMA and the M-ENSO (Figures 2-5c, 26c) exhibit higher stream flows with greater ranges from January to March, while the fall months are not as different from neutral conditions, as has been shown in south Florida already [ Schmidt et al. 2001]. Generally, while winter and fa ll precipitation is lower than summer precipitation on average, in El Ni o years these seasons rainfall can double, resulting in higher stream flows. In south Florida during La Nia condition s, January through Ma rch precipitation and flow tends to be significantly lessened [ Schmidt et al. 2001] compared to neutral or El Nio phases, which can be seen in Figures 2-5b and 2-6b. In comparing the JMA and 72

PAGE 73

M-ENSO trends overall, the main difference is again in the La Nia phase, which, like precipitation, has June with a much larger range using the JMA index (Figures 2-5b, 26b). The sign of the average June anomaly al so changes from positive in the JMA (Figure 2-5b) to negative in the M-ENSO (Fi gure 2-6b). Again, the finer temporal scale of the M-ENSO index may be better demonstrating south Floridas summer climate and hydrology patterns than the annual JMA. Simulated Monthly P Concen tration and ENSO Phase In the Lake Okeechobee watershed, concent rations of nutrients are based on both climate and land management factors, and we would not expect them to show ENSO trends as clearly as those in precipitation and stream flow. Depending on the lag from application of fertilizers to entering bodi es of water, one could expect a month or two lag of peak concentrations behind mont hs with maximum precipitation. A more detailed correlative lag analysis of different observed data is found in Chapter 3 and Appendix A. In the analysis of simulated TP concentra tions separated by month in the annual JMA ENSO designation (Figure 2-7), some patterns can be distinguished that are more similar to simulated stream flow patterns than precipitation patterns. Like precipitation and flow, TP concentration in neutral years tends to have more consistent, although wide, ranges with means closer to zero (Fi gure 2-7a). Concentrations in El Nio years mimic flow trends as well, with the anomalous high stream flow and precipitation in the winter months of January to March caus ing a corresponding possible wide range of positive concentration anomalies (Figure 2-7c). In La Nia years, the low January to March flow and precipitation causes corres ponding negative concentration anomalies in the winter (Figure 2-7b). Comparatively, TP concentration separated by the monthly M73

PAGE 74

ENSO index (Figure 2-8) shares the sa me general neutral, El Nio, and La Nia patterns. However, the finer M-ENSO scale cr eates a more distinct separation between the warm and cool phases. The ranges of P concentration in La Nia months have been restricted to a tighter and sm aller band (Figure 2-8b). while the possibility of positive concentration anomalies in El Nio months has increased, with the ranges favoring the positive side, especially in March (Figure 28c). Like comparisons between the JMA and M-ENSO precipitation and stream flow series, the M-ENSO separation of P concentration also may demonstrate more accu rate fine scale hydrological dynamics in south Florida, especially in wetter El Nio winter months and drier La Nia summers. Simulated Monthly P Load and ENSO Phase Figure 2-2b shows the simulated monthly total P load anomalies, averaged across all reaches for all years and separated by ENSO phase. Trends in P loads mostly followed the precipitation trends shown in Figure 2-2a. Figure 2-2 suggests a strong El Nio association with peak loads in the winte r months in Florida, and an even stronger association with peak loads in the summer of La Nia years. Error bars are not shown on this graph, as the small ENSO phase sa mple sizes make error bars large enough to obscure the visible patterns. Instead, the wit hin-period variability is summarized in the statistics of Table 2-2 (by month) and Table 2-3 (by JMA and M-ENSO phase). The degrees of variability were analyzed relative to each other, to ascertain which months or ENSO phases were associated with extr eme events in the available data. The month with the largest variability (as determined by the standard deviation) in load anomalies was June (Table 2-2). Sept ember had the greatest range between maximum and minimum loads, and the outlier maximum September load is clearly visible (Figure 2-4a). The m onth with the smallest stand ard deviation was December. 74

PAGE 75

While these results suggest that Septem ber and June conditions may be the most favorable for high nutrient loading, the la rge degree of variability of the data must be kept in mind. Conversely, the low variabi lity associated with loads in December suggests that a nutrient load forecast in December of a La Nia year could be considered more reliable. For both the JMA annual and M-ENSO monthl y classifications, the El Nio and neutral phase summary statistics stayed virt ually the same (Table 2-3). The largest difference between the monthly and annual EN SO indices was again seen in the La Nia phase. Using the JMA i ndex, La Nia years have the largest standard deviation and range of the ENSO phases, as well as the absolute maximum load anomaly (Table 2-3). This suggests that despite the lower precipitation and stream flow inherent in La Nia years, there is still a large risk for P loading. When ENSO designation is examined by the M-ENSO index, however, La Nia months have by far the lowest standard deviation, range, and mean, as well as t he absolute minimum anomaly when compared to the other phases (Table 2-3), which make s more physical sense with what is known about the decreased rainfall and stream fl ow during La Nia years, and corresponds with research using the M-ENSO classification [ Grard-Marchant et al. 2008]. Figures 2-9 and 2-10 show monthly tota l P load anomalies averaged across all reaches, and present P load averaged mont hly and separated by JMA and M-ENSO phase. Using the JMA index, El Nio years (F igure 2-9c) exhibited more variability and a larger range, while La Nia (Figure 2-9b) and neutral years (Figure 2-9a) had more consistent patterns. The most visible peak load trend occurred in June of La Nia years. 75

PAGE 76

Figure 2-9b shows the lar ge range of P load anomalies in June in JMA La Nia years, which is quite differ ent from neutral and El Nio year s, and is similar to trends seen in precipitation, stream flow, and concentration. Majo r differences between El Nio and La Nia years were again apparent in the larger loads during January to March in El Nio years, and the larger loads in May-July of La Nia years. P load anomalies that were below average in the summer were of similar magnitude in both El Nio and La Nia events, despite large differences in nutrient runoff. These patterns are similar to those found in precipitation and stream flow analyses previ ously performed in south Florida [ Schmidt et al. 2001]. Like the other hydrological variables examined, when P load is separated by the M-ENSO index rather than the JMA annual index, neutral phases look mostly the same (Figure 2-10a) with M-ENSO showing larger loads from January to March of El Nio months (Figur e 2-10c), and smaller loads in June through August of El Nio months (Figure 2-10c) and June and July of La Nia months (Figure 2-10b). Simulated Seasonal P Load, Concentrati on, Stream flow, and JMA ENSO Phase Simulations showed that El Nio and La Nia years generally exhibited greater nutrient load range and variability, while neutral years stayed closer to average (Figure 2-11). El Nio years tended to produce greater P load runoff in spring (February-April). La Nia years tended to have larger P loads in the summer (May-July), but the variability was quite large, as indicated by the very strong seasonal signal. Simulated seasonal groupings of nutrient loading temp orally match with seasonal simulated stream flow (Figure 2-12) and concentration (Figure 2-13) time series. Seasons with positive flow anomalies and wide ranges of stream flow and P concentrations seem to match with positive P load anom alies and wide ranges of values, such as February to 76

PAGE 77

April (FMA) El Nio flow and concentration (Figure 2-12a and 2-13a) as compared to FMA El Nio P load (Figure 2-11a) Simulated nutrient loading in south Florida is very similar to observed precipitation and str eam flow in different ENSO phases. The greatest increases in flow may cause an increased flush of nutrients, increasing P concentration and consequently nutrient load compared to neutral conditions that occurred in El Nio winter and spring, while onl y winter precipitatio n was significantly depressed in La Nia years. As in the literature, during the other seasons, levels were lower than average, but not significant [ Schmidt et al. 2001]. Seasonal nutrient loads were analyzed both in box plots and in contingency table significance testing. Table 24 summarizes which tercile wa s dominant in which season, and to what percentage significance level the actual P load for a given season and JMA ENSO phase differed from random chance as calculated (Equation 21) with the relevant input parameters. Only one tercile is listed per season, as Fisher's exact test results conveniently had only one tercile si gnificant above 95% in each season and phase. In El Nio years from August-October, no tercile dominated, that is, all were equally likely. Figures 2-11a through 2-11d, whic h help visualize the tendency of low, typical, and high load anomalies, show t he normalized total P load averaged across all reaches, grouped by season and ENSO phase. The potential P runoff or negative load anomaly in each ENSO phase varied widely by season. The use of terciles has been criticized wi dely in research as being a physically meaningless metric, as the designation of thir ds is only relevant to itself. A possible solution to this issue is to use the same contingency table methodology with P limits as defined by BMAPs or TMDLs for basin S191. The FL DEP has created a BMAP for 77

PAGE 78

basin S-191 that designates a TP concentra tion limit of 0.18 mg/L and 91,300 kg/year. However, when examined either monthl y or seasonally, t here were never enough months that were below the 0.18 mg/L limit to be statistically significant at any level. In fact, there were never any springs, summers, or fall seasons below the limit in any of the ENSO phases, and only 4 out of 17 were below the limit during the winter (NDJ) season of both La Nia and Neutral years, making this relatively useless as a management tool to learn which seasons hav e relatively more risk than others. When the TP concentration limit is increas ed to just 0.22 mg/L (merely as an investigatory measure), (Table 2-5), one can better assign some meaning to the temporal analysis of P load risk in different c limatic regimes. Since the tables display results for P load and concentration, they are not directly comparable, however, interestingly, the NDJ winter season in a ll JMA ENSO phases showed a significant chance of being under a 0.22 mg/L P concent ration limit, as did the FMA spring season in La Nia years (Table 2-5). All other seasons and ENSO phases did not have a significant chance of being unde r the P concentration limit. February-March-April Positive P load anomalies from Febr uary to April (Figure 2-11a) were overwhelmingly associated with El Nio years, which are characterized by higher than normal precipitation, whereas La Nia y ears showed negative P load anomalies and a tight range. Neutral year anomalies in t he February-April season remained centered around zero, as expected, although several large outliers were noted. Both flow and concentration anomalies had the largest ranges in El Nio years as well (Figures 2-12a and 2-13a), implying the possibility of very variable conditions, although their median values were centered on zero or negativ e anomalies. P Load contingency analysis 78

PAGE 79

(Table 2-4) was consistent with the trend s een in Figure 2-11a, with the highest load tercile being dominant and significant at the 99% level during El Nio years compared to the chance of observing that result generated from a random sample of the underlying population, and low loads dominating during both La Nia and neutral phases. Concentration contingency analysis (Table 2-5) with a 0.22 mg/L investigatory concentration limit showed that only duri ng La Nia years was t here a significant chance of coming in under the limit in the FMA season, corresponding with the dominance of T1 during La Nia years. These results are consistent with other analyses of seasonal trends in south Fl orida, wherein winter precip itation and ENSO phase are strongly linked, and weather stations in south Florida experienced as much as 50% to 150% more rain than typical [ Schmidt et al. 2001]. In this case, total average FebruaryApril nutrient load in El Nio years was 119% greater than in neutral years and 156% greater than La Nia years. May-June-July Conversely, the May-July season (Figures 2-11b, 2-12b, 2-13b) had maximum P load, flow, and P concentration anomalies in La Nia years, with a maximum load surpassing that of any El Nio event. T he range of P load, flow, and concentration anomalies in La Nia years was also large, indicating a high degree of variability in May-July loads, which is also consistent wit h precipitation trends in the present study and in Schmidt et al. (2001). The trends seen (Figure 2-11b) are again consistent with load contingency table analysis (T able 2-4), with the high load tercile significant at the 99% level in La Nia years, and the low load tercile significant at the 99% level during both El Nio and neutral years. El Nio y ears were associated with negative P load anomalies, and neutral years were centered on zero (Figure 2-11b). May-July total 79

PAGE 80

average P loads were 132% greater in La Nia years than in neutral years and 147% greater than El Nio years. Contingency analysis using the theoretical 0.22 mg/L concentration limit (Table 2-5) showed that there was no significant chance of being under the limit in any of the ENSO phases dur ing the May to July season, even though the load contingency table sug gested that relatively lowe r loads (by tercile) were possible in El Nio and Neutral years. This contradiction s hows the relative strength of each method of statistical cont ingency table analysis (tercile or practical policy based), and that nutrient risk assessm ent in the summer season of La Nia months or years should be a priority. August-September-October August-October is the season with the lowest precipitation in Florida, and this is reflected in the resulting low P load anomal ies across all ENSO phases (Figure 2-11c). The low P loads may be due to heavy rains in the previous season having removed much of the soluble P. There was a slight visible trend that La Nia years had a greater negative load than the other EN SO phases and slightly more variability and average higher load in El Nio years. Neutral y ear total average P load anomaly was 120% greater than El Nio years and 260% gr eater than La Nia years. Phosphorus concentration (Figure 2-13c) and stream flow (Figure 2-12c) during the fall season were more variable than P load, however still had negative anomaly median values in El Nio and La Nia phases. Load contingency table analysi s (Table 2-4) did not identify any tercile as dominant over any other to a signi ficant extent during El Nio years, although it did show that the low tercile was signifi cant at the 99% leve l during La Nia and the high tercile was significant at the 99% level during neut ral years. Concentration contingency analysis did not identify any ENSO phase as having a significant probability 80

PAGE 81

of being under the investigator y 0.22 mg/L TP limit (Table 2-5), meaning that the fall season has a significant possibility of exceeding concentration regulations during all ENSO regimes. November-December-January November-January showed a potential for higher P load (Figure 2-11d) and flow (Figure 2-12d) in both El Nio and neutra l years, while La Nia years demonstrated more negative load and flow anomalies with a tight range and small variability. These findings are consistent with those of Schmidt et al. (2001); 57% of weather stations in Florida received significantly greater precipit ation in November-January of El Nio years than in neutral years, and 65% received signifi cantly less precipitation during La Nia events. Total November-January season nutri ent load anomalies averaged 85% greater in El Nio years than neutral years, and 200% greater than La Nia years. Load contingency table analysis (Table 2-4) found that the mid tercile was significant at the 99% level during El Nio years compared to random chance, indicating slightly above average nutrient loading. Looking further into the contingency data analysis, there were zero events tabulated in the lowest load te rcile for this season in El Nio years, indicating that, while not overwhelming, the nu trient runoff was consistently higher than average. La Nia and neutral years also had significant mid load terciles and zero events tabulated in the low load tercile, which may suggest that the November-January season is of slightly higher nutrient loadi ng risk for all ENSO phases. This is in contradiction to the concentrati on contingency table results (T able 2-5) that consistently shows that across all ENSO phases, winter is the only season that has a significant probability of meeting a concentration upper limit of 0.22 mg/L. When concentration is combined with flow data to get nutrient load values, there is a possibility of higher loads 81

PAGE 82

than desirable. Risk assessment using bot h standard load and concentration based policies is made comple x by these dynamics. M-ENSO Statistics (ANOVA) of Monthly Pr ecipitation, Flow, P Concentration and Load In single-factor ANOVA analysis performed at the p = 0.05 level for all variables using the monthly M-ENSO classification s ystem, analysis was first performed with all ENSO phases to identify if there were any differences in the variables, i.e., if Nio = Nia = Neutral. ANOVA identified that there was a si gnificant difference in the three-way means for all variables tested: precipitation, stream flow, P concentration and P load. To specifically identify which pairs of ENSO phases encompa ssed significant differences for each variable, single-factor ANOVA wa s repeated for each comb ination of ENSO phases: El Nio and Neutral, La Nia and Neutral, and El Ni o and La Nia (Table 2-6). Results showed that while no differenc es significant at p= 0.05 were found between El Nio and Neutral phases, all variabl es were significantly different between La Nia and Neutral phases. Finally, all variabl es except for precipitation were also found to be significantly different from each other between El Nio and La Nia phases. This suggests that using the M-ENSO index can in fact find statistical differences between phases that the JMA index could not, and that the major differences between ENSO phases are based upon La Nia months bei ng more different from the other two phases in south Florida. Accordingly, in a ll the plots and statistics in this research, neutral and El Nio patterns on a whole tend to be more sim ilar to each other than they are to La Nia patterns. Taken as a whole, however, the contingency table analysis and box plots by month and season provide mu ch more information about the timing of loads than the fact that phases are significant ly different from each other do. As this 82

PAGE 83

pertains to nutrient risk management in the southeast, La Nia months and years have consistently been identified in this research as having more potential for high summer loads, perhaps suggesting that BMPs t hat concentrate on summer month load reductions should be a research target. Summary and Discussion In Chapter 2, the relationships between tw o different discrete classifications of ENSO phase, the annual JMA and the new monthly M-ENSO indices, and simulated P loads for a fixed pattern of land use and management in basin S-191 of the Lake Okeechobee watershed are explored. So me ENSO phases tended to produce significantly greater seasonal P loads (Februar y-April of El Nio ye ars, May-July of La Nia years, and August-September of neutral years) or lower seasonal P loads (MayJuly of El Nio years, February-April and August-September of La Nia years, and February-April and May-July of neutral years). The greater P load potential in certain months was mostly consistent with documen ted trends in greater precipitation. The determination that ENSO phase was significantly related to P loads suggests the need for future study on use of short-term climat e forecasts of ENSO phase to help guide management efforts to minimize nutrient loading not only into Lake Okeechobee but in any area in the southeast similarly affected by ENSO phase. While this study does not provide an accurate simulation of historical nutrient loads in S-191, it is a valid indicator of relationships that can be expected in the southeast United States between ENSO phase and P load. The comparison between a monthly M-E NSO index and t he annual JMA index suggested across P load, concentration, stre am flow, and precipitation, that the MENSO classification reduces both the range and average of the La Nia summer 83

PAGE 84

months. Consequently, this monthly classifi cation which is more representative of current SST conditions instead of those of the previous Oc tober has a more accurate temporal scale that suggests that the prev ious months (October to April) of an annual JMA La Nia may be erroneously creating pattern s in the summer months. Additionally, this finer monthly ENSO scale used to cla ssify even simulated data shows that more accurate local patterns may be discovered using even more temporally downscaled ENSO indices, such as a continuous record of the NIO 3.4 temp eratures themselves (see Appendix A). Large flow and nutrient flushes generally followed prolonged periods of lack of precipitation or negative loads. For example, El Nio summers are generally drier than average, producing a nutrient fl ush during the rainy winters with high stream flows. On the other hand, La Nia winters are generally drier, producing a nutrient flush in the summer. The relationship between previous precipitation and the P load in the following months could be a strong one and is invest igated using non-simulated nutrient data in correlation analyses in Chapter 3. Neutral years, which have less consistent trends, showed a smaller range about the normalized average, and more load variability in both summer and winter. Several of the maximu m P loads, concentrations, and flows seen, especially during neutral events, could be explained by looki ng at historical hurricanes affecting Florida. In looking at hurricanes that affected Florida in the past 40 years, there was a relationship between months that hurricanes and peak P loads occurred. The possibility that hurricanes could explain t he outlier points and variability seen in several of the neutral year box plots is an interesting question possibility. 84

PAGE 85

Land and water managers can use these data to help make decisions about how to reduce nutrient load runoff in predict ed ENSO conditions. Because certain ENSO phases correlate more strongly with incr eased P runoff in different seasons, the contingency table data could hel p meet nutrient load limits se t by TMDLs, BMAPs, and Florida BMPs. The combination of the re lative load based seasonal probabilities using terciles, and the concentration limit contingency tables could be a powerful tool, allowing farmers and other stakeholders to identify par ticularly risky months for either high P loads or concentrations. The relative load tercile probabilities indica te that the seasons of highest P loading risk include La Nia from Ma y to July and El Nio from February to April, as previously indicated. However, from a policy viewpoint, the November to December season is the only one across all ENSO phases that is close to meeting current BMAP standards. From a land management perspective, this may mean that under realistic conditions (where changing land use is also a factor), the current restoration plans for the Lake Okeechobee watershed area, especially S-191 are not sufficient to meet stated concentration and load goals, and that the winter season may be a good time for BMP staging so that summer loads across all ENSO phases can be reduced. Focusing on changing land use and BMP effectiveness in different ENSO phases in this area are both questions that are addr essed in Chapter 3, to help researchers and stakeholders in making specific management decisions. In the Florida Everglades area specifically, restoration processes focus on allocating the correct quantity and quality of water at the correct time. K nowing that temporally coar se categorical ENSO based climate variability does allow classification an d quantification of significant differences in 85

PAGE 86

nutrient loads and local hydrological variabl es provides a basis for finding a more accurate method of predicting loads from act ual sea surface temperature time series in Chapter 5. 86

PAGE 87

Table 2-1. Example of a 3 2 contingency table used for verification of a binary forecast within an El Nio ENSO phase. In this table, n is the total number of years; r is the number of El Nio, La Nia, or neutral years; x is the number of T1 events within a specific season and ENSO phase out of b total T1 events in all seasons within that ENSO phase; and y is the number of T3 events within a specific seas on and ENSO phase out of a total T3 events in all seasons within that ENSO phase. El Nio Years Tercile Yes No Total T1 x b-x b T2 r-x-y n-r-b-a+x+y n-b-a T3 y a-y a Total r n-r n Table 2-2. Total P load (kg) anomaly summary statistics separated by month. Numbers in bold type are the largest and smallest values in each column. All statistics represent anomaly values, except for means. Event Std. Dev. Max. Min. Range Mean[a] January 147 483 -75 558 77 February 291 1479 -103 1582 106 March 245 725 -165 890 168 April 124 390 -76 466 80 May 252 1011 -139 1150 143 June 572 1591 -561 2153 570 July 397 937 -459 1396 474 August 377 878 -469 1346 484 September 489 2467 -447 2914 461 October 510 1732 -358 2090 362 November 285 1110 -133 1243 139 December 108 440 -43 483 48 [a] Non-normalized. Table 2-3. Total P load (kg) anomaly summary statistics separated by JMA and MENSO phase. All statistics represent anomaly values, except for means. JMA Index M-ENSO Index El Nio La Nia Neutral El Nio La Nia Neutral Std. Dev. 350 342 346 330 235 382 Maximum 1479 1591 2467 1479 1071 2467 Minimum -561 -557 -546 -561 -551 1622 Range 2041 2149 3013 2041 1622 3024 Mean[a] 263 245 263 287 143 297 [a] Non-normalized. 87

PAGE 88

88 Table 2-4. Seasonal contingency table sign ificance level (compared to random chance) results for P loads for each JMA ENSO phase. Terciles are low (T1), mid (T2), and high (T3). This table repr esents the results from 12 computed contingency tables. El Nio La Nia Neutral Season Sig. [a] Tercile Sig.[a] Tercile Sig.[a] Tercile Feb-Apr 99 T3 99 T1 95 T1 May-July 95 T1 99 T3 99 T1 Aug-Oct --99 T1 99 T3 Nov-Jan 99 T2 95 T2 99 T2 [a] Sig. = significance level (%). Table 2-5. Seasonal contingency table sign ificance level (compared to random chance) results for P concentrations for each ENSO phase by chance of being under or over a 0.22 mg/L limit in S-191. This table represents the results from 12 computed contingency tables. El Nio La Nia Neutral Season Sig.[a] <0.22mg/L Sig.[a] <0.22mg/L Sig.[a] <0.22mg/L Feb-Apr Over 99 Under Over May-July Over Over Over Aug-Oct Over Over Over Nov-Jan 90 Under 99 Under 99 Under a] Sig. = significance level (%). Table 2-6. Single factor ANOVA results between M-ENSO phases (p = 0.05). H0 : 1 = 2, and HA : 1 2. Variables that were found to be significantly different are highlighted in bold text. El Nio/Neutral La Nia/Neutral El Nio/La Nia Variable Accept H0? Accept H0? Accept H0? Precipitation Yes No Yes Flow Yes No No P (mg/L) Yes No No P (kg) Yes No No

PAGE 89

89 # Y # Y # Y # Y # Y # Y # Y # Y # Y # Y # Y # Y # Y # Y TCNS201 T C N S 2 0 4 T C N S 2 09 TCNS206 TCNS207 T C N S 2 1 3 TCNS214 TCNS217 TCNS219 TCNS218 TCNS222 TCNS228 TCNS230 TCNS233" 4 7 "51" 8 4 "16"22"20"24"18"14" 6 0 6 9 "44"35"6"9"36"39"40"5"3"2"11"17"10"78"37"15"38"45"70"76"28"4"7"8"12"13"19"21"23"25"26"27"29"30"31"32"33"34"41"42"43"46"48"49"50"52"53"54"55"56"57" 5 8 5 9 "61"62"63 6 4 6 5 "66"67" 6 8 7 1 "72"73"74"75" 7 7 "79"80"81"82"83"85Reach 2(a) (b) Figure 2-1. (a) Location of Lake Ok eechobee, basin S-191, and t he surrounding basins (from t he South Florida Water Management District), and (b) weather and wa ter quality stations in S-191. Prec ipitation was spatially averaged over all stations in the basin.

PAGE 90

-4 -2 0 2 4 6 1 2 3 4 5 6 7 8 9 10 11 12 Precipitation Anomaly (cm) El Nino La Nina Neutral (a) -200 -100 0 100 200 300 400 Month Load Anomaly (kg) El Nino La Nina Neutral 1 2 3 4 5 6 7 8 9 10 11 12 (b) Figure 2-2. Monthly total (a) precipitat ion (cm) and (b) simulated P load (kg month-1) anomalies, averaged over all reaches for each ENSO phase (1965-2001). Month 1= January to month 12= December. 90

PAGE 91

1 2 3 4 5 6 7 8 9 10 11 12 -15 -10 -5 0 5 10 15 20 25 Precipitation Anomaly (cm) Month(a) 1 2 3 4 5 6 7 8 9 10 11 12 -15 -10 -5 0 5 10 15 20 25 Precipitation Anomaly (cm) Month(b) 1 2 3 4 5 6 7 8 9 10 11 12 -15 -10 -5 0 5 10 15 20 25 Precipitation Anomaly (cm) Month(c) Figure 2-3. Box and whisker plots of observe d precipitation anomalies (cm) for JMA (a) neutral years, (b) La Nia years, and (c) El Nio years. Box lines are at the lower quartile, median, and upper quartile values. In all boxplots, whiskers extend from box ends to upper and lowe r adjacent values, defined as the largest or smallest observation that does not exceed the upper or lower quartile 1.5 interquartile range. Outlier points indicate data that fall outside the whiskers, and are indicated by crosses. 91

PAGE 92

1 2 3 4 5 6 7 8 9 10 11 12 -150 -100 -50 0 50 100 150 200 250 Precipitation Anomaly (mm) Month (a) 1 2 3 4 5 6 7 8 9 10 11 12 -150 -100 -50 0 50 100 150 200 250 Precipitation Anomaly (mm) Month (b) 1 2 3 4 5 6 7 8 9 10 11 12 -150 -100 -50 0 50 100 150 200 250 Precipitation Anomaly (mm) Month (c) Figure 2-4. Box and whisker plots of observed precipitation anomalies (mm) for MENSO (a) neutral months, (b) La Nia months, and (c) El Nio months. Box lines are at the lower quartile, median, and upper quartile values. 92

PAGE 93

1 2 3 4 5 6 7 8 9 10 11 12 -20 0 20 40 60 80 Flow Anomaly (m3/sec) Month (a) 1 2 3 4 5 6 7 8 9 10 11 12 -20 0 20 40 60 80 Flow Anomaly (m3/sec) Month (b) 1 2 3 4 5 6 7 8 9 10 11 12 -20 0 20 40 60 80 Flow Anomaly (m3/sec) Month (c) Figure 2-5. Box and whisker plots of si mulated monthly stream flow anomalies (m3/sec) for JMA (a) neutral years, (b) La Nia y ears, and (c) El Nio years. Box lines are at the lower quartile, m edian and upper quartile values. 93

PAGE 94

1 2 3 4 5 6 7 8 9 10 11 12 -20 0 20 40 60 80 Flow Anomaly (m3/sec) Month (a) 1 2 3 4 5 6 7 8 9 10 11 12 -20 0 20 40 60 80 Flow Anomaly (m3/sec) Month (b) 1 2 3 4 5 6 7 8 9 10 11 12 -20 0 20 40 60 80 Flow Anomaly (m3/sec) Month (c) Figure 2-6. Box and whisker plots of si mulated monthly stream flow anomalies (m3/sec) for M-ENSO (a) neutral months, (b) La Nia months, and (c) El Nio months. Box lines are at the lower quartile median and upper quartile values. 94

PAGE 95

1 2 3 4 5 6 7 8 9 10 11 12 -0.2 -0.1 0 0.1 0.2 0.3 0.4 P Concentration Anomaly (mg/L) (a) 1 2 3 4 5 6 7 8 9 10 11 12 -0.2 -0.1 0 0.1 0.2 0.3 0.4 P Concentration Anomaly (mg/L) (b) 1 2 3 4 5 6 7 8 9 10 11 12 -0.2 -0.1 0 0.1 0.2 0.3 0.4 P Concentration Anomaly (mg/L) (c) Figure 2-7. Box and whisker plots of simulated TP concentration anomalies (mg/L) for JMA (a) neutral years, (b) La Nia years, and (c) El Nio years. Box lines are at the lower quartile, median, and upper quartile values. 95

PAGE 96

1 2 3 4 5 6 7 8 9 10 11 12 -0.2 -0.1 0 0.1 0.2 0.3 0.4 P Concentration Anomaly (mg/L) (a) 1 2 3 4 5 6 7 8 9 10 11 12 -0.2 -0.1 0 0.1 0.2 0.3 0.4 P Concentration Anomaly (mg/L) (b) 1 2 3 4 5 6 7 8 9 10 11 12 -0.2 -0.1 0 0.1 0.2 0.3 0.4 P Concentration Anomaly (mg/L) (c) Figure 2-8. Box and whisker plots of simulated TP concentration anomalies (mg/L) for M-ENSO (a) neutral months, (b) La Nia months, and (c) El Nio months. Box lines are at the lower quartile, median and upper quartile values. 96

PAGE 97

1 2 3 4 5 6 7 8 9 10 11 12 -500 0 500 1000 1500 2000 2500 P Load Anomaly (kg) Month(a) 1 2 3 4 5 6 7 8 9 10 11 12 -500 0 500 1000 1500 2000 2500 P Load Anomaly (kg) Month(b) 1 2 3 4 5 6 7 8 9 10 11 12 -500 0 500 1000 1500 2000 2500 P Load Anomaly (kg) Month(c) Figure 2-9. Box and whisker plots of monthl y P load anomalies (kg) for JMA (a) neutral years, (b) La Nia years, and (c) El Nio years. 97

PAGE 98

1 2 3 4 5 6 7 8 9 10 11 12 -500 0 500 1000 1500 2000 2500 P Load Anomaly (kg) Month (a) 1 2 3 4 5 6 7 8 9 10 11 12 -500 0 500 1000 1500 2000 2500 P Load Anomaly (kg) Month (b) 1 2 3 4 5 6 7 8 9 10 11 12 -500 0 500 1000 1500 2000 2500 P Load Anomaly (kg) Month (c) Figure 2-10. Box and whisker plots of monthly P load anomalies (kg) for M-ENSO (a) neutral years, (b) La Nia year s, and (c) El Nio years. 98

PAGE 99

El Nino La Nina Neutral -400 -200 0 200 400 600 800 Seasonal P Load Anomaly (kg) (a) -400 -200 0 200 400 600 800 (b)El Nino La Nina NeutralSeasonal P Load Anomaly (kg) -400 -200 0 200 400 600 800 (c) El Nino La Nina NeutralSeasonal P Load Anomaly (kg) -400 -200 0 200 400 600 800 (d)El Nino La Nina NeutralSeasonal P Load Anomaly (kg) Figure 2-11. Seasonal P load anomaly box plots separated by JM A ENSO phase: (a) February-April, (b) May-July, (c) A ugust-October, and (d) November-January. Box lines are at the lower quartile, m edian, and upper quartile values. In all boxplots, whiskers extend from box ends to upper and lower adjacent values, defined as the largest or sm allest observation that does not exceed the upper or lower quartile 1.5 interquartile range. Outlier points indicate data that fall outside the whiskers, and are indicated by crosses. 99

PAGE 100

El Nino La Nina Neutral -20 -10 0 10 20 30 40 Flow Anomaly (m3/sec) (a) El Nino La Nina Neutral -20 -10 0 10 20 30 40 Flow Anomaly (m3/sec) (b) El Nino La Nina Neutral -20 -10 0 10 20 30 40 Flow Anomaly (m3/sec) (c) El Nino La Nina Neutral -20 -10 0 10 20 30 40 Flow Anomaly (m3/sec) (d) Figure 2-12. Seasonal stream flow anomal y box plots separated by JMA ENSO phase: (a) February-April, (b) May-July, (c ) August-October, and (d) NovemberJanuary. 100

PAGE 101

El Nino La Nina Neutral -0.1 -0.05 0 0.05 0.1 0.15 0.2 0.25 Concentration Anomaly (mg/L) (a) El Nino La Nina Neutral -0.1 -0.05 0 0.05 0.1 0.15 0.2 0.25 Concentration Anomaly (mg/L) (b) El Nino La Nina Neutral -0.1 -0.05 0 0.05 0.1 0.15 0.2 0.25 Concentration Anomaly (mg/L) (c) El Nino La Nina Neutral -0.1 -0.05 0 0.05 0.1 0.15 0.2 0.25 Concentration Anomaly (mg/L) (d) Figure 2-13. Seasonal TP concentration anomaly box plots separated by JMA ENSO phase: (a) February-April, (b) MayJuly, (c) August-October, and (d) November-January. 101

PAGE 102

CHAPTER 3 EL-NIO/SOUTHERN OSCI LLATION (ENSO) INFLUENCES ON MONTHLY NO3 LOAD AND CONCENTRATION, STREAM FLOW AND PRECIPITATION IN THE LITTLE RIVER WATERSHED, TIFTON, GEORGIA. Introduction In Chapter 2, we used model simulations of hydrological and nutrient data to investigate ENSO related clim ate variability. The categori cal ENSO classifications, however, while they were able to find significant differences in phase, were not on a scale fine enough to capture the inherent variability and dynamics of the hydrology. Additionally, the goal of cr eating a predictive model for nutrient loads based on ENSO indices would not be as accurate using ca tegorical labels. Chapter 3 explores the relationship of continuous measures of ENSO variability with hydrologic data observed over the last 35 years. Climat e variability and change are areas of high priority research both locally and globally. Recent research has shown that ex tremes of climate such as heat waves, droughts, floods and tropica l cyclones may be becoming more common [ IPCC, 2007]. Climate variability directly affects a range of hydrological variables, such as precipitation and water quantity. By extension, climate variability also may affect water quality as seen in pollutants, although these concerns are often overlooked in favor of focusing on stream fl ow amounts and timing. However, volumes of stream flow may not be the most important factor if water quality is lowered to a point where the utility in natural or manmade ecosystems is compromised. There is a need to identify climate non-stationarities and their links to wa tershed outcomes. In particular, for risk management, inter-annual modes of climate variability and their seasonal expression are of interest. 102

PAGE 103

The analyses in Chapter 3 are some of the first analyses to link water quality parameters to these non-st ationary climate modes. The recent IPCC technical paper Climate Change and Water suggests with high c onfidence that the increase in global climate variability and extremes may exacerbat e some types of water pollution, leading to declines in water quality that could a ffect food availability, human health, water infrastructure operating costs, and ecosystem health [ Bates et al., 2008]. This research is targeted at exploring the relationship bet ween climate variability at the inter-annual level via the El Nio/Southern Oscillation (ENSO) and hydrol ogy and water quality at the small basin scale in the Little River watershed, Georgia. By quantifying the relationships between inter-annual climate and wa ter quality, it may be possible to use this information to reduce pollutant loads into target watersheds during high risk months using short-term climate predictions. The ENSO is a periodic ocean and atmo spheric phenomenon with strong effects on the climate of the southeast United States. The tri-st ate region especially (FL, GA, AL) experiences strong ENSO effects. El Ni o winters tend to be cooler and wetter than those in neutral years; whereas La Ni a winters tend to be warmer and drier [Kiladis and Diaz 1989; Hanson and Maul 1991; Schmidt et al. 2001]. Different ENSO phases are identified using indices of sea surface temperatures (SST) in the equatorial Pacific Ocean and associated pressure and wind patte rn changes. Generally, an ENSO year is classified as El Nio if the 6-month spatia lly averaged mean is at least +0.5C greater than average, La Nia if it is -0.5C lowe r than average, and neutral otherwise, although the precise definition has changed ov er time and varies by study [ Trenberth 1997]. In this research, categorical definitions su ch as the Japan Meteorological Agencys El 103

PAGE 104

Nio or La Nia will not be used to define entire years [ Japan Meteorological Agency 1991]. Instead, the continuous SST anomaly ti me series averaged over the region bounded by 5N to 5S, from 170W to 120W and called the NINO 3.4 index is used to encompass finer temporal variab ility within the ENSO signal [ Trenberth 1997]. ENSO phase has previously been shown to have predictable effects on stream flow, precipitation, monsoon occurr ence, crop yield, cholera o ccurrence, flood frequency, and simulated water quality data in different regions around the world [ Chiew et al. 1998; Rajagopalan and Lall 1998; Hansen et al. 1997; Pascual et al. 2000; Piechota and Dracup 1999; Keener et al. 2007]. In monitoring and resear ch efforts during the 20th century, ENSO indices have emerged as one of the most consistent for describing inter-annual climate variability on both global and regional scales [ Ropelewski and Halpert 1986]. Often in research, hydrological processes over long periods of time are assumed to be stationary. Observed data for many di fferent variables prov ide evidence to the contrary. Spectral analysis methods have been used to investigate the frequency relationships between large scale climate i ndices such as the ENSO, North Atlantic Oscillation (NAO) and Pacific Decadal Osci llation (PDO) and hydrologic variables such as stream flow and precipitation [ Rajagopalan and Lall 1998; Enfield et al. 2001; d'Arrigo et al. 2001]. Spectral wavelet methods al low one to explore non-stationary aspects of time series within the frequency domain. The ENSO phenomenon is recognized as having an approximate per iodicity of 3-7 years [ Rasmusson and Wallace 1983], which has various affects on global climate. If relationships shown in other locations between observed precipitation, stream flow, and 104

PAGE 105

the ENSO index [ Rajagopalan and Lall 1998; Labat 2008; Piechota and Dracup 1999] are valid in the Little River watershed in Geor gia, it may be possible to use the NINO 3.4 anomalies as a predictor for high risk monthly nutrient loading, and thus as part of a best management practice (BMP) or BMP sele ction tool. The wavelet analysis allows the characterization of the low-frequency osc illations and amplitudes associated through time with geophysical data such as rainfall, SST [ Torrence and Compo 1998; Y. Wang 1996] and nutrient loading. Coherence and cro ss wavelet analysis then allows the direct comparison of two time series, SST and precip itation, stream flow, or nutrients, for the purpose of identifying areas in wh ich they co-vary with high power [Grinsted et al. 2004]. Investigating and identifying significant oscillation periods in precipitation, stream flow, and nutrient loads in the southeast United States that correspond to ENSO oscillations is a crucial first step in ulti mately reducing risks associated with climate variability in managing water resources and agricultural systems. The result is an understanding of both recurrent, episo dic phenomena, and also a method of quantitatively describing climate risk that is changing in time. Hydrological relationships involving sto rm runoff and infiltration rates that apply in much of the US do not appl y for the coastal plain [ Sheridan et al. 2001], and it is therefore often necessary to create new model s with specific hydraulic relationships for the region. Existing models often rely on em pirical algorithms to represent processes that are not well understood, making wavelet analysis appealing, in which long term, high, or low frequency patterns can be found and used to advance understanding of these processes. Inclusion of the impacts of the large riparian zones in the LRW on flow and water quality is critical to accura te physical hydrological modeling [ Sheridan et al. 105

PAGE 106

1983]. Evidence exists that seasonal precip itation in the region may be shifting, with more rainfall from September to March, and less during the growing season from April to August [ Baigorria et al. 2007; Knisel et al. 1985]. The southeast United States is an important region both economically, agriculturally, and socially, because of its rapidly growing population that will add water demand and environmental stress to currently stressed or degraded ecosystems. To address the multitude of issues associated with the changing climate in the southeast, finding new methods of making predictions for hydro-climatic variables based on periodic non-stationary phenomena such as the El Nio Southern Oscillation is an important tool. In this resear ch, wavelet analysis explores the link between nutrient loading, hydrology, and ENSO in the Little River Watershed near Tifton, Georgia. This study is a step in ultimately modeling how flushes of nutrient loading in the southeast may be reduced via management based on current or short-term predictions of ENSO phase. Data and Methods Field Site: Little River Watershed The Little River Watershed (LRW) in Tifton, southeast G eorgia, is an example of a stereotypical rural coastal plain watershed. Occupying a large area in the southeast US, the LRW is part of the Suwannee River Ba sin of Florida and Georgia, characterized by broad, flat alluvial flood plains with low-gradient, poorly defined channels, sandy soils, and slow moving streams (Figure 3-1a). The LRW covers 334 km2, of which approximately 40% of the s outhern half and 30% of the nor thern half are agricultural cropland [ Bosch et al. 2004]. It is comprised of eight s ub-basins of size varying from 9 to 115 km2 (Figure 3-1b), and has extensive ripar ian buffers and gentle stream slopes ranging from 0.1 to 0.5%. The Little River drai ns into the Withlacoochee River, which in 106

PAGE 107

turn drains into the Suwannee River and ultimately empties into the Gulf of Mexico. All analyses were done on sub-basin K, a 16.8 km2 area with 106-146 m elevation in the watershed uplands with the most comprehensive nutrient data available of the basins. Maintaining environmental quality in rura l landscapes often depends on buffering natural ecosystems from non-poi nt pollution derived from agriculture. In the Suwannee River Basin and the LRW, thes e non-point pollutants include irrigated crops that receive additional nutrient fertilizer and pesticides, anima l production that releases high levels of nutrient and pathogen rich waste, and animal production that mechanically impacts the condition of riparian, stream, or wetland ecosystems through trampling or grazing. History and monitoring In 1967, the United States Agricultural Research Service (ARS) installed hydrologic and climate monitoring systems in t he LRW, so as to create a data-rich site to better research and understand the hydr ology and agricultural impacts on the southern Coastal Plain. The instrumentation was put in throughout the early 1960s, and has been continuously monito red and upgraded since that time, leading the LRW to become the primary experimen tal agricultural watershed in the Coastal Plain [ Bosch et al. 1999]. The extensive rain gauge network was designed to provide a correlation coefficient of 0.9 between measurement s of the nearest gauges, indicating the predictability of one measurement bas ed on one from another location [ Bosch et al. 1999]. Since they were installed, some gauges that were deemed redundant were removed to save costs, and the digita l weighing-type gauges have been replaced by tipping-bucket type gauges, which tip after eac h 0.254 mm of precipitation, and record each accumulated measurem ent after 5 minutes [ Bosch et al. 1999]. 107

PAGE 108

Eight stream flow measurement device s were constructed throughout the entire LRW watershed between 1967 and 1972. Because of the slow-moving and low gradient nature of streams within the LRW, engineers used Virginia V-notch weirs to provide the needed accuracy of measurement without upstream ponding [ Bosch and Sheridan, 2007]. Each weir was constructed with steel-s heet piling cutoff walls across the stream channel, wing walls to direct flow across the measurement device, a concrete apron to dissipate the energy downstream from the notch, and stilling wells on either side of the weir [ Bosch and Sheridan, 2007]. The stilling wells compensate for potential periods of no-overfall over the weir crest, measuring elevation every 5-minutes on two FischerPorter digital stage recorders synchronized across the entire LRW hydrologic network until 1993, and strain gauge pressure-transducer digital loggers after that [ Bosch and Sheridan 2007]. Sub-basin K in particular uses a horizontal weir with a V-notch center section measuring 17.8 m in lengt h with a notch depth of 44.2 cm [ Bosch and Sheridan 2007]. Data were checked for accuracy by SEWRL USDA-ARS researchers, and are maintained in public databases. The water-quality monitoring program was started in the LRW in 1973 in basin K, with an automatic sampler taking three samp les per day from the flow monitoring location to test for in-stream chloride, nitrate and nitrite [ Lowrance et al. 1984]. In 1974, an additional limited duration rainfall waterquality sampling program was started to gather baseline nutrient concentrations [Lowrance and Leonard 1988]. In 1979, total kjeldahl nitrogen and total phosphorus were added to the list of analytes monitored [ Feyereisen, Lowrance, et al. 2007]. Over the years, sample collection methodology and timing has varied according to research objectives, funding, and technological 108

PAGE 109

advances. Collection methodology has vari ed from manual grab-sample, automated timed discrete, and automated flow-weighted composite refrigerated or non-refrigerated, while the place of collection has varied from weir overfall to the upstream stilling well area [ Feyereisen, Lowrance, et al. 2007]. Sampling times range from every 5-minutes, to several hours, to weekly. Analyte loads were calculated by summing the product of concentration and stream flow for a given time step, and are also available in the website database. Soils and geology All of the Little River Watershed study area is in the outcrop era of the Hawthorn Formation, a Miocene Age geological structure that can obstruct regi onal interactions between surface and gr oundwater systems [Stringfield 1966]. In the northern LRW, including basin K, the Hawthorn Formation c onsists mainly of non-marine, cross-bedded gravelly sands mixed with indur ated sandy clays, with a total thickness of approximately 21 meters [ Stringfield 1966]. The lower 6.5 meters are mainly sandy claystone, with a second indentified rock type of sandy gray-g reen clay in the lowest 3 meters. The Hawthorn Formation was subsequently over lain by a 0-6 meter highly permeable Quaternary and Recent eolian and fluvial sediments over older marine sediments [ Asmussen 1971]. The soils of sub-basin K are mainly sandy loam and loamy sand with characteristically high infiltration capacity [ Rawls et al. 1976]. In fact, 85% of the soils within the LRW have infiltration rates of 5 cm/hr or greater [ Calhoun 1983]. Infiltration at the surface laye r from 0-6 meters is restricted by a plinthic layer of clay/sandy clay loam wit h low hydraulic conductivity generally less than 0.01 cm/hr [ Feyereisen et al. 2008]. Upland soils generally have good internal drainage, while lowland drai nage can be very poor, with isolated standing water for 109

PAGE 110

extended times in winter and spring [ Calhoun 1983]. Phosphorus tends to be fixed in the soils, making it less leachable than nitrat e. As a result, while neither pollutant is specifically problematic in the LRW, phosphorus levels are extremely low. Climate and hydrology Climate in the LRW region is humid subtropical, characterized by long and hot summers, short and mild winters, and an average growing season of 245 days [ Sheridan 1997]. Annually, the area's average pr ecipitation (1922-1988) is 1208 +/214 mm [ Sheridan and Knisel 1989], with variable monthly di stribution. As in much of the southeast US, summer rainfall is characte rized by convective storms with shorter individual events that occur with greater frequency and intens ity than in other seasons. In this regard, summers in the LRW usua lly have greater total rainfall than other seasons [Sheridan 1997]. Typically, the fall months hav e the least total rainfall, but are more likely to have extended storms with l onger average duration bet ween precipitation events [ Feyereisen et al. 2008]. Late summer and fall months may also receive heavy rainfall from tropical storms. Winter and spring seasons receive more total rainfall than the fall season, and continue to be characterized by variable low-intensity storm events. Average temperatures range fr om a monthly minimum of 4.2C in January to a maximum of 32.7C in August, with an annual average of 19.1C [Sheridan 1997]. Shallow, groundwater flow is the primary runoff compon ent in uplands, accounting for up to 80% of stream flow [ Sheridan 1997]. Groundwater flow is also the primary flow route for soluble nutrients [Lowrance et al. 1984]. Groundwater recharge to aquifers below the surficial aquifer is restrict ed by the Hawthorn formation, which makes up the confining layer for the artesian Florida Aquifer under the watershed [ Stringfield 1966]. In areas of the LRW, the Hawthorn formation is cont inuous, creating relatively 110

PAGE 111

impermeable subsurface horizons t hat impede deep seepage and groundwater recharge, instead promoting lateral movement. The shallow lateral subsurface flow and groundwater flow from the surficial aquifer contribute seasonally to stream flow as shallow return flow to surface drainage [ Sheridan 1997]. There have been water balance studies at both the field and watershed scale in the LRW. Basin K specifically was m easured from 1967-1977 as having mean surface runoff of 42.8 cm with 75% of that taki ng place between January and May, and a consistent but virtually negligible sub-su rface runoff of 0.0008 cm, while 33% of the recorded precipitation flowed from the basin as surface flow and relative subsurface flow only 1/100th of a percent of rainfall input [ Lowrance et al. 1983]. Land use, management and policies Sub-basin K in 1968 was defined as mi xed-use agricultural, and vegetative mapping showed land use as 39% agricultura l (mostly peanuts, corn, cotton, soybean, tobacco), and 61% woodland (mostly sl ash and longleaf pine with wiregrass or broomsedge, some hardwood pine, swamp hardwood, and pine/oak) [ Sheridan et al. 1983]. In 2003, the land use had changed to approximately 29% agricultural and 57% woodland [ Feyereisen, Strickland, et al. 2007].The Southeast Watershed Research Lab (SEWRL) in Tifton has collected hydraulic and climatic data continuously for the entire watershed since 1965, and water quality data in sub-basin K beginning in the mid-1970s [ Feyereisen et al. 2008]. There are also historical l and-use and fertilizer records starting in the late 1970s. Historically, the environment and water quality of the LRW has not been overly adversely affected by agricultural practices, due to the effect of large riparian filters throughout the heavily forested watershed. Howeve r, the cultivated agricultural area in 111

PAGE 112

sub-basin K has been increasi ng since the early 1990s with a related decline in upland and riparian areas, meaning that the ecology and water quality are being increasingly impacted. Because of the large amount of data available for the LRW, analysis of this impact is possible. Within the 30% of cult ivated area in sub-basin K in 2004, cotton is the major crop grown, comprising 68% of a ll cultivated area with peanut next with only 29%. As one of the more upland sub-basins, field studies in 1984 showed that K was accumulating the most nutrients in the LRW, which suggested that more intensive cropping of the upland fields could lead to more efficient use of nutrients by having actively growing crops for mo re of the growing season [ Lowrance et al. 1984]. Currently, almost year round pr oduction of row crops has led to extensive and sustained use of fertilizer and pesticides on the waters hed (Bosch et al., 2004) On the other hand, less intensive agriculture coupled with incr eased forest and ripar ian area could have similar effects on efficient use of nutrients. In past studies done by SEWRL, the LRW has shown the highest overall stream flow concentrations of nitrate (NO3) and total phosphorus (P) from January to March, and minimums from July to September [ Lowrance et al. 1984]. Low rainfall and low evapotranspiration, decreas ed denitrification and decre ased uptake in the upland and riparian zones may contribute to higher nutrient concentrations in the winter, while crop nutrient uptake and low leaching may caus e low concentrations in the summer [ Lowrance et al. 1984]. Decreased NO3 and P concentrations were observed during April to June even though runoff was still high and fertilizer application occurred, in part because of biological effects such as nutri ent uptake, transpirat ion, and utilization. Nitrate loads are highest when stream flow is highest, from December through 112

PAGE 113

February, and lowest when flow is least, from September to November. Nutrient loads are disproportionately low from June to August wh ile flow is still relatively high, as crops are utilizing the most nutrients in this season [ Lowrance et al. 1984, 1985]. Over the past 25 years, there have been numerous studies by SEWRL documenting the hydrology, ecology and stream chemistry of the LRW [ Bosch et al. 2004; Feyereisen et al. 2008], yet nothing investigating the relationship between the hydrology and local ENSO effects. Little River Watershed Data Observed precipitation and stream flow dat a from sub-basin K of the LRW used in this study ranges from February of 1968 to June of 2005, and have been recorded and compiled by researchers at SEWRL in Ti fton, Georgia (Table 3-1). Details on the daily flow and precipitation data collection, quality, and processing are thoroughly discussed in Bosch et al., 1999, 2007a and 2007b. Flow in the LRW is measured at eight concrete v-notch weirs built betw een 1967 and 1971, one of which lies at the outlet of sub-basin K. There are currently 46 active rain gauges in the LRW, 13 of which are within and surrounding sub-basin K, and were used to calculate the weighted watershed area daily precipitation [ Bosch, Sheridan, and Marshall 2007]. Daily precipitation data wit hin sub-basin K were created from watershed weighted data using the methods of Dean and Snyder [ Bosch, Sheridan, and Marshall 2007; Dean and Snyder 1977], summed into monthly cumulati ve values, and then normalized by monthly average to remove annual cycles for each month of the record. Processed daily stream flow data [ Bosch and Sheridan, 2007] for all 37 years were summed into monthly cumulative values and norma lized by monthly average as well. 113

PAGE 114

Nitrate and total phosphorous concentration datasets contain varied collection frequencies from 1974 to the present, ranging from every five minutes to once a week. Researchers at SEWRL have in-filled the nutrient data to the daily level using a variety of methods based on the sampling method t hat was used (i.e. automatic sample, manual grab). The in-filling provided complete daily time series of 24 years of total P concentration (from January, 1979 to December, 2003), and 29 years of NO3 concentration (from January, 1974 to December, 2003). The detai led methods and reasoning behind stream chem istry in-filling are discussed in Feyereisen et al [ Feyereisen, Lowrance, et al. 2007] and additionally located on the LRW public database website, locat ed at ftp://www.tiftonars. org/, under the file streamchemistry_readme.txt. Daily concentration data were converted to loads by multiplying daily flow by nutrient concentra tion. Nutrient concentrations and loads were averaged into basin-wide monthly values and normalized by median monthly values, creating time series of average monthl y anomalies with the annual cycle removed. Median values were used to normalize nutrient concentrations and loads to minimize difficulties associated with the different methods of nutrient data in-filling. Sea Surface Temperature anomalies were calculated from the Kaplan extended NINO 3.4 index (27.5E to 22.5E, 87.5S to 87.5N), wh ich is available from the Lamont-Doherty Earth Observator y data library (Table 1) [ Kaplan et al. 1998]. The SST data are gridded by 5 5 delineations, and range from 1856 to the present, although only the 1968 to 2005 data are used in this st udy. Temperatures were converted to monthly average and seasonal av erage anomalies by being norma lized against the total 114

PAGE 115

months average. All variables raw time series were plotted against time (month and year) to provide an initial visualizat ion of the data trends (Figure 3-11). All hydrological and climatic data norma lized by monthly medians are shown in boxplots (Figures 3-6 to 3-10) separated by M-ENSO phase (see Chapter 2 introduction for M-ENSO details) and month. Wavelet Analysis Wavelet analysis is a spectral method of decomposing a time series into time and frequency space, allowing t he identification and analysis of dominant localized variations of power, i.e., wher e the variance of the time se ries is largest for a given frequency. In this context, a main purpose of using the wavelet analysis technique is to quantify and visualize statistically significant changes in the ENSO SST and nitrate load variance over a multi-decadal time scale. The windowed Fourier transform is typically used to analyze a signal in frequency space at a global level, however, it is scaledependant, and is hence inefficient and inaccura te when attempting to perform an analysis on non-stationary envir onmental data sets with diffe rent signal powers that evolve through time [Torrence and Compo 1998]. A well known and relevant example is that of SSTs in the equatorial Pacific Ocean, where the dominant mode of variability is ENSO, shown by frequency signals on a time scale of 3-7 years [ Rasmusson and Wallace 1983]. Superimposed on this signal are mu ch longer inter-decadal fluctuations, visible in the wavelet power spectrum over time as the evolut ion of the SST signal power. The inter-decadal fluctuations have t he effect of modulating the amplitude and frequency of occurrence of El Nio events. Torrence and Compo (1998) have written a comprehensive guide to wavelet use for geoph ysical data, the basic theory they discuss will be summarized here. 115

PAGE 116

With any time series, xn ( n = 0N-1) with time spacing t there is a corresponding wavelet function, o ( ), with zero mean and localized in time and frequency space, that depends on a non-dimensional time parameter, and the nondimensional frequency, 0. In this analysis, the Morl et wavelet (Equation 3-1), consisting of a plane wave modulated by a G aussian, is used for all time series, the frequency of which, in practice, is given the default value of 0 = 6 for all analyses [ Farge 1992]: 2/ 4/1 0exp exp )(0 i (3-1) A Morlet wavelet is a non-orthogonal, comple x function that can be used with the continuous wavelet transform, Wn( s). The continuous wavelet transform (Equation 3-2) of a discrete sequence xn, is the convolution of xn with a scaled and translated version of o ( ): ]/)'[( )(1 0' 'stnnxsWN n n n (3-2) Where (*) is the complex conjugate, s is the wavelet scale, n is the localized time index, n is the translated time index, and is the normalized wavelet. An approximate value of Wn can be found by performing the convolut ion N times for each scale, where N is the number of points in the time series. All convolutions can then be done in Fourier space using a discrete Fourier transform of xn, (Equation 3-3), which in our case, is: 1 0 /2exp)/1( N n Nikn n kxNx (3-3) 116

PAGE 117

Where k= (0N-1) is the frequency index. By the convolution theorem, the wavelet transform (Equation 3-4) is the inverse Fourie r transform of the pr oduct of the discrete Fourier transform of xn and with angular frequency k: tni k N k k nksxsWexp)( )(1 0 (3-4) Finally, the wavelet power spectrum is defined as |Wn ( s)|2, and the amplitude at each point, |Wn( s)| and phase, tan-1 [I{ Wn ( s )} / R{ Wn ( s)}] can be found. Significance levels for wavelet spectra are found by comparison against a random background distribution, which, for geophysical spectra, are modeled as either white or red noise (increasing power with decreasing frequency) In this study, red noise is used [ B. Wang 1995; Torrence and Compo 1998], and modeled as a univariate lag-1 autoregressive (AR-1) process. The wavelet variance is indicated by the Global Wavelet Spectrum (GWS), calculated by the integration of the squared transform coeffi cients at different scales for all data points. Areas of global wavelet significance are shown as a dashed blue line, above which indicates a 95% confidence li mit above red noise. Continuous Morlet wavelet transforms and wavelet power spectra were computed and visualized for all monthly normalized time series in Table 1 to find significant periodicities that correspond to ENSO signals. All univariate wavelet analysis was done using the WAVETEST Matlab script for WAVELET, [ Torrence and Compo 1998], which can be downloaded from: http://paos.colorado.edu/ research/wavelets/. 117

PAGE 118

Cross Wavelet and Coherence Transforms A detailed explanation of cross wa velet transform (XWT) can be found in Torrence and Compo [ Torrence and Compo 1998]. Given two different time series, X and Y, with different wavelet transforms WX n( s) and WY n( s), the cross wavelet transform is WXY n( s) = WX n(s) WY* n( s), where (*) denotes the complex conjugate. The spectrum is complex, thus the cross-wave let power can be defined as |WXY n( s)|. The cross wavelet transform finds regions in time frequency sp ace where the two time series show high common power, and thus, significance. In particular, this transform examines whether regions in time frequency space with lar ge common power have a consistent phase relationship, and therefore are suggestive of causality between the time series [ Grinsted et al., 2004]. Computing the wavelet coherence trans form (WTC) finds regions in time frequency space where the two time series co -vary, but do not necessarily have high power. For this reason, both the cross wavelet transform and the wavelet coherence transform are necessary when analyzing two ti me series to assess both causality and local covariance. Generally, a WTC will hav e more significant areas than an XWT spectrum, as the significance is sacrificed for the desirable visualization of shared power in the XWT. The wavelet coher ence transform of tw o time series [ Grinsted et al. 2004] is defined as: )))((()))((( ))(( )(2 1 2 1 2 1 2sWSsSsWSsS sWsS sRY n X n XY n n (3-5) Where S is a smoothing operat or defined by the wavelet type used and the entire expression is similar to that of a tradition al correlation coefficient localized in time118

PAGE 119

frequency space. The statistical significance le vel of the wavelet coherence is estimated using Monte Carlo methods, and then the significance level for each scale is calculated using only values outside the cone of influence. Cross-wavelet and coherence transforms were done on normalized NINO 3.4 SST time series with precipit ation, stream flow, NO3 concentration and NO3 load, all calculated to 95% significance levels. Wa velet coherence and cross wavelet software used was via the XWT and WTC scripts writ ten by Aslak Grinsted for Matlab, 20022004, downloadable at http://www.pol.ac.uk /home/ research/waveletcoherence/. Cross-Correlation Time Series Analysis The lagged cross correlation between the ENSO index and the hydrological time series, x(i) and y(i) at different delays (d = 0N-1) were computed as: i x i y i y xix diy diyix R ))(())(( ))(()(2 2 (3-6) In which N is the length of the time series and x and y are the respective means. A positive correlation is associated with the series correlated in their current configuration, while a negative correlation indicates a re lationship with the inverse of one of the time series. The delay at which maximum correlation is seen describes the most significant time lag relationship between the two data series. Cross correlations were calculated between the NINO 3.4 SST time series and pr ecipitation, stream flow, NO3 concentration and NO3 load for delays from zero to 24 months. Significance was calculated at 95% confidence corrected for time series autocorrelation [ World Meterorological Orginization 1966]. 119

PAGE 120

Results Wavelet Analysis The monthly wavelet power spectra for the NINO 3.4 SST, precipitation, stream flow, and NO3 concentration and load time series are shown in Figures 3-2a-e. As previously demonstrated [ Y. Wang 1996; Torrence and Compo 1998], when the annual cycle is removed, SST power (Figure 3-2a) is concentrated within the ENSO periodicity band of 3-7 years, although the am plitude and dominant modes tend to shift through time. From 1911-60, a 5-7 year per iod is strongest, while a 4-5 year period dominates from 1972-92 [ Y. Wang 1996]. Longer decadal variations cannot be assessed with significance giv en the limited length of reco rd. The annual cycle of ENSO is associated with observing strong La Ni a power in the wavelet spectrum [ Y. Wang 1996], as well as normal seasonal variations or possible manifestat ions of the QuasiBiennial Oscillation (QBO) [ Baldwin et al. 2001; Reed et al. 1961]. The enhancement of annual power by La Nia, and respective weakening during El Nio, has been hypothesized to be due to the thermocline becoming shallower and the trade winds strengthening, which reinforces the ocean up welling effect that modulates the annual SST cycle [Gu and Philander 1995]. Therefore, by remo ving the annual cycle, we are able to more easily visualize powerful El Nio events in the 3-7 year band. The most pronounced SST variability within the wavele t power spectrum is observed during the strongest El Nio phenomena enc ompassed within this time series, namely the 1982-83 and 1997-98 eventsthe latter of which is th e strongest El Nio recorded in the 20th century [ Chavez et al. 1999]. In the time domain, NINO 3.4 SST anomaly trends are easy to discern, as they form the basis for cl assification of ENSO phases (Figure 3-6). All ENSO phases have relatively low variability, while neutral months are close to zero 120

PAGE 121

(Figure 3-6a); La Nia years are slightly below zero (Figure 3-6b) and El Nio years are slightly above zero (Figure 3-6c). Precipitation in the southeast US and the LRW has distinct seasonality. In the coastal plain region and more specifically the LRW, the most in tense precipitation events are in the spring and summer months, associated with convective or cyclonic storms [ Sheridan 1997]. Summer events are shorte r, smaller in area, and more frequent and intense, while fall and winter events are frontal in nature, milder, but longer in duration. The time domain boxplots and wavelet power s pectrum of monthly precipitation anomalies from 1968-2004 is shown in Figures 3-7 and 3-2b, respectively. Normalized precipitation time series show la rge variability for all months in the neutral phase (Figure 3-7a), more summer month posit ive anomalies and September to December negative anomalies dominate in the La Nia phase (Figure 3-7b), and a larger Sept-Dec and summer variability in El Nio phase (Figure 3-7c). Low frequency precipitation information corresponding to the 3-7 year ENSO signal has been demonstrated in the Florida Everglades via wa velet analysis, as well as in the western US [Kwon et al. 2006; Rajagopalan and Lall 1998], and the Little River Watershed shares that signal. Regions of high power relative to the noise background are seen in the precipitation record in the same 3-7 ye ar periodicity as for the NINO 3.4 series (Figure 3-2b), although the signal is not power ful enough for statistica l significance, nor is it as clear as the SST record in Fi gure 3a. Since a 4-5 year quasi-periodicity dominates ENSO from 1972-92 as seen in longer SST wavelet analyses [ B. Wang 1995; Y. Wang 1996], this may explain the 4-5 year period visible in the precipitation time series. Again, the strongest power at that frequency is seen from 1981 to 1992 and 121

PAGE 122

surrounding the 1997-98 El Nio event, which im plies that ENSO variability clearly is associated with precipitation in basin K of the LRW (Figure 3-2b). The high power still visible at the 1-2 year period, despite re moval of the annual cycle, may represent the variability from a seasonal anomaly. Anal yzing summer and winter seasons separately with each hydrological variable showed slightly stronger 3-7 year ENSO wavelet power in winter months, although not enough to be statistically significant. In basin K of the LRW, average annual stream flow depth is approximately onethird of annual precipitation [ Sheridan, 1997; Feyereisen et al. 2008], a figure comparable to similar statisti cs from other coastal plain wa tersheds. As such, the interannual component of the wavelet power spectr um (Figure 3-2c) for the stream flow record for 1974-2003 shares much of the power and variability pattern seen in the precipitation spectrum, although the signal is actually stronger as it is less noisy than the precipitation record. The M-ENSO monthly plots of stream flow show similar trends in the neutral and El Nio phase (Figures 3-8a and 3-8c) of particularly low flows from June to October, with most flow from De cember through April. The La Nia M-ENSO anomalies show relatively higher flows from May to September, an d greater January to April variability (Figure 3-8b). In the wavelet spectrum, th e ENSO periodicity is seen again as high power in stream flow for the 3-7 year period, surrounding the 1982-83 and 1997-98 El Nio events (Figure 3-2c). High power is again seen in the 1-2 year period, perhaps related to intense summer storms or the QBO. Clearly, t he observed 3-7 year periodicity in the stream flow spectrum demonstrates that while ENSO signals have been found in large rivers around the world [ Chiew et al. 1998; Labat 2008], the 122

PAGE 123

relationship is valid for small, slow-moving c oastal plain streams and rivers in the LRW as well. Wavelet power spectra of nitrate concent ration (mg/L) and load (kg) are shown in Figures 3-2d and 3-2e, respectively. While it has been shown that categorical ENSO signals exist in model simulated nu trient loads in south Florida [ Keener et al. 2007], observed and continuous dat a have not been analyzed in the literature. Both nutrient series extend from 1974 to 2003. M-ENSO boxplots of NO3 concentration show a more uniform spread and lack of trend through all months during t he neutral phase (Figure 39a), increased potential for positive anomalies from May to July and low variability negative anomalies from August to Decem ber in the La Nia phase (Figure 3-9b). During El Nio, anomalies are present from June to August larger than those seen in La Nia, with upward anomalies trending from January to May (Figure 3-9c). NO3 concentration exhibits less marked high power in the ENSO periodicity than either precipitation or stream flow, which is expec ted due to a less direct relationship with the atmosphere. There is also high power c entered on the 1988-1989 La Nia, at 1-2 year periodicity, and may signify an anomalous event. These regions of high variability imply that NO3 concentration is more sensitive to the high intensity rains characteristic of La Nia summers or hurricane in the southeast, although no periods are significant at the 90% level as manifested in the global wave let spectra. The minimal concentration power visible indicates that the concentrati on signal does not change in a significant or systematic way, and that flow and load should remain the main focus. Nitrate load (Figures 3-10 and 3-2e) is a combination of stream flow and concentration time series. Accordingly, t here is a broader band of ENSO-related high 123

PAGE 124

power in the 3-7 year per iod centered on the 1982-83 El Nio, and only medium power visible in the cone of influence around the 199798 El Nio. A longer time series may reveal more about the load structure during the latter El Nio. High power is seen in nitrate load in the 1-2 y ear periodicity centered around 1986-89, which may reflect the stream flows modulation of the concentration, and the La Nia event of 1998-99, the pattern of which is also reflected in high po wer seen in precipitation, stream flow, and SST spectra. This may again be related to the high annual power noted in some strong La Nia events [ Y. Wang 1996]. Load boxplots show the main differences in M-ENSO phase are increased January to March variabili ty in both El Nio and La Nia phases as compared to neutral (Figure 3-10a-c), wit h very high load anomalies possible in February and March of El Nio (Figure 3-10c ). Summer loads in La Nia are not higher than usual according to these metrics (Figure 3-10b). Cross-Wavelet Analysis Although one can see regions in the wave let power spectra in the previous section where two time series show high co mmon power, a more direct analysis of two series is needed. Significance levels of the cross-spectra power are calculated against a red noise background, indicated by thick bl ack outlines in the cross wavelet transform spectra (Figure 3-3) to the 5% level. The cross wavelet tran sform between SST and precipitation (Figure 3a) show s that areas that were pi cked out as possibly sharing power in the single wavelet spectra, do indeed share significant power in the 3-7 year periodicity surrounding the 1982-83 and 1997-98 El Nio. The significant areas within the 3-7 year period are phaselocked positively, which when considered with the fact that ENSO is a main driver of global climate variation, suggests a causal relationship between SST and precipitation in the LRW t hat was already suspected. The same 124

PAGE 125

patterns of shared power, phase, and 3-7 year periodicity are seen in the cross wavelet transform between SST and stream flow (Figure 3-3b). The cross wavelet transform relationship between NO3 concentration and SST (Figure 3-3c) shows less significant shared power, although there is a small area from 1985-90 around 3-4 year periodicity. The gener al 3-7 year period area shows high power, but too little for significance at the 5% level. The lack of definitive relationship between SST and concentration is explained by nutrient concentrations being very much an extraneous variable, d ependent on both local agricultural activities and rainfall or river flow. Nutrient load, on the other hand, takes stream flow into account. The cross wavelet transform of NO3 load and SST (Figure 3-3d) show s the same significant area as the cross wavelet transform with stream flow and precipitation, a shared 3-7 year inphase periodicity extending fr om 1980-90, and a 1-2 year per iodicity in the area around the 1998-99 La Nia. Wavelet Coherence Analysis Compared with the cross wavelet transfo rm, a larger area in the wavelet coherence spectra is marked as significant. However, neither the significance of the wavelet coherence transform nor the cro ss wavelet transform on their own do not necessarily imply causality, as any two va riables can be significantly correlated by chance. Small areas of wavelet coherence transform significance are unlikely to be causal taken on their own. More extensive ar eas of significance, however, are less likely to be due to chance, and should be examined ca refully for additional relationships to a physical mechanism. The wavelet coherence transform between SST and precipitation (Figure 3-4a) and SST and stream flow (Fi gure 3-4b) show very similar sustained significant areas from 19792000 in the 3-7 year period, indicating high correlation 125

PAGE 126

between SST and each time series in the area relating to ENSO signal. Longer time series would be desirable for every vari able to further investigate the changing periodicities; however within the cone of in fluence, there is a su stained ENSO signal. The wavelet coherence transform between NO3 concentration and SST is shown in Figure 3-4c. There are arguably no areas of causal significance, especially when considered with respect to the importance of variables outside of this analysis such as fertilizer application or land-use changes. Practica lly, this means that the two time series do not show any significant correlation at any periodicities of interest to this study, although there is a small area of ENSO related significance in the XWT. The concentration significance in the XWT may be a remaining artifact from the much stronger stream flow signal (Figures 4b and 5b), showing up despite the stochastic nature of the concentration series. When analyzed in conjunction with the WTC, it becomes clear that this is probably an artifact and should not be assigned any causal explanation. Nitrate load, however, (Figure 3-4d) shows a very high correlation with SST throughout all the years within the cone of influence at 3-7 year periodicity. This correlation reflects a combination of the st rong stream flow ENSO signal with the weak concentration ENSO signal, but should not be discounted as non-causal, as stream flow response is directly related to precip itation and SST. The in-phase correlation and power of SST and NO3 load suggests that a model bas ed upon the 3-7 year periodicity of ENSO could have predictive skill for LRW nitrate loads. Across all the wavelet power analyses, the ENSO signal appears in sustained and common 3-7 year frequency bands. This frequency structure is replicated ac ross the NINO 3.4, precipitation, stream flow, and NO3 load time series, and is moreover found within the cross-wavelet and 126

PAGE 127

wavelet coherence spectra, supporting this ENSO structure and dem onstrating that the ENSO climate teleconnection extends to pr ecipitation, hydrology, and observed NO3 loads in the small streams of the LRW. Cross Correlation Analysis As common frequency bands and high or significant power were found in wavelet analysis of SST and precipitation, stream flow, and NO3 load, time series cross correlation analysis was performed to tempor ally quantify how these relationships actually manifest during the year. After t he removal of annual trends from each variable considered, significant monthly corre lations were found between SST and all hydrological variables (Figure 3-5a-c). T he correlation of highest magnitude is between SST and NO3 load (Figure 3-5c), which is re inforced by the wavelet coherence transform analysis. All cross-correlations were dominated by positive significant correlations with the ENSO 3. 4 index. Precipitation an d stream flow had maximum cross-correlation function values of 0.139 and 0.342 with SST at a four month and two month lag, respectively, and NO3 load a maxi mum of 0.360 at a three month lag. This means that high or low precipitation and str eam flow are correlated with high or low sea surface temperature obs erved four and two mont hs earlier, while NO3 load is positively correlated with SST three months priorobser vations that are supported by the cross wavelet transform and wavelet coherence transform spectral analyses. Additional cross-correlation analyses we re done to determine t he strength of the lag relationship between precipitation and stream flow, pr ecipitation and NO3 load, and stream flow and NO3 load. In each relationship, the most significant lag was at zero months, suggesting that a smaller time-st ep may be needed to assess the correlative lag more accurately. However, the str ongest CCF value of 0.663 (Figure 3-5d) was 127

PAGE 128

seen between precipitation and flow at lag zero, followed by stream flow and NO3 load with a CCF of 0.472, and finally precipitat ion and load with 0.268. These values show that the NO3 load is most directly correlated wit h flow, which supports the previous wavelet analyses. Summary and Discussion Chapter 3 uses wavelet analysis to i dentify and quantify the significance of a teleconnection between sea surface temperat ures associated with the NINO 3.4 index and precipitation, stream flow, and nitrate concentration and load in the Little River Watershed in Georgia. We found common areas of high power and time series interannual variability manifested in the ENSO signal for 36 years of LRW monthly precipitation data, and 29 years of stream flow, nitrate conc entration and nitrate load data. Areas of the highest pow er for all hydrological variabl es were observed in the 3-7 year periodicity known to be related to ENSO modes of variability. Temporally, the area of greatest variability was c entered on the 1997-98 El Nio event, which is on record as the strongest anomaly in the 20t h century. Sea surface te mperatures dramatically decreased from the transition of the 1997-98 El Nio to t he 1998-99 La Nia, which may be reflected in the anomalous variability visible in the wavelet power spectra of the hydrologic variables. Nitrate concentrati on was the variable with the weakest ENSO signal power, which is due to it being more dependent on extraneous variables such as human-caused agricultural activities. High or significant power was seen in precipitation, stream flow, and nutrient loads in the 1-2 year period cent ered on the 1998-99 La Nia, and may be related to a signal from the QB O and strong seasonal signals. The stronger power seen in nitrate load ti me series, rather than conc entration or precipitation, suggest that stream flow variability dominates the trends seen in loads. 128

PAGE 129

Regions of shared power were f ound between the NINO 3.4 index and the hydrological variables of interest. There was common high power and phase in the 3-7 year mode of variability for strong El Nio ev ents, as well as in the 1-2 year mode of variability for strong La Nia episodes for SST and precipitation, stream flow, and nitrate load. High time series covariance exists in the same 3-7 year period for SST and precipitation, stream flow, and nitrate loads. The cross wavelet transform and wavelet coherence transform confirm that the known physical mechanism of ENSO teleconnection in the southeast United States [ Schmidt et al. 2001; Ropelewski and Halpert 1986] is causally linked to inter-annual variability within prec ipitation, stream flow, and nitrate load signals in the LRW. The high shared power and significant correlation between these variables confirms that the ENSO tele connection seen in the precipitation and stream flow signals in large river and watershed systems around the world [ Chiew et al. 1998; Rajagopalan and Lall 1998; Handler 1990; Kulkarni, 2000; Hansen et al. 1997; Piechota and Dracup 1999; Pascual et al. 2000] extends to the hydrology and, more interestingly, nitrate loads in a small basin of the Little River Watershed. Mechanistically, El Nio (La Nia) event s result in increased (decreased) sea surface evaporation. The mid-latitude jet is displaced equatorially (poleward), increasing (decreasing) winter frontal precipitation in the southeast United St ates. Additional winter moisture is advected into the southeast from the tropical Pacific by the subtropical jet stream [ Ropelewski and Halpert 1987]. Precipitation in El Nio winters typically increase as a result, and as seen in the large El Nio events centered around 1985 and 1997-98 in the LRW, resulted in increased river discharge and pollution transport. It is 129

PAGE 130

unusual that the ENSO signal is more visible in the stream flow and nitrate loads of the LRW than in the precipitation signal. Although the thirteen weat her stations used to form the precipitation series in this research are well distributed across the 16.8 km2 area of basin K in the LRW, spatial variability of ra infall in the southeast United States is great, and still does not encompass all of the variab ility inherent in the watershed. For this reason, the ENSO signal present in the pr ecipitation record may be somewhat damped, especially when consider ed next to the stream flow record. In addition to being a smoothed function of precipitat ion, stream flow in the Little River Watershed may be strengthening the correlation wi th the ENSO signal. As has been previously discussed, groundwater flow in the LRW is responsible for up to 80% of the total stream flow [ Sheridan 1997], as well as comprising the main route for movement of soluble nutrients [ Lowrance et al. 1984]. The ENSO signal in st ream flow may increase in power due to the significance of the role of groundwater, and additionally due to the presence of the Hawthorn confining layer, restricting flow into the aquifer system and increasing the amount of lateral groundwater flow represented in the stream flow record. The nature of the strong ENSO po wer visible in the wavelet analyses demonstrates that in the LRW during anomalous ENSO phases, BMPs and crop management practices could be more effect ively applied to reduce the risk of high nutrient runoff resulting from the increas ed precipitation and stream flow. Best Management Practices in the LRW have mainly been comprised of crop rotations designed to more effectively use available water and prevent nutrient leaching. At this point, conservation tillage or crop cover practices have not been implemented in subbasin K. Some BMP modeling work with SW AT has been done in sub-basin K as a part 130

PAGE 131

of the national Conservation Effect Assessm ent Project (CEAP), in which a simulated increase of 30% cultivated area with bette r crop management show ed a 17% reduction in total nitrogen load [ Cho et al. 2010], suggesting that BMPs could in fact be quite effective in generally reducing loads. While sometimes it is both economically as well as environmentally in a farmers best interest to implement BMPs, it can be difficult to quantify t he effectiveness of the results and for farmers to know when best to apply them The strong ENSO signal in the nitrate load and stream flow, coupled wit h the cross-correlation analysis of SSTs suggests that when an anomalous ENSO phase is predicted or is occurring, BMPs could be put in place to control the risk of high nutrient runoff. The ability to make management decisions months in advance with reduced uncertainty of the BMPs not being needed could save the farmer both time and money, and result in environmental and water quality improvement s. Implementing and monito ring the effectiveness of different BMPs in the LRW in differ ent seasons and ENSO phases would take additional time and effort, but would be wort h verifying if BMP application by ENSO phases would actually save time and money for watershed managers, farmers, or other stakeholders. In this study, more than 29-36 years of data would have been informative as to how inter-decadal variability affects hydro-c limatic events. However, short-term interannual climate variability is a crucial factor affecting climate in the southeast US, and therefore must be understood in how it ultimate ly relates to water quality. Additional large El Nio and La Nia anomalies within the hydrological time series record will also help to confirm patterns in the time and frequency domain. As a correlation has been 131

PAGE 132

established linking ENSO to precipitation, stream flow, and nutrient load in the LRW, the significant 3-7 year modes of variabi lity can be extracted from the signal and reconstructed into a predictive time series model for nitrate load based on the NINO 3.4 sea surface temperatures. By analogy, time seri es of different pollutant loads affecting water quality could be analyzed using these methods. Generating a forecast system for nutrient loads or other pollutants based on low frequency ENSO phenomena will be useful, feasible, and could avoid uncertainty associated with possible misrepresentation of physical processes and used in conjunction with hydrological models for the southeast coastal plain. 132

PAGE 133

133 Table 3-1. LRW time series data and sources used. Variable Years Source Comments NINO 3.4 SST (C) 19682005 Kaplan extended index, http://ingrid.ldeo.columbia.edu/ SOURCES/ .KAPLAN/ .EXTENDED/.ssta/ Anomaly data averaged monthly Precipitation (mm) 19682005 ftp://www.tiftonars.org/databas es/LREW/Precipitation/Daily Summed, averaged and normalized monthly. Some in-filled values. Stream Flow (m3) 19682005 ftp://www.tiftonars.org/databas es/LREW/streamflow/daily Daily values averaged and normalized monthly. Some in-filled values. NO3 Concentration (mg/L) 19742003 ftp://www.tiftonars.org/databas es/LREW/stream_water_ quality/long_term_water_ quality Daily/weekly values averaged and normalized by median monthly. Some In-filled data. NO3 Load (kg) 19742003 ftp://www.tiftonars.org/databas es/LREW/stream_water_ quality/long_term_water_ quality Daily/weekly values averaged and normalized by median monthly.

PAGE 134

134 (a) (b) Figure 3-1. The Little River Watershed and s ub-basins, Tifton, Georgia. (a) Position relative to the entire Suwannee River Basin [ Bosch, Sheridan, Lowrance, et al. 2007]. (b)The total area is 334 km2. All analysis was done with data from upland sub-basin K, with an area of 16.8 km2. Triangles indicate weir/stream flow measurement locations for each sub-basin.

PAGE 135

Figure 3-2. Significant Wavelet Power Spec tra are shown within the cone-of-influence, which depends on time series length and degrees of freedom. Figures are color-mapped to indicate high wavelet power with reds and oranges, and low powers in blue and white. The Global Wave let Spectrum (GWS) at the right of each figure shows power integrated over all scales and times. The 95% confidence limit is shown on the GWS (dashed blue line), the periodicities above which show significance. (a) M onthly NINO 3.4 ( C) Sea Surface Temperatures, (b) Monthly precipitation anomaly (mm ), (c) Monthly stream flow anomaly (m3), (d) Monthly NO3 concentration anomaly (mg/L), (e) Monthly NO3 load anomaly (kg). 135

PAGE 136

Figure 3-2. Continued. 136

PAGE 137

Figure 3-3. Cross Wavelet Spectrum betw een (a) monthly SST and Precipitation, (b) monthly SST and stream flow (m3), (c) monthly SST and NO3 concentration (mg/L), (d) monthly SST and NO3 Load (kg). Black figure outlines indicate areas significant to 95% confidence, wh ile arrows represent variables phase relationship. Arrows pointing clockwise indicate in-phase behavior, while counter clockwise arrows indicate anti-phase behavior. 137

PAGE 138

Figure 3-4. Wavelet Coherence Analysis bet ween (a) monthly SST and Precipitation, (b) monthly SST and stream flow (m3), (c) monthly SST and NO3 concentration (mg/L), (d) monthly SST and NO3 load (kg). Black figure outlines indicate areas significant to 95% confidence, while arrows represent variables' phase relationship. Arrows pointing clockwise indicate in-phase behavior, while counter clockwise arrows indicate anti-phase behavior. 138

PAGE 139

(a) (b) (d) (c) Figure 3-5. Cross-correlation analysis between NINO 3.4 (C) and LRW sub-basin K (a) precipitation (mm) (b ) stream flow (m3) (c) NO3 (kg), and between (d) precipitation (mm) and stream flow (m3). A negative lag indicates months that the NINO 3.4 SST leads the variable in question. The strongest CrossCorrelation Function (CCF) relationship within the NINO 3.4 index is SST leading stream flow by two months, wh ile the strongest hydrologic relationship overall is between precipitation and stream flow. Values above or below dashed lines indicate significant correla tion above 95% confidence, corrected for autocorrelation. 139

PAGE 140

1 2 3 4 5 6 7 8 9 10 11 12 -5 0 5 10 SST (deg. C) Month (b) 1 2 3 4 5 6 7 8 9 10 11 12 -5 0 5 10 SST (deg. C) Month (a) 1 2 3 4 5 6 7 8 9 10 11 12 -5 0 5 10 SST (deg. C) Month (c) Figure 3-6. Box and whisker plots of observed NINO 3.4 SST anomalies (C) for MENSO (a) neutral years, (b) La Nia year s, and (c) El Nio years. Box lines are at the lower quartile, median, and upper quartile values. Whiskers extend from box ends to upper and lower adjacent values, defined as the largest or smallest observation that does not exceed the upper or lower quartile 1.5 interquartile range. Outlier points indicate data that fall outside the whiskers, and are indicated by crosses. 140

PAGE 141

1 2 3 4 5 6 7 8 9 10 11 12 -150 -100 -50 0 50 100 150 200 Precipitation Anomaly (mm) Month (a) 1 2 3 4 5 6 7 8 9 10 11 12 -150 -100 -50 0 50 100 150 200 Precipitation Anomaly (mm) Month (b) 1 2 3 4 5 6 7 8 9 10 11 12 -150 -100 -50 0 50 100 150 200 Precipitation Anomaly (mm) Month (c) Figure 3-7. Box and whisker plots of obs erved precipitation anomalies (mm) for MENSO (a) neutral years, (b) La Nia year s, and (c) El Nio years. Box lines are at the lower quartile, medi an, and upper quartile values. 141

PAGE 142

1 2 3 4 5 6 7 8 9 10 11 12 -2 -1 0 1 2 x 106 Flow Anomaly (m3/sec) Month (a) 1 2 3 4 5 6 7 8 9 10 11 12 -2 -1 0 1 2 x 106 Flow Anomaly (m3/sec) Month (b) 1 2 3 4 5 6 7 8 9 10 11 12 -2 -1 0 1 2 x 106 Flow Anomaly (m3/sec) Month (c) Figure 3-8. Box and whisker plots of observed stream flow anomalies (m3/sec) for MENSO (a) neutral years, (b) La Nia year s, and (c) El Nio years. Box lines are at the lower quartile, medi an, and upper quartile values. 142

PAGE 143

1 2 3 4 5 6 7 8 9 10 11 12 -10 0 10 20 30 40 NO3 Concentration Anomaly (mg/L) Month (a) 1 2 3 4 5 6 7 8 9 10 11 12 -10 0 10 20 30 40 NO3 Concentration Anomaly (mg/L) Month (b) 1 2 3 4 5 6 7 8 9 10 11 12 -10 0 10 20 30 40 NO3 Concentration Anomaly (mg/L) Month (c) Figure 3-9. Box and whisker plots of observed NO3 concentration anomalies (mg/L) for M-ENSO (a) neutral years, (b) La Nia year s, and (c) El Nio years. Box lines are at the lower quartile, medi an, and upper quartile values. 143

PAGE 144

1 2 3 4 5 6 7 8 9 10 11 12 -200 0 200 400 600 800 1000 1200 NO3 Load Anomaly (kg) Month (a) 1 2 3 4 5 6 7 8 9 10 11 12 -200 0 200 400 600 800 1000 1200 NO3 Load Anomaly (kg) Month (b) 1 2 3 4 5 6 7 8 9 10 11 12 -200 0 200 400 600 800 1000 1200 NO3 Load Anomaly (kg) Month (c) Figure 3-10. Box and whisker plots of observed NO3 load anomalies (kg) for M-ENSO (a) neutral years, (b) La Nia years, and (c) El Nio years. Box lines are at the lower quartile, median and upper quartile values. 144

PAGE 145

Precipitation (1968 2005) 0 100 200 300 19681973197819831988199319982003 Years after January, 1968Monthly Preci p (mm) (a) Stream Flow (1974 2004) 0.0E+00 1.0E+06 2.0E+06 3.0E+06 1974 1979 1984 1989 1994 1999 2004 Years after January, 1974Monthly Flow (m3/se c (b) NO3Concentration (1974 2004) 0 10 20 30 40 50 1974 1979 1984 1989 1994 1999 2004 Years after January, 1974Monthly Concentrati o (mg/L)(c) NO3Load (1974 2004) 0 400 800 1200 1974 1979 1984 1989 1994 1999 2004 Years after January, 1974Monthly Loa d (kg)(d) Figure 3-11. Raw monthly time series of (a) precipitation (mm), (b) stream flow (m3/sec), (c) NO3 concentration (mg/L) and (d) NO3 load (kg) 145

PAGE 146

CHAPTER 4 INTEGRATION OF ENSO SIGNAL POWER THROUGH HYDROLOGICAL PROCESSES IN THE LITTLE RIVER WATERSHED Introduction When the relationship of the El-Nio /S outhern Oscillation (ENSO) to hydrology is commonly discussed, it is typically in te rms of the ability to separate significantly different hydrologic variable responses depending on the type of ENSO anomaly that has taken place. Using this kind of analysis, ENSO phase has been shown to have predictable effects on precipit ation, stream flow, mons oon occurrence, crop yield, cholera occurrence, flood frequency, phyt oplankton growth, and water quality data in different regions around the world [ Chiew et al. 1998; Rajagopalan and Lall 1998; Hansen et al. 1997; Pascual et al. 2000; Piechota and Dracup 1999; Keener et al. 2007, 2010; Nezlin and Li 2003; Chiew and McMahon 2003]. From a research standpoint, most of the work relating ENSO trends to proxy variables had been done on precipitation records until the mid 1990s, at which point in creasing numbers of studies started to focus on ENSO relationships with stream flow as well as other environmental variables [ Cayan et al. 1999; Charles et al., 1997; Fraedrich and Muller 1992; Gutierrez and Dracup 2001; Ropelewski and Halpert 1987; Zorn and Waylen 1997]. The signals in stream flow are typically complex, repres enting the integration of both climatic, landscape, and anthropological respons es that are able to increase the power of the inherent ENSO signal in regional precipitation data. With increasing research on ENSO and stream flow relationships, it has been observed in case studies around the world that the complex relationships in a hydrological system can serve to increase the inherent ENSO signal power within precipitation in other hydrologic variables For example, increased atmospheric CO2 146

PAGE 147

levels representing greenhouse gas globa l warming scenarios have been shown to have a doubling effect on winter stream flow signal in El Ni o years in north California [ Maurer et al. 2006], while stream flow responses to ENSO in the western United States have generally been shown to be ac centuated over precipitation due to differences in the duration of wet phases in different ENSO regimes [ Cayan et al., 1999]. Multiple studies at both the regional and global scale have discovered that the relationship between ENSO and stream flow response can be stronger than that with its relationship to precipitation, likely due to str eam flow effectively spatially and temporally integrating the signal in rain fall and reducing the noise [ Chiew and McMahon 2003; Keener et al. 2010; Whiting et al. 2004; Dettinger and Diaz 2000]. This non-intuitive relationship strength could prove to be quite useful in terms of making seasonal climate predictions based on climate indices such as ENSO, as a geographic area with a seemingly weak EN SO precipitation signal could have a stronger stream flow or other proxy vari able climate signal that could be used for predictive purposes. The use of seasonal forecasts of both flows and other hydrologic variables could also more immediately benefit water resource managers or farmers, as decisions on irrigation and allocation would be able to incorporate additional climate information, and would not be predictions made so far in the future as to render them virtually unusable in typical policy-based management plans. In Chapter 3, both univari ate and multivariate wavelet analysis were used to quantify the significance of a teleconnection between the continuous NINO 3.4 index of sea surface temperatures (SST) and observed precipitation, stream flow, and nitrate concentration and load in the Little River Watershed (LRW) in Georgia. Mechanistically, 147

PAGE 148

El Nio (La Nia) events result in increas ed (decreased) sea surface evaporation. The mid-latitude jet stream is displaced equator ially (poleward), increasing (decreasing) winter frontal precipitation in the southeast Un ited States. Additional winter moisture is advected into the southeast from the tropical Pacific by the subtropical jet stream [ Ropelewski and Halpert 1987]. Precipitation in El Nio winters typically increases as a result, and as seen in the large El Ni o events centered ar ound 1985 and 1997-98 in the LRW, resulted in increased river di scharge and nutrient pollution transport [ Keener et al., 2010]. In Chapter 4, hydrological simulation using the Soil and Water Assessment Tool (SWAT) and observed data is used to explore the ENSO signal power in stream flow time series compared to prec ipitation. Although the thirte en weather stations used to form the precipitation series in this re search are well distributed across the 16.8 km2 area of basin K in the LRW, spatial variability of rainfall in the southeast United States is high, and the weather stations still do not enc ompass all of the prec ipitation variability inherent in the watershed [ Baigorria et al. 2007]. Specifically, it is shown how: (1) the observed and SWAT simulated stream flow signal in the LRW has a more powerful relationship with ENSO than that of obser ved precipitation, (2) validated SWAT simulations of groundwater dynamics can help us identify what is causing the climate and stream flow signal power increase, and (3) wavelet analysis can be used in a novel way to elucidate not only changes in spectral strength over time, but also to identify mechanistic similarities in simulated data. We hypothesize that there are several me chanisms of ENSO/stream flow signal power increase: If groundwater flow in the LR W is responsible for up to 80% of the total 148

PAGE 149

stream flow [ Sheridan 1997], as well as comprising the main route for movement of soluble nutrients [ Lowrance et al. 1984], then the power of the ENSO signal in stream flow may be strengthened by the significant ro le of groundwater and the presence of the Hawthorn confining layer, both restricting flow into the deep aquifer system and increasing the amount of interflow in the root zone represented in the st ream flow record [ Stringfield 1966; Sheridan 1997]. Another possibility is that in anomalous ENSO phases, the potential evapotranspiration (E T) is affected enough to significantly influence stream flow. For example, if in an El Nio winter in southern Georgia and the LRW, temperature decreases and precipitation significantly increases; then conversely, ET may significantly decrease due to less so lar radiation and less of a vapor pressure deficit. In turn, this could in crease the stream flow signal. To elucidate what is mechanistically happ ening to the climate signal in the LRW, a SWAT model of basin K that is calibrat ed and validated for hydrology, phosphorus and total nitrogen nutrient cycles was used to l ook at simulated surface and groundwater movement and runoff from 1979 to 2005, with observed land use input from LandSat images and observed crop rotations and management [ Bosch et al. 2004, 2006; Cho et al. 2010; Feyereisen, Strickland, et al. 2007]. Using a validated SWAT model of the LRW enables us to analyze detailed continuous simulations of the entire water budget, and more specifically the groundwater budget re cords to look for ENSO trends that may be visible in the groundwater hydrology time series. Additionally, simulated and observed potential ET records c an be analyzed in the same manner. Simulated hydrological cycle components ar e visualized using an average annual water budget and as a percentage of observed annual average precipitation. They were 149

PAGE 150

then analyzed using exceedance curves to show the 10, 50, and 90% probabilities that a specific variable was equaled or exceed ed, separated by monthly ENSO phase dynamics and compared to the LRW observ ed climatology during all phases. The exceedance probabilities in the observed EN SO precipitation were found to be even more pronounced in the SWAT simulated flow and groundwater variables, demonstrating the various flows increasing inter-annual variability signal power as compared to rainfall. Wavelet analysis, a spectral method of dec omposing a time series into time and frequency space that allows the identification and analysi s of dominant localized variations of power, was used as in Chapter 3. There, it was used to analyze the NINO 3.4 SST index and observed hy drological and nutrient data in the LRW basin K. Here, wavelet analysis is used to analyze and compare various simulated SWAT components of the hydrological cycle wit h the NINO 3.4 index. In this way, we can quantify components of the hydrological cycle that are strengthening the ENSO related stream flow signal. Patterns found in the observed rainfall are more spectrally powerful in the observed and simulated str eam flow, and these shared patterns are identified as powerful signals of interest in the SWAT si mulated groundwater dynamics that increase the oscillatory climate-related signals. Data and Methods Field Site and Data: The Little River Watershed The same field site, the Little River Wa tershed (LRW), basin K, is used as is described in Chapter 3, Data and Methods, se ction Field Site: Little River Watershed. The same observed precipitati on from 13 spatially distributed weather stations, stream flow, and nutrient data from basin K is also used as described in Chapter 3, Data and 150

PAGE 151

Methods, section Little River Watershed Data as input and calibration data for SWAT. Additional weather input, in cluding minimum and maximum daily air temperatures, solar radiation, and relative humidity data were obtained from a University of Georgia weather station that is 1 km northeast of LRW-K [ Bosch et al. 2004; Cho et al. 2010]. A digital elevation model (DEM) with a 30 meter grid size was used in SWAT to specify subbasins, stream networks, and topography, and digital soil data from the USDA-NRCS Soil Data Mart were used to der ive the soil related parameters [ Cho et al. 2010]. Agricultural land boundaries and non-row cr op land cover were input from one meter resolution USGS orthoquadrangle aerial pictures, and while the land boundaries were assumed to be constant during the peri od of simulation, changes in crop rotations and management within the fields were i nput using observed harvested crop area data from the University of Georgia Cooperative Extension service [ Bosch et al. 2004; Cho et al., 2010; Feyereisen, Strickland, et al. 2007]. Delineations of ENSO phase are done via the categorical monthly M-ENSO index [ Grard-Marchant et al. 2008], and from the continuous NINO 3.4 SST monthly index as described in the introductions of Chapter 2 and Chapter 3, respectively. SWAT Model Hydrologic Calibration and Validation The Little River Watershed is a part of the USDA-ARS Watershed Assessment Study and the national Conservation Effect s Assessment Program (CEAP), which has the aim of quantifying the e ffects and successes of soil and water conservation and management procedures around the country to support sound policy implementation [ USDA-ARS 2005]. To gather data for this nati onal assessment, researchers at the SEWRL in the Little River Watershed have performed a calibration and sensitivity analysis on basin Ks hydrolog ic and subsequently its nutrient parameters in SWAT that 151

PAGE 152

could provide guidance for future SWAT calibration in t he southeastern U.S [ Feyereisen, Strickland, et al. 2007; Cho et al. 2010]. SWAT is a versatile agrohydrologic, geochemical process model that can simulate hydrologic budgets and nutrient processes at either field or watershed scales [ Neitsch et al. 2005]. SWAT is spatially semi-distributed, as it first delineates sub-catchments via topography and a designated threshold area, and then further divides the sub-catchments into Hydrological Response Units (HRU) bas ed on unique inputs of management, soil, and vegetation parameters. Each HRU is modeled individually, then aggregated at the subcatchment level and routed to the associated stream reach. Groundwater flow is also spatially semi-distributed, and considers sp ecific hydraulic conductivities and storage coefficients at the HRU level only. Full theoret ical documentation of SWAT is available online in Neitsch et al, 2005. In its vari ous incarnations, SWAT has been successfully tested and modified for different environment al conditions throughout the world [ White and Chaubey 2005]. The version of SWAT used in these modeling studies is AVSWATX, which is the same process based model with a GIS interface implemented in ArcView 3.2. For hydrological calibration between 19791994 and validation from 1995-2004, the manual calibration procedure outlined in the 2000 SWAT user manual was followed to maximize the Nash-Sutcliffe Efficiency and minimize the sum-of-squared differences of the water budget calculation (SSDWBC) [Feyereisen, Strickland, et al. 2007; Neitsch et al., 2002]. First, annual surface flow was balanced by adjustment of the NRCS runoff curve number, followed by groundwater bas eflow adjustment to observed values [ Shirmohammadi et al., 1986]. Model evapotranspiration (ET) output was calibrated to 152

PAGE 153

match observed ET, minimizing the SSDWBC, and finally, parameters influencing baseflow speed, shallow aquifer depth and su rface runoff lag were adjusted to maximize the Nash-Sutcliffe efficiency (NSE) [ Feyereisen, Strickland, et al. 2007]. These calibrated parameters were used as t he base values for the sensitivity analysis on the 16 selected SWAT hydrologic paramet ers covering surface, sub-surface, and overall basin flow response. Input parameters chosen for calibration and sensitivity analysis were based on a combination of know ledge from the results of simulations in different basins adjacent to K within the Little River Watershed [ Van Liew et al. 2007; Bosch et al. 2004], and several investigatory soil parameters of interest. Local sensitivity was calculated using the sensitivity coefficient, S (Equation 4-1), which quantifies the ratio between t he rate of change of a model output ( O ) and the rate of change of model input parameter ( P ) of interest. P O S (4-1) Relative sensitivity, the unit-less variable Sr, was then calculated as equation 4-2 [ Haan 2002], where OP+/P are parameter outputs plus or mi nus 25% of the input parameter base value, normalized by the model output with the parameter at the base value, OP, and the absolute change in the va lue of the input parameter, P with respect to its initial value, P PP OOO SrPPPPP/2 / (4-2) This measure of local relative sensit ivity, while it does measure parameter influence on model output, does not take param eter interactions into account, and does not simultaneously vary parameters in a global sense. Feyereisen et al (2007) achieved a final model monthly NSE of 0.88 and a daily NSE of 0.56. Relative sensitivity analysis 153

PAGE 154

results showed that surface response paramet ers were much more sensitive than basin and sub-surface parameters. T he five most sensitive hydr ologic input parameters for total water yield, storm flow, and base flow proved to be mainly the CN for cropped land (CN = 76), followed by the CN for fo rested land (CN = 50), soil evaporation compensation factor, so il available water content, and soil bu lk density, while insensitive to the other 11 parameters tested [ Feyereisen, Strickland, et al. 2007]. Overall, the CN for cropped land and the soil available wate r content were t he parameters that dominated the SWAT calibration for t he Little River Watershed basin K. The dominance of the curve number fo r cropped land on the model output was surprising, given that 30% of the area in basin K is agricultural in nature [ Bosch et al. 2004; Feyereisen, Strickland, et al. 2007]. However, a change in only 1% of the base value of the CN for cropped land would result in a 74% increase in total water yield output, and a 422% increase in storm flow output [ Feyereisen, Strickland, et al. 2007]. This sensitivity is in part due to the incr eased surface runoff from even average-sized precipitation events on cropped land. Studies have been done on the entire LRW to evaluate the hydrologic impacts of observed land-use change over the past 35 years. As part of the USDAs Conservation Reserve Program (CRP), over 13 million ha of erodible and environ mentally sensitive agricultural land has been converted into non-tilled land since 1985 [ Bosch et al. 2006], including 261,511 ha of trees planted in Georgia alone [Moorhead and Dangerfield 1996]. The potential for the CRP program to cause dramatic changes in forested landuse within many watersheds is large, with resulting change in ET, stream flow, and infiltration regimes. Using SW AT on LRW-K, it was found t hat the land-use changes that 154

PAGE 155

occurred between 1975 and 2003 did not significant ly affect the str eam flow or other hydrological regimes of the basin, although the land-use percentages did not drastically change from agriculture to forest duri ng the dates monitor ed (Figure 4-1) [ Bosch et al. 2006]. In the whole time period analyzed, ther e was only a 1% decrease in total upland and riparian forest acreage, with a 3% in crease between 1985 and 2003 when the CRP effects would have been expected to appear. As seen in Figure 4-1, a 12% decrease in fallow, 7% increase in agriculture, and 5% increase in pasture acreage was observed over the whole watershed [ Bosch et al. 2006]. SWAT partitions groundwater into tw o aquifers: a shallow, unconfined aquifer that may interact with surface soil layers, and a deep, confined aquifer in which the contributed flow is assumed to exit the wa tershed. In the SWAT representation of LRWK, deep aquifer recharge is only 1% of total precipitation due to t he confining Hawthorn layer, and therefore almost negligible [ Rawls et al. 1976]. To investigate certain processes in the watershed, specific variabl es were isolated withi n SWAT (Figure 4-2). For the LRW in SWAT, simulated surface r unoff is calculated from the empirical SCS Curve Number method, based on precipitation, soil parameters and water content [ USDA 1972]. Simulated potential ET is computed via the Priestly-Taylor method [ Priestley and Taylor 1972]. Percolation of groundwater occurs when the field capacity of the soil layer is exceeded and the layer below is not saturated. If percolated water passes through the last root zone layer, it enters the unsaturated vadose zone before becoming shallow aquifer recharge. In SWAT, interflow is called lateral flow and is defined as sub-surface flow within the root zone. Interflow is signific ant in areas with soils having high hydraulic 155

PAGE 156

conductivity in the surface layers, and an impermeable layer at shallow depths, which occurs in some areas of the LRW in which the Hawthorn Formation is either at or very close to the surface. Groundwater flow repr esents baseflow that or iginates from the vadose zone and shallow aquifer of the SWAT model, and is dependent on the particular watersheds storage capacity and physica l characteristics. All flows are routed to the open stream channel, at which point they are routed using Mannings Equation and a variety of open channel hydrodynamic methods [ Neitsch et al. 2005]. Water Budget and Exceedance Curves An annual hydrologic budget of relevant vari ables is a simple method of comparing relative amounts of water simulated by SWAT in the LRW. In this case, since we are interested in the relationship between hy drology and ENSO climatology, the annual average water budget for each variable has been calculated for all years, and for each defined M-ENSO year. The SWAT simulated variables SUR_Q (surface runoff), PERC (percolation past the root z one), GW_Q (baseflow contributi on to stream flow), LAT_Q (root zone flow contribution to stream flow ), Q (total stream flow), and total ET (evapotranspiration) were compared as single yearly average values over the simulation period 1979 to 2004 in differ ent categorical ENSO phase s. The observed actual precipitation record was also used as a point of comparison. The water budget was represented as the total annual average water yi eld for each variable for all years, and as a percentage of total annual observed prec ipitation for each of the three ENSO phases. The same SWAT simulated variables listed above and observed precipitation were also utilized in probability of exceedance gr aphs. An exceedance curve is a backwards cumulative form of a probability density curv e, showing what percentage of area under 156

PAGE 157

the probability density curve lies to the right of the value of t he variable on the x-axis. In this case, the probability of exceeding eac h value for each variable was calculated for four different cases: all avai lable data, or the c limatology (black dots), El Nio (blue dots), La Nia (red dots) and Neutral ENSO phase (green dots). One then can compare the curves in each ENSO phase with bot h the whole climatology and against each other. For example, a tendency toward drier t han normal precipitation conditions in La Nia would be indicated if the red curve lies to the left of the black curve. Exceedance curves should not be read as forecasts. If the red and black curves overlapped, it would not indicate a forecast for average conditi ons in a La Nia phase, but that there are equal probabilities in a La Nia phase for anything to occur that has happened during the total climatological period. The values for each variable at the 10, 50, and 90% probability of exceedance were compared across ENSO phases and climatologies. Wavelet Analysis All technical wavelet methodology is the same as detailed in Chapter 3, Data and Methods sections Wavelet Analysis and Cross Wavelet and Coherence Transforms. Typically, as in Chapter 3, wavelet analysi s is used in conjunction with geophysical data to visualize how dominant frequency structures in non-stationary data change over time. In a more novel use, here we use wavelet analysis to observe whether SWAT simulated components of the hydrological cycle share the climate-based significant spectral variability we identified in the observed stream flow record. By analyzing SWAT simulated surface runoff, groundwater and inte rflow, percolation pa st the root zone and evapotranspiration, patterns can be identified that are shared with those seen in the wavelet spectra of observed surface runoff. 157

PAGE 158

Hydrological components t hat may be strengthening th e ENSO signal power in stream flow time series as compared to precipitation wo uld be expected to share more spectral power with the NINO 3.4 SST index. Simulated individual variables with significant visible ENSO in ter-annual power were then analyzed by cross-wavelet and wavelet-coherence analysis with the NINO 3.4 index to confirm areas of shared significant power and localized correlation. In this way, we were able to visualize components of the hydrological cycle that strengthen the ENSO related stream flow signal. Univariate wavelet analysis, and mult ivariate cross-wavelet (XWT) and wavelet coherence (WTC) analyses were done on obse rved NINO 3.4 SST records and the simulated SWAT variables total flow, groundw ater flow, interflow, percolation past the root zone, and evapotranspiration from 1979-2005. Processed daily SWAT simulated flow, groundwater components, and ET were su mmed into monthly cumulative values and normalized by monthly average for input into the wavelet analysis. Results Observed Precipitation and Stream Flow In Chapter 3, an interesting result f ound through wavelet analysis was that the significant spectral climate-based signal in the observed stream fl ow data was stronger than that seen in the correspond ing precipitation data (Figur e 4-3). This can be seen in the fact that the areas of high spectral power, areas of red and orange in the wavelet spectra, are more widespread and cohesiv e throughout the wavelet spectrum of observed stream flow records (Figure 4-3b) than pr ecipitation (Figure 4-3a), as well as in known ENSO periodicities sustained ar ound 3-7 years and surrounding times of high El Nio abnormalitie s in 1997 and 1983. 158

PAGE 159

Additionally, the global wavelet spectrum (GWS) on the right side of each wavelet spectrum figure shows the more standard spec tral visualization of time-integrated variance of the energy coefficients at ev ery scale throughout the wavelet transform (Figure 4-3). In the GWS, sca les that are significant abov e the 95% level against a red noise background exceed the dashed line. While the observed precipitation time series only shows 95% significance within the coneof-influence from t he 1.2 to 1.7 year period, the observed stream flow signal shows significance both from the 1.5-1.8 and 3.34.4 year period. Observed stream flow also shared more significant areas in crosswavelet and wavelet coherence analysis with the NINO 3.4 index than it did with precipitation (Chapter 3), reinforcing that the inter-annua l climate signal is indeed stronger in flow. SWAT Water Budget and Exceeda nce Curves by ENSO Phase Simulated hydrologic variables were compared in the water budget as single yearly average values over the simulati on period from 1979 to 2004, and compared to their values when separated by M-ENSO phase. The SWAT simulated variables examined in detail were SUR_Q, PERC, GW_ Q, LAT_Q, Q, and total ET. The observed precipitation record was also used as a poi nt of comparison. The water budget is first visualized as the total annual average water yiel d for each variable for all years, and for each of the three ENSO phases (Figure 4-4). Figure 4-4 shows that in the El Nio phas e in the LRW-K, there is more total precipitation than in other EN SO phases, while there is less precipitation in the La Nia phase. When this is visualized in the water budget, we see that the trend of greater precipitation in El Nio years is also reflect ed in greater simulated surface runoff, stream flow, interflow, groundwater flow, and percolation in El Ni o years, with the greatest 159

PAGE 160

disparity in groundwater flow and percola tion. On the other hand, total average evapotranspiration remains fairly consistent through each ENSO phase. When the same data are analyzed in terms of the simulated hydrologic variables as a percentage of the total observed precip itation (Figure 4-5), the picture can be explained further. As a percentage of total precipitation, ET in El Nio years was less than all other ENSO phases, whil e it was greater in La Nia y ears. This implies that the increased (decreased) precipit ation and decreased (increased) temperature in El Nio (La Nia) ENSO phases does in fact decrease (i ncrease) the relative percentage of total ET. Combined with the observation that the relative increase or decrease in precipitation during El Nio or La Nia years is mirror ed in the simulated groundwater flow and percolation, it appears that mec hanistically, both groundwater flow and ET amounts respective to precipitation are contri buting to the increased power of the ENSO signal seen in observed st ream flow records. The probability of exceedance curves are s hown in Figure 4-6. The general trends plotted as monthly values separated by M-ENSO phase follow the same patterns that are present in the water budgets (Figure 4-4 and 4-5). In the curves for all variables (Figure 4-6 a-g), the total c limatological distribution over lays with the neutral M-ENSO year distribution very well, meaning that neutral and average observed conditions have the same basic probabilistic distribution fo r all variables. The values of the given variables at 10, 50, and 90% probability of exceedance during the different M-ENSO phases are listed in Table 4-1. Quantitatively as well, the climatology and neutral values for all variables are either ex actly the same or very close to being the same, for both 10, 50 and 90% probabilities (Table 4-1). It should be noted, however, that the 160

PAGE 161

climatological distribution is only based on the available data, and therefore may not capture extreme historical events if they were not in this data record. For observed precipitation (Figure 4-5a), there was a general pattern of greater probability of observing lower(higher) than normal values in La Nia(El Nio) phases, and slightly higher(lower) than average valu es in El Nio(La Nia) phases. These patterns are reflected in the distribution ske w, that at 10% probabi lity of exceedance El Nio rainfall was 107% of and La Nia was 74% of neutral rainfall, while at 90% probability of exceedance El Nio was 64% and La Nia was 101% of neutral rainfall (Table 4-1). The observed precipitation patterns were reflected in the SWAT simulated total stream flow (Figure 4-5b), gr oundwater flow (Figure 4-5c), interflow (Figure 4-5d), and percolation (Figure 4-5e) to more significant extents, agai n showing the increase of the relative flows signal power. In the case of total flow, groundwater and interflow and percolation, the exceedance curves during El Nio were skewed towards higher values. The most extreme differences at the 10% probability of exceedance between El Nio and La Nia years were seen in simulated total flow (141% and 54% of neutral flow) and even more in groundwater flow (143% and 44% of neutral flow), in which both magnitudes of difference are greater than those seen in obs erved precipitation. Less extreme climatic differences were seen in interflow, percolation, surface runoff and ET records (Table 4-1). Practically, this means that historically, there was a 10% probability in El Nio of exceeding high rela tive values of total stream flow and groundwater flow relative to neutral conditions. 161

PAGE 162

Univariate Wavelet Analysis The monthly wavelet power spectra for the NINO 3.4 SST and SWAT simulated groundwater flow, interflow, surface runoff, per colation, and ET time series are shown in Figure 4-7a-f. As previously demonstrated [ Y. Wang 1996; Torrence and Compo 1998; Keener et al. 2010], when the regular annual cycle is removed, SST power (Figure 47a) is concentrated within the ENSO per iodicity band of 3-7 years, although the amplitude and dominant m odes tend to shift through time From 1911-60, a 5-7 year period is strongest, while a 4-5 year period dominates from 1972-92 [Y. Wang 1996]. Longer decadal variations cannot be assessed with significance given the limited length of record. Precipitation in the southeas tern United States has distinct seasonality. In the coastal plain region and more specifically the LRW, the most in tense precipitation events are in the spring and summer months, associated with convective or cyclonic storms [ Sheridan 1997]. Summer events are shorter, smaller in area, and more frequent and intense, while fall and winter events are frontal in nature, milder, but longer in duration. Spectral precipitation info rmation corresponding to the 3-7 year ENSO signal has been demonstrated in the Florida Everglades via wavelet analysis, as well as in the western US [ Kwon et al. 2006; Rajagopalan and Lall 1998], and the Little River Watershed shares that signal (Figure 4-3a). R egions of high power relative to the noise background were seen in the observed precip itation, flow, and nutrient record in the same 3-7 year periodicity as for the NINO 3.4 series (Chapter 3, [ Keener et al. 2010]). In basin K of the LRW, total average annual stream flow depth is approximately one-third of annual precipitation [ Sheridan 1997; Feyereisen et al. 2008], a figure comparable to similar statisti cs from other coastal plain wa tersheds. In a typical cool, 162

PAGE 163

wet winter in the LRW, which is more associated with El Nio events, the shallow aquifer and the vadose zone are nearly fully sa turated, with the water table directly below the root zone [ Bosch et al. 2004]. Additionally, detailed field studies in the watershed have found the shallow aquifer to va ry from being 0 to 7 meters below the surface throughout the year [ Bosch et al. 2003], making it quick to become fully saturated. As a consequence, almo st all precipitation in El Nio winters is received as stream flow (Figure 4-7b), enhancing the overall ENSO signal in the groundwater (Figure 4-7c). This pattern is especially enhanced by the confining Hawthorn formation underlying the entire area of bas in K, which under these satu rated conditions allows even the interflow component (LATQ) to exhibit a strong signal powe r (Figure 4-7d) that joins the total stream flow. Conversely warm and dry winters associated with La Nia events exhibit measured water table condit ions that are well below saturation [ Shirmohammadi et al., 1986; Bosch et al. 2004]. Precipitation events received in these conditions in the LRW can infiltrate ve ry quickly, with little to none reappearing as stream flow, as no-flow condi tions can occur in the LRW [ Shirmohammadi et al., 1986; Bosch et al. 2004; Feyereisen, Strickland, et al. 2007]. As seen in the wavelet power spectra of these SWAT simulated time series, 3-7 year periodic ENSO power significant at the 95% level against a red noise background is visible in the stream flow and mirrored through the groundwater and interflow spectra (Figure 4-7c and d). This strong inter-annual power in the groundwater and interflow, denoted by reds and oranges in the wavelet s pectrum, have both a more powerful interannual periodicity than the stream flow si gnal and a larger integrated area of 95% significance in the global wavelet spectra (GWS) to the right of each wavelet spectra 163

PAGE 164

figure, despite the interflow component having a much smaller magnitude than the other flows. In the GWS, simulated total flow has significant peak spectra at 3.2 to 4.0 years (Figure 4-7b), while groundwater has peaks fr om 1.0-1.6 and 2.4-3.6 years (Figure 47c) and interflow from 1.1-1.8 and 2.8-3.6 years (Figure 4-7d) both of which show more significant periodicities than observ ed precipitation as discussed earlier. Simulated percolation past the root zone (Figure 4-7e) also exhibits a significant 37 year ENSO signal; however, t he strength of the signal is not as strong (signified by strong patterns of reds and yellows) as the flow or groundwater components. This is possibly because during wet El Nio winters, the soil remains consistently saturated, obscuring dynamics that could enhance the ENSO signal, although percolation still exceeds the 95% GWS significance from 1.11.8 and 3.3-4 years. Finally, the wavelet power spectrum of the SWAT simulated evapotranspiration time series (Figure 4-7f) does not show any significant periodicities in the GWS graph, nor does it show the familiar powerful patterns in the interannual climate based peri odicities that are recognized in the other spectra of Figure 4-6. This fact implies that while ET is in fa ct different in El Nio and La Nia years from neutral conditions, it does not exhibit a spectr al pattern showing a strong teleconnection between ENSO sea surface temperatures and evapotranspiration in the LRW. The observations of patterns of strong inte r-annual groundwater and interflow power with greater significant 95% periodicities in t he GWS lend support to the hypothesis that groundwater and interflow are more responsible for increasing the power of the total stream flow ENSO signal rela tive to precipitation, than ET dynamics and percolation. To 164

PAGE 165

determine whether NINO 3.4 SST and these SWAT simulated variables varied together and shared spectral power, cross wavelet and wavelet coherence analyses were done. Cross Wavelet and Wavelet Coherence Analysis To directly analyze whether or not two time series share common wavelet power and localized correlation, cross wavelet (X WT) and wavelet coherence (WTC) analyses were performed on the NINO 3.4 SST record and the SWAT simulated variables of interest, groundwater flow, interflow, perco lation, and evapotranspiration. Significance levels of the cross-spectra power are calculated against a red noise background, indicated by thick black outlines in the cr oss wavelet transform spectra (Figure 4-8a-d) to the 5% level. The cross wavelet transform between SST and simulated groundwater flow (Figure 4-8a) shows that areas that were selected as possibly sharing power in the single wavelet spectra also share significant power in the 3-7 year periodicity centered around 1982-83 (it is difficult to see the exact center since it is outside of the cone of significance) and 1997, two El Nio, high prec ipitation and high flow years in the LRW. These same areas of high common power we re also seen between SST and interflow (Figure 4-8b), and SST and percolation (Figure 4-8c). The relationship between SST and interflow had the most areas of signifi cant high shared power. Evapotranspiration and SST had the least shared signi ficant power (Figure 4-8d). The areas with 95% significance were phase-locked positively in the case of SST and groundwater flow, interflow, and percolation (Figure 4-8a, b, c), meaning they vary together. The significant areas in Figure 4-8d, SST and evapot ranspiration, were out of phase, implying an opposite re lationship between the two variables. Compared to the XWT between NINO 3.4 SST and observed stream flow seen in the results of Chapter 165

PAGE 166

3, (Figure 3-8b), there was le ss significant area in the simulated groundwater flow and interflow cross-spectra. However, there were also fewer visible years in the analysis which narrows the cone of influence, and while groundwater flow and interflow account for nearly 80% of stream flow in the LRW [ Sheridan 1997; Lowrance et al. 1984], they do not account for all of it, which may slightly weaken the oscillatory signal. Based on the cross-wavelet transforms, we are able to say that the NINO 3.4 index shares significant 3-7 year spectral power with SWAT simulated groundwater and interflow time series, some power with percolati on, and little explanatory power with evapotranspiration. This furthers the hypothesis that the ENSO signal in stream flow is enhanced by the role of groundwater and inte rflow as restricted by the Hawthorn formation. As was the case with the observed LRW time series in Chapter 3, a larger area in the WTC spectra of SWAT simulated data (Fi gure 4-9a-d) was marked as significant as compared to the XWT spectra. These areas of significa nt wavelet coherence are indicative of localized correlation strength, no t shared time series wavelet power. With a short record of simulated hydrological data, it is difficult to at tribute even areas of sustained significance visible in the WTC as implying causality; however, it is still possible to compare them to significant regions in the XWT to verify a simulated variables covariance with the NINO 3.4 index Compared to the observed stream flow record visualized via WTC in Chapter 3 (F igure 3-4b), all simulated variables had less significant power, which is expected given t hat these are components of the stream flow itself. 166

PAGE 167

Of the SWAT simulated variables, NINO 3.4 SST and groundwater flow had the most significant wavelet coherence pattern in the 3-7 year oscillatory period (Figure 49a). This was followed closely in significant area by simulated interflow (Figure 4-9b), with much less significant covariance betwe en SST and percolation (Figure 4-9c) and evapotranspiration (Figure 4-9d). The areas of sustained WTC significance in the groundwater and interflow spectra were bot h most likely center ed around 1982-84, and also showed a positive in-phase relationship with SST. It is curious that the 1997 El Nio event does not show up on any of the WTC spectra, as it does on both the observed data from Chapter 3 (Figure 3-4), and in the XWT (Figure 4-8). The lack of a signal in 1997 means that these variables we re not showing localized covariance with SST, which may be due to the nature of the SW AT simulation, or may have implications for the covariance of the groundwater dynamics with climate variability during this large El Nio event. The shared wavelet significance of the NINO 3.4 index seen in the XWT and the WTC for groundwater and interflow in the 3-7 year ENSO periodicity, and lack thereof in the evapotranspiration time series, again supports the hypothesis of groundwater and interflow enhancing the stream flow ENSO signal in basin K of the Little River Watershed. Summary and Discussion Hydrological simulation of the Little River Watershed basin K via the Soil and Water Assessment Tool (SWAT) and observed data are used to determine the mechanism that is strengthening the corre lation of the spectr al ENSO signal with observed stream flow and load ing time series compared to observed precipitation. Using a calibrated and vali dated SWAT model of LRW-K, the detailed surface and 167

PAGE 168

groundwater hydrology was continuously simulated on a daily basis from 1979 to 2004, and aggregated into monthly average va lues for use in analysis. The clear increase in the power of the inter-annual climate signal was demonstrated by shared patterns in water budget and exceedance curves, as well as qualitatively and quantitatively in high ENSO related energy in the wavelet spectral analyses and the 95% significant periodiciti es against a red noise background for each variable both analyzed alone and with the NINO 3.4 SST index. We found that in the Little River Watershed basin K, in Tifton, Ge orgia, the power of the climatological teleconnection to the El-Nio/ Southern Oscillation was incr eased in both the observed and simulated stream flow th rough the mechanisms of groundw ater and interflow, as they are confined by a geological layer, the Hawthorn Formation. The Hawthorne formation in the Little River Watershed forms a nearly perfect aquaclude, so although infiltration is very fast, soil saturation can be reached relatively quickly while recharge into the deep aquifer is extremely slow [Rawls et al. 1976; Sheridan 1997]. This results in the increased si gnificance of groundwater and interflow contributions to both soluble nut rient movement into outlet stream flow, and the stream flow itself. In the LRW, perhaps much mo re than other watersheds without a ubiquitous confining layer and lack of topography, the power of the El-Nio/Southern Oscillation 37 year signal that was found in precipitatio n is increased in the observed and simulated stream flow signal. The ent ire explanation of the str engthened ENSO power may have as much to do with the groundwater, interflo w and Hawthorn Formation, as with the land use of the watershed itself. In turn, this raises the question of what the actual physical properties of the watershed are that lead to this potentially un ique increased ENSO 168

PAGE 169

signal power in the simulated and observed su rface and sub-surface flow regimes of the LRW-K. By gathering physical pr operties of this relatively small coastal plain watershed such as the total area and depth to the surfic ial aquifer, transmissivity and porosity in the sandy soils, residence times of groundw ater and surface water, and hydraulic conductivities, it may be possible to create an analytic function describing exactly how the precipitation based integr ation of stream flow through time and space in the entire hydrological system occurs. Calibration and validation of the SWAT model for LRW-K generally showed that while the model predicted the hydrologic trends in the observed data, there were inconsistencies in the magnitude and duratio n of simulated daily stream flows. Generally, in La Nia-like conditions, flow was over-predicted in summer months when zero-flow conditions were observed, as was persistence of interflow [ Bosch et al. 2004; Feyereisen, Strickland, et al. 2007]. This discrepancy in st ream flow may help explain why the ENSO related SWAT simulated str eam flow wavelet spectrum (Figure 4-7b) was not as strong as that of the observed data (Figure 43b). El Nio winters in the LRW tend to be cooler and wetter, resulting in near saturation of the shallow aquifer and vadose zone, and an increased stream flow si gnal from precipitation events. SWAT simulated storm flow rose and fell too r apidly in response to winter storms, while interflow did not recede rapidly enough rela tive to observed events. Despite these discrepancies, however, the univariate and cro ss-wavelet spectral ENSO signals were still visible and significant in the groundwater and interflow SWAT simulations, which suggest that we could expect to see an even more powerful signal if the flow data were 169

PAGE 170

simulated more precisely with a fully spatiall y-distributed model or there was continuous detailed observed groundwater data. Relative sensitivity analysis and calibrati on of SWAT simulated LRW-K hydrology previously showed that surface response param eters were significantly more sensitive than those pertaining to sub-surface hydrology [Feyereisen, Strickland, et al. 2007]. The most sensitive parameter to overall wate r yield, storm flow, groundwater flow and interflow was the curve number for agricultural land, partly due to the increased surface runoff from even average prec ipitation events on cropped land. In relationship to the future of the Little River Watershed, this implies that an increase in land area used for agricultural purposes may serve to strength en the ENSO signal in stream flow even more than it currently is, while a signifi cant decrease in cropped area could cause the teleconnection and predictive relationship between ENSO and flow, and consequently other variables such as pollutant loads, to weaken. As such, this ENSO teleconnection and its relationship with agricultural area co uld hold for other Coastal Plain watersheds as well. The specific mechanism by which the ENSO signal power is increased via ground water hydrology and a confining layer in the Little River Watershed is by no means the only way in which a watershed could increa se a climate signal delivered primarily through precipitation [ Cayan et al. 1999; Maurer et al. 2006]. However, the spatial integration and reduction of noise in precipitation that the str eam flow signal effectively accomplishes is most likely a primary met hod of signal power increase in different watersheds, as research on river flows fr om around the world have shown significant spectral signals corresponding to multiple climate indices [ Labat 2008]. This non170

PAGE 171

intuitive relationship between ENSO signal strength and stream flow could prove to be helpful for making seasonal climate predicti ons in a geographic area with a weaker than desirable ENSO/precipitation signal, as a predictive relationship could be found between stream-flow or other proxy hydro-climatic vari ables. The exploration of additional non-precipitation hydrologic vari ables as they correlate to ENSO could expand how climate data could be used for more practical hydrologic prediction and for municipal water supply management, as well. The use of seasonal forecasts based on ENSO teleconnection with both surface and ground water flows and pollutant and nutrient loads could also more immediately benefit water resource managers or farmers, as decisions on irrigation, allocation, and runoff management would be based on actual climate information, and would not be predictions made so far in the future as to render them virtually unusable in normal management plans. In Chapter 5, this question is explored using a time series model to predict nitrate-N loads in the Little River Watershed based on the NINO 3.4 s ea surface temperature index. 171

PAGE 172

Table 4-1. Reference 10, 50, and 90% probabilities from all M-ENSO separated exceedance curves. This table indicate s the probability that during a given ENSO phase, a certain variable will be above the number indicated. The largest differences between ENSO phases are seen in the 10% exceedance probabilities for stream flow and groundwat er flow; compared to neutral, EN stream flow is 141% great er, while LN is only 54% of neutral flow, while groundwater flow is 143% greater in EN and 44% of neutral groundwater flow in LN. Exceedance Probability Climatology Neut ral El Nio La Nia Precipitation 10% 188.4 190.0 203.7 141.2 (mm) 50% 91.4 89.6 106.9 68.3 90% 29.5 29.9 19.1 30.2 Stream Flow 10% 89.7 89.7 126.4 48.1 (mm) 50% 8.9 8.1 11.0 3.6 90% 0.8 0.8 0.5 1.1 Surface 10% 16.6 16.6 20.3 14.1 Runoff (mm) 50% 2.1 2.2 3.0 0.8 90% 0.1 0.2 0.1 0.1 Groundwater 10% 68.9 69.3 99.2 30.6 Flow (mm) 50% 5.0 5.0 6.4 1.6 90% 0.2 0.2 0.1 0.3 Interflow 10% 6.0 6.0 6.2 4.3 (mm) 50% 2.4 3.7 2.6 1.7 90% 0.6 1.9 0.5 1.0 Percolation 10% 87.9 87.9 101.2 46.8 (mm) 50% 5.9 6.6 6.1 3.3 90% 0.2 0.2 0.1 0.3 ET 10% 126.7 126.7 133.2 123.2 (mm) 50% 61.5 61.5 66.4 39.6 90% 20.5 21.5 19.1 21.9 172

PAGE 173

Figure 4-1. Observed land-use from 19752003 at 10 km resolution from LandSat images in Little River Watershed basin K [ Bosch et al. 2006]. ET Figure 4-2. Cartoon cross-section of relev ant SWAT simulated surface and sub-surface hydrological processes in an HRU of the LRW. SURQ is non-infiltrated surface runoff to the open stream channel ET is potential evapotranspiration, PERC is percolation past the root zone soil layers into the vadose zone, LATQ is interflow from the root z one to the open stream channel, and GWQ is baseflow from the vadose zone and shallow aquifer to the open stream channel. The Hawthorn Formation is a confining layer above the deep aquifer, which receives negligible recharge from the area. 173

PAGE 174

(a) (b) Figure 4-3. Significant Wavelet Power Spec tra are shown within t he cone-of-influence, which depends on time series length and degrees of freedom. Figures are color-mapped to indicate high wavelet power with reds and oranges, and low powers in blue and white. The Global Wave let Spectrum (GWS) at the right of each figure shows power integrated over all scales and times. The 95% confidence limit is shown on the GWS (dashed blue line), the periodicities above which show significance. (a) Observed precipitation (mm) and (b) observed stream flow (m3). [ Keener et al. 2010] 174

PAGE 175

Figure 4-4. Average annual water yield (mm) of selected SWAT hydrologic variables in different M-ENSO phases, including: obs erved precipitation (Precip_Obs) and SWAT simulated surface runoff (S UR_Q), interflow (LAT_Q), total groundwater flow (GW_Q), percola tion past the root zone (PERC), evapotranspiration (ET), and total flow (Q). Figure 4-5. Average annual water yield as percentage of observed annual average precipitation of selected SWAT hydr ologic variables in different M-ENSO phases, including: SWAT simulated su rface runoff (SUR_Q), interflow (LAT_Q), total groundwater flow (GW_ Q), percolation past the root zone (PERC), evapotranspiration (ET), and total flow (Q). 175

PAGE 176

Figure 4-6. Probability of exceedance curv es in different M-ENSO phase and observed climatological record for monthly (a) Observed Precipitation, and SWAT simulated (b) stream flow (c) groundwat er flow (GWQ) (d) interflow (LATQ) (e) percolation past the root zone (PERC), (f) evapotranspiration (ET), and (g) surface runoff (SURQ). An exceedance curve is a backwards cumulative probability density curve, showing w hat percentage of area under the probability density curve lies to the right of the value of the variable on the xaxis. For example, a tendency toward drier than normal precipitation conditions in La Nia would be indicated if the red curve lies to the left of the black curve. Exceedance curves s hould not be read as forecasts. 176

PAGE 177

Figure 4-6. Continued. 177

PAGE 178

(b) (c) Figure 4-7. Significant Wavelet Power Spec tra are shown within t he cone-of-influence, which depends on time series length and degrees of freedom. Figures are color-mapped to indicate high wavelet power with reds and oranges, and low powers in blue and white. The Global Wave let Spectrum (GWS) at the right of each figure shows power integrated over all scales and times. The 95% confidence limit is shown on the GWS (dashed blue line), the periodicities above which show significance. (a) NINO 3.4 SST index (C) (b) SWAT Simulated Stream Flow (mm) (c) Groundw ater Flow (mm) (d ) Interflow (mm) (e) Percolation (mm) (f) Evapotranspiration (mm) 178

PAGE 179

(d) (e) (f) Figure 4-7. Continued. 179

PAGE 180

1980 1985 1990 1995 2000 0.25 0.5 1 2 4 8 1/8 1/4 1/2 1 2 4 8 1980 1985 1990 1995 2000 0.25 0.5 1 2 4 8 1/ 8 1/ 4 1/ 2 1 2 4 8 YearPeriod 1980 1985 1990 1995 2000 0.25 0.5 1 2 4 8 (c) (d) (c) (b) Period 1980 1985 1990 1995 2000 0.25 0.5 1 2 4 8 (a) (b) (d) Figure 4-8. Cross Wavelet Spectrum betw een monthly NINO 3.4 SST (C) index and (a) groundwater flow (mm) (b) Inte rflow (mm) (c) Percolation (mm) (d) Evapotranspiration (mm). Black figure outlin es indicate areas significant to 95% confidence, while arrows represent variables phase relationship. Arrows pointing clockwise indicate in-phase behavior, while counter clockwise arrows indicate anti-phase behavior. 180

PAGE 181

1980 1985 1990 1995 2000 0.25 0.5 1 2 4 8 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1980 1985 1990 1995 2000 0.25 0.5 1 2 4 8 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Period 1980 1985 1990 1995 2000 0.25 0.5 1 2 4 8 (c) Period 1980 1985 1990 1995 2000 0.25 0.5 1 2 4 8 (a) (b) (d) Figure 4-9. Wavelet Coherence Transform be tween monthly NINO 3. 4 SST (C) index and (a) groundwater flow (mm) (b) Inte rflow (mm) (c) Percolation (mm) (d) Evapotranspiration (mm). Black figure outlin es indicate areas significant to 95% confidence, while arrows represent variables phase relationship. Arrows pointing clockwise indicate in-phase behavior, while counter clockwise arrows indicate anti-phase behavior. 181

PAGE 182

CHAPTER 5 AN ENSO BASED MULTIVARIATE TI ME-SERIES MODEL FOR FLOWS AND NITROGEN LOADS IN THE LITTLE RIVER WATERSHED Introduction In socio-economic and political affairs of the late 20th and 21st centuries, the concept of using Global Cir culation Models (GCMs) to ex trapolate the Earths changing climate in the next 50 to 200 years has been a highly controversial issue. While still thoroughly discussed, there is not nearly as mu ch debate in the scientific community as in the public arena about the core issue: that the Earths climate is changing and we as a species urgently need to address and ultimate ly adapt to these changes to ensure the well being of as many people as possible. What has become very clear, especially in the wake of recent global meetings such as the United Nations Climate Change Conference in Copenhagen (December, 2009), is that despite mostly good intentions, it is virtually impossible for the global comm unity to come together and make meaningful changes in their day-to-day policies using the predictions from these 50-200 year forecasts of global climate using model ensembles. Aside from the socio-political obstacles to using long-range GCM predictions in a practical manner, there are scientific uncerta inties as well. For example, it has been stated that using outputs fr om averaged ensembles of GCM runs to extrapolate a nonlinear system with corresponding non-linear interactions, such as Earths climate, into the future only increases uncertainty and introduces bias shifts into model outputs [ Rajagopalan et al. 2002]. The use of these outputs as boundary conditions and inputs to other models such as hydrological packages for making futu re simulations of hydrology changes may be very biased and unce rtain. Even the estimate of global climate sensitivity, an impor tant climatological refer ence number defined as the 182

PAGE 183

estimated change in global mean surface air temperature that would result from a doubling of atmospheric CO2 concentration, changes regular ly, with the most recent IPCC estimate ranging from 2 to 4.5C [ IPCC, 2007], a relatively uncertain range. For both political and scientific reasons, it may be more beneficial and practical to concentrate more research efforts on predicting hydrologica l consequences using seasonal and inter-annual climate variability, as we can learn from our mistakes by observing if we are correct, predict and prepare for catastrophic events in the near term, and perhaps have an easier time effecting faster policy changes. In Chapter 3, a known clim atological signal was identif ied in the 3-7 year ENSO spectral range through univariate and multivaria te wavelet analysis which was reflected in the observed monthly precipit ation, stream flow, nitrate concentration and nitrate load of the Little River Watershed (LRW) in Tifton, GA [ Keener et al. 2010]. This periodic ENSO signal power that is visible in the hydrologic variables in the LRW is representative of a climatol ogical teleconnection that is known to exist between sea surface temperatures in t he equatorial Pacific Ocean, ENS O, and the southeast United States [Ropelewski and Halpert 1986, 1987; Schmidt et al. 2001]. Chapter 3 was the first study to extend the observation of the EN SO signal down the hydrologic chain to instream nutrient loads. It was found that the strongest wavele t power was seen in ENSO correlations with the monthly stream flow and monthly nitr ate load. Both of these relationships were stronger than that betw een ENSO and monthly pr ecipitation, which was shown in Chapter 4 to be because of spatia l integration of the stream flow signal as opposed to the chaotic precipitation time series, and because of the Hawthorn Formation, a nearly perfect aquaclude that forms a confining layer under LRW-K 183

PAGE 184

[ Feyereisen, Strickland, et al. 2007; Stringfield 1966]. In addition to representing a smoothed function of precipitat ion, stream flow in the Little River Watershed has a strengthened correlation with the ENSO signal (Chapter 4). Groundwater flow in the LRW is responsible for up to 80% of the total stream flow [Sheridan 1997], as well as comprising the main route for mo vement of soluble nutrients [Lowrance et al. 1984]. The ENSO signal power in stream flow is strengthened by the significance of the role of groundwater, and additionally increased due to the presence of the Hawthorn confining layer, restricting flow into the deep aqui fer and increasing the amount of interflow represented in the stream flow record (C hapter 4). As nutrient concentration is dependent on both natural climat ological phenomena as well as anthropological effects such as agriculture and urbanization, a less significant ENSO correlation was seen in its wavelet power spectrum relative to stream flow. For these reasons, stream flow and nutrient load were targeted as the main time series of predictive interest. Wavelet analysis in both its univariate and mu ltivariate forms has been accused of being comprised of interesting, colorful pi ctures and little substantial, quantitative data [ Grinsted et al. 2004]. In this research, we have bui lt upon the identified significant relationships in the hydrologic variables of the LRW and then isolated, extracted, and rebuilt the wavelet oscillatory signals for use in a local monthly predictive time series stream flow and nutrient load model. Time se ries models are used in municipal water supply management such as by Tampa Bay Water [ Asefa and Adams 2007], and forecasting of flows and precip itation in different location s, although they are not widely published in academic journals. This may be bec ause of the relative subjectivity of determining the appropriate parameterization fo r time series models, or simply because 184

PAGE 185

they are not able to answer larger questions having to do with land-use changes or long term simulation. In our case, however, a time series model that simply provides shortterm flow and nitrogen load fore casts that can be easily updat ed with additional data is a useful option for a decision support tool. This study investigates the periodicity and variability in str eam flow, nitrate and total nitrogen loading as a non-stationary pr ocess, by quantifying and comparing the effects of spectrally significant component s on the predictabilit y of flows and water quality in the LRW. Our hypothesis is that the wavelet time series models will outperform standard time series models in fore casting one to three month nutrient loads in the LRW, while SWAT will be more accura te in simulating hydrology on an annual basis. The wavelet decomposition models will also contain higher-orders of parameters than the time-domain models. However, for this research we are examining these models for future pragmatic forecasting, making minimizing the number of parameters less crucial. Chapter 5 evaluates the effectiveness of a series of wavelet (W-VAR) and nonwavelet based (VAR and VARX) vector time se ries models with different types of inputs in predicting flows, total nitrogen, and nitrate loads in the LRW basin K, including possible inputs of observed NO3 load, wavelet reconstructed components (RC) of NO3 load, observed stream flow, wavelet RC of stream flow, observed and SWAT simulated TN load, observed and SWAT simulated wa velet RC of TN load, SWAT simulated wavelet RC of stream flow and an exogenous variable in the non-wavelet models of either observed or forecasted NINO 3.4 SST All models were trained using data from 1968-2002, and validated using predicted and observed data from 2003. 185

PAGE 186

The resulting slew of time series models gives insight into the different advantages and disadvantages of different types of time series models, and into using spectral inputs as opposed to those in the time domain. In theory, the time series models that are based on the wavelet extracted significant oscillations will retain both the marginal distributions of all involved variables, as well as the spectral signature including the amplitude and frequency over time of the variables. These spectral time series models may excel at reproducing quasi-periodi c long memory behavior and non-linear dynamics that give rise to both persistent regimes of behavior and stochastic regime transitions without a priori modeling specifications [ Kwon et al. 2007], such as the various ocean-atmosphere climate indices af fecting the southeast Coastal Plain. While SWAT and time series models theoretically se rve very different purposes, in conjunction their abilities are magnified. A time series model is not meant to and cannot simulate spatially distributed, long-term and theoretical future condi tions based on changing land uses, a use at which SWAT excels. T he W-VAR models can provide shorter term monthly and seasonal predictions of high risk nutrient loads and high or low flows with very minimal input once the initial model has been set up. The forecasts of one to three month outlooks of stream flows and nutrient loads can provide general advice for stakeholders and natural resource managers that are concerned wit h or trying to actively manage and reduce their loading risk in certain seasons and months as part of a BMP or even an adaptive management plan. 186

PAGE 187

Data and Methods Field Site and Hydro-Climate Data The same field site, the Little River Wa tershed (LRW), basin K, is used as is described in Chapter 3, Data and Methods, se ction Field Site: Little River Watershed. The same precipitation, stream flow, and nut rient data from basin K are also used as described in Chapter 3, Data and Methods, section Little River Watershed Data, for obtaining spectral reconstructed components. All data delineating continuous observed measures of the El-Nio/Southern Oscillation are from the observed NINO 3.4 sea surface temperature index, as described in Chapter 3, Data and Methods section Little River Watershed Data. For evaluation of time series models, in some cases a year of hindcast values of the NINO 3.4 SST index from 2003 were used for comparison as an exogenous variable input. These past forecasts of anticipated NINO 3.4 were obtained from the IRI/LDEO Data Library (http://iri dl.ldeo.columbia.edu/SOURCES /.IRI/.FD/.SSTA_FCST/.ASST/ .version01/.sst/L/data.html), and are from the NCEP coupl ed ENSO forecast model; they are presented as the mean of the NOAA NCEP coupl ed ENSO Climate Forecast System (CFS) [ Saha et al. 2006], LDEO model [ Cane et al. 1986; Zebiak and Cane 1987], and CPC constructed analogue (CA) [ van den Dool and Barnston 1995; van den Dool et al. 2003] models, meant to give the best ensemble estimate of expected NINO 3.4 SST behavior at the time. SWAT Nutrient Calibration and CP Implementation Using the results of the hydrologic parameter SWAT calibration as described in Chapter 4, Data and Methods, section SWAT Model Hydrologic and Nu trient Calibration and Validation, a subsequent calibration and validation of nutri ent parameters and 187

PAGE 188

effects of BMP implementation and conservation practices (C P) on agricultural areas in basin K was performed [ Cho et al. 2010]. Again, AVSWAT-X, SWAT 2005 with a GIS interface, was used to simulate the hydrology of the Little River Watershed basin K, and calibrated for water quality and conservati on practice scenarios. Based on the SWAT settings, 26 sub-basins and 452 hydrologic response units (HRUs) were created for LRW-K. As a spatially semi-distributed m odel, SWAT individually models each HRU response, aggregates them at the sub-basin level, and then routes each sub-basin to the appropriate stream reach within the watershed. The water quality and CP were calibrated for eight years from 1979 to 1986, and validated from 1987 to 2004 [ Cho et al. 2009, 2010]. Calibration was manually performed as outlined in the SWAT manual [ Neitsch et al. 2002]. While the hydrological parameters remained as they had under the pr evious analysis, additional parameters affecting biomass of crops, upland sediment erosion, and in-stream processes were calibrated. Crop rotations that were calibrat ed and validated were three year rotations of corn-corn-peanut and cotton-cotton-peanut, and a six year rotation of corn-peanut-cornpeanut-cotton-peanut, while non-cropped area m anagement was kept constant. An observed phenomenon in the LRW is the si gnificant removal and transformation of nitrogen from nitrate-N to organic-N in the riparian forested buffers [ Lowrance 1992; Lowrance et al. 1984]. Since SWAT is unable to simulate dynamic nutrient transformations within riparian zones [ Li et al. 2004], only total nitrogen (TN) and phosphorus (TP) at the watershed outlet were calibrated [ Cho et al. 2009]. Previous detailed studies of nutrient processes within the LRW have shown that overall, approximately 40% of annual natural and anthropological TN inputs are 188

PAGE 189

removed during harvest, while approximately 57% either remain in the watershed or are lost in an unspecified way [ Lowrance et al. 1985]. In other basins of the LRW, 94% of TN was observed to be transported from the uplands to the stream channel through sub-surface flow, while 67% of groundwater TN and 37% of TN in surface runoff was removed through the riparian filters [ Lowrance et al. 1984, 1983; Lowrance and Sheridan 2005]. Since there is no option in SWAT to calibrate nitrogen reduction separately within surface and sub-surface ri parian regimes, all nitrogen reduction was simulated through adjusting the ri parian filter width (FILTERW) and the half-life of nitrate in the shallow aquifer (HLIFE_NGW) variables [ Cho et al. 2009]. Within the simulation, nutrient processes were calibrated such that simulated in-stream TN reduction at the outlet matched observed values for basin K of 3.9% of TN input, or 3.34 kg/ha of TN load [ Cho et al. 2009; Lowrance et al. 1985]. Percent error (PE) was used to quantit atively compare obse rved and simulated total runoff and pollutant load for the entire simulation period. Monthly RMSEobservations standard deviation ratio (RSR) [ Moriasi et al. 2007] and Nash-Sutcliffe Efficiency Index (NSE) [ Nash and Sutcliffe 1970] were used to calculate error statistics and a monthly correlation statistic [ Cho et al. 2009]. Simulation performance is considered satisfactory if m onthly NSE > 0.5, and monthly RSR 0.7, which Cho et al. achieved, with a PE of 14.4%, RSR and NSE of 0.62 for TN load (Table 5-1). From 1985 to 2003, approximately 11% of the total land area in the LRW had undergone a conservation practice of some kind [ Sullivan and Batten 2007]. Of those implemented conservation practices, the most common were nutrient management (13.1%), pest management (12.9%), and grassed waterways (9.6%) [ Sullivan and 189

PAGE 190

Batten 2007]. The entire SWAT simulated stream flow (Figure 5-1) and TN load (Figure 5-2) from 1979 to 2004 was converted to m onthly average values and can be seen as compared to observed flows and TN loads. For the purposes of this research, daily SWAT simulations for 2003 were grouped into monthly values to be compared to the forecasts made by the vari ous time series models. It should be noted that the simulations produced by SWAT use observed weather data, and are not forecasts. By comparing the time series models with the SW AT simulated data, we are able to see how well a simple time series model can fo recast flows and loads in the short-term, and if looked at over several years of forecast and simulation, could identify years in which the SWAT simulations are not capturing climate variability adequately. Wavelet Reconstruction Every vector in a vector space can be wri tten as a linear combination of the basis vectors in that vector space, specifically by multiplying t he vectors by some constants, and then taking the summation of the products. The analysis of the signal involves the estimation of these constants (Fourier coeffi cients or wavelet coefficients for example). Reconstruction of the original signal amounts to extracting the relevant coefficients and computing the linear combination equation. The wavelet transform is analogous to a band-pass filter with a response function corresponding to the original mother Morlet wavelet function used (Equation 3-1). Since the mother wavelet is known, it is possible to reconstruct the entire time series usi ng deconvolution or t he inverse filter [ Torrence and Compo 1998]. In the case of the continuous wavelet transform used in this analysis, the original time series, Xn, can be reconstructed using a different synthesizing wavelet, the easiest of which is a delta ( ) function [ Farge 1992]. The 190

PAGE 191

reconstructed time series, Rjn, is then the sum of the real part of the wavelet transform over all scales of interest, a (Equation 5-1): J j j jn tj jna aX C R0 5.0 0 5.0)( )0( (5-1) The factor 0(0) (Morlet = -0.25) removes the energy scaling, aj 0.5 converts the wavelet transform to energy density, and C is a constant from the re constructed delta function from the particular mother wavelet used (Morlet = 0.776), and can be looked up in appropriate tables [ Farge 1992; Torrence and Compo 1998]. For this study, the 95% significant rec onstructed components as identified from the Global Wavelet Power (GWP) were extract ed from the wavelet transforms of the observed data described in Chapter 3 in t he Little River Watershed, and summed into a single time series representing inter-annual o scillatory variability. The scales for which the GWP spectrum is higher t han the 95% red noise significance level were selected for time series reconstruction. This process was done for monthly precipitation, stream flow, SWAT simulated total nitro gen (TN) load and observed nitrate (NO3) load. Nitrate concentration was not considered as a relev ant reconstructed time series in this analysis, as it did not meet relevant significance criteria in Chapter 3. Wavelet Time Series Model: VARX(p,s) In many research studies involving geophy sical data, there are observations from several different variables that relate to the overall system being a nalyzed. In our case, we have identified several hydrologic and c limate based time series in Chapters 2 through 4 that have a significant relationship with the nitrate nutrient loads we want to forecast, primarily, stream flow and the NINO 3.4 SST index [ Keener et al. 2010]. 191

PAGE 192

Expanding on the general in troduction to time series anal ysis in Chapter 1, Literature Review section Time Series Models, this se ction will more specifically discuss vector time series models. The vector time series model is a natural extension of a univariate model to multivariate dynamic time seri es, while an exogenous variable outside the system of interest ( X ) can be added to increase forecast ing power. For example, if a single variable is denoted by Yt = ( y1, y2yt) to form an autoregressive AR( p ) model of lag p then by extension, a vector model with Yt = (y1t, y2tynt) denotes a ( n x 1) dimensional vector of variables in a vector autoregressive (VARX( p,s )) model of lag p with exogenous variable X : Ttfrom X XY YYtsts tptp tt...1 ... ...11 11 (5-2) In which i are ( n x n ) coefficient matrices, Xt is a ( m x n ) matrix of exogenous variables, s is a parameter matrix es timated via regression, and t is a zero-mean white noise vector process. A VAR(p ) model without the exogenous variable simply lacks the corresponding matrix of parameters that include X The following assumptions are made: E( t) = 0, E( t, t ) = a positive-definite, m x m matrix of the residual covariances, and E( t, t ) = 0 for t not equal to s. The VAR and VARX processes are st ationary, i.e. the roots of det[ (x) ] = 0 are outside the unit circle. Explanatory (independent) variable(s) xt are not correlated with the residuals t. In abbreviated summation notation, the VARX( p,s ) model (Equation 53) can be written as: tit p i i s i itity xY 1 0 (5-3) 192

PAGE 193

Optimal lag selection for each model is done by minimizing the Akaike information criterion (AIC) (Equation 5-4) and/or the Baye sian information criterion (BIC) (Equation 5-5). Each method of m odel selection has advantages and disadvantages. Typically, the AIC will choose the best model even if the number of parameters is unreasonably large, which may result in over-fitting. The BIC incorporates a penalty term for the number of parameter s in the model, often resulting in optimal fits with fewer terms than the AIC, but less overall predictive accuracy. Theoretically, the AI C for vector models gives a quantifiable estimate of the relati ve amount of information lost when balancing a statistical models bias versus variance, in which p is the number of parameters in a given mdimensional model of n time points, and is the residual covariance matrix without degree-of-freedom co rrections from the VAR( p) model, representing the estimated error variance t. p pn pm p n likelihood n pAIC ln 2 )( 2 ) ln(max 2 )(2 (5-4) The BIC penalty for additiona l parameters is in the sec ond term (Equation 5-5), making BIC an increasing function of both p and the variance of the error term. n t tt p p en where n npm n n p pBIC1 '1 2 2 )ln( detln )ln( ln()( (5-5) 193

PAGE 194

In this research, the vector comp onents include observed raw and wavelet reconstructed LRW monthly stream flow, and monthly nitrate-N and total-N load time series, while both observed and hindcast NINO 3.4 SST index values are used as the exogenous variable. To achieve stationarit y and variance stabilization, the raw stream flow and TN load data were transformed via first-order differencing (i.e., yit = yit yit-1), while the NO3 load was log transformed and then fi rst-order differenced. Wavelet RC and NINO 3.4 exogenous variable time series were not transformed. AIC was used to select the best lag for VAR( p ) and VARX( p,s ) models, while BIC was used for WVAR( p ) models to keep the maximum num ber of parameters in check. Forecasting and Uncertainty Forecasting is a main objective of time series analysis, and forecasting from a vector model is very similar in theory to fo recasting from a univariate model. The optimal minimum mean squared error (MSE) h -step-ahead forecast of Yt+h at time t is (Equation 5-4) when the parameters have been estimated as is found recursively with a forecast error covariance matrix (Equation 5-5), j j ht tpht p thtthtY Y Y ... 1 (5-4) o j jjoo h o oowhere h1 1 0 )( (5-5) The hstep forecast error (Equation 5-6) now incl udes the part of the forecast error due to estimating the parameters of the VAR or VARX model. 194

PAGE 195

1 1 1 0 | p j jjoo tht htoht h s o tht htmatricesthewhere YY YY (5-6) In this study, confidence intervals for the forecasts are provided at the 95% level. The confidence level is inferred from the empirica l forecast error cova riance matrix (Equation 5-5). Cones of forecast uncertainty are also typically generated by adding multiples of the prediction standard error estimates at each time step, illustrating the growing uncertainty the farther one get s from the last observed data point. This method is not used to illustrate the uncertainty in this research. Tercile Analysis After generating one to twelve month fore casts for nitrate load, TN load, and stream flow anomalies in the year 2003, forecast accuracy was tallied using terciles. In this case, monthly forecasts made us ing different VAR and VARX models with exogenous input of the NINO 3.4 index and monthly SWAT simulations are counted if they are in the correct observ ed tercile of the observed time series variable in question. Counts of the correct tercile pr ediction of each model were m ade at one, three, six, and twelve month intervals to observe how each models accuracy changed over time. Results Reconstructed components (RC) were extracted from the univariate wavelet transform of normalized observed monthl y precipitation, stream flow, NO3 load and SWAT simulated TN load and stream flow The reconstructed components in the spectrum for each variable that were abov e the 95% GWS significance level were isolated and summed into a single component representing t he wavelet-extracted inter195

PAGE 196

annual climate variability within a given time series (Figure 5-3). The number of significant RC for each variable are seen in Table 5-2, as well as the percentage of variance that the extracted RC comprise relative to the original anomaly time series and the optimal lag chosen by AIC and BIC for each time series model type. Observed stream flow and nutrient load ano malies were chosen as vector inputs, as they exhibit the most significant RCs as well as t he largest percentage of representative variance and cross-correlative power with the NINO 3.4 index [ Keener et al. 2010], while observed and hindcast NINO 3.4 SSTs were used as the exogenous input variable. From both the conclusions drawn in Chapter 3, the relevance of the significant RC calculated here, and time se ries modeling results that are not shown with poor predictive power in the vector m odels, observed precipitation and NO3 concentration are not included in the subsequent time series mode ling section. As such, the vector inputs to the model are observed monthly NO3 or TN load, stream flow, and SST. VAR, VARX, and W-VAR Time Series Models Two main classes of time series m odels predicting monthly N-nutrient load anomalies in the LRW basin K are compared and contrasted in this study: VAR models with and without an exogenous input (VARX) of the NINO 3.4 index, and wavelet-VAR (W-VAR). Theoretically, t he W-VAR models should not require the addition of the exogenous variable of NINO 3.4 SST to perform better, as they inherently encompass inter-annual climate-based variability as m anifested in the significant reconstructed components. More interestingly, the WVAR models do not only include NINO 3.4 based climate signals, as all significant re constructed components were included in the analysis, not only those in the 3-7 year o scillatory period corresponding to ENSO. The reconstructed W-VAR model could encom pass inter-annual oscillatory climate 196

PAGE 197

information from any of a number of periodi c ocean-atmospheric indices that affect different aspects of the southeast United States, some of which have been initially identified as affecting cott on yields more than ENSO [ Pathak 2010], such as the Tropical Northern Hemi sphere (TNH) index [ Mo and Livezey 1986] and the North Pacific Pattern (NPP). This gi ves the W-VAR models a distin ct advantage in the amount of climate-based information they are abl e to incorporate without the addition of exogenous variables which can introduce addi tional model or forecast uncertainty. The optimal number of lags chosen vi a AIC and BIC was one month for all VAR and VARX models using different exogenou s input types, five for W-VAR NO3 load and stream flow, and seven for W-VAR TN load and stream flow (Table 5-2). By necessity, the wavelet based models contain significantly more parameters th an those constructed in the time domain, as they are expect ed to capture the inter-annual oscillatory character in the respective time seri es data, given a re presentative group of reconstructed components. While the number of parameters selected in the W-VAR models is not excessive, it does intr oduce another source of potential model uncertainty. Figures 5-4 through 5-11 show the differ ent time series model types and the monthly observed and predicted values of nut rient load and stream flow for 2003. With any model used for forecasting, it is expect ed that prediction accuracy decreases as the model gets further from the la st observed data point. Time series models are generally used to forecast only a few steps in advanc e at a given time, and are routinely updated with new data to generate new forecasts through time. In these time series models, 12 months of forecasts were made for exploratory analysis, however, practically, only 197

PAGE 198

forecasts at the one to three month level woul d likely be used for their potential decision making abilities. It should also be noted that using the monthly M-ENSO categorical index, January and February of 2003 were classified as El Nio months, while the rest of the calendar year was N eutral. This was preceded by nine months of El Nio conditions from April to December of 2002. These ENSO categorizations could have effects on the models prediction abilities, as transition regimes from wetter to drier seasons can have effects on watershed st orage potential and saturation levels, and further forecasts and evaluations need to be done on years representing different climate regimes. Upon visual inspection of the VAR(1) models of NO3 load and stream flow (Figure 5-4) and TN load and stream flow (Figure 5-8), it is clear that these models are not adequately simulating the dynamics of either vari able, as the entire y ear of predictions remains nearly static around the mean with wi de confidence intervals. The VAR models also show the fewest correct predictions of load and flow terciles, especially in the 1-3 month forecasts (Table 5-3), and some of the highest RMSE estimates, especially for NO3 and TN loads (Table 5-4). Methods to impr ove the fit of using only the raw time series for forecasting could include incorpor ating a seasonal dimension to the model to add dynamics that the current model is missi ng. In this research, however, we are interested in looking at the ways in which a wavelet model compares with a time-domain model, not in methods of produc ing a best-fit time series model. For these reasons, these simple VAR models are used without any climate information incorporated as a baseline to compare with our models that use climate data either directly or implicitly. 198

PAGE 199

The next incarnation of models are VARX( 1) forecasts using observed NINO 3.4 sea surface temperatures as an exogenous input. Theoretic ally, these models create a more dynamic simulation by directly inco rporating climate data. Indeed, visual inspection of the NO3 load and flow (Figure 5-5) and TN load and flow (Figure 5-9) shows a more dynamic forecast, which does much better during the first three months of forecasting in 2003. The VARX models using observed SSTs show reduced RMSE values for all variables as compared to the straight VAR model, es pecially in months 13. The number of correctly pr edicted load and flow terciles also increases from the VAR model, again concentrated in months one th rough three, with 2/3 months correctly forecast for the TN load and flow model, and 1/3 and 2/3 months respectively for NO3 load and flow (Table 5-4). To recreate a more realistic representat ion of how this type of model would be used in forecasting situations, we then used VARX(1) models to predict flows and loads in 2003 using predicted NINO 3.4 sea surf ace temperatures as an exogenous input. Once again, the forecasts flatten out and lose their dynamic nature for both the NO3 load model (Figure 5-6) and the TN load model (Figure 5-10), and begin to resemble the VAR models. The numbers of correctly predict ed flow and load terciles are almost the same as the numbers seen fo r the VAR model (Table 5-3), and the RMSE estimates (Table 5-4) again increase, and are even slightly greater than those seen in the VAR model for some variables. The lack of dynam ic simulation seen by using the predicted SSTs is odd, as we would expect to see the same dynamic results as the observed NINO 3.4 model, only less accurate. Similar results have been seen in regional cotton yield data [ Baigorria et al. 2008].Comparing the observed and predicted monthly NINO 199

PAGE 200

3.4 SST values for 2003 produces an R2 of 0.6. These results show that uncertainty in predicted values of SST degrade forecast s of flow and loads, and in actual management scenarios, observed SSTs for model input will not be available. Finally, the W-VAR models are shown for NO3 load and flow (Figure 5-7) and TN load and flow (Figure 5-11). It is immediat ely apparent that the W-VAR models are more dynamic and accurate in their forecasts of bot h load and flow. In the suite of exploratory time series models, the W-VAR models have the lowest RMSE estimates for NO3 load, TN load, and stream flow a ssociated with each model (Table 54). With the exception of the flow forecast associated with TN load, the W-VAR models also generate the correct tercile of flow and load for 3/3 months (Tabl e 5-3). Given these statistics and trends, the W-VAR models are seemingly doing a much better job of inco rporating climate information into the seasonal forecasting models, despite not directly using any external climate data. The immediate adv antage of this is that forecasts based on this wavelet method are not dependent on an additional variable that must be forecast, introducing another source of uncertainty and error, and at the same time this method manages to use implicit inter-annual climate data. A reason that the W-VAR models do bette r than the VARX models using observed NINO 3.4 SST data is that the spectral inform ation included in the model is not limited to ENSO relationships, but any significant frequ encies that were identified in the global wavelet spectra. This is also advantageous in the sense that even the ENSO signal can be considered a non-stationary variable, as there are arguments that anthropogenic climate forcing may in fact be changing the nature of ENSO oscillation regularity and strength [ Park and Mann 2000; Tsonis and Swanson 2009]. By not limiting the 200

PAGE 201

reconstructed components to a narrow band of pre-determined valid frequencies, we allow both different climate i ndices and a possibly changing EN SO itself to be included in the signal reconstruction. In the case of nitrate load and stream fl ow, this included five RCs, and seven from the total nitrogen load. However, the number of total significant RCs was limited by the length of record and degrees of associ ated freedom with the analyzed data. Although over 30 years of nutrient and load data were used, the cone-of-influence of the wavelet power spectra only reached half of that, about 15 years, as the maximum periodicity that was able to be analyzed in a significant way. As a result, only reconstructed components from 1-15 years were able to be included as candidates for reconstruction, and any lower-oscillation climate signals were not captured in the W-VAR forecast time series models. It should also be noted that while these models are based on relatively temporally extensive data, they are not spatially distributed, only simulating loads and flows in one placethe stream outlet of a small forested upland basin of the Little River Watershed. Bi-monthly updated W-VAR for ecasts and SWAT simulations To represent how the W-VAR models woul d be used more realistically, forecasts were made for every two months of 2003, updated with the past two months of observed data and then re-forecast for the following two months. This updating and reforecasting of the flows and nitrate loads (F igure 5-12) and total nitrogen loads (Figure 5-13) was done until December of 2003. Since SWAT was unable to simulate the NO3 loads, we cannot directly compare t he updated W-VAR model with it. However, compared to the other time series models, the bi-monthl y updates for the nitrate do much better in both the short and long te rm. After 12 months, the updated W-VAR 201

PAGE 202

model had predicted 10 nitrate load and 8 fl ow terciles correctly, while the un-updated W-VAR model only predicted 6 months correc tly for both flow and load (Table 5-3). Additionally, the RMSE values through the entire year as compared to the un-updated W-VAR model were slightly decreased (Table 5-4). When compared to SWAT simulated flow and TN load for 2003, the updated WVAR models are quite competitive for the first three months of prediction, with tallies of correctly forecast terciles equal for the first three months (Table 5-3), and RMSE values that are only slightly differ ent than the SWAT estimates for overall stream flow, and comparable with respect to TN load. In fact, the updated W-VAR models match SWAT in the number and distribution of correctly fo recast TN loads through all of 2003, with 10/12 months in the correct tercile for both models, and 6/10 correct flow W-VAR updated flows compared to 8/10 correct flow terciles for SWAT. The ability of the WVAR models to match a calibrated and validated simulation of flows and loads through SWAT shows that the use of these conceptually simpler statistical models for short and middle term management and decision making can save time and provide equally accurate information. Although SWAT replicates the entire year of flows more accurately than any of the time series suite of models, this is to be expected, given that a properly calibrated and validated physically based model with extensiv e input land use data, and topography is a model that is expected to perform well for at least a year. It should also be noted that the SWAT input weather data used to si mulate 2003 was observed. Together, the SWAT and time series models theoretically se rve very different purposes. A time series model is not meant to and cannot simula te spatially distributed, long-term and 202

PAGE 203

theoretical future conditions based on changing land uses, a use at which SWAT excels. The W-VAR models can provide shor ter term monthly and se asonal predictions of high risk nutrient loads and high or low fl ows with minimal input once the initial model has been set up. Using the tercile methodology one to three month outlooks of stream flows and nutrient loads can provide general advice for stakeholders and natural resource managers that are concerned with or trying to actively manage and reduce their loading risk in certain seasons and m onths as part of a BMP or even an adaptive management plan. Summary and Discussion Using the knowledge from Chapters 2 th rough 4, we have utilized the ENSO teleconnection to the Little River Watershed bas in K in Tifton, Geor gia, that extends from climate to a strong inter-annual oscillatory signal in the stream flow and nutrient loads in several time series models that can forecast and identify months of high loading risk. As wavelet analysis can be likened to a band-pass filter, this can be exploited to extract and rebuild the component signals at di fferent scales according to the desired specifications. We found that stream flow and nutrient loads were identified as having the most significant NINO 3.4 SST correlation and spectral signature [ Keener et al. 2010]. Although precipitation is the major driver of the hydrological cycle, it was not included as an input in these time series models. The non-inclusion of rainfall as a variable can be considered an advantage, as in the southeast, precipitation has high spatial variability even within a relatively sm all area such as basin K of the LRW. This high degree of variability and corresponding noi se in the signal can add uncertainty to an already uncertain signal, the exclusion of which gives us interesting insight into hydro-climate processes in the LRW wit hout a direct precipitation input. 203

PAGE 204

For these reasons, observed NO3 and TN load and stream flow monthly anomalies were used as inputs to a multivariate time series model. Data in the time-domain was combined with observed NINO 3.4 values input as an exogenous independent variable to create models that incor porated ENSO information in a non-spectral way. This was contrasted with wavelet time series model s that used significant reconstructed frequency components of NO3 and TN load and stream fl ow to inherently encompass ENSO climate variability without the addition of an exogenous variable. In the VARX models, it was interesting that the inclusion of an ensemble hindcast of NINO 3.4 SSTs versus the observed SSTs in most cases made the prediction accuracy worse, yet in the case of forecast nitrate loads, actually made it better. This may be an artifact resulting from possibl y erroneous 1992-1995 observed nitrate data, however, it decreases the am ount of trust one is able to put in the VARX models as effectively including climate information into the time series model framework. In almost all cases, though, forecast error was reduc ed even by using the VARX and hindcast NINO 3.4 data as compared to the straight VAR models. In general, the W-VAR models did a much better job than time-domain VAR and VARX models of reproducing and encompassing the monthly and seasonal climate variability trends in the LRW basin K, while theoretically maintaining the spectral signature of the inter-annual clim ate-based oscillations that are within each time series. The bi-monthly updated W-VAR models, more si milar to how such a forecasting system might actually be implement ed, demonstrated increasingl y accurate monthly load predictions, correctly predicti ng 10/12 load terciles for both NO3 and TN, and 6/12 and 8/12 correct flow terciles with reduced RMSE values. However, the modeling adage of 204

PAGE 205

garbage in, garbage out applies here, as pr oblems with the nutrient data could be revealed by SWAT that are ignored by statistical modeling. Indeed, it appears as if the observed nitrate loads between 1992 and 1995 ma y exhibit measurement error, as the anomalies do not make sense with the observed or SWAT simulated total nitrogen loads (Figure 5-3c, 5-3d). Additionally, it is impossible for any of th e time series models to specifically incorporate land-use change in formation in their predictions other than in the most inherent way, although SWAT simulati ons reflecting land-use changes in the last three decades in the LRW did not show any major changes or influences on the hydrology or nutrient loads. In a watershed in which there were major land-use changes in a short period of time, the W-VAR method may not perform as accurately. In all non-updated time series models, forecast accuracy declined precipitously after the first three months of prediction, as would be expec ted from this type of model. These time series models are designed to be shorter-term tools for natural resource managers and farmers, who could use them to implement BMPs more quickly and appropriately as to reduce nutrient runoff risk by managing periods of high stream flows. This method has an advantage over time co nsuming processes of continually recalibrating, adding huge amounts of weather and land-use data, and validating large and complex agro-hydrological models su ch as SWAT to make management predictions that are ne cessarily more short term. As of 2007, 16% of the land area in the entire LRW had implemented a BMP of some sort, of which 13.1% was devoted to nutrient management [ Sullivan and Batten 2007]. The implementation of these management practices were either establis hed as a component of a federally funded cost share conservation program, or were paid by an individual who was provided 205

PAGE 206

technical support free of charge. Nutr ient management and water quality BMPs implemented within the watershed include gr assed waterways, contour farming and terraces, seasonal residue management, improved grazing and animal management, and fertilizer and crop rotation optimization [ Sullivan and Batten 2007]. While some of these BMPs are implemented once (such as gr assed waterways or fencing cattle out of stream tributaries), some are also implemented seasona lly or annually, and could be positively influenced by one to three month advance knowledge of climate-based risk for high stream flows and the associated risk of flushing out high nutrient loads into the greater watershed area. These prediction tools do not address the ro ot causes of nutri ent pollution, which is application of chemical fertilizers on agric ultural land, or increasing animal production operations. In fact, the predict ion tools are based on high or low stream flows flushing out nutrients from the watershed, whic h is in turn dependent on anthropogenic agricultural practices, surf ace and groundwater residence time s, and nitrogen residence time in the LRW. In choosi ng to more carefully manage ti mes of high stream flow, we are effectively managing how mu ch nutrient pollution is let into the watershed during a period of time, not reducing t he total amounts of pollutants that are input to the system. However, by using these models to more effectively manage stream flows and their nutrient pollution, the overa ll health of the water body and surrounding ecosystem would still be positively affected. If by managing a hi gh flow season predicted using the climate based W-VAR model the amount of nutrient load is reduced, t here would be less chance of eutrophication and ecosystem damage. While the current W-VAR model is based on data from one outlet in basin K of the LR W, it is feasible that climatic trends in 206

PAGE 207

the region would hold similar patterns, and t hat warnings from one location could be extended to many. This would need to be tested in several other basins of the LRW, however. Ideally, monthly time series models of hi gh stream flows and nutrient loading risk would be used for shorter term seasonal and annual management, in conjunction with a larger hydrological package su ch as SWAT for any longer term simulation studies about more general land-use change or hydrology e ffects. This kind of analysis could easily be extended to other chemical pollutants in runoff, to create models of how other indicators of water quality and determi nants of human health are influenced and predicted by inter-annual or l onger climate variability. 207

PAGE 208

Table 5-1. Summary of model performance measures for calibration period from 1979 to 1986 on LRK total runoff, sediment yield, and total nitrogen (TN) [ Cho et al. 2009]. Monthly RSR refers to the r oot mean squared error-observations standard deviation ratio, while NSE refers to the Nash-Sutcliffe Efficiency. Evaluation Measures Runoff (mm/year) Sediment (ton/year) TN (kg/year) Observed total 342 60.0 4,975 Simulated total 354 62.2 5,691 Percent Error (%) 3.6 3.8 14.4 Monthly RSR 0.28 0.68 0.62 Monthly NSE 0.92 0.53 0.62 Table 5-2. Time series descriptions and attributes for both wavelet-VAR and VARX models for precipitation, nitrate load and concentration, stream flow, and total nitrogen load. Dashes (--) indicate that ca lculations were not done because of lack of significance of variable. Flow series with n/a do not have optimum lag indicated because the lag is shared with the associated nutrient load. Precipitation (mm) NO3 (kg) NO3 (mg/L) Stream Flow (mm) TN (kg) Duration (mo/yr) 1/74-12/02 1/74-12/02 1/74-12/ 02 1/74-12/02 1/79-12/02 Number of sig. RC 3 6 0 5 6 % RC Variance 34.5% 54.5% 20.8% 45.1% 48.8% Optimum lag W-VAR -5 -n/a 7 Optimum lag VARX -1 -n/a 1 208

PAGE 209

Table 5-3. Number of times each vector autoregressive (VAR) or SWAT model correctly predicted the observed tercile of 2003 monthly TN load, NO3 load, and corresponding stream flow, at diffe rent time steps away from t0. O-SST refers to observed NINO 3.4 values, P-SST refers to predicted hindcast NINO 3.4 values, W-VAR to wavelet based VAR, and updated W-VAR to the model updated with observed data every two mont hs. SWAT values that are not applicable (n/a) either we re not simulated (NO3), or only have one simulated value (flow). VAR VARX + O-SST VARX + P-SST W-VAR Updated W-VAR SWAT TN (kg): 1 month 0 1 1 1 1 1 3 months 0 2 1 3 3 3 6 months 0 4 1 5 6 6 12 months 2 6 3 6 10 10 Flow (mm): 1 month 0 1 0 1 1 1 3 months 0 2 0 2 2 3 6 months 1 3 1 3 4 4 12 months 4 4 4 5 6 8 NO3 (kg): 1 month 0 0 1 1 1 n/a 3 months 0 1 3 3 3 n/a 6 months 0 3 5 4 5 n/a 12 months 4 5 6 6 10 n/a Flow (mm): 1 month 0 1 0 1 1 n/a 3 months 0 2 1 3 3 n/a 6 months 1 4 2 5 5 n/a 12 months 4 5 4 6 8 n/a 209

PAGE 210

Table 5-4. Summary of forecast vector autoregressive (VAR) time series model and SWAT performance by root mean squared error (RMSE) for one, three, six, and twelve month forecasts for NO3, TN load, and stream flow. O-SST refers to observed NINO 3.4 values, P-SST re fers to predicted hindcast NINO 3.4 values, and W-VAR to wavelet based VAR models. For each predicted load model, the predicted stream flow RMSEs are listed under the loading of interest. VAR VARX + O-SST VARX + P-SST W-VAR Updated W-VAR SWAT RMSE at: NO3 (kg) 1 month 59.7 47.7 3.5 2.1 2.1 n/a 3 months 111.1 92.1 52.1 26.5 10.6 n/a 6 months 110.9 81. 2 59.4 62.9 31.7 n/a 12 months 105.8 117. 2 103.0 98.3 47.7 n/a Flow (mm) 1 month 35.9 4.5 47.5 7.8 7.8 n/a 3 months 29.7 14.5 38.1 9.7 11.0 n/a 6 months 26.2 27.6 29.0 13.1 11.9 n/a 12 months 20.4 46.5 23.0 13.8 12.2 n/a TN (kg) 1 month 654.6 199.4 442.1 186.7 186.7 12.6 3 months 499.9 415.4 451.8 167.4 138.8 165.0 6 months 411.2 330.7 402.8 188.8 130.8 132.0 12 months 349.4 380.8 344.7 248.0 138.9 162.5 Flow (mm) 1 month 33.4 4.9 37.9 11.8 11.8 0.2 3 months 28.6 32.1 31.2 12.5 14.5 10.4 6 months 25.8 30.0 26.9 10.6 18.0 8.2 12 months 20.1 37.5 21.6 15.0 19.3 10.2 210

PAGE 211

0 50 100 150 200 1979 1980 1981 1982 1983 1984 1985 1986 Monthly total streamflow (m m Observed Simulated 0 50 100 150 200 250 1987 1988 1989 1990 1991 1992 1993 1994 1995 Monthly total streamflow (m m Observed Simulated 0 50 100 150 200 1996 1997 1998 1999 2000 2001 2002 2003 2004 Monthly total streamflow (m m Observed Simulated Figure 5-1. SWAT simulated total monthl y stream flow (mm) in LRW basin K. Calibration is from 1979 to 1986, and validation is fr om 1987 to 2004. [ Cho et al. 2009] 211

PAGE 212

0 500 1000 1500 2000 2500 3000 1979 1980 1981 1982 1983 1984 1985 1986Monthly total nitrogen loads (k g Observed Simulated 0 1000 2000 3000 4000 5000 6000 1987 1988 1989 1990 1991 1992 1993 1994 1995Monthly total nitrogen loads (k g Observed Simulated 0 1000 2000 3000 4000 5000 1996 1997 1998 1999 2000 2001 2002 2003 2004Monthly total nitrogen loads (k g Observed Simulated Figure 5-2. SWAT simulated total monthly total nitrogen load (kg) in LRW basin K. Calibration is from 1979 to 1986, and validation is from 1987 to 2004. [ Cho et al. 2009] 212

PAGE 213

Figure 5-3. Original observed time series (monthly anomaly) and 95% significant wavelet reconstructed components for (a) precipitation (mm) (b) stream flow (m3) (c) nitrate load (kg) (d) total nitr ogen load (kg). Red-dashed horizontal line indicates zero-line. 213

PAGE 214

-300 -200 -100 0 100 200 300 2003.02003.22003.42003.62003.8Load (kg ) -100 -50 0 50 100 150 2003.02003.22003.42003.62003.8Monthly Flow (m m) Figure 5-4. VAR(1) time series model with 95% CI (red dashed lines) of monthly nitrate load anomaly (kg) and streamflow (mm). Models are trained using data from 1974-2003, and validated using observed 2003 monthly data. Solid black line indicates observed data, while red lines with circles indicate forecast. 214

PAGE 215

-300 -200 -100 0 100 200 300 2003.02003.22003.42003.62003.8Load (kg ) -100 -50 0 50 100 150 2003.02003.22003.42003.62003.8Flow (mm) Figure 5-5. VARX(1) + Obse rved NINO 3.4 SST time seri es model with 95% CI of monthly nitrate load anomaly (kg) and st reamflow (mm). Models are trained using data from 1974-2003, and validated using observed 2003 monthly data. Solid black line indicates observed data, while red lines with circles indicate forecast. 215

PAGE 216

-300 -200 -100 0 100 200 300 2003.02003.22003.42003.62003.8Load (kg ) -100 -50 0 50 100 150 2003.02003.22003.42003.62003.8Flow (mm ) Figure 5-6. VARX(1) + Predi cted NINO 3.4 SST time series model with 95% CI of monthly nitrate load anomaly (kg) and st reamflow (mm). Models are trained using data from 1974-2003, and validated using observed 2003 monthly data. Solid black line indicates observed data, while red lines with circles indicate forecast. 216

PAGE 217

-300 -200 -100 0 100 200 300 2003.02003.12003.22003.32003.42003.52003.62003.72003.82003.9Load (kg ) -100 -50 0 50 100 150 2003.02003.22003.42003.62003.8Monthly Flow (mm ) Figure 5-7. W-VAR(5) time series model with 95% CI of monthly nitrate load anomaly (kg) and streamflow (mm). Models ar e trained using data from 1974-2003, and validated using observed 2003 monthl y data. Solid black line indicates observed data, while red lines wi th circles indicate forecast. 217

PAGE 218

-1500 -1000 -500 0 500 1000 1500 2003.02003.12003.22003.32003.42003.52003.62003.72003.82003.9Load (kg ) -100 -50 0 50 100 150 2003.02003.22003.42003.62003.8Monthly Flow (m m) Figure 5-8. VAR(1) time series model with 95% CI of monthly TN load anomaly (kg) and streamflow (mm). Models are trai ned using data from 1974-2003, and validated using observed 2003 monthly data. Solid black line indicates observed data, while red lines wi th circles indicate forecast. 218

PAGE 219

-1500 -1000 -500 0 500 1000 1500 2003.02003.22003.42003.62003.8Load (kg ) -100 -50 0 50 100 150 2003.02003.22003.42003.62003.8Monthly Flow (mm ) Figure 5-9. VARX(1) + Obse rved NINO 3.4 SST time seri es model with 95% CI of monthly TN load anomaly (kg) and streamflow (mm). Models are trained using data from 1974-2003, and validated using observed 2003 monthly data. Solid black line indicates observed data, while red lines with circles indicate forecast. 219

PAGE 220

-1500 -1000 -500 0 500 1000 1500 2003.02003.22003.42003.62003.8Load (kg ) -100 -50 0 50 100 1502003.02003.22003.42003.62003.8Flow (mm) Figure 5-10. VARX(1) + Predi cted NINO 3.4 SST time seri es model with 95% CI of monthly TN load anomaly (kg) and streamflow (mm). Models are trained using data from 1974-2003, and validated using observed 2003 monthly data. Solid black line indicates observed data, while red lines with circles indicate forecast. 220

PAGE 221

-1500 -1000 -500 0 500 1000 1500 2003.02003.12003.22003.32003.42003.52003.62003.72003.82003.9Load (kg ) -50 0 50 100 150 2003.02003.22003.42003.62003.8Monthly Flow (mm ) Figure 5-11. W-VAR(7) time se ries model with 95% CI of monthly TN anomaly (kg) and streamflow (mm). Models are tr ained using data from 1974-2003, and validated using observed 2003 monthly data. Solid black line indicates observed data, while red lines wi th circles indicate forecast. 221

PAGE 222

-300 -200 -100 0 100 200 300 123456789101112Load (kg ) -100 -50 0 50 100 150 12345678910111Flow (mm) 2 Figure 5-12. Bi-monthly updated W-VAR(5) time series model for 2003 with 95% CI of monthly nitrate anomaly (kg) and stream flow (mm). Mo nth 1:12 (Jan: Dec) is indicated on the x-axis. Solid black li ne indicates observed data, while red lines with circles indicate forecast. 222

PAGE 223

-1500 -1000 -500 0 500 1000 1500 123456789101112Load (kg ) -100 -50 0 50 100 150 123456789101112Flow (mm) Figure 5-13. Bi-monthly updated W-VAR(7) time series model for 2003 with 95% CI of monthly total nitrogen anomaly (kg) and streamflow (mm). Month 1:12 (Jan: Dec) is indicated on the x-axis. Solid black line indicates observed data, while red lines with circles indicate forecast. 223

PAGE 224

CHAPTER 6 CONCLUSIONS Global climate change has been called the socio-economic, environmental, and energy challenge of our lifetime. The possible consequences of a changing climate are both far-reaching and severe, making resear ch that can help us better understand, adapt to, mitigate, or stop and reverse its consequences a crucial part of modern environmental science. Warming of the aver age global climate ov er the past century has been proven unequivocally [ IPCC et al. 2007], and because the hydrological cycle is linked closely to variations in climate, citizens and water resource managers both will have to deal with new challenges associated with both water quantity and quality. Over the past several decades, research into global warming has revealed significant changes in precipitation patte rns, weather extremes su ch as floods, droughts and storms, snow-pack duration and amount, increas ed evaporation, increased wildfire risk, and changes in soil moisture and runoff [ Mote et al. 2005; Stewart et al. 2005; Westerling et al. 2006]. These relatively sudden changes make it increasingly difficult to efficiently adapt current anthropological practi ces such as large-scale agriculture and maintaining a municipal water supply, while continuing to provide clean, safe drinking water and adequate food to an ev er increasing population. Specifically, it is difficult to say exac tly what climate change effects will be, as climate signals are fairly chaotic and noisy, encompassing annual, inter-annual, decadal, or much longer periods of variability. This climate variability combined with the effect of exogenous variables and the la ck of extensive monitoring systems, both spatially and temporally, can make extracti ng short or long term climatic patterns an uncertain process. For both short and mi ddle term risk managem ent planning, inter224

PAGE 225

annual modes of climate variabi lity and their seasonal expression are of interest. There is a need to identify climate non-stationarities and their links to watershed water quality outcomes. In this dissertation, we isolated, modeled and forecasted the effects of interannual climate variability on hydrology and water quality in the southeast United States. To explore whether there was a basis fo r correlating ENSO phase, hydrology, and water quality in the southeast United States, we used over 30 years of hydrological data simulated by the Watershed Assessment Mo del (WAM) for basin S-191 of the Lake Okeechobee watershed in south Florida. By doing both monthly and seasonal contingency table analyses of hydrology and phosphorus (P) loads in the watershed as delineated by annual JMA categor ical ENSO phase, we found that some ENSO phases tended to produce significantly greater seas onal P loads (FebruaryApril of El Nio years, May-July of La Nia years, and August-September of neutral years) or lower seasonal P loads (May-July of El Nio year s, February-April and August-September of La Nia years, and February-Apr il and May-July of neutral y ears). The greater P load potential in certain months was mostly consistent with documented trends in greater precipitation. As it was difficult to assign statistica l significance using annual designations of ENSO, a newer monthly M-ENSO index was also used to examine the results on a finer temporal scale. Across all variables of simulated P load, concentration, stream flow, and observed precipitation, the M-ENSO classification reduced both the range and average of the La Nia summer months, and allow ed more standard methods of identifying formal significance. Consequently, this m onthly classification which is more representative of current SST conditions inst ead of those of the previous October has a 225

PAGE 226

more accurate temporal scale that suggests t hat the previous months (October to April) of an annual JMA La Nia may be erroneously cr eating patterns in the summer months. Because of the discrepancies in signifi cance between annual and monthly ENSO phase, we concluded that using a conti nuous record of NINO 3.4 sea surface temperatures themselves woul d allow the ident ification of more accurate local trends and patterns using observed data. In Chapter 3, univariate and multivariate wavelet analysis is used to identify and quantify the significance of a teleconnection between SST associated with the NINO 3.4 index, and observed precipitation, stream flow, and nitrate concentration and load in the Little River Watershed in Georgia. The re sults confirm that the known physical mechanism of ENSO teleconnection in the southeast United States [ Schmidt et al. 2001; Ropelewski and Halpert 1986] is causally linked to inter-annual variability within precipitation, stream flow, and nitrate load signals in the LRW. The high shared power and significant correlation between these variables confirms that the ENSO teleconnection seen in the precipitation and st ream flow signals in large river and watershed systems around the world [ Chiew et al. 1998; Rajagopalan and Lall 1998; Handler 1990; Kulkarni, 2000; Hansen et al. 1997; Piechota and Dracup 1999; Pascual et al. 2000] extends to the hydrology and nitr ate loads in a small basin of the Little River Watershed. We found common areas of high, shared power and time series inter-annual variability manifested in the NI NO 3.4 SST index and 36 years of LRW monthly precipitation data, and 29 years of st ream flow, nitrate co ncentration and nitrate load data. Areas of the highest power for all hydrological variables were observed in the 3-7 year periodicity known to be related to ENSO modes of variability. Temporally, the 226

PAGE 227

area of greatest variability was cent ered on the 1997-98 El Nio event. Nitrate concentration was the variable with the weak est ENSO signal power, which is due to it being more dependent on extraneous variables such as human-caused agricultural activities. High or significant power was seen in precipitation, stream flow, and nutrient loads in the 1-2 year period centered on the 1998-99 La Nia, and may be related to other climate indices and str ong seasonal signals. The str onger power seen in nitrate load time series, rather than concentration or precipitation, sugges t that stream flow variability dominates the trends seen in loads. A finding of note from the work in Chapter 3 was that the ENSO signal is more visible in the stream flow and nitrate loads of the LRW t han in the precipitation signal. Although the thirteen weat her stations used to form the precipitation series in this research are well distributed across the 16.8 km2 area of basin K in the LRW, spatial variability of rainfall in the southeast United States is great, and still does not encompass all of the variabilit y inherent in the watershed. For this reason, the ENSO signal present in the precip itation record may be somewhat damped, especially when considered next to the stream flow record In addition to being a smoothed function of precipitation, stream flow in the Li ttle River Watershed may be strengthening the correlation shared with the ENSO signal. To determine what the mechanism for this increased ENSO signal power correlation with flow rather than precipitation is, in Chapter 4 we again turned to a physical hydr ology simulation model, the Soil and Water Assessment Tool (SWAT). Using a calibrated and validated SWAT m odel of LRW-K, the detailed surface and groundwater hydrology was continuously simulated on a daily basis from 1979 to 2004, 227

PAGE 228

and aggregated into monthly average values for use in analysis. The clear increase in the power of the inter-annual climate signal is demonstrated by shared patterns in water budget and exceedance curves, as well as qua litatively and quantitatively in high ENSO related energy in the wavelet spectral analyses and the 95% significant periodicities against a red noise background for each variable both analyzed alone and with the NINO 3.4 SST index. As has been previously discussed, groundwater flow in the LRW is responsible for up to 80% of the total stream flow [ Sheridan 1997], as well as comprising the main route for mo vement of soluble nutrients [Lowrance et al. 1984]. In the Little River Watershed basin K, in Tifton, Georgia, the power of the climatological teleconnection to the El-Nio/ Southern Oscillation is str engthened in both the observed and simulated stream flow th rough the mechanisms of groundw ater and interflow, as they are confined by a geolog ical layer, the Hawthorn Formation. Both the univariate and multivariate spectral ENSO signals were visible and significant in the groundwater and interflow SWAT simulations, which suggest that we could expect to see an even more powerful signal if the flow data were si mulated more precisely with a fully spatiallydistributed model or there was continuous detailed observed groundwater data. The Hawthorne formation in the Little River Watershed forms a nearly perfect aquaclude, so although infiltration is very fast, soil saturation can be reached relatively quickly while recharge into the deep aquifer is extremely slow [ Rawls et al. 1976; Sheridan 1997]. This results in the increased si gnificance of groundwater and interflow contributions to both soluble nutrient movement into outlet stream flow, and the stream flow itself. In the LRW, perhaps much more than other watersheds without a ubiquitous confining layer and lack of t opography, the power of the 3-7 year ENSO signal that is 228

PAGE 229

found in precipitation is increased in the observed and simulated stream flow signal. The entire explanation of t he strengthened ENSO power may have as much to do with the groundwater, interflow and Hawthorn Formation, as with the land use of the watershed itself. In turn, this raises the questi on of what the actual physical properties of the watershed are that lead to this potent ially unique increased ENSO signal power in the simulated and observed surface and sub-su rface flow regimes of the LRW-K. By gathering physical properties of this relative ly small coastal plain watershed such as the total area and depth to the surficial aquifer, transmissivity and porosity in the sandy soils, residence times of groundwater and surfac e water, and hydraulic conductivities, it may be possible to create an analytic function describing exactly how the precipitation based integration of stream flow through time and space in the entire hydrological system occurs. Relative sensitivity analysis and calibrati on of SWAT simulated LRW-K hydrology previously showed that surface response param eters were significantly more sensitive than those pertaining to sub-surface hydrology [Feyereisen, Strickland, et al. 2007]. The most sensitive parameter to overall wate r yield, storm flow, groundwater flow and interflow was the curve number for agricultural land, partly due to the increased surface runoff from even average prec ipitation events on cropped land. In relationship to the future of the Little River Wa tershed, this implies that an increase in land area used for agricultural purposes may serve to strength en the ENSO signal in stream flow even more than it currently is, while a signifi cant decrease in cropped area could cause the teleconnection and predictive relationship between ENSO and flow, and consequently other variables such as pollutant loads, to weaken. As such, this ENSO teleconnection 229

PAGE 230

and its relationship with agricultural area co uld hold for other Coastal Plain watersheds as well. The specific mechanism by which the ENSO signal power is increased via ground water hydrology and a confining layer in the Little River Watershed is by no means the only way in which a watershed could increa se a climate signal delivered primarily through precipitation [ Cayan et al. 1999; Maurer et al. 2006]. However, the spatial integration and reduction of noise in precipitation that the str eam flow signal effectively accomplishes is most likely a primary met hod of signal power increase in different watersheds, as research on river flows fr om around the world have shown significant spectral signals corresponding to multiple climate indices [ Labat 2008]. This nonintuitive relationship between ENSO signal strength and stream flow could prove to be helpful for making seasonal climate predicti ons in a geographic area with a weaker than desirable ENSO/precipitation signal, as a predictive relationship could be found between stream-flow or other proxy hydro-climatic vari ables. The exploration of additional non-precipitation hydrologic vari ables as they correlate to ENSO could expand how climate data could be used for more practical hydrologic prediction and for municipal water supply management, as well. The use of seasonal forecasts based on ENSO teleconnection with both surface and ground water flows and pollutant and nutrient loads could also more immediately benefit water resource managers or farmers, as decisions on irrigation, allocation, and runoff management would be based on actual climate information, and would not be predictions made so far in the future as to render them virtually unusable in normal management plans. 230

PAGE 231

As a significant hydrologic and water quality teleconnection between ENSO and the LRW basin K has been identified, quantif ied, and mechanistic ally explained, Chapter 5 explored using a novel wavelet-bas ed time series model (W-VAR) to predict monthly stream flows and nitrogen (N) loads in the Little River Watershed based on the NINO 3.4 sea surface temperat ure index. As wavelet anal ysis can be likened to a bandpass filter, this can be exploited to extract and rebuild the component signals at different scales according to the desired specifications. Although precipitation is the major driver of the hydrological cycle, it was not include d as an input in these time series models. The non-inclusion of rainfall as a variabl e can be considered an advantage, as in the southeast, precipitation has high spatial variability even within a relatively small area such as basin K of the LRW. This high degree of variability and corresponding noise in the signal can add uncertainty to an already uncertain signal, the exclusion of which gives us interesting insight into hydro-climate processes in the LRW without a direct precipitation input. For these reasons, observed NO3 and TN load and stream flow monthly anomalies were used as inputs to a multivariate time series model. Data in the time-domain was combined with observed NINO 3.4 values input as an exogenous independent variable to create models that incor porated ENSO information in a non-spectral way. This was contrasted with wavelet time series model s that used significant reconstructed frequency components of NO3 and TN load and stream fl ow to inherently encompass ENSO climate variability without the addition of an exogenous variable. In general, the W-VAR models did a much better job than time-domain VAR and VARX models of reproducing and encompa ssing the monthly and seasonal climate 231

PAGE 232

variability trends in the LRW basin K, while theoretically maintaining the spectral signature of the inter-annual clim ate-based oscillations that are within each time series. The bi-monthly updated W-VAR models, more si milar to how such a forecasting system might actually be implemented, demonstrated increasingly accurate monthly load predictions, correctly predicti ng 10/12 load terciles for both NO3 and TN, and 6/12 and 8/12 correct flow terciles with reduced RMSE values. However, the modeling adage of garbage in, garbage out applies here, as pr oblems with the nutrient data could be revealed by SWAT that are ignored by statistical modeling. Indeed, it appears as if the observed nitrate loads between 1992 and 1995 ma y exhibit measurement error, as the anomalies do not make perfect sense with the observed or SWAT simulated total nitrogen loads (Figure 5-2c, 5-2d). Additionally, it is impossible for any of the time series models to specifically incorporate land-use change information in their predictions other than in the most inherent way, although SWAT simulations reflecting land-use changes in the last three decades in the LRW did not show any major changes or influences on the hydrology or nutrient loads. In a wate rshed in which there were major land-use changes in a short period of time, the WVAR method may not perform as accurately. In all non-updated time series models, forecast accuracy declined precipitously after the first three months of prediction, as would be expected from this type of model. For that reason, it should be remembered that these models should not be used to make predictions more than three months in the future, as the uncertainty is magnified each month farther from where the observati ons end, even beyond what is shown in the 95% confidence intervals in the forecasts. Th ese time series models are designed to be shorter-term tools for natural resource managers and farmers to use to implement 232

PAGE 233

BMPs more quickly and appropriately as to reduce nutrient runoff risk by managing periods of high stream flows, instead of co ntinually re-calibrating, adding huge amounts of weather and land-use data, and validating large and complex agro-hydrological models such as SWAT to make management predictions that are necessarily more short term. These prediction tools do not address the ro ot causes of nutri ent pollution, which is application of chemical fertilizers on agric ultural land, or increasing animal production operations. In fact, the predict ion tools are based on high or low stream flows flushing out nutrients from the watershed, whic h is in turn dependent on anthropogenic agricultural practices, surf ace and groundwater residence time s, and nitrogen residence time in the LRW. In choosi ng to more carefully manage ti mes of high stream flow, we are effectively managing how mu ch nutrient pollution is let into the watershed during a period of time, not reducing t he total amounts of pollutants that are input to the system. However, by using these models to more effectively manage stream flows and their nutrient pollution, the overa ll health of the water body and surrounding ecosystem would still be positively affected. If by managing a hi gh flow season predicted using the climate based W-VAR model the amount of nutrient load is reduced, t here would be less chance of eutrophication and ecosystem damage. While the current W-VAR model is based on data from one outlet in basin K of the LR W, it is feasible t hat climatic trends in the region would hold similar patterns, and t hat warnings from one location could be extended to many. This would need to be tested in several other basins of the LRW, however. 233

PAGE 234

Ideally, monthly time series models of hi gh stream flows and nutrient loading risk would be used for shorter term seasonal and annual management, in conjunction with a larger hydrological package su ch as SWAT for any longer term simulation studies about more general land-use change or hydrology e ffects. This kind of analysis could easily be extended to other chemical pollutants in runoff, to create models of how other indicators of water quality and determi nants of human health are influenced and predicted by inter-annual or l onger climate variability. In the course of researching effects of inter-annual climate variability as manifested in the ENSO phenomenon on hydrol ogy and water quality in the southeast United States, we have arrived at predictive models to forecast the near future. Through these models, we thereby add knowledge to how we understand global non-stationary climate modes and their regional consequences at the watershed level. However, this work does bring about questions of how we as a society will deal with the fact of environmental and hydrological nonstationarity as a whole. A recent topic generating much discussion in the academic environmental sciences is that of The Death of Stationarity [Pielke Jr. 2009; Rial et al. 2004]. In this case, the general term stationarity refers to our ability to predict the future based on recorded or reconstructed averages, variances, and statistics accurately describing the past. If global climate change is in fact partly or significantly ant hropogenic in nature, then humans are changing the globa l climate cycle at a faster rate and in different ways than the Earth has ever experi enced before. Additionally, t he Earths climate system in geologic time has always been non-linear in nature [ Rial et al. 2004], with chaotic inputs and outputs, fuzzy boundaries, and multiple equilibria that can cause seemingly 234

PAGE 235

unforeseeable sudden shifts and abrupt changes. In light of these realizations, we can reasonably say that environmental stationar ity has always been dead, and perhaps it is just recently, with anthropogenic influences putti ng the final nail in it s coffin, that the research community as a whole is beginning to accept it and deal with it in models and forecasts. There are those advocating the use of models to predict a wide range of future conditions so as to be able to enable robust policy related decision-making and test the limits of human adaptation as a method of dealing with system nonstationarities [ Dessai et al. 2009]. There are also thos e who, equally importantly, continue to advance the understanding of ph ysical systems informing the models in order to make forecasts increasingly accurate despite unforeseen non-linearities. In the end, however way we as a globa l community choose to deal with climate and hydrological non-stati onarities, we must first recognize that they exist, both in our short-term planning of municipal water s upplies and reservoirs, environmental flows and restoration, agricultural syst ems, and even coastal management. The original research presented in this dissertation is a step to wards recognizing and incorporating climate and hydrological non-stat ionarities into management on th e regional watershed scale, and demonstrates that it c an be done both practically and informatively. By using inherent climate variability to our advantage in hydrological forecasts, we will increase our ability to effectively adapt to the unpredictable challenges of the future. 235

PAGE 236

APPENDIX A WAVELET ANALYSIS OF LAKE OKEECHO BEE BASIN S-191 MONTHLY OBSERVED PRECIPITATION AND WAM SIMULATED STREAM FLOW, TOTAL P LOAD AND CONCENTRATION Introduction To provide continuity to the analyses performed on the data in Chapter 2 and the methods in Chapter 3, Appendix A discusses the results of univariate and multivariate wavelet analyses performed on the Watershed Assessment Model (WAM) simulated data in S-191 of the Lake Okeechobee watershed. The objective in analyzing the WAM simulated data in this way is to ascertai n if the ENSO variability and predictability inherent in the hydrological variables di scovered via contingency table analysis [ Keener et al., 2007]; Chapter 2) is also present in th e frequency structure of the WAM simulated data. While hydrological simulation may be able to accurately re-create and predict magnitudes and general timing of stream flows and water quality, we are interested to see if data simulated by hydrology mechanisms and soil physics as currently understood are able to maintain certain climate based spectral signatures such as the 3-7 year periodicity of ENSO. In Chapter 3, wavelet analysis was used to extract the 3-7 year ENSO periodicities as reflected in the NINO 3.4 sea surface temperatur e index from observed precipitation, stream flow and nutrient parameters in the Little River Watershed in Georgia. Taken by themselves, the stronges t ENSO signal in the LRW mirrored in the monthly hydrological variables was seen in stream flow, then subsequently in nitrate load, nitrate concentration, and finally, least of all in precipitation. In the multivariate cross-wavelet (XWT) and wavelet coherence (WTC) analyses, the most shared inphase high power was seen between the NINO 3.4 index and stream flow, while the 236

PAGE 237

highest covariance between series was bet ween the NINO 3.4 and nitrate load. Physically, this is explained by the inherent noi se in these variables time series. While we may have expected to see the most ENSO signal in the variable most directly related to the El Nio phenomenon, precipitation, it is also an extremely noisy series (Chapter 3, Figure 11a). While stream flow is an additional step away, mechanistically, it ultimately has a stronger ENSO signal in the frequency spectrum than that of precipitation, as it is a smoothed version of the noisier rainfall signal. As nutrient instream loads are mainly dependent on stream flow, and flow variability is largely based on precipitation in the LRW, it is not surprising that nutrient load has the next strongest ENSO power. Since nutrient concentrati on may be more dependent on agricultural activities than natural or climatic processe s, the least visible spectral signal in the concentration variable is also unsurprising. ENSO signal is generally regarded as being more strongly manifested in south Florida, where Lake Okeechobee is located, than in southern Georgia, where the LRW is found [ Ropelewski and Halpert 1986, 1987; Schmidt et al. 2001]. Therefore, if the spectral signature of the hydrologic variables in maintained in the simulated data, we could expect even more 3-7 year power r epresentative of ENSO trends in the Lake Okeechobee data than in the LRW. From the wavelet results for the simulated data in S-191, however, it appears as if the relative spectral signatures of the simulated variables is not maintained as powerfully as they are in observed data in the LRW, where the ENSO signal is known to be weaker. 237

PAGE 238

Data and Methods Data: WAM, Basin S-191 and the NINO 3.4 Index Site description and history of Lake Okeechobee, basin S-191 are discussed in Chapter 2, Data and Methods section T he Lake Okeechobee Watershed. Data used from Lake Okeechobee is identical to that in Chapter 2, Da ta and Methods section Data: observed daily precipitation from 1967-2001 over 14 weather stations was summed into monthly cumulative averages and normalized by monthly average [ Keener et al. 2007]. Variables simulated by WAM from 1967-2001 include daily stream flow; total P concentration (mg/L) and total P load (kg). WAM calibration and validation procedures are also identical to those described in C hapter 2, Data and Methods section Watershed Assessment Model [ Jacobson 2002; Keener et al. 2007; Soil and Water Engineering Technology, Inc. 2004]. The continuous measure of sea surface temperature (SST) used representing ENSO phase, the NINO 3.4 index, is ident ical to that discussed in Chapter 3, Data and Methods, section Little River Watershed Data, with the exception that the SST data used was extended back to 1967. Wavelet Analysis Univariate wavelet analysis was perfo rmed on monthly obser ved precipitation data and simulated total P concentration, to tal P load, and stream flow. The methods are identical to those as indicated in C hapter 3, Data and Methods section Wavelet Analysis [ Torrence and Compo 1998]. Multivariate cro ss-wavelet (XWT) and wavelet coherence (WTC) analysis was performed on monthly NINO 3.4 SST and observed precipitation, simulated total P concentration, total P load, and stream flow. These methods are identical to those described in Chapter 3, Data and Methods section Cross Wavelet and Coherence Transforms [ Grinsted et al. 2004]. 238

PAGE 239

Cross-Correlation Analysis Cross-correlation analysis was perform ed on the monthly NINO 3.4 index and observed precipitation, simulated P load, P concentration, and stream flow. It was also performed between all hydrologic variables: prec ipitation, stream flow, P concentration, and P load. The methodology used is as described in Chapter 3, Data and Methods section Cross-Correlation Time Series Analysis. Results Univariate Wavelet Analysis The monthly wavelet power spectr a for the NINO 3.4 SST, observed precipitation, and simulated stream flow, and total P concentration and load time series are shown in Figure A-1. As discussed in Chapter 3, main power in the NINO 3.4 spectrum is concentrated within the 3-7 year period (Figure A-1a). Precipitation in the Lake Okeechobee region is seasonal, and previous wavelet analysis on Everglades annual rainfall [ Kwon et al. 2006] shows a figure of similar signal strength to that done with the S-191 basins observed precipitation data (Figure A1b). Precipitation shows a relatively high power within the 3-7 year period, centered around 1980-1985. This relative period of signal power is visible in precipitation from 1980-1985 in the LRW as well (Figure 3-2b), and may be indicative of the large 1982-1983 El Nio effects. Now analyzing the WAM simulated data, we can see from the stream flow wavelet spectrum that while the signal st rength has about the same magnitude and 3-7 year periodicity as that of the observ ed precipitation surr ounding 1980-1985 (Figure A1c), it does not share the strengthened ENSO spectral signat ure that was observed in the LRW. While the flow spectrum in t he LRW demonstrated a stronger ENSO signal than precipitation in the LRW because of a less noisy, smoothing effect, this is not seen 239

PAGE 240

in the simulated flow to the same extent. Ho wever, the flow 3-7 y ear periodicity is still powerful and statistically significant above t he red noise background, indicating that the ENSO signal is very much present in the si mulated data. This finding is not directly comparable to the LRW results, as the hydrologic system is most definitely different in each site. Strangely enough, although the observed nitrate concent ration (Figure 3-2d) had the least wavelet ENSO power of all hydrolog ic variables examined, the simulated total P concentration in basin S-191 (Figure A-1d) demonstrates both the highest 3-7 year ENSO signal power, and the most tempora lly prolonged in all the simulated Lake Okeechobee data. While in the LRW, it wa s reasoned that concentration had the least correlation with the NINO 3.4 index because of its dependence on anthropological activities such as agriculture, this is not the demonstrated case in the simulated S-191 concentration spectrum. This could be for several reasons: it may be impossible to compare the two field sites in any respect, or the anthropological fertilizer inputs may be timed differently in either the model simulation or in actuality, despite the fact that the LRW as a general rule has a much smaller problem with water quality, while basin S191 is a nutrient hot-spot for provid ing phosphorus input s to the Lake. In basin S-191, the opposite relationship wi th nutrient load holds than in the Little River Watershed; while the strongest ENSO wavelet relationship was seen in nitrate load in the LRW, in S-191, total simulated P load (Figure A-1e) shows the weakest 3-7 year signal power, although it is still statistically significant against the background noise. In Chapter 2, we found that simu lated load trends seemed to follow observed trends in precipitation in the Lake Okeechobee watershed. The lack of signal strength in 240

PAGE 241

total P load here could again be related to the fact that the tw o field sites may be impossible to compare in terms of signal st rength. However, it could also signify a weakening spectral signal that is inherent in the WAM simulations of hydrological data. Since the flow signal and the P concentra tion signal both appear to have greater spectral 3-7 year power than t he P load, it is unlikely that this is the case. It must be noted, though, that the precipitation used in this analys is (Figure A-1b) is observed actual data, not simulation. T herefore, it is possible that the WAM simulations are simply maintaining the ENSO spectral signal in the flow and water quality data, rather than recognizing or generating them. Cross-Wavelet and Wavelet-Coherence Analysis The cross-wavelet transform (XWT) identifies shared frequencies between two variables with high common power. R egions with shared high power, and thus, significance against a red noise background also indicate a consistent in or out of phase relationship. In the XWT between the NINO 3.4 SST index and observed S-191 precipitation (Figure A-2a), si gnificant areas of shared in-phase high power are visible centered around the 3-7 year per iod from 1980-90, and slightly less power around the 2-3 year period from 1970-75. In fact, all the other XWT analyses performed between NINO 3.4 and simulated stream flow (Figure A-2b), P concen tration (Figure A-2c), and P load (Figure A-2d) show basically the same strength, length, and periodicity of shared power with the same phase relationship. Th is basically identical relationship seems unlikely, and, while the univariate wavelet analyses implied that there were differences in the ENSO signal strength in the simu lated S-191 hydrological variables, these multivariate analyses show that the variabili ty is less than we were lead to believe. 241

PAGE 242

What the XWT analyses demonstrate is t hat the shared time series significant modes of ENSO variability established between the NINO 3. 4 index and observed precipitation is replicated almost exactly in the simulated hydrological data. In real, observed data, such similar areas of common power and significance are rarely seen. Of course, it is possible that this could be a real relationship, although, given the simulation circumstances, it seem s more likely to assume that the variability inherent in the rainfall data was replicated in the stream flow and P concentration and load time series almost identically using empirical an d deterministic relationships in WAM, and that some measure of ENSO related vari ability is missing from the simulated data. The wavelet coherence transform (WTC) analyses, however, do not measure shared power, but time series co-variance. In Chapter 3, it was shown that more area was given significance in WTC than in XW T analysis, reflecting the likely causal mechanism between ENSO and precipitation, stream flow, and nutrient data in the Little River Watershed. In the S-191 WTC analysis, however, less area is significant than in the XWT figures (Figure A-3). It has been previously noted that the less area that is marked significant in a WTC analysis, the less likely that the shared power is causal in nature [ Grinsted et al. 2004]. In fact, while the WTC relationship between Lake Okeechobee observed precipitat ion and NINO 3.4 shows a weakly significant covariance in 3-7 year ENSO periodicities (Figure A-3a), both simu lated stream flow (Figure A-3b) and total P load (Figure A-3d) show similar and stronger co-variance relationships, while total P concentration (Fi gure A-3c), the strongest 3-7 year univariate wavelet signal, has almost no areas of co-v ariance with the NINO 3.4 SST record. 242

PAGE 243

In Chapter 3, in the Little River Wa tershed, on the other hand, stronger 3-7 year ENSO bands of high power were reinforc ed even more strongly in wavelet coherence transform analyses. A relatively weak WT C coherence between SST and precipitation in the LRW gave way to very strong and sustained co-variance for flow and especially for nitrate load. The collective lack of significant high power areas in the WTC (Figure A-3) imply that while univariate ENSO power and some shared multivariate power may be present in simulated hydrologic variables in the Lake Okeechobee ba sin S-191, there is a lack of co-variance between the NINO 3. 4 SST index and the simulated hydrological data that may be representative of a deficiency in replicati ng climate variability in the multi-decadal timescale. Cross Correlation Analysis Time-series cross-correlation analysis was performed between monthly hydrological data and with NINO 3.4 SST to investigate the maximum lag relationships throughout the simulation. Although significant monthly lag correlations were identified between SST and all hydrological variables, they were only very weakly above the confidence bounds. The NINO 3. 4 SST and observed precipitation had a significant but small maximum correlation of 0.156 at zero months lag (Figure A-4a). Sea surface temperatures lead stream flows by 2 m onths, with a maximum correlation of 0.125 (Figure A-4b). Total P concentration lead SS T by 1 month, with a maximum correlation of 0.179 (Figure A-4c), while SST lead P load by one mont h, with a maximum correlation of 0.110 (Figure A-4d). While these cross-correlations are relatively small, the results again show the greatest connecti on of SST with P concentration, as was seen in the univariate wavelet analysis (Figur e A-1d). However, the confusing maximum lag times that alternate between SST leading or following a hydrologic variable tend to 243

PAGE 244

make me skeptical of these results, as a causal relationship of nutrients leading sea surface temperatures does not make sense. Finally, cross-correlation analyses were performed between the hydrologic variables themselves, separate from the infl uence of SST, of precip itation and stream flow (Figure A-4e), stream flow and P load (Figure A-4f), and P load and P concentration (not shown). All relationships had a maximum correlation at zero months lag, indicating that a smaller time step would be needed to identify precise hydrologic relationships. The strongest correlation was between flow and P load (0.977), followed by P concentration and load (0.805), and finally precipitation and flow (0.704). The high correlations between precipitation, flow and load are to be expected, based on the mechanistic hydrology simulations done by WAM, which uses precipitation as its driving input. Summary and Discussion It is difficult to make sensible co nnections between the Lake Okeechobee and Little River Watershed field sites, not onl y because of the difficult in comparing simulated and observed data from two different places, but also because of the differences in the field sites themselves, despi te their presence in the Coastal Plain. In and of itself, however, univariate wavelet analysis of basin S-191 shows that probably due to the input of the drivi ng force of the model, observed precipitation, the 3-7 year ENSO modes of wavelet variability are c onserved in simulated stream flow, P concentration, and P load. It is a credit to WAMs validity that it can maintain the spectral signature of low-frequency ENSO precip itation variability in its simulated stream flow and water quality data. 244

PAGE 245

Given that the ENSO signal in south Flori da is known to be stronger than it is in Georgia and in the Little River Watershed, it is probably not wise to put too much faith in the univariate or multivariate wavelet analys es performed on the simulated data, as they show a stronger signal in the LRW than in the Lake Okeechobee region. Specifically for our cases, the nutrient load in basin S191 is shown to have a relatively weak correlation to the NINO 3.4 index as com pared to the other simulated hydrologic variables and to the LRW analyses done in Chapter 3. This may be because the field site is simply different, or it may be bec ause of the increased mechanistic distance of load from observed precipitation as compared to stream flow, which is highly mechanistically correlated, or concentration, which is less re lated to natural climate and more dependent on agricultural activities. It is questionable whether or not a simulat ed precipitation time series produced by re-sampling or a weather generator w ould retain ENSOs spectral signature throughout water quality WAM simulations. Forcing WAM with different generated precipitation inputs in order to test which one maintained spectral ENSO signals would in fact be an interesting experiment that c ould be of some use to climatologists and modelers in places that are affected si gnificantly by phenomena such as ENSO or different climatological indices. 245

PAGE 246

Figure A-1. Significant wavelet power spec tra are shown within t he cone-of-influence, which depends on time series length and degrees of freedom. Figures are color-mapped to indicate high wavelet power with reds and oranges, and low powers in blue and white. The Global Wave let Spectrum (GWS) at the right of each figure shows power integrated over all scales and time. The 95% confidence limit is shown on the GWS (dashed blue line), the periodicities above which show significance. Mont hly (a) NINO 3.4 (C) SST, (b) precipitation anomaly (cm), (c) stream flow anomaly (m3), (d) total P concentration anomaly (mg/L), and (3) total P load anomaly (kg). 246

PAGE 247

Figure A-1. Continued. 247

PAGE 248

Perio d 1970 1980 1990 2000 0.25 0.5 1 2 4 8 1/8 1/4 1/2 1 2 4 8 (a) 1970 1980 1990 2000 0.25 0.5 1 2 4 8 1/1 6 1/8 1/4 1/2 1 2 4 8 16 (b) YearPerio d 1970 1980 1990 2000 0.25 0.5 1 2 4 8 1/1 6 1/8 1/4 1/2 1 2 4 8 16 (c) 1970 1980 1990 2000 0.25 0.5 1 2 4 8 1/ 8 1/ 4 1/ 2 1 2 4 8 (d) Figure A-2. Cross Wavelet Spectrum betw een (a) monthly SST and Precipitation (cm), (b) monthly SST and stream flow (m3/sec), (c) monthly SST and total P concentration (mg/L), (d) monthly SST and total P Load (kg). Black figure outlines indicate areas significant to 95% confidence, while arrows represent variables phase relationship. 248

PAGE 249

Perio d 1970 1980 1990 2000 0.25 0.5 1 2 4 8 0 0.2 0.4 0.6 0.8 1 (a) 1970 1980 1990 2000 0 .25 0.5 1 2 4 8 0 0.2 0.4 0.6 0.8 1 (b) Perio d 1970 1980 1990 2000 0.25 0.5 1 2 4 8 0 0.2 0.4 0.6 0.8 1 (c) 1970 1980 1990 2000 0 .25 0.5 1 2 4 8 0 0.2 0.4 0.6 0.8 1 (d) Figure A-3. Wavelet Coherence Analysis between (a) monthly SST and Precipitation (cm), (b) monthly SST and stream flow (m3/sec), (c) monthly SST and total P concentration (mg/L), (d) monthly SST and total P load (kg). Black figure outlines indicate areas significant to 95% confidence, while arrows represent variables' phase relationship. 249

PAGE 250

Figure A-4. Cross-correlation analysis between NINO 3.4 (C) and Basin S-191 (a) precipitation (cm) (b) stream flow (m3) (c) total P (mg/L), (d) total P (kg), and between (e) precipitation (cm) and stream flow (m3), (f) stream flow and P load. A negative lag indicates months that the NINO 3.4 SST leads the variable in question. The strongest Cross-Correlation Function (CCF) relationship within the NINO 3.4 index is between SST and total P concentration at zero months (no lag), although it is still very weak, while the strongest hydrologic relationship overall is between stream flow and total P load at zero months. Values above or below dashed lines indicate significant correlation above 95% confidence. 250

PAGE 251

LIST OF REFERENCES Aceituno, P. (1988), On the functioning of the Southern Oscillation in the South American sector. Part I: Surface climate, Mon. Weather Rev. 116 (3), 505-524. AchutaRao, K., and K. Sperber (2002), Simula tion of the El Nio Southern Oscillation: Results from the coupled m odel inter-comparison project, Clim. Dyn. 19(3), 191209. Asefa, T., and A. Adams ( 2007), Short-term urban water demand forecast models in action: Challenges from model developm ent to implementation to real-time operations, pp. 1-7, Orlando, FL. Askar, A., A. E. Cetin, and H. Rabitz ( 1996), Wavelet transform for analysis of molecular dynamics, J. Phys. Chem. 100 (49), 19165-19173. Asmussen, L. (1971), Hydrolog ic effects of Quaternary sediments above the marine terraces in the Georgia coastal plain, Southeast. Geol. 12, 189-201. Baigorria, G. A., J. W. Jones, and J. J. O' Brien (2007), Understanding rainfall spatial variability in southeast USA at different timescales, Int. J. Climatol. 27(6), 749 760. Baigorria, G. A., J. W. Hansen, N. Ward J. W. Jones, and J. J. O'Brien (2008), Assessing predictability of cotton yields in the southeastern United States based on regional atmospheric circulat ion and surface temperatures, J. App. Meteorol. 47(1), 76. Baldwin, M. P. et al. (2001), T he quasi-biennial oscillation, Rev. Geophys. 39(2), 179 229. Bates, B. C., Z. W. Kundzewicz, S. Wu, and J. P. Palutikof (2008), Climate Change and Water. Technical Paper of the Inte rgovernmental Panel on Climate Change IPCC Secretariat, Geneva. Baudrillard, J. (1985), Simulacra and Simulation (The Body In Theory: Histories of Cultural Materialism) University of Michigan Press, Ann Arbor, MI. Bingner, R. L., and F. D. Theurer (2007), AnnAGNPS Technical Processes Documentation, Version 4.0 U.S. Dept. Agriculture, NRCS, Washington, D.C., USA. Bjerknes, J. (1969), Atmospheric teleconnections from the equatorial Pacific, Mon. Weather Rev. 97(3), 163-172. Bollerslev, T. (1986), Gener alized autoregressive conditional heteroscedasticity, J. Econometrics 31, 307-327. 251

PAGE 252

Bonham, S. G., A. M. Haywo od, D. J. Lunt, M. Collins, and U. Salzmann (2009), El Nio-Southern Oscillation, P liocene climate and equifinality, Philos. Transact. A Math. Phys. Eng. Sci. 367 (1886), 127-156. Bosch, D. D., R. Lowrance, J. M. Sheri dan, and R. G. Williams (2003), Ground water storage effect on streamflow for a s outheastern coastal plain watershed, Ground Water 41(7), 903. Bosch, D. D., and J. M. Sheridan (2007), Stream Discharge Database: Little River Experimental Watershed, G eorgia, United States, Water Res. 43, W09473. Bosch, D. D., J. M. Sheridan, H. L. Batten, and J. G. Arnold (2004 ), Evaluation of the SWAT model on a Coastal Plain agricultural watershed, Trans. ASABE 47(5), 1493-1506. Bosch, D. D., J. M. Sheridan, and F. M. Davis (1999), Rainfall characteristics and spatial correlation for the Georgia Coastal Plain, Trans. ASAE 42(6), 1637-1644. Bosch, D. D., J. M. Sheridan, R. Lowrance, R. K. Hubbard, T. C. Strickland, G. W. Feyereisen, and D. G. Sull ivan (2007), Little River Experimental Watershed database, WRR 43, W09470. Bosch, D. D., J. M. Sheridan, and L. K. Marshall ( 2007), Precipitation, soil moisture, and climate database, Little River Experimental Watershed, Georgia, United States, Water Resour. Res. 43, W09472. Bosch, D. D., D. G. Sullivan, and J. M. Sheridan (2006), Hy drologic impacts of land-use changes in Coastal Plain watersheds, Trans. ASABE 49 (2), 423-432. Bottcher, A. B., N. B. Pickering, and A. B. Cooper (1998), EAAMOD-FIELD: A flow and phosphorus model for high water tables, in Proceedings of the 7th Annual Drainage Symposium pp. 599-606, ASAE, St. Joseph, MI. Burkett, V., R. Ritschard, S. McNulty, J. J. O'Brien, R. Abt, J. W. Jones, U. Hatch, B. Murray, S. S. Jagtap, and J. Cruise (2001), Potential Consequences of Climate Variability and Change for the Southeastern United States in National Assessment Synthesis Team for the US Gl obal Change Research Program Cambridge University Press, Cambridge, UK. Cai, X., and M. W. Rosegrant (2002), Global water demand and supply projections, Part 1: a modeling approach, Water Int. 27(2), 159-169. Calhoun, J. (1983), Soil Survey of Tift County USDA-SCS, Washington, D.C. Campbell, K., J. Capece, and T. Tremwel (1995), Surface/ subsurface hydrology and phosphorus transport in the Kissi mmee River Basin, Florida, Ecol. Eng. 5 (2-3), 301-330. 252

PAGE 253

Cane, M. A. (1986), El Nio, Annu. Rev. Earth Planet Sci. 14(1), 43-70. Cane, M. A. (2005), The evolutio n of El Nio, past and future, Earth Planet Sci. Lett. 230 (3-4), 227-240. Cane, M. A., R. J. Patton, S. E. Zebiak, and S. C. Dola n (1986), Experimental forecasts of El Nio, Nature 321 (6073), 827. Carpenter, S. R., N. F. Caraco, D. L. Correll, R. W. Howarth, A. N. Sharpley, and V. H. Smith (1998), Non-point pollution of surf ace waters with phosphorus and nitrogen, Ecol. Appl. 8 (3), 559-568. Cayan, D., K. T. Redmond, and L. G. Riddl e (1999), ENSO and hydrologic extremes in the western United States, J. Clim., 12(9), 2881. Chan, T. F., and J. Shen (2005), Image Processing and Analysis : variational, PDE, wavelet, and stochastic methods Society for Industrial Mathematics. Charles, C. D., D. E. Hunter, and R. G. Fairbanks (1997), In teraction between the ENSO and the Asian monsoon in a cora l record of tropical climate, Science 277 (5328), 925. Chavez, F. P., P. G. Strutton, G. E. Friederich, R. A. Feely, G. C. Feldman, D. G. Foley, and M. J. McPhaden (1999), Bi ological and chemical re sponse of the equatorial Pacific Ocean to the 1997-98 El Nio, Science 286 (5447), 2126. Chiew, F. H. S., and T. A. McMahon (2003), El Nio/Southern Oscillation and Australian rainfall and streamflow, Aust. J. Wat. Resour. 6 (2), 115. Chiew, F. H. S., T. C. Piechota, J. A. Dracup, and T. A. McMahon (1998), El Nio/Southern Oscillation and Australian ra infall, streamflow, and drought: Links and potential for forecasting, J. Hydrol., 204 138-149. Cho, J., Bosch, D.D., Vellidis, G., Lowrance, R.K., and Strickland, T.C. (2009), Calibrating and validating SWAT for cons ervation practice assessment at the watershed scale, Trans. ASABE in press Cho, J., Bosch, D.D., Vellidis, G., Lowrance, R.K., and Strickland, T.C. (2010), Evaluation of hydrologic and water quality impacts of crop management in the Little River watershed using SWAT, J. Am. Water Resour. Assoc., accepted. Collins, M. (2005), El Nio-or La Nia-like climate change?, Clim. Dyn. 24(1), 89-104. Dahl, T. E. (1990), Wetlands losses in the United Stat es, 1780's to 1980's. Report to the Congress National Wetlands Inventory, St. Petersburg, FL (USA). d'Arrigo, R., R. Villalba, and G. Wiles ( 2001), Tree-ring estimates of Pacific decadal climate variability, Clim. Dyn. 18(3), 219. 253

PAGE 254

Daubechies, I. (1990), The wavelet transfo rm, time-frequency localization and signal analysis, IEEE Trans. Inf. Theory 36(5), 961-1005. Dean, J., and W. Snyder (1977), Tempor ally and aerially distributed rainfall, J. Irrig. Drain. Div. Am. Soc. Civ. Eng. (103), 221 229. Dessai, S., M. Hulme, R. Lem pert, and R. A. Pielke Jr. ( 2009), Climate prediction: a limit to adaptation? in Living with climate change: are there limits to adaptation? edited by W. Adger, I. Lor enzoni, and K. O'Brien, Camb ridge University Press. Dettinger, M., and H. F. Diaz (2000), Global char acteristics of stream flow seasonality and variability, J. Hydrometeorol. 1 (4), 289-310. DHI (2004), MIKE-SHE, An Integrated Hy drological Modeling System Technical Manual, Danish Hydraulic Institute, Hrsholm, Denmark. van den Dool, H., and A. Barnston (1995), Fore casts of global sea surface temperature out to a year using the constructed analogue method, in Proceedings of the 19th Annual Climate Diagnostics Workshop pp. 416-419, College Park, Maryland. van den Dool, H., J. Huang, and Y. Fan (2003), Performance and analysis of the constructed analogue method applied to US soil moisture over 1981, J. Geophys. Res 108 8617. Baigorria, G. A., J. W. Hansen, N. Ward J. W. Jones, and J. J. O'Brien (2008), Assessing predictability of cotton yields in the southeastern United States based on regional atmospheric circulat ion and surface temperatures, Journal of Applied Meteorology 47(1), 76. Duda, P., J. Kittke Jr., A. D onigan Jr., and R. Kinerson (2006), Better Assessment Science Integrating Point and Nonpoint Sources (BASINS) Watershed Models, edited by V. Singh and D. Frevert, C.R.C. Press, Taylor & Francis Group, Boca Raton, FL. Enfield, D. B., A. M. Mestas -Nunez, and P. J. Trimble (2001), The Atlantic Multi-decadal Oscillation and its relationship to rainfall and river flows in the Continental US, Geophys. Res. Lett 28, 2077. Farge, M. (1992), Wavelet transforms and their applications to turbulence, Annu. Rev. Fluid Mech. 24(1), 395. Federico, A. C., F. E. Davi s, and J. Milleson (1981), Nutri ent retention in an unmodified flow-through marsh, Proceedings of a Symposium on Progress in Wetlands Utilzation and Management, Orlando, FL., p. 432. Fedorov, A. V., and S. G. Philander (2001) A stability analysis of tropical ocean atmosphere interactions: Bridging measurements and theor y for El Nio, J. Clim. 14(14), 3086-3101. 254

PAGE 255

Feyereisen, G. W., R. Lowrance, T. C. Strickland, D. D. Bosch, and J. M. Sheridan (2008), Long-term stream chemistry trends in the southern Georgia Little River Experimental Watershed, J. Soil Water Conserv., 63(6), 475-486. Feyereisen, G. W., R. Lowrance, T. C. Strickland, J. M. Sheridan, R. K. Hubbard, and D. D. Bosch (2007), Long-term wate r chemistry database, Little River Experimental Watershed, southeast Coastal Plain, United States, Water Resour. Res. 43, W09474. Feyereisen, G. W., T. C. Strickland, D. D. Bosch, and D. G. Sullivan (2007), Evaluation of SWAT manual calibration and input param eter sensitivity in the Little River Watershed, Trans. ASABE 50(3), 843-855. Fisher, R. A. (1935), The logic of inductive inference (with discussion), J. Royal Stat. Soc. 98(1), 39-82. Flaig, E. G., and K. R. Reddy (1995), Fate of phosphorus in the Lake Okeechobee watershed, Florida, USA: overview and recommendations, Ecol. Eng. 5 (2-3), 127142. Florida DEP (2005), Floridas Total Maximum Daily Lo ad Program: the First 5 Years. A Report to the Legislature and Governor Division of Water Resource Management. Mimi Drew, Director. Florida DEP (2001), Total Maximum Daily Load for Total Phosphorus Lake Okeechobee, Florida U.S. EPA, Tallahassee, FL, USA. Fox, D. D., S. Gornak, T. D. McCall, D. W. Brown, and C. J. Morris (1993), Lake Okeechobee fisheries investi gations completion report Florida Game and Fresh Water Fish Commission, Tallahassee, FL, USA. Fraedrich, K., and K. Muller (1992), Climate anomalies in Europe associated with ENSO extremes, Int. J. Climatol., 12(1). Fraisse, C. W., and K. Campbell (1997), FHANTM Version 2.0 Users Manual, Report to South Florida Water Management District West Palm Beach, FL. Gakstatter, J. H., A. F. Bartsch, and C. A. Callahan (1978), The impact of broadly applied effluent phosphorus standards on eutrophication control, Water Resour. Res. 14, 1155. Grard-Marchant, P. G. F., D. E. Stooksbury, and D. Zierden (2008), Alternative characterization of ENSO episodes and applicat ions to precipitation patterns in the Southeastern US, submitted to J. Service Climatol. Gioia, G., and F. A. Bombardelli (2002), Scaling and similarity in rough channel flows, Phys. Rev. Lett. 88(1), 6465. 255

PAGE 256

Goddard, L., S. J. Mason, S. K. Zebiak, C. F. Ropelewski, R. Basher, and M. A. Cane (2001), Current approaches to seasonal-t o-interannual climate predictions, Int. J. Climatol. 21, 1111. Graham, A. (1989), Late tertiary paleoaltitude s and vegetational zonation in Mexico and Central America, Botanica Neerlandia 38, 417-424. Grinsted, A., J. C. Moore, and S. Jevrejev a (2004), Application of the cross wavelet transform and wavelet coherenc e to geophysical time series, Geophysics 11, 561566. Gu, D., and S. G. H. Philander (1995), Se cular changes of annual and interannual variability in the tropics during the last century, J. Clim., 8 864. Gutierrez, F., and J. A. Dr acup (2001), An analysis of the feasibility of long range streamflow forecasting for Colombia usin g El-Nio/Southern Oscillation indicators, J. Hydrol., 246 181-196. Haan, C. (2002), Statistical methods in hydrology 2nd ed., Ames: The Iowa State University. Halpert, M. S., and C. F. R opelewski (1992), Surface temper ature patterns associated with the Southern Oscillation, J. Clim. 5 (6), 577-593. Handler, P. (1990), USA corn yields, t he El Nio and agricul tural drought: 1867-1988, Int. J. Clim. 10, 819-28. Hansen, J. W., A. Irmak, and J. W. J ones (1997), El Nio-Southern Oscillation influences on Florida crop yields, Soil Crop Sci. Soc. Florida 57, 12-16. Hansen, J. W., J. W. Jones, C. F. Kiker, and A. W. Hodges ( 1999), El NioSouthern Oscillation Impacts on Winter Vegetable Production in Florida, J. Clim., 12(1), 92102. Hanson, K., and G. A. Maul (1991), Florida precipitatio n and the Pacific El Nio, 1895 1989, Fla. Sci. 54(3/4), 160. Harmel, R. D., R. J. Cooper R. M. Slade, R. L. Haney and J. G. Arnold (2006), Cumulative uncertainty in measured st reamflow and water quality data for small watersheds, Trans. ASABE 49(3), 689-701. Havens, K. E., and D. E. Gawlik (2005) Lake Okeechobee conceptual ecological model, Wetlands 25(4), 908-925. Haywood, A. M., and P. J. Valdes (2004), Modeling Pliocene warmth: contribution of atmosphere, oceans and cryosphere, Earth Planet. Sci. Lett. 218 (3-4), 363-377. 256

PAGE 257

Hefner, J. M., B.O. Wilen, T. E. Dahl, and W. E. Frayer (1994), Southeast wetlands; status and trends, mid-1970's to mid-1980's. U.S. Dept. Int., Fi sh and Wildl. Serv., Atlanta, Georgia. Hoskins, B. J., and D. J. Karoly (1981), The steady linear response of a spherical atmosphere to thermal and orographic forcing, J. Atmos. Sci. 38(6), 1179-1196. Huang, Y., B. Shuman, Y. W ang, T. Webb, E. C. Grim m, and G. Jacobson (2006), Climatic and environmental controls on the variation of C3 and C4 plant abundances in central Florida for the past 62,000 years, Palaeogeogr. Palaeoclimatol. Palaeoecol. 237 (2-4), 428-435. Huber, M., and R. Caballero (2003), Eocene El Nio: Evidence for robust tropical dynamics in the hothouse, Science 299 (5608), 877-881. Ice, G. (2004), History of innovative best ma nagement practice development and its role in addressing water quality limited water bodies, J. Environ. Eng. 130 684. IPCC (2007), Climate Change. 2007. Synthesis R eport. A Contribution of Working Groups I, II, and III to the Fourth Asse ssment Report of the Intergovernmental Panel on Climate Change IPCC, Geneva, Switzerland. IPCC, M. L. Parry, O. F. Canziani, J. P. Palutikof, P. J. van der Linden, and C. E. Hanson (2007), Climate Change 2007: Impacts, A daptation and Vulnerability. Contribution of Working Gr oup II to the Fourth A ssessment Report of the Intergovernmental Panel on Climate Change Cambridge University Press, Cambridge, UK. Jacobson, B. (2002), Hydrologic/water quality charac terization of the watershed: Results of Phase 2 analytical approach. Central and Southern Florida project comprehensive Everglades restoration plan (CERP) SWET, Inc., Gainesville, Florida. Jacobson, B., and A. B. Bottcher (1998), Unique routing algorithm for watershed assessment modeling, ASAE Conference Paper Paper #982237 Jagtap, S. S., J. W. Jones, P. Hildebrand, D. Letson, J. J. O'Br ien, G. Podesta, D. Zierden, and F. Zazueta (2002), Respondi ng to stakeholder's demands for climate information: from research to applications in Florida, Agric. Syst. 74 (3), 415-430. Japan Meteorological Agency (1991), Climate Charts of Sea Surface Temperatures of the Western North Pacific and the Global Ocean Marine Department. Kaiser, G. (1994), A Friendly Guide to Wavelets Birkhuser Boston, Cambridge, MA. Kaplan, A., M. A. Cane, Y. Kushnir, A. C. Clement, M. B. Blumenthal, and B. Rajagopalan (1998), Analyses of globa l sea surface te mperature 1856-1991, J. Geophys. Res., 103 (18), 567. 257

PAGE 258

Keener, V. W., G. W. Feyereis en, U. Lall, J. W. Jones, D. D. Bosch, and R. Lowrance (2010), El Nio/Southern Oscillati on (ENSO) influences on monthly NO3 load and concentration, stream flow, and precipitation in the Littl e River Watershed, Tifton, Georgia., J. Hydrol., 381 (3-4), 352-363. Keener, V. W., K. T. Ingram, B. Jacobson, and J. W. Jones (2007), Effects of ElNio/Southern Oscillation on simulated phosphorus loading, Trans. ASABE 50(6), 2081-89. Kiladis, G. N., and H. F. Diaz (1989), Global climate anomalies associated with extremes in the Southern Oscillation, J. Clim. 2 1069-1090. Knisel, W. G., P. Yate s, J. M. Sheridan, T. K. Woody III, J. Allen, and L. E. Asmussen (1985), Hydrology and hydro-geology of Upper Taylor Creek watershed, Okeechobee County, Flor ida: data and analysis. USDAAgricultural Research Service, Washington, D.C. Koutavas, A., P. B. Demenocal, G. C. O live, and J. Lynch-Stieglitz (2006), MidHolocene El Nio-Southern Oscillation (E NSO) attenuation revealed by individual foraminifera in eastern tr opical Pacific sediments, Geology 34(12), 993-996. Kulkarni, J. R. (2000), Wavelet analysis of the association between the southern oscillation and the Indian summer monsoon, Int. J. Climatol. 20(1). Kwon, H. H., U. Lall, and A. Khalil (2007), Stochastic simulation model for nonstationary time series using an autoregressive wavele t decomposition: Applications to rainfall and temperature, Water Resour.Res. 43. Kwon, H. H., U. Lall, Y. Moon, A. F. K halil, and H. Ahn (2006), Episodic interannual climate oscillations and their influenc e on seasonal rainfall in the everglades national park, Water Resour. Res. 42. Labat, D. (2008), Wavelet analysis of the annual discharge records of the worlds largest rivers, Adv. Water Resour. 31, 109. Latif, M., D. Anderson, T. Bar nett, M. A. Cane, R. Kleeman, A. Leetmaa, J. O'Brien, A. Rosati, and E. Schneider (1998), A review of the predictability and prediction of ENSO, J. Geophys. Res. 103 (14), 375. Lau, K. M., and P. J. Sheu (1991), Telec onnections in global rainfall anomalies: seasonal to inter-decadal time scales, in Teleconnections Linking Worldwide Climate Anomalies: Scientif ic Basis and Societal Impact pp. 227, Cambridge University Press. Leonard, R. A., W. G. Knisel, and D. A. Still (1987), GLEAMS: Groundwater Loading Effects of Agricultur al Management Systems, Trans. ASABE 30, 1403-1418. 258

PAGE 259

Li, X., R. B. Ambrose, and R. Araujo ( 2004), Modeling mineral nitrogen export from a forest terrestrial ec osystem to streams, Trans. ASABE 47(3), 727-739. Lins, H. F., and J. R. Slack (1999), St reamflow trends in the United States, Geophys. Res. Lett. 26(2). Livezey, R., and T. Smith (1999), Covariability of aspects of North American climate with global sea surface tem peratures on interannual to interdecadal timescales, J. Clim. (12), 289-302. Lowrance, R. (1992), Groundwater Nitrate and denitrification in a coastal plain riparian forest, J. Environ. Qual. 21, 401-405. Lowrance, R., and R. A. Leonard (1988), Streamflow nutrient dynamics on coastal plain watersheds, J. Environ. Qual. 17(4), 734. Lowrance, R., R. A. Leonard, and L. E. Asmussen (1985), Nutrient budgets for agricultural watersheds in the southeastern coastal plain, Ecology 66(1), 287-296. Lowrance, R., and J. M. Sheridan (2005), Surface runoff water quality in a managed three zone riparian buffer, J. Environ. Qual. 34(5), 1851. Lowrance, R., Sheridan, J.M., Yate s, P., and L. E. Asmussen (1983), Hydrology and Sediment Movement from south eastern Coastal Plain Watersheds Nutrient Cyclling in Agricultural Ecosystems, Special Publication, University of Georgia, SEWRL. Lowrance, R., R. L. Todd, and L. E. Asumss en (1984), Nutrient cycli ng in an agricultural watershed: Streamflow and artificial drainage, J. Environ. Qual. 13, 27-32. Malik, A. S., B. A. Larson, and M. Ribaudo (1994), Economic incentives for agricultural nonpoint source pollution control, J. Am. Water Resour. Assoc. 30 (3), 471-480. Mason, S. J., and L. Goddard (2001), Probabilistic precipitat ion anomalies associated with ENSO, Bull. Am. Met. Soc. 82(4), 619-638. Maurer, E. P., S. Gibbar d, and P. B. Duffy (2006), Amplific ation of streamflow impacts of El Nio by increased atmospheric greenhouse gases, Geophys. Res. Lett. 33(2), L02707. Messina, C. D., J. W. Jones and J. W. Hansen (2001), Un derstanding ENSO effects on tomato yields in Florida: a modeli ng approach, Florence, Italy, pp. 155. Mo, K., and R. Livezey (1986), Tropica l-extratropical geopotential height teleconnections during the Northern Hemisphere winter, Mon. Weather Rev. 114 (12), 2488-2515. 259

PAGE 260

Molnar, P., and M. A. Cane ( 2007), Early Pliocene (pre-Ice Age) El Nio-like global climate: Which El Nio?, Geosphere, 3 (5), 337. Moorhead, D., and C. Dangerfield (1996), Shifting annual rowcrop production to Conservation Reserve Program tree planting: Impacts on agricultural and personal income, and employment in Georgia, in Conference Abstracts p. 202, The Pennsylvania State University Dept. of Agricultural Economics & Rural Sociology and The School of Forest Resources, University Park, PA. Moriasi, D. N., J. G. Arnold, M. W. Van Liew, R. L. Bingner R. D. Harmel, and T. L. Veith (2007), Model evaluat ion guidelines for systematic quantification of accuracy in watershed simulations., Trans. ASABE 50 (3), 885-900. Mote, P. W., A. F. Hamle t, M. P. Clark, and D. P. Le ttenmaier (2005), Declining mountain snowpack in western North America, Bull. Am. Met. Soc. 86(1), 39-49. Mouri, H., M. Takaoka, and H. Kubotani (1999), Wavelet identification of vortex tubes in a turbulence velocity field, Phys. Lett. A 261 (1-2), 82-88. Nash, J. E., and J. V. Sutcliffe (1970), River flow forecasting through conceptual models, Part I: A discussion of principles, J. Hydrol. 10(3), 282-290. Neitsch, S. L., J. G. Arnold, J. R. Kiniry, R. Srinvas an, and J. R. Williams (2002), Soil and water assessment tool: Users manual: Version 2000 USDA-ARS GSWRL and Blackland Research Center, Temple, TX. Neitsch, S. L., J. G. Arnold, J. R. Kiniry, J. R. Williams, and K. W. King (2005), Soil and Water Assessment Tool: Theoretical documentation, Version 2005. Blackland Research Center, USDA-ARS Grassland Soil and Water Research, Temple, TX. Nezlin, N. P., and B. Li (2003) Time-series analysis of remote-sensed chlorophyll and environmental factors in the Santa MonicaSan Pedro Basin off Southern California, J. Mar. Syst. 39(3-4), 185-202. Ouyang, Y. (2003), Simulating dynamic load of naturally occurring TOC from watershed into a river, Water Res., 37, 823. Park, J., and M. Mann (2000), Interannual Temperature Event s and Shifts in Global Temperature: A Multiwavelet Correlation Approach, Earth Interactions 4 (1), 136. Pascual, M., X. Rodo, S. P. Ellner, R. Colwell, and M. J. Bouma (2000), Cholera dynamics and El Nio -Southern Oscillation, Science 289 1766-1769. Pathak, T. (2010), Cotton Yield Forecasting for the Southeastern United States, Ph.D. Dissertation, University of Florida. 260

PAGE 261

Phillips, N. A. (1956), T he general circulation of t he atmosphere: A numerical experiment, Q. J. Roy. Meteorol. Soc. 82(352). Piechota, T. C., and J. A. Dracup (1999), Long-range streamflow forecasting using El Nio-Southern Oscillation indicators, J. Hydrol. Eng. 4 (2), 144-151. Pielke Jr., R. A. (2009), Collateral damage from the death of stationarity, GEWEX Newsletter (May), 5-7. Postgate, J. (1998), Nitrogen Fixation Cambridge University Press, Cambridge, UK. Priestley, C., and R. Taylor (1972), On the assessment of surface heat flux and evaporation using large-scale parameters, Mon. Weather Rev. 100 (2), 81-92. Purdum, E. D. (2002), Florida Waters: A Water Resources Manual Institute of Science and Public Affairs, Florida State University, Tallahassee, Florida. Purdum, E. D., L. C. Bur ney, and T. M. Swihart (1998), History of Water Management In E.A. Water Resources Atlas of Florida Florida State University, Tallahassee, Florida. Rabalais, N. N., R. E. Turner, and D. Scavia (2002), Beyond science into policy: Gulf of Mexico hypoxia and the Mississippi River, Bioscience 52 (2), 129-142. Rajagopalan, B., and U. Lall (1998), Interannual variability in western US precipitation, J. Hydrol., 210 51-67. Rajagopalan, B., U. La ll, and S. E. Zebiak (2002), Categorical climate forecasts through regularization and optimal combinat ion of multiple GCM ensembles, Mon. Weather Rev. 130 (7), 1792. Rasmusson, E. M., and J. M. Wallace (1983), Meteorological aspects of the El Nio/southern oscillation, Science 222 (4629), 1195-1202. Rawls, W., P. Yates, and L. Asmussen (1976), Calibration of selected infiltration equations for the Georgia Coastal Plain, SEWRL, Tifton, GA. Reed, R., W. Campbell, L. Ra smussen, and R. Rogers (196 1), Evidence of a downward propagating annual wind reversal in the equatorial stratosphere, J. Geophys. Res. 66, 813. Rial, J. A. et al. (2004), Nonlinearities, feedbacks and critical thresholds within the Earth's climate system, Climatic Change 65(1), 11. Ropelewski, C. F., and M. S. Halpert (1987), Global and r egional scale precipitation patterns associated with the El Nio/Southern Oscillation (ENSO), Mon. Weather Rev. 115 1606-1626. 261

PAGE 262

Ropelewski, C. F., and M. S. Halpert (1986), North American precipitation and temperature patterns associ ated with the El Nio/Sout hern Oscillation (ENSO), Mon. Weather Rev. 114 2352-2362. Rosen, B. H., S. Gr ay, and E. Flaig (1996), Implementation of Lake Okeechobee Watershed Management Strategies to Control Phosphorus Load South Florida Water Management Di strict (SFWMD). Rosensweig, C., and D. Hillel (1995), Potential impacts of climate change on agriculture and food supply, Consequences 1 (2), 22-23. Saha, S., S. Nadiga, C. Thiaw, J. Wang, W. Wang, Q. Zhang, H. M. Van den Dool, H. L. Pan, S. Moorthi, and D. Behringer ( 2006), The NCEP climate forecast system, J. Clim. 19(15), 3483. Schmidt, N., E. K. Lipp, J. B. Rose, and M. Luther (2001), ENSO influences on seasonal rainfall and river discharge in Florida, J. Clim. 14, 615-628. Serbruyns, I., and S. Luyssaert (2006), Acceptance of sticks, carrots and sermons as policy instruments for direct ing private forest management, Forest Policy Econ. 9 (3), 285-296. Serrano, E., R. Compagnucci, and M. Fadio (1992), The use of the wavelet transform for climatic estimates, in Fifth International Meeti ng on Statistical Climatology vol. Toronto, ON. Canada, pp. 259-262, Am erican Meteorological Society. Sheridan, J. M. (1997), Rainfall-streamflow relations for coas tal plain watersheds, Appl. Eng. Agric. 13(3), 333-344. Sheridan, J. M., and W. G. Knisel (1989), Rain fall trends in the Georgia Coastal Plain, in Proceedings of the 1989 Georgia Water Resources Conference University of Georgia, pp. 9-11, Universi ty of Georgia, Athens, GA. Sheridan, J. M., W. H. Merkel, and D. D. Bo sch (2001), Peak rate factors for flatland watersheds, Appl. Eng. Agric. 18(1), 65-69. Sheridan, J. M., P. Yates, and L. E. Asmussen (1983), Hydrology and nutrient and sediment movement from South eastern Coastal Plain watersheds University of Georgia College of Agricultural Experiment al Stations, University of Georgia. Shirmohammadi, A., J. M. Sheridan, and L. E. Asmussen (1986), Hydrology of alluvial stream channels in southern coastal plain watersheds, Trans. ASABE 29(1), 135142. Sittel, M. C. (1994), Marginal probabilities of the extremes of ENSO events for temperature and precipitation in the southeaster n United States Center for Ocean-Atmospheric Studies, Florida St ate University, Tallahassee, FL. 262

PAGE 263

Skaggs, R. W. (1981), DRAINMOD-Reference report: Methods for design and evaluation of drainage-water management systems for soils with high water tables National Resources Conservation Servic e (NRCS), Wetland Science Institute, Washington, D.C., USA. Soil and Water Engineeri ng Technology, Inc. (1996), EAAMOD, Technical Manual Everglades Agricultural Area MODel Field (11/96 revision) SWET, Inc., Gainesville, Florida. Soil and Water Engineeri ng Technology, Inc. (2000), EAAMOD, User's Manual, Everglades Agricultural Area MODel Fi eld (11/00 revision). Interface Version 3.0.0, Model version 14.6.0. SWET, Inc., Gainesville, FL. Soil and Water Engineeri ng Technology, Inc. (2004), WAM Model Documentation and User Manual SWET, Inc., Gaines ville, Florida. South Florida Water Managemen t District (SFWMD) (2001), Surface water improvement and management (SWIM) plan-update for Lake Okeechobee. South Florida Water Management District (SFWMD), West Palm Beach, FL. Stewart, I. T., D. Cayan, and M. Dettinger (2005), Changes toward earlier streamflow timing across western North America, J. Clim. 18(8), 1136-1155. Stringfield, V. T. (1966), Artesian water in Tertiary limestone in the Southeastern States Geological Survey Professional Paper U.S. Government Printing Office, Washington, D.C. Sullivan, D. G., and H. L. Batten (2007), Little River Experimental Watershed, Tifton, Georgia, United States: A hi storical geographic database of conservation practice implementation, Water Resour. Res. 43. Tilman, D., J. Fargione, B. Wolff, C. D'Ant onio, A. Dobson, R. Ho warth, D. Schindler, W. H. Schlesinger, D. Simberloff, and D. Swackhamer (2001), Forecasting agriculturally driven glo bal environmental change, Science 292 (5515), 281-284. Torrence, C., and G. P. Co mpo (1998), A practical guide to wavelet analysis, Bull. Am. Met. Soc. 79(1), 61-78. Trenberth, K. E. (1997), The Definition of El Nio, Bull. Am. Met. Soc. 78(12), 27712777. Tsonis, A. A., and K. Swanson (2009), Has the climate recently shifted?, Geophys. Res. Lett., 36, doi:doi:10. 1029/2008GL037022. Tudhope, A. W., C. P. Chilcott, M. T. McCulloch E. R. Cook, J. Chapp ell, R. M. Ellam, D. W. Lea, J. M. Lough, and G. B. Shimmi eld (2001), Variability in the El NioSouthern Oscillation Through a Gl acial-Interglacial Cycle, Science 291 (5508), 1511-1517. 263

PAGE 264

U.S. EPA (2008), State adoption of Numeric Nutrient Standards: 1998-2008 U.S. Environmental Protection Agency, Office of Water, Washi ngton, D.C., USA. [online] Available from: http://www. epa.gov/waterscience/criteria/nutrient/ files/report1998-2008.pdf U.S. EPA (1996), Environmental indicators of wa ter quality in the United States U.S. Government Printing Office, Washington, D.C., USA. U.S. EPA (1998), National Strategy for the Developm ent of Regional Nutrient Criteria U.S. Environmental Protecti on Agency, Office of Water, Washington, D.C., USA. UN (2006), World Water Development Report 2: Water, a Shared Responsibility UNESCO, Paris, France. USDA (1999), Census of Agriculture National Agricultural Statistics Service, Montgomery, Alabama. USDA (1972), National Engineering Handbook, Hydrology Section National Engineering Handbook, Soil Conservation Service. USDA-ARS (2005), Conservation Effects Assessment Project: The Watershed Assessment Study, USDA Agricultural Research Service, Washington, D.C. Van Liew, M. W., T. L. Veith, D. D. Bosch, and J. G. Arnol d (2007), Suitability of SWAT for the conservation effects assessment pr oject: Comparison on USDA agricultural research service watersheds, J. Hydrol. Eng. 12, 173. Veith, T. L., M. L. Wolfe, and C. D. Heatwole (2003), Optimization procedure for cost effective BMP placement at a watershed scale, J. Am. Water Resour. Assoc. 39(6), 1331-1343. Wang, B. (1995), Interdecadal changes in El Nio onset in the last four decades, J. Clim. 8 267-285. Wang, C. (2001), A unified osci llator model for the El NioSouthern Oscillation, J. Clim. 14(1), 98-115. Wang, Y. (1996), Temporal structure of the Southern Oscillat ion as revealed by waveform and wavelet analysis, J. Clim., (9), 1586-1598. Westerling, A. L., H. G. Hidalgo, D. Cayan, and T. W. Swetnam (2006), Warming and earlier spring increase Western U. S. forest wildfire activity, Science 313 (5789), 940-943. White, K. L., and I. Chaubey (200 5), Sensitivity analysis, calibration, and validations for a multisite and multivariable SWAT model, J. Am. Water Resour. Assoc. 41(5), 1077. 264

PAGE 265

Whiting, J. P., M. F. Lam bert, A. V. Metcalfe, P. T. Adamson, S. W. Franks, and G. Kuczera (2004), Relationships between the El-Nio Southern Osc illation and spate flows in southern Africa and Australia, [online] Available from: http://halsde.archives-ouvertes.fr/hal00304986/ (Accessed 12 January 2010) Wilks, D. S. (1995), Statistical methods in the atmospheric sciences, Academic Press, San Diego, CA. Wolter, K., and M. S. Timlin (1993), Monitoring ENSO in COADS with a seasonally adjusted principal component index, in Proc. 17th Climate Diagnostics Workshop pp. 52. World Meteorological Organization (1966), Climatic Change, Technical Note No. 79, WMO, Geneva, Switzerland. Yuan, T. L. (1965), A survey of alum inum status in Florida soils, in Proc. Soil Crop Sci. Soc. Fla vol. 25, pp. 143-152. Zebiak, S. E., and M. A. Cane (1987), A model El Nio-Southern Oscillation, Mon. Weather Rev. 115 2262-2278. Zhong, C., and T. Yang (2007), Wavelet analysis algorithm for synthetic pulsar time, Chin. Astron. Astrophs. 31(4), 443-454. Zorn, M. R., and P. R. Waylen (1997), Seasonal response of mean monthly streamflow to El Nio/Southern Oscillati on in north central Florida, Prof. Geog. 49(1), 51-62. 265

PAGE 266

BIOGRAPHICAL SKETCH At the old age of 30, Victor ia Keener is finally done with her dissertation. Despite the many stressful nights, mornings, a fternoons, and evenings, she made it through mostly intact, minus an Anterior Cruciate Li gament in her left knee, with the help of her feline sidekicks Bun-Bun and Frey, and her superhero husband Keith. Victoria grew up in Alexandria, Virginia, a suburb of Washington, D.C., and received her B.S. in Bioengineering from Rice University in Houston, Texas, in 2002. After working in a genetics lab for two years, she decided that animal research was not for her, and went back to school to get a Ph.D. in Agricultural and Biological Engineering at the University of Florida. Her current ex perimental organism is The Earth, which suffers less acutely than a mouse, yet is much more diffi cult to control for an experiment. In her spare time, Victoria enjoys numer ous non-mathematically centered hobbies, including gardening, cooking, rocking out, and hiking. She is ready to again embrace the seasons after living in the s outh for the last eleven years. 266