<%BANNER%>

Change Detection and the Application of Spectroscopic Techniques in the Sediments of Lake Okeechboee, FL

Permanent Link: http://ufdc.ufl.edu/UFE0024077/00001

Material Information

Title: Change Detection and the Application of Spectroscopic Techniques in the Sediments of Lake Okeechboee, FL
Physical Description: 1 online resource (144 p.)
Language: english
Creator: Vogel, William
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2008

Subjects

Subjects / Keywords: change, chemometrics, concentration, detection, distribution, lake, mud, okeechobee, phosphorus, rpd, sediment, spatial, spectroscopy, structure
Soil and Water Science -- Dissertations, Academic -- UF
Genre: Soil and Water Science thesis, M.S.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: Spatial and temporal changes in sediment chemical concentrations and their spatial structure were investigated, as well as changes in mud sediment distribution and depth. Data from 2006 was compared against data from previous surveys in 1988 and 1998. Average mud depth has declined in Lake Okeechobee since 1988, while increasing in area and decreasing in total volume. The largest changes were observed between 1998 and 2006, a period of time when several hurricanes passed over the lake. Changes in sediment chemical parameters were also seen over time, with Ca, Mg, and Fe increasing in concentration. P increased between 1988 and 1998, and then decreased between 1998 and 2006. For most analytes, change signals (p values) were more strongly significant when considering sites whose basic substrate type had not changed between surveys, suggesting that this should be taken into account when monitoring for change in Okeechobee's sediments. Spatial structure of most sediment chemistry properties was in the moderate to low range, and spatial structure of change between years was usually low. Also investigated was the use of chemometrics to analyze samples from the lake and augment mapping efforts. Spectral models constructed from sediment spectral absorbance patterns in the visible and near-infrared part of the spectrum were shown to be useful in the construction of predictive models for sediment properties. Mg, Ca, TN, TC, LOI were well-predicted, and Fe, Al, TP, HCl-P, and KCl-P were moderately predicted. SRP was poorly predicted by spectral models, as were all porewater properties. Scanning the sediments wet resulted in only a 7% decrease in average RPD score (the ratio of population standard deviation to the root mean square error of the predictive model), suggesting that wet scanning in the field may be a viable option for faster and less costly data acquisition. Interpolation maps based on values predicted from wet scans appear to capture the same spatial trends and patterns for Ca, Mg, TC, TN, and TP that maps derived from wet chemistry do, suggesting that the accuracy afforded by spectral methods may be sufficient for use in that type of application. The RMSE for spectral models compared favorably with the RMSE from kriging models based on wet chemistry derived values: 5.8 spectral to 11.2 g/kg kriging for TN, 239 to 226 mg/kg for TP, 9780 to 17030 mg/kg for Mg, 12.3 to 26.7% for loss-on-ignition, and 26320 to 55640 mg/kg for Ca. These results suggest that higher density spatial sampling using spectroscopy could potentially increase the accuracy of interpolation maps for some sediment analytes.
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 William Vogel.
Thesis: Thesis (M.S.)--University of Florida, 2008.
Local: Adviser: Osborne, Todd Z.
Local: Co-adviser: Cohen, Matthew.
Electronic Access: RESTRICTED TO UF STUDENTS, STAFF, FACULTY, AND ON-CAMPUS USE UNTIL 2009-06-30

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2008
System ID: UFE0024077:00001

Permanent Link: http://ufdc.ufl.edu/UFE0024077/00001

Material Information

Title: Change Detection and the Application of Spectroscopic Techniques in the Sediments of Lake Okeechboee, FL
Physical Description: 1 online resource (144 p.)
Language: english
Creator: Vogel, William
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2008

Subjects

Subjects / Keywords: change, chemometrics, concentration, detection, distribution, lake, mud, okeechobee, phosphorus, rpd, sediment, spatial, spectroscopy, structure
Soil and Water Science -- Dissertations, Academic -- UF
Genre: Soil and Water Science thesis, M.S.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: Spatial and temporal changes in sediment chemical concentrations and their spatial structure were investigated, as well as changes in mud sediment distribution and depth. Data from 2006 was compared against data from previous surveys in 1988 and 1998. Average mud depth has declined in Lake Okeechobee since 1988, while increasing in area and decreasing in total volume. The largest changes were observed between 1998 and 2006, a period of time when several hurricanes passed over the lake. Changes in sediment chemical parameters were also seen over time, with Ca, Mg, and Fe increasing in concentration. P increased between 1988 and 1998, and then decreased between 1998 and 2006. For most analytes, change signals (p values) were more strongly significant when considering sites whose basic substrate type had not changed between surveys, suggesting that this should be taken into account when monitoring for change in Okeechobee's sediments. Spatial structure of most sediment chemistry properties was in the moderate to low range, and spatial structure of change between years was usually low. Also investigated was the use of chemometrics to analyze samples from the lake and augment mapping efforts. Spectral models constructed from sediment spectral absorbance patterns in the visible and near-infrared part of the spectrum were shown to be useful in the construction of predictive models for sediment properties. Mg, Ca, TN, TC, LOI were well-predicted, and Fe, Al, TP, HCl-P, and KCl-P were moderately predicted. SRP was poorly predicted by spectral models, as were all porewater properties. Scanning the sediments wet resulted in only a 7% decrease in average RPD score (the ratio of population standard deviation to the root mean square error of the predictive model), suggesting that wet scanning in the field may be a viable option for faster and less costly data acquisition. Interpolation maps based on values predicted from wet scans appear to capture the same spatial trends and patterns for Ca, Mg, TC, TN, and TP that maps derived from wet chemistry do, suggesting that the accuracy afforded by spectral methods may be sufficient for use in that type of application. The RMSE for spectral models compared favorably with the RMSE from kriging models based on wet chemistry derived values: 5.8 spectral to 11.2 g/kg kriging for TN, 239 to 226 mg/kg for TP, 9780 to 17030 mg/kg for Mg, 12.3 to 26.7% for loss-on-ignition, and 26320 to 55640 mg/kg for Ca. These results suggest that higher density spatial sampling using spectroscopy could potentially increase the accuracy of interpolation maps for some sediment analytes.
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 William Vogel.
Thesis: Thesis (M.S.)--University of Florida, 2008.
Local: Adviser: Osborne, Todd Z.
Local: Co-adviser: Cohen, Matthew.
Electronic Access: RESTRICTED TO UF STUDENTS, STAFF, FACULTY, AND ON-CAMPUS USE UNTIL 2009-06-30

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2008
System ID: UFE0024077:00001


This item has the following downloads:


Full Text

PAGE 1

1 CHANGE DETECTION AND THE APPLICATION OF SPECTROSCOPIC TECHNIQUES IN THE SEDIMENTS OF LAKE OKEECHOBEE, FL By WILLIAM JUSTIN VOGEL A THESIS PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLOR IDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF SCIENCE UNIVERSITY OF FLORIDA 2008

PAGE 2

2 2008 W. Justin Vogel

PAGE 3

3 To my parents, because my successes are also thei rs and because they represent that which is and has always been most important to me my family.

PAGE 4

4 ACKNOWLEDGMENTS First and forem ost, I thank my parents for their love, support, and encouragement to succeed, both academically and in other arenas of life. An unbroken line can be traced from their influence to my accomplishments, and I hope that wh en all is said and done that I will be worthy of their pride. I would like to acknowledge Yu Wang, Gavin Wilson, Shao-wei Yu-bao, Evan, Keith, and Richard for their help with laboratory work. I w ould like to thank Rex Ellis, Sanjay Lamsal, and Casey Schmidt for their help with navigating the stormy and treacherous waters of ArcGIS. Also, much thanks to Kevin Ratkus, Casey Schmidt, Monroe Neuman, and John Lindhoss for making the Dungeon a fun place to toil. I would like to thank my co-chair and co mmanding officer, Todd Osborne (aka Captn Rackham), for teaching a quiet physics major the nuts and bolts of getting environmental science done under tough field conditions, for making a true pirate out of me, and for never reporting me to school counselors for my drinking habits. Finally, I would like to thank my advisor, Matt Cohen, for his incredible patience and exemplary academic and scientific guidance, wit hout which this project would not have been completed. Id also like to thank him for giving me a chance to work with him in spite of my Smith & Wesson hat.

PAGE 5

5 TABLE OF CONTENTS page 0ACKNOWLEDGMENTS...............................................................................................................4 LIST OF TABLES................................................................................................................. ..........7 LIST OF FIGURES.........................................................................................................................8 ABSTRACT...................................................................................................................................12 CHAP TER 311 INTRODUCTION TO LAKE OKEECHOBEE AND STUDY GOALS.............................. 14 1Phosphorus in Lake Okeechobee............................................................................................ 16 1Hurricane Activity............................................................................................................. .....21 1Visible/Near-Infrared Spectroscopy.......................................................................................22 1This Study...............................................................................................................................23 42 SPATIAL AND TEMPORAL CHANGE DETE C TION IN LAKE OKEECHOBEE, FL...25 1Introduction................................................................................................................... ..........25 1Methods..................................................................................................................................30 Statistical Methods.......................................................................................................... 32 2Geostatistical Methods....................................................................................................33 1Results.....................................................................................................................................34 Change Detection............................................................................................................ 34 2Geostatistics.....................................................................................................................35 1Discussion...............................................................................................................................36 1Conclusion..............................................................................................................................42 53 SPECTRAL DETERMINATION OF SEDIMENT PROPERTIES ......................................62 1Introduction................................................................................................................... ..........62 Hypothesis 1:.................................................................................................................. .65 Hypothesis 2:.................................................................................................................. .65 Hypothesis 3:.................................................................................................................. .66 2Methods..................................................................................................................................66 2Sample Procurement and Preparation............................................................................. 66 2Data Analysis...................................................................................................................68 2Results.....................................................................................................................................70 2Discussion...............................................................................................................................70 2Conclusions.............................................................................................................................74 64 PROJECT REVIEW AND SYNTHESIS ...............................................................................92

PAGE 6

6 7APPENDIX....................................................................................................................................95 8LIST OF REFERENCES.............................................................................................................141 9BIOGRAPHICAL SKETCH.......................................................................................................144

PAGE 7

7 LIST OF TABLES Table page 2-1 Mud area and volume for surveys in 1988, 1998, and 2006. .............................................43 2-2 The p values for dependent t-tests by anal yte, and summary of di rection of change. ....... 43 2-3 Geostatistics for each year and each pair of years. ............................................................ 44 3-1 Sediment and porewater pr operties based on standard m ethods of analysis (USEPA, 1993)..................................................................................................................................76 3-2 Sediment attribute spectral re sults using scans of dry sam ples......................................... 77 3-3 Sediment attribute spectral resu lts using scans of wet sam ples......................................... 77 1-4 Porewater attribute spectral re sults using scans of dry sam ples........................................ 78 3-5 Porewater attribute spectral re sults using scans of wet sam ples........................................78 3-6 Spectral results for 5 analytes based on spatial selection of the validation set. ................. 79

PAGE 8

8 LIST OF FIGURES Figure page 1-1 Lake Okeechobee and its hydrologic input s and outputs. Red circles indicate established sam ple stations................................................................................................ 24 2-1 Map of lake and sample site s, with sedim ent types from 1988 and 1998.......................... 45 2-2 Principal components anal ysis results for the 3 majo r substrate classifications. Separation between classes indicates that the classification schem e has validity............. 45 2-3 Interpolations of mud depth from 1 988, 1998, and 2006 studies. All units in cm ............ 46 2-4 Interpolations of TP from 1988, 1998, and 2006 studies. .................................................47 2-5 Interpolations of Ca from 1988, 1998, and 2006 studies. ................................................. 48 2-6 Interpolations of Fe from 1988, 1998, and 2006 studies. ................................................. 49 2-7 Interpolations of Fe change between 1988-1998 and 1998-2006 studies. ....................... 50 2-8 Interpolations of Mg from 1988, 1998, and 2006 studies. ................................................ 51 2-9 Interpolations of porewater SRP from 1988, 1998, and 2006 studies. ............................ 52 2-10 Interpolations of Ca change between 1988-1998 and 1998-2006 studies. ...................... 53 2-11 Interpolations of Mg change between 1988-1998 and 1998-2006 studies. ..................... 54 2-12 Interpolations of TN fr om 1988, 1998, and 2006 studies. ............................................... 55 2-13 Interpolations of HCl-P in 1998 and 2006, and change between those years. .................. 56 2-14 Interpolations of TC in 1998 and 2006, and changes between those studies. ................. 57 2-15 Interpolations of porewater SRP change between 1988-1998 and 1998-2006 studies. ............................................................................................................................................58 2-16 Interpolations of mud depth cha nge between 1988-1998 and 1998-2006 studies. ............ 59 2-17 Interpolations of TP change between 1988-1998 and 1998-2006 studies. ........................ 60 2-18 Interpolations of TN change between 1988-1998 and 1998-2006 studies. ....................... 61 3-1 Map of sampling sites from 2006 sa mpling effort used in study .......................................79 3-2 Spectrally predicted vs. observed total P for hold out validation sam ples (outliers removed) based on spectra from dry samples.................................................................... 80

PAGE 9

9 3-3 Spectrally predicted vs. observed total P for hold out validation sam ples (outliers removed) based on spectra from wet samples................................................................... 81 3-4 Spectral TP scatterplot of predicted vs observed values from wet scans, including outliers................................................................................................................................82 3-5 Spectral TP scatterplot of predicted vs observed values from dry scans, including outliers................................................................................................................................82 3-6 Spectral TN scatterplot of predicted vs observed values from dry scans.......................... 83 3-7 Spectral TN scatterplot of predicte d vs observed values from wet scans.......................... 83 3-13 Comparison of dry vs wet scan RPD for sedim ent spectral results. All units in %; anything above 100% indicates that dry scanning performed better than wet scanning. Anything below 100% indicates that wet scanning performed better............... 84 3-14 Flowchart of methodol ogy for chemom etrics.................................................................... 85 3-15 Comparison of cost for different an alysis m ethods. All units in dollars........................... 86 3-8 Maps of observed Ca, predicted Ca, e rror between predicted and observed, and the standard error of prediction for the sam e spatial region based on the lakewide wet chemistry-derived values. ................................................................................................. 87 3-9 Maps of observed LOI, predicted LOI, error between predicted and observed, and the standard error of prediction m ap for the same spatial region based on the lakewide traditionally-deriv ed values...............................................................................................88 3-10 Maps of observed TP, predicted TP, e rror between predicted and observed, and the standard error of prediction m ap for the same spatial region based on the lakewide wet chemistry-derived values. .......................................................................................... 89 3-11 Maps of observed TN, predicted TN, e rror between predicted and observed, and the standard error of prediction m ap for the same spatial region based on the lakewide wet chemistry-derived values. .......................................................................................... 90 3-12 Maps of observed Mg, predicted Mg, e rror between predicted and observed, and the standard error of prediction m ap for the same spatial region based on the lakewide wet chemistry-derived values............................................................................................ 91 A-1 1988 Ca distribution...........................................................................................................95 A-2 1998 Ca distribution...........................................................................................................96 A-3 2006 Ca distribution. .........................................................................................................97 A-4 Change in Ca distribution between 1988 and 1998. .........................................................98

PAGE 10

10 A-5 Change in Ca distribution between 1998 and 2006. .........................................................99 A-6 1988 Fe distribution. .......................................................................................................100 A-7 1998 Fe distribution. .......................................................................................................101 A-8 2006 Fe distribution. .......................................................................................................102 A-9 Change in Fe distribution between 1988 and 1998. ........................................................103 A-10 Change in Fe distribution between 1998 and 2006. ........................................................104 A-11 1998 HCl-P distribution. ................................................................................................. 105 A-12 2006 HCl-P distribution. ................................................................................................. 106 A-13 Change in HCl-P distribution between 1998 and 2006. .................................................107 A-14 1988 Mg distribution. ......................................................................................................108 A-15 1998 Mg distribution. ......................................................................................................109 A-16. 2006 Mg distribution. .......................................................................................................110 A-17 Change in Mg distri bution between 1998 and 2006. ..................................................... 111 A-18 Change in Mg distri bution between 1998 and 2006. ..................................................... 112 A-19 1988 Mud depth...............................................................................................................113 A-20 1998 Mud depth...............................................................................................................114 A-21 2006 Mud depth...............................................................................................................115 A-22 Change in mud depth between 1988 and 2006................................................................116 A-23 Change in mud depth between 1998 and 2006................................................................117 A-24 1988 porewater SRP distribution.....................................................................................118 A-25 1998 porewater SRP distribution.....................................................................................119 A-26 2006 porewater SRP distribution.....................................................................................120 A-27 Change in porewater SRP distribution between 1988 and 1998.. .................................... 121 A-28 Change in porewater SRP distribution between 1998 and 2006.. .................................... 122 A-29 1998 TC distribution........................................................................................................123

PAGE 11

11 A-30 2006 TC distribution........................................................................................................124 A-31 Change in TC distribution between 1998 and 2006.........................................................125 A-32 1988 TKN distribution.....................................................................................................126 A-33 1998 TN distribution. ......................................................................................................127 A-34 2006 TN distribution........................................................................................................128 A-35 Change in TKN/TN distribution between 1988 and 1998............................................... 129 A-36 Change in TN distribution between 1998 and 2006........................................................130 A-37 1988 TP distribution. ......................................................................................................131 A-38 1998 TP distribution. ......................................................................................................132 A-39 2006 TP distribution. ......................................................................................................133 A-40 Change in TP distribution between 1988 and 1998. .......................................................134 A-41 Change in TP distribution between 1998 and 2006. .......................................................135 A-42 Location of spectral/kr iging m apping area for TP........................................................... 136 A-43 Location of spectral/kr iging m apping area for Ca........................................................... 137 A-44 Location of spectral/kr iging m apping area for LOI......................................................... 138 A-45 Location of spectral/kr iging m apping area for Mg.......................................................... 139 A-46 Location of spectral/kr iging m apping area for TN.......................................................... 140

PAGE 12

12 Abstract of Thesis Presen ted to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Master of Science CHANGE DETECTION AND THE APPLICATION OF SPECTROSCOPIC TECHNIQUES IN THE SEDIMENTS OF LAKE OKEECHOBEE, FL By William Justin Vogel December 2008 Chair: Todd Z. Osborne Cochair: Matthew J. Cohen Major: Soil and Water Science Spatial and temporal changes in sediment chemical concentrations and their spatial structure were investigated, as well as changes in mud sedime nt distribution and depth. Data from 2006 was compared against data from previous surveys in 1988 and 1998. Average mud depth has declined in Lake Okeechobee since 1988, while increasing in area and decreasing in total volume. The largest changes were observe d between 1998 and 2006, a period of time when several hurricanes passed over the lake. Changes in sediment chemical parameters were also seen over time, with Ca, Mg, and Fe increasing in concentration. P increased between 1988 and 1998, and then decreased between 1998 and 2006. For most analytes, change signals (p values) were more strongly significant when considering site s whose basic substrate type had not changed between surveys, suggesting that this should be taken into acc ount when monitoring for change in Okeechobees sediments. Spatial structure of most sediment chemistry properties was in the moderate to low range, and spatial structure of change between y ears was usually low. Also investigated was the use of chemometri cs to analyze samples from the lake and augment mapping efforts. Spectral models constructed from sediment spectral absorbance patterns in the visible and near-i nfrared part of the spectrum we re shown to be useful in the

PAGE 13

13 construction of predictive models for sedime nt properties. Mg, Ca, TN, TC, LOI were wellpredicted, and Fe, Al, TP, HCl-P, and KCl-P were moderately predicted. SRP was poorly predicted by spectral models, as were all po rewater properties. Sca nning the sediments wet resulted in only a 7% decrease in average RPD score (the ratio of populat ion standard deviation to the root mean square error of the predictive model) suggesting that wet scanning in the field may be a viable option for faster and less cost ly data acquisition. Interpolation maps based on values predicted from wet scans appear to capture the same spatial trends and patterns for Ca, Mg, TC, TN, and TP that maps derived from wet chemistry do, suggesting that the accuracy afforded by spectral methods may be sufficient for use in that type of a pplication. The RMSE for spectral models compared favorably with the RMSE from kriging models based on wet chemistry derived values: 5.8 spectral to 11.2 g/kg kriging for TN, 239 to 226 mg/kg for TP, 9780 to 17030 mg/kg for Mg, 12.3 to 26.7% for lo ss-on-ignition, and 26320 to 55640 mg/kg for Ca. These results suggest that higher density spatial sampling using spectroscopy could potentially increase the accuracy of interpolation maps fo r some sediment analytes.

PAGE 14

14 CHAPTER 1 INTRODUCTION TO LAKE OKEECHOBEE AND STUDY GOALS Lake Okeechobee is a large, shallo w, sub-trop ical lake located in south Florida (Figure 1-1). It has a surface area of ~1730 km2 and an average depth of 2.7 m. The lake is of major importance to the state of Florida. It supp lies drinking water to surrounding communities and irrigation water to agricultural operations to the s outh, as well as providing water that eventually enters into the Everglades (James et al., 2006; South Florida Water Ma nagement District, 2002). It acts as a large store of water for the state, and an auxiliary supply for the large urban population to the southeast. It is a popular recr eational site, and supports economically important recreational and commercial fish ing activities (James et al. 2006) It is important habitat for much native flora and fauna, including alligators, bass and other sport fish, and many species of wading bird and water fowl (Havens and Ga wlik, 2005; South Florida Water Management District, 2002). Lake Okeechobee has been dramatically affect ed by human activities in its recent history. In response to the loss of life caused by flooding associated with two hur ricanes in the early 20th century, construction on the Herbert Hoover Dike began in the 1930s by the Army Corps of Engineers, and continued over the next 30 year s (South Florida Water Management District, 2002). This resulted in a marked re duction in the size and depth of the lake, and affected the flow of water southward out of it. In the years after the dike was built, agricultural development began to rapidly expand in the area, with dairy a nd cattle operations occurring to the north and sugarcane being grown on the ric h, drained organic soils to the south. Runoff from farmlands to the north whose soils had accumulated high levels of nutrients from animal waste and fertilizers resulted in elevated loading of nitrogen (N) and phosphorus (P) into the lake, affecting the trophic status and ecology of this important reso urce. The main ecological issues facing Lake

PAGE 15

15 Okeechobee now are elevated P loads, high and low extremes in water levels, and the encroachment of invasive plants in its littora l region, including exotic s such as torpedograss ( Panicum repens ) and melaleuca ( Melaleuca quinquenervia ) as well as the native cattail ( Typha sp.), which is spreading in the lakes shallow littor al zones due to its ability to take advantage of high nutrient levels. In 2001, in response to eutrophication c oncerns, the Florida Department of Environmental Protection established a total maxi mum daily load (TMDL) goal for the lake of 140 metric tons of P per year, w ith the intent of bringing P con centrations in the water of the pelagic (off-shore) zone to 40 ppb (FL DEP, 2001). In the period between 1995 and 2000, the average P loading rate into the lake was 641 metric tons (mtons) per year (FL DEP, 2001); between 2001 and 2005, the average loading was 580 mton/yr and the average water column P was 142 ppb, far above the goal of 40 ppb (James et al. 2006). There are conc erns that the large amount of P stored in the sediments may provide internal loading that will maintain high water column P concentrations even if these extern al loads are significantly reduced (South Florida Water Management District, 2002). Pursuant to addressing this concern, as well as collecting general limnological data on the lake, the South Florida Water Management District (SFWMD) arranged for studies on Lake Okeechobees sedi ments to be performed in 1988 and 1998. These studies focused on the mapping of many physical a nd chemical characteristics of the sediments on a lake wide basis, including nutrient and ma jor cation concentratio ns, P fractionation, bulk density, and substrate type. Between 2005-2006 the lake received direct hi ts from three hurricanes Jeanne and Francis in September 2004 and Wilma in Octobe r 2005 raising concerns that its P-rich sediments may have been impacted by wind-induced mixing. Both in response to this and also as

PAGE 16

16 part of its continuing monitoring efforts, the SFWMD decided to conduct a third survey to gauge the status of the sediments. The research descri bed herein was conducted as a part of the most recent survey. With the Lake Okeechobee Protection Act of 2000 calling for the TMDLs to be met by 2015, there is legal and political pressure on mana ging entities to address the nutrient problems within the lake. While the source of the troublesome P lies outside the lake, the fact that its sediments act as both a large store of P and as a large source of internal loading means that an understanding of the physical and chemical propert ies of the sediments, as well as changes in these properties, is important for understanding the behavior of P in this large system. The importance of the sediments becomes even clearer when considered in the context of evidence that their capacity to sorb P and act as a sink ma y be impaired as they reach retention capacity (James et al. 2006). This could result in a larger fraction of the external P load remaining in the water column due to saturation of the sediments, increasing the risk of harmful algal blooms (Havens and Gawliks, 2005). Phosphorus in Lake Okeechobee High levels of P loading, both ex ternally and internally, have been identified as the m ajor cause of eutrophication of Lake Okeechobee (Reddy et al., 2002; South Florida Water Management District, 2002). In a paleolimnologi cal study on the lakes sediments, Brezonik and Engstrom (1998) found that yearly accumulation of P had increased by factor of four over the past century. This change in nutrient accumula tion has occurred concurrently with dramatic changes to the lakes watershed and in Okeechob ee itself. Unfortunately, detailed limnological data did not begin to be colle cted on the lake until around 197 0 (Brezonik and Engstrom, 1998), well after human impacts would have begun to be felt due to increased population, the engineering projects begun in th e 1930s, and intensification of agricultural activities in the

PAGE 17

17 watershed. This has made it difficult to chronicle th e changes occurring in the system or establish baseline values for water qual ity or sediment parameters. Much effort has been made to understand how P behaves in the system, especially in the sediments where much of it is stored and from wh ich it can subsequently be released back into the water column. There are two ma in thrusts of the research in this area: P flux, and P form and location. The latter is important because it is useful to know wher e in the lake P is accumulating, whether it is in a biologically available form, what differences exist between the P pools in different sediment types, and because it aids in constructing a P budget for the lake. Flux is important because it describes how P moves between various storage pools, how it gets from the sediments to the water column where it can be ut ilized by algae, and what effect various in-lake conditions may have on its movement and bi oavailability. The importance of P in the Okeechobee system warrants a discussion about its behavior in the lake. The lake sediments can be broadly categorized into several different varieties: sand, mud, peat/littoral, marl, and limestone (Moore et al. 1998). Each sedime nt type has different physical and chemical characteristics, and influences P storage and transport dynamics in a different manner. As most of the flux and concentration data that is used to construct lake-wide budgets is derived from small subsamples from the various sediment zones and then extrapolated over the estimated area of the substrate type, the spatial distribution of sediment s, as well as spatial gradients of the associated physical and chemical characteristics, is important for determining accurate estimates of internal nutrient loading, sediment stor age capacity, and total amounts of elements or compounds of interest. Accurate de termination of substrate extent and spatial structure, as well as analyte di stributions, is therefore important for management purposes, as is monitoring for changes in these qualities. Accumulation or redistribution of substrate due to

PAGE 18

18 events such as hurricanes has the potential to ch ange the nature of the mosaic of the sedimentwater interface throughout the lake basin, and thus how the two pools interact. While the sediments act as a net sink for tota l P due to detrital organic matter continually settling out of the water column, they act as a so urce of soluble reactive P (SRP), which is much more available for uptake by organisms (Reddy et al. 2002, Fischer et al 2005). For the mud sediments, Reddy et al. (2002) determined that within the top 56 cm, 83% of the P is Ca-Mg bound, and a further 12.4% is in residual organic form, both types being stable and generally unavailable to biota. This implies that only a small amount of the total P present is typically available for release back into the water column (though the Ca-Mg fraction could become available under certain circumstances, such as a decrease in the pH of lake waters). Even though the potentially-labile pool may be a small fraction of the total, it is dramatically affected by chemical conditions in the sediments and has the potential to release large amounts of soluble P into the water. Moore and Reddy (1994) studied the effect of Eh and pH on P availability in Lake Okeechobee sediments. Their study looked at SRP concentrations in th e porewater of mud samples which were subjected to a range of i nduced pH and Eh levels. They found that both parameters had an effect on P flux, especially redox potential. Fluxes were much higher under anaerobic conditions than unde r aerobic ones: 2.78 mg P m-2 d-1 compared to 0.04 mg P m-2 d-1. They determined that under oxidized conditions, such as exist in the usually aerobic water column and at the sediment-water interface, Fe reactions controlled P behavior. Within a few centimeters of the surface, reducing anaerobic cond itions are present in the sediment and Fe is utilized by microorganisms as an electron acceptor, causing the breakdown of Fe-P minerals and therefore resulting in higher levels of soluble phosphorus in deep er regions of the sediment. As

PAGE 19

19 the SRP migrates into higher, oxidize d regions, it binds with available Fe3+ and its concentration in the water column is thus regulated. Microbial action is not the onl y explanation put forward for increases in flux rate under anaerobiosis, however. In another study, Moore et al. (1998) suggested that the response of P fluxes to changes in O2 concentrations in experiments was too rapid to be attributed to microbial utilization of Fe for respiration, and instead pos tulated that abiotic m echanisms may be more important, with amorphous Fe-phosphate minera ls precipitating out of solution under aerobic conditions. They found that peat sediments, which are comparatively low in Fe, did not have the same clear reaction to redox stat us that other sediment types displayed with P flux increasing markedly under anoxic conditions. There was also a notable lack of correlation between P flux and dissolved reactive P (DRP) in the porewater pr ofiles of the sediment cores, further indicating that redox conditions are a major controller of P flux in those sediments containing significant amounts of iron. The mud sediments have been iden tified as having the highest concentrations of Fe (Fisher et al. 2005). Fisher et al. (2005) investigated internal loading of N and P from Lake Okeechobees sediments. Among their findings was that internal loading of labile forms of P and N exceeded external inputs; 1.09 vs 0.60 mg m-2 day-1 for inorganic P and 10.6 vs 2.07 mg m-2day-1 for inorganic N, respectively. Moore and Reddy (1994) found that SRP concentrations in the water column of intact sedime nt cores were 0.035 mg P L-1 under aerobic conditions, and over 0.45 mg P L-1 under anaerobic conditions over an order of magnitude larger, again highlighting the importance of redox status. Reddy et al. (2002) found an accumulation of P in the upper regions of mud sediments, with the top 10 cm having 33% more TP than the underlying 10 cm. This is attributed to the upward diffusi on of Fe-bound P from reducing to oxidizing regions, where the P

PAGE 20

20 re-binds with oxidized Fe. The fact that comp aratively bioavailable P is accumulating in the upper region of the sediment where resuspension by wind can occur is a cause of concern for lake managers. Removing all of the mud could actually worsen the situation, however. In the Reddy et al. (2002) study, experime nts were performed that were designed to simulate dredging at different depths and with di fferent water column P loading le vels. Results generally suggested that removing the top 30 cm of mud would have the best effect on internal loading, lowering both the amount of P released from the sediments and the equilibrium water column P concentration. They found that the sand sediment s underlying the mud had a lower capacity to sorb and retain P, indicating that completely sc ouring the bottom of mud c ould very likely result in higher concentrations of P in water column due to lack of ability to sorb, and thus mitigate, externally loaded P. P flux rates from the lake sediments are fairly modest when compared to other examples of eutrophic systems. Fisher et al (2005) measured a P flux rate from the mud zone of Okeechobee of 0.83 mg m-2day-1, a rate which they cite as comparativ ely low, given such rates as 2.7 mg m-2 day-1 for hypereutrophic Lake Apopka and 6.47 mg m-2 day-1for a nutrient-impacted site in the Everglades. The shallowness of the lake leaves it vulnerable to even low levels of P release such as this, however, due to the ease with which th e water column can be mixed. They also found that there was no statistically significant difference between flux rates found at various sampling sites in Okeechobee in 1999, and rates measured at the same sites ten years earlier. Much of the attention in the literature ha s been focused on the mud sediments of Lake Okeechobee; less has been given to describing the spatial structur e and extent of the various sediment types, which is related to the internal load and how nutrien ts are stored in the substrate. Temporal change in how P is stored is of cr itical importance for unders tanding how it might flux

PAGE 21

21 into the water column, and as substrate changes in space (due to hurricane mixing, for example) there is potential to also change the flux characteristics. Since P flux dynamics from the different substrates are known approximately, maps of the present sediment-water interface could be used to indirectly estimate fluxes and internal loading of P. Hurricane Activity Florida is often subjected to hurrican e st rikes, and on occasion Lake Okeechobee has received direct or glancing blows from thes e storms. In the period from 1990-1996, the six counties surrounding the lake had annual probabilit ies of seeing hurricane force winds (speeds >74 mph) ranging from 8-16% (F loridaDisaster.org). The wind action associated with these storm events can result in the resuspension of phosphorus-rich mud and flocculent material from the upper sediment layers into the water column; even less dramatic wind events can resuspend sediment, since the lake has such a shallow mean depth (2.7 m) and large fetch (Havens et al., 2001). The movement of the top sediment layers back into the water column is a concern because it has been found that they possess higher concentrations of P than the deeper sediments, and that larger fractions of th is pool are in more bioavailable forms (Brezonik and Engstrom, 1998; Moore and Reddy, 1994). Thus, major storm ev ents have the potential to interrupt the process of sequestration of P into the lake sedi ments and instead increase internal loading of P, impacting water quality and complicating mana gement of the lake (Havens et al., 2001). The effects of a hurricane on Lake Okeechobee have been previously observed. Havens et al. (2001) describe how Hurricane Irene in 1999 uprooted aquatic vegetation and suspended finer sediments in the water column. The resultin g turbid conditions made it difficult for aquatic vegetation to reestablish itself, and the lack of roots stabilizing the sediment made subsequent sediment resuspension even easier, creating a positive feedback loop that suppressed plant growth and allowed phytoplankton to rise in dominance. The study dealt with the shallow littoral

PAGE 22

22 regions where vegetation occurred; in the deeper central part of the lake where much of the Penriched mud is located, there is conflicting evidence about how much of an impact on the sediment a hurricane can have. Brezonik and Engstrom (1998) found evidence from 241Am activity that there had been no ill effect on the st ratigraphic record due to vertical mixing of the sediments by wind activity or benthos. Follow-up studies using such diverse methods as pollen analysis, radioisotopes, PCBs, and heavy metal concentrations have reached similar conclusions (Shottler and Engstrom 2006; Engstrom et al. 2006) If this is accurate, it could imply that resuspension of sediments by hurri canes is a somewhat rare event, or is at least limited in its severity. In 2002, Reddy et al. reported that the area of the lakebed covered by mud was in excess of 80,000 hectares, and that there was about 200 million m3 of the material in the lake. Since the mud has been identified as both a major store of P and a potential threat for drastically increased internal loading of P under certain conditions, the current state of its distribution within the lake and how this might be changing ove r time are important questions. Visible/Near-Infrared Spectroscopy The high cost and labor intensive m ethods involv ed with sampling a lake of such large size make it difficult to sample at spatially and temp orally dense levels. A possible means of reducing the cost and effort involved with sampling Lake Okeechobee is the use of visible/near-infrared spectroscopy. This is a field-portable, non-destru ctive, and multi-componen t technology that is used to correlate the wavelengths of light a substance absorbs over a certain region of the electromagnetic spectrum to its physical and chemical charac teristics through the use of statistical models. This process is referred to as chemometrics.and has been used to model sediment properties of a lake in Canada (Ma lley and Williams 1997), to predict water column

PAGE 23

23 chemical properties from sediment spectra (Nilsson et al. 1996), a nd in many cases of terrestrial and wetland soil analysis. This Study The aim s of the research extended along two lines. One was to look for spatial and temporal changes in sediment distribution and chemical properties over time. The other was to investigate the use of spectroscopy as a means of enabling cheaper and de nser sampling in order to augment the current monitoring efforts. The goals were to: Evaluate potential changes in sediment chemistry over time, especially P, and see if they a) are detectable, and b) match up with expectations. Quantify changes in mud distribution, as this par ticular material has been of special concern Quantify the spatial structure of sediment chemical parameters. Determine if spectroscopy can be used to accurately analyze Okeechobees sediments Compare dry vs wet spectral scanning results Compare mapping efforts based on tr aditional methods to those derived from spectral methods

PAGE 24

24 Figure 1-1. Lake Okeechobee a nd its hydrologic inputs and output s. Red circles indicate established sample stations.

PAGE 25

25 CHAPTER 2 SPATIAL AND TEMPORAL CHANGE DETECTION IN L AKE OKEECHOBEE, FL Introduction Lake Okeechobee is a large, shal low, sub-trop ical lake locate d in south Florida. It has a surface area of ~1730 km2 and an average depth of 2.7 m. It has been subjected to years of anthropogenic eutrophication, specifically high phos phorus (P) loads, to the point where water quality in the lake has been degraded. The importance of the la ke to the state of Florida has resulted in extensive efforts to reduce nutrient loads to it, maintain water quality at levels sufficient to meet designated us es (fishing and swimming, Class III), and attempt to restore ecosystem functions and services to more natural levels (DEP 2001). Lake Okeechobees sediments store significant quantiti es of P, acting as a significant internal load (Fisher et al. 2005). Of particular importance is the presence of P-enriched mud that covered approximately 40% of the lakebed in 1998 (Reddy et al. 2002). This flocculent mud has the potential to dramatically control water column P and light dynamics when entrained in the water column (Havens et al. 2001), a process that occurs regularly due to the sh allowness of the lake and its large fetch. Change detection. An important aspect of management is detection of change in key ecological attributes over relevant spatial and te mporal scales. For Lake Okeechobee, there is an institutional and public desire to restore the lake to a condi tion more resembling its predisturbance state, and to maintain or improve its values as a resource both economic values as a recreational asset and source of wa ter, as well as environmental values as a vital part of the hydrology and overall health of th e south Florida ecosystem. As tim e and money are invested in developing and implementing best management pr actices, setting total maximum daily loads, and other efforts intended to rest ore the health of the lake, the ability to detect changes in

PAGE 26

26 performance indicators (such as P concentration) becomes vital in order to objectively evaluate restoration success and potentially revise rest oration strategies (Hellawell and Holloway 1977). The ability to accurately detect change is al so important for forecasting lake conditions, as well as estimating the time and costs required to achieve management goals. Change detection provides quantitative evidence need ed to judge the effectiveness of expensive or difficult efforts to change land use practices in the watershed, set more stringent water qu ality requirements for inflows, or develop guidelines for lake stag e management (Parr et al. 2003). Whether the management goal is to restore the system or to sustain current conditions, the ability to detect change is required to justify a course of action, since stakeholde rs are unlikely to support and sustain burdensome measures in the absence of evidence of their effectiveness (Walter and Holling 1990). Detection of change is, therefore, an integral part of adaptive management, which seeks to treat management prac tices as experiments and learn systematically from observed effects of management actions in the system. It is also important for determining the impacts of non-anthropogenic stresses on the lake such as hurricanes and droughts. Change that is of restorat ion interest can occur in th e magnitude, rate, or spatial distribution of key attributes a nd processes. Which type is of most interest depends on the situation, and sometimes multiple parameters are e qually essential. Internal nutrient loading rate from the sediment is a key concern in Okeechobe e, but spatial distribution of mud is also important for two reasons: 1) mud has a direct association with high le vels of P accumulation (Reddy et al. 2002) and 2) because flocculent mud can retard or disrupt SAV growth by acting as a poor rooting medium or by reduc ing light penetration in the ev ent of resuspension (Havens et al. 2001). Also of interest is the required spat ial and temporal resolution necessary to detect

PAGE 27

27 changes in the Okeechobee sediment system, given the dual considerations of a desire for quality data and the high cost associated with monitoring a lake of this size. Research question 1 : Have mud sediments changed in depth, distribution or volume since previous surveys? Hypothesis 1-1: Total mud volume will have increased since 1988. Mud accumulation is an ongoing process that has been shown to be accelerating in the time since humans began to impact the lake (Brezonik and Engstrom 1998), so some increase between 1988 and 2006 should be expected. Reddy et al. (2002) reported that the total vol ume of mud in the lake of approximately 200 million m3. Hypothesis 1-2: Mean mud depth will be shallower in 2006 than in 1998 and spread over a larger area. In the absence of strong disturbing for ces, over time flocculent material such as mud should gradually settle and accumulate in the lowest-energy regions of the lake (i.e., the deeper and less easily disturbed parts). In the presence of a po werful enough perturbance, this process could be upset and the material transported to other regions of the lake basin. Since there was a finite amount of mud at the time of the hur ricanes, removal of material from the mud zone would result in shallower substrate depths as it is moved into other sediment areas. Reddy et al. (2002) reported that mud covered an area around 80,000 ha. Hypothesis 1-3: The differences in mud depth and distribution between the 1998 and 2006 studies will be larger than the differences seen between the 1988 and 1998 studies. Lake Okeechobee was impacted by a number of hurricanes between 1998 and 2006, several of them shortly before the current study was undertak en. Since there were a larger number of highenergy wind events affecting th e lake between the second and th ird studies, there is a greater possibility that the changes in mud depth and ex tent will be more pronounced due to these storm

PAGE 28

28 effects in the 1998-2006 time pe riod than the 1988-1998 period, when there was less hurricane activity. In addition to determining changes in the spatial extent and amount of the mud, changes in sediment chemistry are also important for unde rstanding nutrient dynamics and storage in the lake basin. Fe is of interest since Fephosphates are highly susceptible to dissolution and release into the water column under anaerobic, reducing conditions (Moore and Reddy 1994). Ca and Mg are important because the majority of P stored in the mud sedime nts are Ca-Mg bound (Reddy et al. 2002), and because these fractions are generally unavailable for biological uptake they may act as mitigating factors in the eutrophication issue facing Okeechobee. The distribution of sediment Mg is of added interest due to recent work s uggesting that the Mg (spe cifically palygorskite, a filamentous magnesium silicate mineral) is a substantial fraction of a su rprisingly large mineral fraction in the mud, a fraction that may be of allocthonous or igin (Harris et al. 2007) and therefore under watershed control. C and N are indicative of lake productivity and organic matter quality. Finally, P is of partic ular importance since it is the limiting nutrient and major pollutant in Lake Okeechobee. Data on sediment P concen trations may allow managers to estimate the amount of P being sequestered in the sediments over time, and determine whether preferential accumulation may be occurring in some parts of the lake basin. Certain chemical parameters have been associ ated with sediment types, such as P with mud (Reddy et al. 2002), and the sediment type s have been described as occurring in zones (Moore et al 1998, Fisher et al. 2005), suggesti ng some amount of spatial structure to the distribution of chemical concentr ations. This has not been quan tified for the lake, however. One way of quantifying spatial structure is the Q statistic defined as the ratio of the sill to the sum of

PAGE 29

29 the sill and the nugget of a semivariogram for an analyte (Lyons et al. 1998). The values for Q can range from 0 to 1 (or 0 to 100%, analogously), with 0 representing weak spatial structure and 100% strong spatial structure. Research Question 2: Has there been detectable change in the concentrations or spatial structure of sediment chem ical parameters since 1988? Hypothesis 2-1 TP, HCl-P (mineral P fraction) porewater SRP, TN, and TC concentrations will have increased in the sedi ments. Ca, Mg, Fe will not have experienced significant changes. The sediments have been shown to be a net sink for P, which is continuing to be loaded into Lake Okeechobee from human activities in its watershed, so it is expected that P concentrations will continue to increase in the lake over time. Nitrogen-fixing cyanobacteria have been able to take advantage of excess P in the water column and bloom more frequently (Havens and Gawlik 2005), and as this biomass se ttles out N should increa se in the sediments. Cyanobacteria, algae, and SAV/macrophyte detritus in the littoral zones s hould continue to load carbon into the sediments. In the absence of st rong external loads or high exports, the other analytes are not expected to change significantly. Hypothesis 2-2 TP, HCl-P, TN, TC, Fe, and mud depth will show strong spatial structure. Ca, Mg, and porewater SRP will show m oderate or weak spatial structure. Fe and P have been shown to be dependent on sediment type (Moor e et al. 1998), and so these quantities should vary in space along with the distribution of substrate. The mud has been observed largely in the pelagic zone in previous surveys and absent in other areas, leading to an expectation that it will continue to have strong spatial dependen ce with more mud occurring in the deeper central region and less on the lake frin ge. The patterns of mud should have associated

PAGE 30

30 effects on the spatial pattern of elemental constituents for which mud has distinct concentrations. Marl sediment areas should be higher in Ca, but with areas of exposed limestone bedrock acting as a source of Ca to the water of the lake and no obvious strong preference for substrate type, Ca is expected to have less spatial structure. Mg may or may not be associated with mud, depending on the muds origin. Porewater SRP concentratio ns are dependent on local conditions, and so while they might be expected to be correlated to sediment characteristics like sediment P in general, small spatial-scale differences in conditions may result in there being less of a dependence on substrate type (and hence substrate distribution). As such, the resulting spatial pattern is expected to be weak. Hypothesis 2-3 Changes in TP and HCl-P (1988 to 1998; 1998 to 2006) will show strong spatial structure. The change maps of the other analytes will show moderate spatial structure. P is expected to increase faster in the fine mud sediments with which P has been associated with in the past (Reddy et al. 2002 ). Since the mud is located in what could be described as a spatially-explicit area in the pelagic zone (Fisher et al. 2005), differential P change is expected to occur in line with sediment spatial distribution an d thus show strong spatia l structure. Porewater SRP may not be as spatially coherent due to dependence on local circumstances. Changes in other analytes may not be as strongly associated with sediment type, resulting in less spatiallydriven structure to changes. However, at leas t moderate structure is expected, on basis of difference in physico-chemical characteristics between substrate types alone. Methods Sites (n = 174) (Figure 2-1) were previously selected during lake surveys in 1988 and 1998. These were rev isited during the summer of 2006; sediment cores were successfully obtained at 156 stations on the la ke with the remainder either too dry to access or where the bottom substrate was rock. Duplicate cores were taken at 19 randomly ch osen sites from which

PAGE 31

31 short range variability could be determined. Intact cores were ta ken either by piston corer or by hand to the point of refusal using 10 cm polycarbonate tubes. Mud depths were measured in each core in the field using a meter stick; the bottom of the mud was defined as a sediment discontinuity with rock, marl or sand while the to p was defined visually after a settling period of several minutes. A minimum of 10 cm of site water was left standing above the surface of each core, and each was capped and kept on ice until transported to the lab, where it was refrigerated at 4 C until analysis. The cores were extruded inside of an N2-purged hood to maintain anaerobic conditions, and the surface sediment (0-10 cm) was sectioned for chemical analysis. Porewater was extracted by spinning subsamples fr om the core in airti ght centrifuge tubs at 10,000 RPM for 10 minutes; syringes were inserted through a rubber stopper in the cap of the tube in order to remove the porew ater without exposing it to air. Sediments from cores were analyzed for a suite of physical and chemical properties according to standard U.S Environmental Protection Agency methods (United States Environmental Protection Agency, 1993). Sediment was analyzed for total P (EPA 365.1), total C and N with a Carlo Erba NA-1500 CNS anal yzer, bulk density, HCl-P (EPA 365.1), HClextractable Ca, Mg, Fe, and Al (EPA 200.7), KCl-P (EPA 365.1), KCl-NH4-N (Mulvaney 1996), and NaHCO3-P (EPA 365.1). Porewater was analyzed for total P (EPA 365.1), soluble reactive P (EPA 365.1), total Kjeldahl N (EPA 351.2), NH4-N (Mulvaney 1996), pH (EPA 150.1), conductivity (EPA 120.1), as well as Ca, Mg, Fe, and Al (EPA 200.7). Subsamples were dried in an oven at 70 C for 3 days to determine mois ture content and bulk de nsity. These subsamples were then ground in a mortar and pestle and ball-milled until fine enough to pass through a 20 m sieve.

PAGE 32

32 Variable methodology or classification between surveys introduced several difficulties for the comparison of observations between dates. Samples were collected by piston corer in 1988, by SCUBA diver in 1998, and by pi ston corer again in this 2006 study. Substrate classification was similar but not identical between surveys, an d efforts were taken to standardize designations across years. This was further complicated by th e fact that multiple sediment types may be observed in one core; sites were classified base d on the sediment type dominating the top 10 cm of the core, both for this study a nd retroactively for the prior ones. Principal components analysis was used to test whether the se diment classifications that were used in 2006 were meaningful. Specifically, based on the full array of sedime nt chemical properties, PCA allowed the examination of whether sediment types (e.g., m ud) fell within the same ordination space. Another important difference in methodology was core sectioning in previous studies was done based on substrate type boundaries, with se parate chemical anal yses performed on each section. In 2006, only the proper ties of the bulk, homogenized t op 10 cm were determined. In those cases from previous studies where a s ubstrate break occurred before 10 cm and two analyses were performed, physical and chemical characteristics were averaged based on volume fraction to create a composite valu e equivalent to the top 10 cm. Statistical Methods. Differences between surveys were investigated using paired t-tests (w here pairs were the early and later observations for a particular site) for dependent samples using Statistica (Statistica 8.0, StatSoft, Inc.). Nine different sediment properties were chosen for statistical analysis: total P, N, C, Ca, Mg, and Fe, 1-M HC l-extractable P, porewater SRP, and mud depth. No sediment TC measurements existed for the 1988 data set, and only to tal Kjeldahl nitrogen and 0.5 M HCl-P extractions were available from that survey. TKN was deemed likely to be comparable to TN measurements due to lack of nitrate existing in the anaerobic conditions found

PAGE 33

33 throughout most of the sediment cores (M oore and Reddy, 1994). HCl-P for 1988 was not compared to other years since it was not clear wh ether differences in acid concentration would invalidate comparis ons between years. In addition to conducting t-tests on a lake-wide ba sis, tests were repeated on the subset of sites whose substrate type did not change be tween surveys. Given the markedly different characteristics of the various sediment types in the lake, sites whose sediment type changed between surveys, due to redeposition or high local heterogeneity, ma y confound detection of shifts in the lake sediment occurring due to change s in internal or external drivers. Shifts due to change in substrate are not necessarily less meani ngful when considering variability in general in the lake, but by only including sites whose subs trate remained unchanged between surveys can trends occurring across sediment types can be discerned. Since change comparisons were not done on a geographic or individual substrate basis, it should be noted that it is possibl e that there could be statistically significant changes in particular areas or sediments that are not being addressed or detected by the statistical tests being performed. Geostatistical Methods. The sam e sediment properties listed above we re chosen for geostatistical analysis and mapping. Concentrations were mapped for the three individual surveys, as were changes between site values for the 1988-1998 surveys and 1998-2006 surveys. Ordinary kriging was used for interpolation for all variables following second or der trend removal; standard lag distances (3500 m) and numbers of lags (10) were used for all an alytes to avoid differenc es due to interpolation model selection. To estimate mud volumes, the in terpolated depth of th e mud at each pixel was multiplied by pixel area, and then summed over the lake area.

PAGE 34

34 Spatial structure was measured by the statistic Q, as describe d in Lyons et al. (1998). Q is the ratio of the partial sill (the sill minus the nugget) to the sill, and is usually expressed as a percentage. The partial sill represents the amount of semivariance explained by the interpolation model; the sill is the total semivariance. As the part ial sill approaches the value of the sill (i.e., as the nugget value approaches zero), Q approaches 100% and indicates that the spatial variability of the analyte in question is explained strongl y by spatial structure. Conversely, as the nugget value increases, Q goes to zero and implicates factors other than sp atial distribution as responsible for the variability. Following the methods of Cambardella et al. (1994), three classes of spatial structure are de fined based on the Q value: 75-100% Strong spatial structure 25-75% Moderate spatial structure 0-25% Weak spatial structure Results Figure 2-2 shows the results of the principal components analysis be tween the three m ain substrate types: sand, peat, and mud. Separation between groups is evident, indicating that the classification scheme has validity. Table 2-1 shows the total mud area and volume for each year as calculated from the GIS maps. The total area covered by mud (irrespectiv e of depth) was stable between 1988 and 1998, but seemed to expand by 2006. This was accompanied by an apparent drop in total mud volume. Change Detection Significant changes in sediment properties were observed at both 10and 20-year intervals (Table 2-2). For some analyses, partic ularly P-related ones, statistically significant changes were seen only when considering sites w hose substrate type did not change between the surveys being compared. Change in sediment TP was dependent on continuity of substrate type, as well as time. When consideri ng all sites, there was no significant change in TP levels. When

PAGE 35

35 looking only at sites whose subs trate type had not changed be tween surveys, there were significant changes. Between 1988 and 1998 TP increase d at these sites at si gnificant levels (p = .047). Between 1998 and 2006, however, ther e was a significant (p = .027) decline in TP. The result of these fluctuations was that when considering the entire time period between 1988 and 2006, there was no significant change in TP. Porewa ter SRP showed the sa me increase/decrease pattern, and HCl-P also showed a significant decrease between 98/06, though da ta was lacking to see if it also showed an increase between 88/98. Geostatistics Table 2-3 summarizes the results for the spatial stru cture of sedim ent chemistry conditions and their changes between surveys. Figures 2-3 through 2-18 show interp olations of mud depth and analyte concentrations from the 1988, 1998, a nd 2006 studies, as well as changes between studies. Larger maps for individual years a nd changes between year s can be found in the Appendix (Figures A-1 through A-41). In general, spatial structure tended to be moderate to weak, with stronger Q scores being seen for individual years (mean struct ure 38%) than for maps of changes between survey s (mean structure 20%). A notable exception to this was mud depth. The mud sediments were deepest in the central pelagic region of the lake in all three surveys, and stretch out towards the eastern shore and northwards towards the Kissimmee River in a to ngue of material (Fig ure 2-3). A noteworthy aspect of the maps is the change in strength of spatial structure between the first two surveys and the one from 2006. For 1988 and 1998, there was str ong spatial structure in the mud distribution pattern, with sharp gradients in mud depth with distance from the center of the lake. In 2006, however, the mud was found to be much less deep (maximum depth 51 cm, compared to broad areas with >50 cm depth in 88/98), and the spa tial structure much weaker (by >30%). The mud appears to be more evenly distributed in the north/north east part of the lake than before, and the

PAGE 36

36 gradient in mud depth greatly reduced. The re duction of mud depth was corroborated by the ttests, which indicated that there were statis tically significant decreases between 88/98, 98/06, and 88/06 (Table 2-2). The changes in mud depth between 1988 and 1998 (Figure 2-16) had essentially no spatial structure according to the geostatistical value (Q = 0%), while the 1998 2006 change map, exhibited moderate spatial st ructure (Q = 46%) suggestive of systematic drivers of change, and not just measurement uncerta inty or local heterogeneity as is the case for 1988 to 1998. Change is dominated by decreases in m ud depth, especially in the center of lake where the deepest deposits are, though increased depths are seen in the north, west-central, and south. Fe concentrations were higher in the central and eastern parts of the lake (Figure 2-6), especially in the area near the central eastern shore, and cons istent with prior observations (Moore and Reddy 1994) seem to be associated to some degree with mud. Large concentrations of Ca were seen in the south-central area of the lake, and expo sure of the underlying limestone bedrock in this area may be the reason for this. Ca spatial structure (Figure 2-5) saw a pronounced increase in Q value from 47% to 71% between 1998 and 2006. TN (Figure 2-12) saw a marked decrease in spatial struct ure between 1988 and 1998, though no statistically significant changes in concentrati on between any of the studies. Lit tle spatial structure was seen for TC (Figure 2-14) across studies, perhaps beca use of generally high or ganic inputs into all sediments resulting from hi gh productivity in the lake. Discussion Changes seen in m ud distribution support the hypotheses about mud depth/area and the magnitude of the changes between surveys, thoug h they did not support the hypothesis that total volume would increase with time. Between 1998 and 2006, the depth was significantly shallower and the areal extent of mud increased by 18%. Gi ven the annual rates of sediment accumulation

PAGE 37

37 in the mud zone reported by Brezonik and Engstrom (1998) of ~600 g m-2 and the conventional wisdom that it has been cons tantly accumulating since anth ropogenic eutrophication began, the apparent large decline in total mud volume is somewhat surprising. Resuspension due to intense wind action could move fine-grained material in to the water column, potentially leading to an increase in export of total susp ended solids and acting as a possi ble explanation for loss of mud. Resuspension of deeper mud sediments could al so expose organic matter typically residing in anaerobic conditions to aerobic ones, potentially resulting in oxidation and loss of material. Another possibility is compaction of the mud sediments, as this could decrease the volume while maintaining (or even increasing) the total mass. Th e density profile of the entire mud column at each site would be required to determine whethe r there has been a net loss of mud mass over time, however. The elongated formation of mud seen extending northwards in all maps is an interesting feature, since two major hydrologic inputs enter th e lake in this area: the Kissimmee River and Taylor Creek. The mud material has been descri bed as resulting from the settling of organic biomass from the increasingly eutrophic water column (Brezonik and Engstrom 1998). The presence of a large amount of settled-out fine an d flocculent material ex tending towards a part of the lake expected to have relatively high wate r flow heading in the opposite direction seem counterintuitive. Harris et al. (2007) put forth the possibility of an allochthonous origin of the mud, perhaps carried in from Hawthorn Formation sediments entrained and transported by the Kissimmee River. As such, the tongue of mud c ould be interpreted as a result of differential settling of heavier fracti ons of mud material that drop out of the water column closer to their source, and this could perhaps be tested by comp arison of mud particle size or bulk density along a north/south transect. The fact th at mud does not seem to be further accumulating with time and

PAGE 38

38 that the spatial distribution was stable prio r to the period between 1998 and 2006 might suggest that mud accumulation has not been as cont inuous a phenomenon as previously thought (Brezonik and Engstrom 1998). Th e potential contribution of hu man activity to transport of suspended solids into Lake Okeechobee, perhaps from projects such as the channelization of the Kissimmee River in the 1960s, mer its further investigation. There appears to have been significant m ovement of mud northwards in the period between 1998 and 2006, resulting in shallower but larger expanses of mud coverage. There was a corresponding northwards migration of higher sedime nt chemical concentration levels, and this may represent more chemically-inert sand se diments being covered by the finer, more chemically rich and complex mud. In spite of so me paleolimnological studi es that suggest that significant sediment mixing either does not occu r in Okeechobee or is re latively rare (i.e. Shottler and Engstrom 2006; Engstrom et al. 20 06; Brezonik and Engstrom 1998), it seems that sediment redistribution should be considered as a possible explanat ion for the recent change in spatial distribution of mud material. The years before the study saw hurricanes Jean, Frances, and Wilma pass over the lake, and these kinds of large storms have been s hown to cause large seic hes along the north-south transect of the lake (Chimney 2005). It was found that the largest changes in water level over the course of the events occurred along transects in a north/south alignment, with a maximum difference in water depth of 3.06 m during Hurricane Frances, and 4.91 m during Hurricane Jeanne; east/west movement of water was less sign ificant. Chimney (2005) reported that damage to the levee and the uprooting and piling up of aquatic vegetation was observed after these storms, and attributed at least in part to the scouring action of wind set-up water. This could potentially be responsible for the movement of mud; lack of any recipr ocal observed southern

PAGE 39

39 expansion due to this sloshing effect might be explained by interception of sediment or impediment of water flow by the rocky outcropp ing immediately south of the lake center. Every hypothesis regarding sediment chemis try change over time was rejected. The analytes expected to remain unchanged (Fe, Mg, and Ca) all increased between 1988 and 2006, and the ones predicted to increase either remained the same (TP and porewater SRP) or didnt change (TN and TC). HCl-P decreased between 1998 and 2006, and data was not available for the 88/98 period. It is expected th at it would have been in line with the other P variables and increased during this time, also resulting in no clear change in the longer term. The spatial structure hypotheses were likewise largely rejected. It was expected, in the presence of both strong sediment physico-chem ical differences and clear boundaries between substrates, that spatial structur e of chemical conditions would be strong; this was not the case. Few sediment properties were even in higher modera te levels of spatial st ructure, as defined by Cambardella et al. (1994). There are multiple wa ys of interpreting this; it may be that the relationship between sediment chemis try and substrate type is weaker than was predicted, or it is possible that the density of spa tial sampling and effectiveness of the kriging models used for interpolation was deficient. Maps of differences between surveys we re generated to see if change was occurring in certain regions or following par ticular patterns (e.g., radial expansion as in mud), which may have highlighted substrate differences or point sources of substances; change maps, with few exceptions, possessed low lower spat ial structure, suggesting that change drivers were either highly local processe s or uniform across the lake. The evidence supports an argument that usef ul information can still be gleaned from mapping, and perhaps from the Q statistic, in La ke Okeechobee even in the absence of strong spatial structure in concentration or change maps. For example, P seemed to be associated with

PAGE 40

40 mud sediments, though hotspots were located in the far east and also in the southwest marsh area. Mg change map spatial structure tracked that of mud depth, with very low structure existing between 88/98 and and large increase between 98 /06, largely due to changes occurring in the central region of the lake. Consistent increase s in Mg concentration have been seen in the southeast part of the lake, and a notable increase occurred in the nor thern area between 98/06, concurrent with increase mud depth in that area Fe also increased markedly in the northern tongue area between 98/06. High N and C values we re associated most with the peat sediments of the southern lakebed. C was also higher in th e central mud area, and increased concentrations of it in the north correspond with mud movement in that area. Southerly in creases in TC content could be a relic of overzealous interpolation of the gradient of largely inorganic mud to high C content peat, or perhaps represent real increases due to ongoing carbon sequestration due to SAV growth, such as the macro-alga Chara that was reported by Havens et al. (2001) to have colonized this area during a low lake stage. A fundamental question is whether the current sampling regime is spatially and temporally sufficient to characterize the propert ies of interest in the lake and detect changes in their status. To what extent are current efforts able to detect changes in attributes that we expect to change? Phosphorus was an example of the sampling regime being fine enough to capture fluctuations in concentration, as statistically significant increases and decrease s were detected over the time span monitored. In the case of P, an 18 year sampling period would have been temporally insufficient to detect these changes, while sa mpling at the 8-10 year range appears to have detected decadal changes in the sediment TP pool Of course, the amount of effort and resources available to expend will be a factor in the determination of sampling density as well.

PAGE 41

41 Greater spatial structure was expected in the sediment chemistry properties, as the sediment types (mud, sand, peat, marl) are physically and chemically distinct from each other, and have been depicted as occurring in more or less discreet zones (Reddy et al. 2002). Whether the lack of observed spatial structure was due to heterogeneity in substrate distribution, homogeneity in sediment chemistry, or spatiall y insufficient sampling density is unclear. The change maps tended to have even less spatial coherence than the distribution maps, and again this could be because changes are occurring more or less uniformly in the lake sediments, or because they are occurring tota lly randomly, or because spatial and temporal aspects of the sampling regime are not insufficient to obs erve and properly model these changes. The fact that some properties changed in one direction between 1988/1998 and then reversed between 1998/2006, resulting in no sign ificant change between 1988 and 2006 (as with TP and porewater SRP), is an encouraging indicator that at least some variability is being picked up at the near-decadal time scale, and which might otherwise be missed at lower temporal resolutions. It is not clear from these three studies whether the increases and decrease s are part of natural cycles, were in response to specific extern al parameters, or are purely random in nature. One clear inference from this analysis is that long-term sampling and monitoring should seek to keep the sampling and analytical prot ocols the same between studies. The lack of contintuity in chemical properties measured, how they were tested, and even how the cores were sampled (by boat and piston corer in 2006 and by SCUBA diver in 1998, for example) introduces possible error and lack of compa tibility. The manipulations and work-arounds sometimes required to address these issues can be less than ideal and may confound our ability to detect changes in the system by introducing error and uncertainty due to methodological changes.

PAGE 42

42 Conclusion The m ud sediments of Lake Okeechobee have incr eased in area while decreasing in depth and volume since 1988. Sediment P concentrations increased be tween 1988 and 1998, then decreased through 2006. Significant changes were also s een in Ca, Mg, Fe, and porewat er SRP; TC and TN did not change between any of the surveys. Spatial structure of most analyte maps was in th e weak or moderate range with spatial structure of change maps between surveys tending to be even lower. Continued monitoring programs are vital for tracking the c ondition of the lake and not repeating the mistakes of the past, which have left important gaps in our knowledge of Lake Okeechobee and so introduced uncertainty regarding efforts to mana ge and protect this important resource. The failure to monitor and track ch anges before the lake was impacted by human activities has led to unce rtainty in what restoration goals ough t to be. In order to appropriately plan for the future, a better understanding of how the lake is changing, and how to best monitor these changes, is required.

PAGE 43

43 Table 2-1. Mud area and volum e for surveys in 1988, 1998, and 2006. Mud Area (ha) Mud Volume (million m3) 1988 80,000 205 1998 79,000 192 2006 93,000 141 Table 2-2. The p values for dependent t-tests by analyte, and summary of direction of change. Unchanged 88-98 98-06 88-06 All 88-98 98-06 88-06 TP 0.0465 0.0270 0.5568 TP 0.2010 0.1536 0.4817 TN 0.4686 0.2563 0.2797 TN 0.8434 0.6161 0.7294 TC 0.1263 TC 0.4271 Mud Depth Mud Depth 0.0073 0.0017 0.0000 Ca 0.0557 0.7214 0.0465 Ca 0.0838 0.9555 0.0412 Mg 0.4572 0.0000 0.0000 Mg 0.6478 0.0000 0.0000 Fe 0.0548 0.0000 0.0000 Fe 0.0033 0.0000 0.0000 HCl-P 0.0009 HCl-P 0.0598 PW SRP 0.0000 0.0001 0.4027 PW SRP 0.0000 0.0000 0.1027 Unchanged 88-98 98-06 88-06 All 88-98 98-06 88-06 TP Increase Decrease None TP None None None TN None None None TN None None None TC None TC None Mud Depth Mud Depth DecreaseDecrease Decrease Ca None None Increase Ca None None Increase Mg None Increase Increase Mg None Increase Increase Fe None Increase Increase Fe DecreaseIncrease Increase HCl-P Decrease HCl-P None PW SRP Increase Decrease None PW SRP Increase Decrease None

PAGE 44

44 Table 2-3. Geostatistics for each year and each pair of years. Analyte Nugget Range Partial Sill Q (%) RMSE 1988_TP 107381 9172 26666 20 366.8 1998_TP 121742 9950 43770 26 393.6 2006_TP 110003 21148 37251 25 375.5 98-88_TP 94613 5701 12209 11 343.3 06-98_TP 172968 34737 13417 7 451.3 1988_Mud_Depth 26 11756 193 88 8.033 1998_Mud_Depth 27 11535 216 89 8.182 2006_Mud_Depth 45 10853 58 56 8.149 98_88_Mud 40 33285 0 0 6.443 06_98_Mud 72 9443 61 46 9.666 1998_1M_HCLP 57624 7537 20977 27 277 2006_1_M_HCLP 54109 28843 19147 26 265.4 06_98_1MHCLP 99879 33285 0 0 330.5 1988_TKN 119 34737 92 44 14.06 1998_TN 80 34737 14 15 9.376 2006_TN 51 34737 14 22 7.818 98-88_TN 1.23E+08 34737 1.02E+08 45 15.93 06_98_TN 61 34737 13 18 8.524 1998_TC 11792 34737 2534 18 114.5 2006_TC 8951 34737 2558 22 103.3 06_98_TC 8374 34737 1909 19 101.5 1988_Ca 7.79E+08 10488 5.99E+08 43 34110 1998_Ca 1.96E+09 14402 1.75E+09 47 57370 2006_Ca 8.16E+08 15327 1.98E+09 71 43310 98_88_Ca 1.89E+09 13302 5.07E+08 21 49790 06_98_Ca 1.59E+09 12251 9.65E+08 38 50210 1988_Mg 3.92E+07 8073 4.60E+07 54 8494 1998_Mg 7.09E+07 10244 7.51E+07 51 12090 2006_Mg 7.70E+07 13171 8.83E+07 53 11780 98_88_Mg 5.00E+07 5701 5.10E+06 9 7331 06_98_Mg 3.11E+07 10610 3.89E+07 56 7701 1988_Fe 1.90E+06 10384 8.82E+05 32 1530 1998_Fe 1.45E+06 11357 8.58E+05 37 1414 2006_Fe 3.04E+06 11672 2.34E+06 44 2115 98_88_Fe 1.12E+06 10585 2.15E+05 16 1165 06_98_Fe 3.67E+06 16908 5.34E+05 13 2163 1988_PW_SRP 0.124 33285 0 0 0.3657 1998_PW_SRP 0.126 13573 0 16 0.3975 2006_PW_SRP 0.077 34737 0 29 0.3867 98_88_PW_SRP 0.186 22160 0 4 0.463 06_98_PW_SRP 0.205 34737 0 21 0.5615

PAGE 45

45 Figure 2-1. Map of lake and sample site s, with sediment types from 1988 and 1998. -3.0 -2.5 -2.0 -1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 2.0 -3.0-2.0-1.00.01.02.03.04.05.0 Mud Peat Sand Figure 2-2. Principal components analysis results for the 3 ma jor substrate classifications. Separation between classes indicates that the classification scheme has validity.

PAGE 46

46 Figure 2-3. Interpolations of mud depth fr om 1988, 1998, and 2006 studies. All units in cm. 1988_Mud 1998_Mud 2006_Mud Nugget 25.8 26.7 45.1 Range 11755 11534 10853 Partial Sill 193 216 58.0 Q (%) 88 89 56

PAGE 47

47 Figure 2-4. Interpolations of TP from 1988, 1998, and 2006 studies. All units in mg kg-1. 1988_TP 1998_TP 2006_TP Nugget 107381 121741 110003 Range 9171 9949 21147 Partial Sill 26665 43769 37251 Q (%) 20 26 25

PAGE 48

48 Figure 2-5. Interpolations of Ca from 1988, 1998, and 2006 studies. All units in mg kg-1. 1988_Ca 1998_Ca 2006_Ca Nugget 7.79E+08 1.96E+09 8.16E+08 Range 10488 14402 15326 Partial Sill 5.99E+08 1.75E+09 1.98E+09 Q (%) 43 47 71

PAGE 49

49 Figure 2-6. Interpolations of Fe from 1988, 1998, and 2006 studies. All units in mg kg-1. 1988_Fe 1998_Fe 2006_Fe Nugget 1897563 1454217 3037058 Range 10384 11356 11672 Partial Sill 882158 858288 2341050 Q (%) 32 37 44

PAGE 50

50 Figure 2-7. Interpolations of Fe change between 1988-1998 and 1998-2006 studies. All units in mg kg-1. 98_88_Fe 06_98_Fe Nugget 1122726 3671386 Range 10584 16907 Partial Sill 214853 534198 Q (%) 16 13

PAGE 51

51 Figure 2-8. Interpolations of Mg from 1988, 1998, and 2006 studies. All units in mg kg-1. 1988_Mg 1998_Mg 2006_Mg Nugget 39184445 70940552 76957131 Range 8072 10244 13171 Partial Sill 46022175 75056754 88264732 Q (%) 54 51 53

PAGE 52

52 Figure 2-9. Interpolations of porewater SRP from 1988, 1998, a nd 2006 studies. All units in mg L-1. 88_PW_SRP 98_PW_SRP 06_PW_SRP Nugget 0.124 0.125 0.076 Range 33284 13572 34736 Partial Sill 0 0.024 0.031 Q (%) 0 16 29

PAGE 53

53 Figure 2-10. Interpolations of Ca change between 1988-1998 and 1998-2006 studies. All units in mg kg-1. 98_88_Ca 06_98_Ca Nugget 1.89E+09 1.59E+09 Range 13301 12250 Partial Sill 5.07E+08 9.65E+08 Q (%) 21 38

PAGE 54

54 Figure 2-11. Interpolations of Mg change between 1988-1998 and 19982006 studies. All units in mg kg-1. 98_88_Mg 06_98_Mg Nugget 50011341 31102165 Range 5701 10610 Partial Sill 5097496 38863276 Q (%) 9 56

PAGE 55

55 Figure 2-12. Interpolations of TN from 1988, 1998, and 2006 studies. All units in g kg-1. 1988_TKN 1998_TN 2006_TN Nugget 119 79.8 50.8 Range 34736 34736 34736 Partial Sill 92.4 14.4 14.1 Q (%) 44 15 22

PAGE 56

56 Figure 2-13. Interpolations of HCl-P in 1998 and 2006, and change between those years. All units in mg kg-1. 1998_HCl-P 2006_HCl-P Nugget 57623 54109 Range 7537 28842 Partial Sill 20976 19146 Q (%) 27 26 06_98_HCl-P Nugget 99879 Range 33284 Partial Sill 0 Q (%) 0

PAGE 57

57 Figure 2-14. Interpolations of TC in 1998 and 2006, and changes between those studies. All units in g kg-1. 1998_TC 2006_TC Nugget 11791 8951 Range 34736 34736 Partial Sill 2533 2557 Q (%) 18 22 06_98_TC Nugget 8373 Range 34736 Partial Sill 1908 Q (%) 19

PAGE 58

58 Figure 2-15. Interpolations of porewat er SRP change between 1988-1998 and 1998-2006 studies. All units in mg L-1. 98_88_PW_SRP 06_98_PW_SRP Nugget 0.186 0.205 Range 22160 34736 Partial Sill 0.007 0.054 Q 4 21

PAGE 59

59 Figure 2-16. Interpolations of mud depth change between 1988-1998 and 1998-2006 studies. All units in cm. 98_88_Mud 06_98_Mud Nugget 39.9 72.1 Range 33284 9442 Partial Sill 0 60.9 Q (%) 0 46

PAGE 60

60 98-88_TP 06-98_TP Nugget 94612 172968 Range 5701 34736 Partial Sill 12208 13416 Q (%) 11 7 Figure 2-17. Interpolations of TP change between 1988-1998 and 1998-2006 studies. All units in mg kg-1.

PAGE 61

61 98-88_TN 06_98_TN Nugget 1.23E+08 61 Range 34736 34736 Partial Sill 1.02E+08 13 Q (%) 45 18 Figure 2-18. Interpolations of TN change between 1988-1998 and 19982006 studies. All units in g kg-1.

PAGE 62

62 CHAPTER 3 SPECTRAL DETERMINATION OF SEDIMENT PROPERTIES Introduction Lake Okeechobee is a large (area ~1730 km2) shallow (mean depth 2.7 m), heavily managed and monitored lake in South Florida. Wate r quality in the Lake has been determined to be impaired due to excessive nutrient loading, an d its sediments are an important component of phosphorus cycling in the lake, sinc e they act as a sink for particul ate P and a large internal load of soluble reactive P (Moore et al. 1998). The size of Lake Okeechobee limits the density of sampling in space and time due to high costs associated with collecting, processing, and analyzing large numbers of samples. In particular traditional chemical analysis of soil/sediment samples is expensive and labor intensive, and is often the limiting factor affecting spatial and temporal detail during project planning. Incr eased sampling frequency or density, which is widely viewed as a constraint on effective large-area monitoring and assessment, requires decreasing the costs associated with sample an alysis while maintaini ng reasonable analytical accuracy. One tool that has been investigated for use in large-area envi ronmental surveillance (Shepherd and Walsh 2002) is visible/near-inf rared (VNIR) diffuse reflectance spectroscopy, wherein patterns of electromagnetic radiation absorbed in the visibl e (350-750 nm) and nearinfrared (750-2500 nm) regions are us ed to predict to the physical and chemical characteristics of a substance through the use of multivariate stat istical models called chemometrics. Different materials absorb and reflect different wave lengths of light, depending on their physical composition (elemental composition, amorphous vs crystalline structure, etc). Whether through the detection of overtones of longer wavelength emissions of certain compounds, or indirect association with materials reflecting in the visible/near-infrared part of the electromagnetic

PAGE 63

63 spectrum, statistical corr elations between the spectral signatu res of substance and their physical and chemical properties have successfully been determined for sediments and soils. This technique, which gained favor with the emergence of low cost spectroradiometric equipment, has been successfully applied over a range of terre strial, aquatic, and wetland systems. Model performance is commonly measured through seve ral metrics, including the coefficient of determination (R2) between predicted and observed values, th e RPD statistic (defined as the ratio of the standard deviation of the population to the RMSE of the model), and the RMSE (Change et al 2001, Dunn et al. 2002). Good model performance is indicated by high R2 and RPD scores, and low RMSE values. Malley and Williams (1997) used VNIR to predic t concentrations of various heavy metals in the sediments of an oligotrophic freshwate r lake in Canada. They found that metal concentrations were linked to sample organic ma tter content, and that this could be predicted effectively using NIR analysis. Nilsson et al. (1996) investigated using spectral reflectance of lake sediments to infer properties of the overlyi ng water column. Looking at a number of lakes in Sweden, they found that NIR-derived models could detect 83% of the variability of total phosphorus, 68% for total organic carbon, and 85% for pH. Similar work has been done on soils. Chang et al (2001) were able to predict total C, total N, cation exchange capacity (CEC), moisture content, sand and silt fractions, and extractable Ca of soils collected from around the U.S. Dunn et al. (2002) were able to predict CEC, exchangeable Ca and Mg, and pH in both tops oil and subsoil samples. Shepherd and Walsh (2002) successfully predicted pH, CEC, Ca, Mg, organic C, and soil texture; Brown et al. 2006 were able to predict total clay content, clay fractionation (kaolinite vs montmorillonite), and CEC, and Lee et al. (2003) found that some soil nutrients could be pred icted in a study across

PAGE 64

64 several soil orders found in Fl orida. Though terrestrial soils are different in many respects from lake sediments, some of the pr operties successfully modeled in them, such as Ca, Mg, C, and N are important in Lake Okeechobee and are currently monitored there. The lake also has a large littoral marsh, where the substrate is more like wetland soils than lake sediment. Cohen et al (2005) determined that spectroscopy can be used to predict a variety of chemical characteristics from wetland riparian soils, includ ing P, which is the key nutrient of interest in Okeechobees sediments. In other P-related VNIR work, Bogrecki and Lee (2005b) have showed that phosphate minerals associated with Fe, Al, Mg, a nd Ca could be spectrall y detected in pure sand samples. The same authors investigated the eff ects of particle size on sp ectral testing of soil P and found that particle size c ould be accurately detected using spectral methods, and that correcting for particle size in the models coul d improve the accuracy of the tests for phosphorus (2005a). Whether or not P concentrations and spa tial trends can be spectra lly predicted in Lake Okeechobees sediments is an important questi on in the larger fram ework of testing the applicability of spectroscopy in this system. Spectral analysis offers significant cost benefits over traditional wet chemistry, in part because multiple sample characteristics can be predicted using spectra. For systems like Okeechobee, whose size limits the spatial and te mporal density of sampling, using spectral predictions as a component of monitoring efforts could increase the attainable resolution by offering a low-cost alternative to traditional an alyses. Samples are usually brought back to the laboratory and undergo preparati on (drying, grinding, sieving) be fore being scanned (Nilson et al. 1996, Malley and Williams 1997). Howeover, fi eld-portable models spectroradiometers are now widely available and could be used to co llect sediment spectra on site. If scans of unprocessed material obtained in th e field were usable for spectral modeling then the time, cost,

PAGE 65

65 and effort associated with sample collection a nd processing could be dimi nished or avoided once an initial spectral library was created. A principa l constraint that has lim ited field application of spectral prediction has been that moisture conten t strongly affects spectra (Kooistra et al. 2003); in environments where moisture contents va ry (e.g., terrestrial soils), this may introduce variability that renders field scanning impractical. However, lake sediments are generally saturated, minimizing the effects of variable moisture content. As such, this study had two objectives: 1) to determine whether spectral mo deling could be used to predict sediment characteristics (including spatial pattern) in La ke Okeechobee, and 2) to investigate the viability of prediction from saturated sediment samples. The following research questions were posed: 1. Are chemometric methods useful for char acterizing sediment properties in Lake Okeechobee? 2. Are predictions of sediment properties from we t spectra sufficiently accurate to replace dry spectral prediction? 3. Are spectral predictions of key sediment properties sufficient to predict spatial pattern in the lake? Hypothesis 1: Statistical analysis of reflectance spectra from dry sediments will be able to predict physical and chemical characteristics to a usable degree. Previous success at statistical prediction in soils and sediments fr om a wide variety of environmental settings suggests that effective prediction in Lake Okeechobees sediments using conventional chemometric methods should be tenable. Hypothesis 2: Sediment reflectance spectra from wet sediment samples will have lower predictive power than spectra from dried samples, but will still be utilizable for accurate spectral modeling. Reflectance spectra from soils change with soil moisture content, even while certain chemical parameters remain the same (Bogrekci a nd Lee 2005b), so drying is

PAGE 66

66 commonly used to reduce the effect of this potentially confounding vari able. Lake sediments by their nature are under constant saturation, however, and so, while total water content may still vary according to the physical characteristics of the sediment, the environment is presumably more homogenous and stable with respect to sa turation and its effects on substrate chemistry. Thus, lake sediments may be a good medium in wh ich to attempt to extract viable scans from saturated samples. Hypothesis 3: Spectral modeling will be able to p redict spatial trends in sediment chemical properties. VNIR spectroscopy has demonstrated u tility for creating predictive models for soils and lake sediments. The next step would be to use spectrally predicted values to augment mapping schemes. If predicted values ar e of sufficient accuracy, it would be expected that spatial interpolations based off of those values would be similar to those based off of values derived from traditional wet chemistry measurements. Methods Sample Procurement and Preparation Cores were taken at 156 stations on the lake (Figure 3-1). Dup licate cores were taken at 19 random ly chosen sites. Intact cores were take n either by piston corer (in deep water) or by pushing the core tube into the substrate by hand (i n shallow water). Cores were taken to the point of refusal. A minimum of 10 cm of site water wa s left standing above the surface of each core, and each was capped and kept on ice during transpor t to the lab, where it was refrigerated at 4 C. The cores were extruded inside of an N2-purged hood to maintain anaerobic conditions, and the surface sediment (0-10 cm) was sectioned for chemical and spectral analysis. Porewater was extracted by centrifugation in airtight containers at 10,000 RPM for 10 minutes; syringes were inserted through a rubber stopper in the cap of the tube in order to remove the porewater without exposing it to air.

PAGE 67

67 Following porewater separation, subsamples of sediments were dried in an oven at 70 C for 3 days to determine moisture content and bulk density. These subsamples were then ground in a mortar and pestle and ball-mille d until fine enough to pass through a 20 m size sieve. Part of the sample was used for spectroscopy, and the rest was analyzed for a suite of physical and chemical properties according to standard me thods: total P (EPA 365.1), total C and N with a Carlo Erba NA-1500 CNS analyzer, bulk density, HCl-P (EPA 365.1), HCl-extractable Ca, Mg, Fe, and Al (EPA 200.7), KCl-P (EPA 365.1), KCl-NH4-N (Mulvaney 1996), and NaHCO3-P (EPA 365.1). Porewater was analyzed for tota l P (EPA 365.1), soluble reactive P (EPA 365.1), total Kjeldahl N (EPA 351.2), NH4-N (Mulvaney 1996), pH (E PA 150.1), conductivity (EPA 120.1), as well as Ca, Mg, Fe, and Al (EPA 200.7). Dried and milled samples were scanned using a using a FieldSpec Pro spectroradiometer (Analytical Spectral Devices, Boul der CO). This device measures diffuse spectral reflectance in the visible (350-750 nm) and near-infrared ( 750-2500 nm) in 1-nm bands using Spectralon (Labsphere, Hutton, NH) as a white reference. Samples were scanned from below through borosilicate glass dishes (Shepherd et al. 2003). Spectra were resampled at 10 nm increments in order to reduce data set dimensionality prior to st atistical analysis. Raw spectra were derivative transformed (Fearn 2000) to highlight spectral response patterns, reduce albedo effects and control for minor differences between batches in li ght source intensity and effects of differential sample grinding. To compare prediction accuracy between wet and dry spectra, scans were taken of the sediment under both conditions. Samples were co llected as part of a larger study on Lake Okeechobees sediments, so it was not feasible to conduct scanning of the samples under fieldwet conditions; instead, after be ing dried and ground samples were rewetted to saturation using

PAGE 68

68 distilled de-ionized water, mixed to a homogene ous consistency, then scanned wet after standing for approximately 20 minutes to ensure adequate saturation and equilibration. Data Analysis Partial least squares (PLS) regre ssion was us ed to develop statistical predictions of sediment and pore-water chemistry spectral da ta (Lee et al. 2003). PLS is a multivariate technique that is useful when there are a large number of highly co-linea r factors and the goal is to create a useful predictive model of a phenom enon rather than to fully understand how each individual factor affects its behavior (Tobias 1996). Spectral data, which often consist of measurements of many wavelengths whose values tend to be str ongly spectrally autocorrelated, is an ideal candidate for PLS, and the primary an alytical tool used for chemometric development in the literature (Nilsson et al. 1996, Dunn et al. 2002, Lee et al. 2003, Bogrekci and Lee 2005b). The software package Statistica v. 8.0 (StatSoft Inc., Tulsa OK) was used to perform the analysis and construct predictive models for the various analytes. Site s were randomly divided into calibration and validation sets su ch that 67% of the available data was used to create the statistical model and the 33% used for validation (Dunn et al. 2002). Separately randomized sets for calibration/validation were used for the wet and dry spectral experiments. Any negative values predicted by the models were assumed to be equal to zero. Outlier removal and natural log transformation we re used with some analytes in order to control for non-normal distributions. Removal of ou tliers from the sediment TP for the wet scans significantly improved the performance of the model, while removal of the same outliers from the dry scan worsened it, high lighting the confounding effect that these anomalous values can have, especially in smaller data sets. The distribution of the outliers in the calibration/validation sets was different between wet and dry due to the way that the sets were randomly chosen; in the dry scan, both outlier sites were in the calibratio n set, while in the wet scan one each was in each

PAGE 69

69 of the sets; this demonstrates that the compositi on of the sets can also have an effect on the apparent success of the model. Model performance was evaluated using multiple metrics, including the coefficient of determination (R2) derived from a linear regression betw een observed and pred icted values; the root-mean-squared error (RMSE) of the predicted values; and the RPD statistic, which is a unitless ratio of the population standard deviation to the RMSE (Chang et al. 2001). Model efficiency is inferred from high R2 and RPD, and low RMSE. While no published accuracy/usability thresholds exist for R2 or RMSE, there are relativel y standardized thresholds for RPD that have been widely reported; RPDs <1.5 are generally considered unsuitable for analytic use, between 1.5 and 2.0 are moderately useful, partic ularly for mapping efforts, and >2.0 are considered very good (Chang et al. 2001, Cohen et al. 2005, Dunn et al. 2002). Janik et al. (1998) and Cohen et al. (2007) used RPD for both spectral and wet chemistry applications to conclude that spectroscopy predictions could r easonably replace wet chemistry for a subset of analytes.If RPDs >1.5 are observed, then hypothe ses 1 and 2 will be tentatively accepted. To test the utility of spectral modeling for use in GIS mapping and trend/hotspot detection, separate calibration and validation sets were created for the sediment properties total phosphorus (TP), total nitrogen (TN), loss on ignition (LOI), and total Ca and Mg. Thes e validation sets were chosen on a spatial rather than random basis, and were selected in locations in the lake where spatial features such as gradients or hotspots were observed to occur. Different areas of the lake were chosen for different analytes, and were of large enough size to accommodate a spectral validation set of roughly one third (~40 sites) of the total observations (see Figures A-42 through A-46 in Appendix). Sample observati ons outside the target locations were used to calibrate the model. After spectrally predicted values were obtained for the validation sites, they were

PAGE 70

70 interpolated using an ordinary kriging m odel in ArcGIS v9.3 (ESRI Inc., Redlands CA) following second order trend removal and otherwise all default settings in the geostatistical wizard. The wet chemistry values for the same site s were interpolated using the same ordinary kriging parameters over the same area; only site s from the validation set were used for the interpolations. Interpolated maps of th e error in prediction were also created. Results Summ ary statistics for the laboratory measurem ents of sediment and porewater analytes are shown in Table 3-1. Summar y statistics for the PLS results of the dry and wet scans for sediment characteristics are shown in Tables 3-2 and 3-3, and results for the porewater predictions are shown in Tables 3-4 and 3-5. In general properti es of the sediment materials were able to be modeled with some degree of accuracy (with RPD of 1.5 or greater) under both dry and wet conditions, while porewater properties were predicted poorly. Sediment Ca, Mg, C, and N were generally effectively predicted, with RPD values over or near 2, as did loss on ignition, bulk density, and moisture content (T able 3-2). Predictions for Fe, Al and total P were less effective, falling in the usable but applicatio n-dependent region (1.5 < RPD < 2.0). Scatter plots of observed vs predicted values are shown for TP and TN in Figures 32 through 3-7. Sediment soluble reactive phosphorus (SRP) was poorly predic ted by all spectral models. Porewater properties were not well pr edicted by spectral methods, with generally low coefficients of determination and RPDs < 1.5 (Tables 3-4 and 3-5). The interpolation maps for the observed and predicted values for TP, TN, LOI, Ca, and Mg are shown in Figures 3-8 through 3-12, and the statistics for model performance in Table 3-6. Discussion VNIR spectroscopy shows prom ise for use as a fa st and relatively inexpensive means of predicting the physical and chemical characteris tics of Lake Okeechobees sediments. Most key

PAGE 71

71 sediment properties had statistical metrics in the range that suggests either excellent or adequate prediction. However, phosphorus, the nutrient of major concern to lake managers and the pollutant held responsible for the eutrophica tion of the lake (Florida Department of Environmental Protection, 2001) was on the low end of spectral modeling utility. Nilsson et al (1996) were able to derive a model that explai ned 83% of the variance in lake water P from spectroscopic analysis of surface sediments, a nd Cohen et al. (2005) observed high RPD (2.40) for wetland soil P, while Dunn et al. (2002) experienced low RPD (1.1) for soil P in their work. Our P models, which exhibit RPD values around 1.5, correspond closely to results of Lee et al. (2003) who saw coefficients of determination (R2) for TP prediction in the range of 0.52 0.66. It is unclear whether this is a function of the calibration set size being small, the spectral signature in the VNIR range not be ing closely related to P concentr ations, or some other factor. Determination of porewater properties was uniformly poor for both wet and dry scans. Low validation and calibration R2 values, as well as low RPD scores suggest that VNIR spectroscopy is not well-suited for predicting porewater chemistry in Lake Okeechobee. This could be due to the transien t nature of porewater chemistry, which is susceptible to environmental factors such as temperature, dissolved oxygen levels, a nd redox status, and not just the composition of the parent substrate matr ix in which they reside. Moreover, porewater chemistry is dominated by labile fractions of se diment nutrients, and not by bulk fractions. The observation that the Dry RPD scores for sediment predictions were on average 7% higher than those of those from wet scans. Figure 3-13 shows a comparison between dry and wet spectral modeling effectiveness as measured by RPD. The fact that saturated samples did very nearly as well as dried ones shows promise for field determinatio n of sediment propertie s, which could save

PAGE 72

72 substantial time and effort and further cut cost s associated with sampling. Further research is needed to determine if accuracy levels can be maintained with wet spectra obtained under field sampling and scanning conditions. Mapping using VNIR Mapping the spatial distribution of sediment properties is an important component of ongoing monitoring activiti es in Lake Okeechobee. It seemed a natural progression to turn spectroscopy to this applic ation, though when attempting to use our spectral models to create such maps with our small data set we were initially caught between having too few validation sites to properly spatially ch aracterize a lake the size of Okeechobee and overstating the predictive powers of the spectrally-derived m odel by including points from the calibration set, which are expected to have unrealistically better concordance between the predicted and observed values. This led to choosing a validation set based on spatial subsetting rather than random subsetting in order to mean ingfully test whether interpolations based on spectral predictions captured the same spatial trends in sediment properties observed in interpolations based on traditional methods. All spectral maps compare favorably to maps derived from laboratory values, both in broad trends and hotspot detection; in particular, the Ca map was nearly identical. This is not entirely surprising for analytes where the prediction util ity of the models was judged to be good; however, the concordance between spectral and raw data interpolations for total P despite lower predictive utility (i.e., low RPD) of chemometrics for TP is encouraging and demonstrates that the application dependent proviso for models with intermediate levels of RPD should be taken at face value. That is, analytes for which RP D values are comparatively low should not be written off as lacking spectral prediction utility. Th e utility of VNIR spectroscopy for lake sediment mapping would likely lie more in either fine-scale boundary de tection and delineation,

PAGE 73

73 or detection of short spatial ra nge extreme values within an undersampled and otherwise uniform matrix, i.e. cases where larger-scale differences in chemical properties exist, rather than in detecting smaller changes in relatively homogeneous areas. For all analytes the mean error (bias) was larger for the spectral models than for the kriging models derived from the lake-wide observati ons, while the RMSE was lower for all spectral models. This means that while the kriging models were more evenhanded in whether they overor under-predicted a value at a given observation point, the spectral models were more accurate, suggesting that denser spatial sampling utiliz ing spectral methods would benefit the spatial characterization of these sediment properties. It should be noted that a relativ ely small data set was used to calibrate and test the spectral models. One would ideally have as large a calibration set as possi ble in order to better capture variation in spectral reflectance a nd other properties. Calib ration libraries need not be of infinite size, however at some point one will see dimini shing returns in adding ca libration points, as the spectral models will approach maximum effectiven ess and explain as much of the variance in the phenomenon as they are capable of doing. A future research priority wi ll be determining where this point lies with Lake Ok eechobees sediments; it is probab ly beyond the number of samples currently available (b etween 105-126, depending on analyte). Once a suitably-sized spectral library is constructed, the process by which se diment properties can be monitored will become cheaper and streamlined (Figure 3-14). The decision whether to employ VNIR met hods for lake assessment should weigh the predictive limitations of the met hod against the savings in both cost and effort, which can be substantial. Considering only the 5 analytes mapped using spectral methods, field sampling constitutes the largest fraction of costs (Figur e 3-15) and would be present regardless of the

PAGE 74

74 method used, but processing and anal ytical costs could be greatly reduced using VNIR in lieu of wet chemistry. In addition to reducing the cost by ~25%, there would also be a significant reduction in time and effort involv ed in obtaining the data, especi ally if samples were scanned and discarded in the field. Because one spectrum can be used to determine multiple characteristics of a sample, the savings in cost can increase if spectral methods are used to determine additional analytes of interest. Conclusions Results suggest that spectroscopy m ay be us eful for lake sediment monitoring and assessment in some settings, particularly where the costs of analysis have constrained the temporal and spatial data density of monitoring. For detection of gradients and hotspots, VNIR spectroscopy can provide a valuable analytical tool that allows for greater spatial and temporal monitoring for far less cost and effort. Conve rsely, where high accuracy observations are required (e.g., for testing of specific legal criter ia or close scrutiny of localized management goals), VNIR technology may not be sufficiently accurate to be a primary analytical tool, and traditional wet chemistry is preferred. Because of the large size of the lake and resulting expense and effort associated with high spatial and temporal resolution sampling, use of VNIR technology could allow for more detailed mapping and monitoring of the system, particularly where a small subset of samples are returned to the laboratory for wet chemistry verification of the spectral predictions. Wet scanning produced models that were somewhat less accurate than dry scanning, but which were still in the usable range. The consid erable savings in cost and effort that would accompany field scanning justify accepting slig htly lower predictive power, and further investigations of these methods are recommended.

PAGE 75

75 The present study was conducted using a smaller data set than would be ideal for construction of spectral models; it would be beneficial for future efforts to use VNIR prediction if additional samples were added to the spect ral library for Okeechobees sediments by scanning samples that will be undergoing traditional analys is anyway. By increasing the size of the spectral library, it is expected that the statistic al models will increase in power, increasing the utility and generality of the tech nique. This relatively inexpensive investment would have long term benefits for lake system surveillance.

PAGE 76

76 Table 3-1. Sediment and porewat er properties based on standard methods of analysis (USEPA, 1993). Analyte Mean Standard deviation Range Total P [sediment] (mg/kg) 593.2 490 16.6 to 2859 Total C [sediment] (g/kg) 140.2 127 1.1 to 490 Total N [sediment] (g/kg) 9.06 8.87 0.07 to 41 Total Ca [sediment] (mg/kg) 64652.8 65919 113 to 347789 Total Mg [sediment] (mg/kg) 13066.0 15538 36.1 to 91451.3 Total Fe [sediment] (mg/kg) 3480.3 2991 39.1 to 10748 Total Al [sediment] (mg/kg) 797.5 602 0 to 2299 Bulk Density (g/cm^3) 0.61 0.68 0.04 to 2.06 Site Moisture Content (%) 62.68 27.2 6.3 to 95 KCl-P (mg/kg) [residual] (mg/kg) 0.66 1.31 0.08 to 12.6 KCl-NH4-N (mg/kg) [residual] (mg/kg) 20.32 43.1 0.05 to 428 NaHCO3-SRP (mg/kg) [residual] (mg/kg) 13.36 9.97 1 to 37.2 NaHCO3-TP (mg/kg) [residual] (mg/kg) 14.18 11.0 1 to 58.6 Loss-on-Ignition [sediment] (%) 27.41 24.3 0.5 to 92.7 HCl-Extractable P [sediment] (mg/kg) 410.3 372.1 2.1 to 2357 pH [Porewater] 7.64 0.42 4.9 to 8.3 Conductivity [Porewater] (uS/cm) 477.6 220.0 178 to 1656 Total P [Porewater] (mg/L) 0.24 0.41 0.02 to 4.1 Soluble Reactive P [Porewater] (mg/L) 0.20 0.40 0.001 to 3.9 Total Kjeldahl N [Porewater] (mg/L) 4.39 4.04 0.9 to 38 Ammonium (NH4-N) [Porewater] (mg/L) 2.25 3.55 0.007 to 34.5 Total Ca [Porewater] (mg/L) 74.84 29.7 19.3 to 243 Total Mg [Porewater] (mg/L) 18.24 7.33 3.8 to 53.1 Total Fe [Porewater] (mg/L) 1.20 11.5 0 to 136 Total Al [Porewater] (mg/L) 0.01 0.07 0 to 0.5

PAGE 77

77 Table 3-2. Sediment attribute spectral results using scans of dry samples. Dry Scan Sediment attribute Calibration efficiency Sample Validation efficiency SampleRPD RMSE ME (bias) Ca (mg/kg) 0.93 126 0.88 46 3.08 20978 -5868 KCl-NH4-N (mg/kg) 0.57 115 0.59 40 2.35 17.4 -8.37 Loss-on-ignition (%) 0.82 126 0.84 46 2.34 10.23 1.59 Bulk Density (g/cm3) 0.46 117 0.62 42 2.28 0.30 -0.13 Total C (g/kg) 0.82 126 0.83 46 2.24 56.1 12.0 Moisture content (%) 0.82 117 0.76 43 2.23 12.32 2.53 Mg (mg/kg) 0.85 126 0.76 46 2.15 7020 290 Total N (g/kg) 0.74 126 0.86 46 1.98 4.41 0.76 Fe (mg/kg) 0.91 126 0.67 46 1.70 1726 120 Al (mg/kg) 0.82 126 0.65 46 1.68 348 -9.09 TP (mg/kg) 0.65 126 0.51 46 1.53 316 -59.5 NaHCO3-TP (mg/kg) 0.48 115 0.33 40 1.53 7.84 -0.80 HCl-P (mg/kg) 0.36 126 0.54 46 1.52 242 -37.8 NaHCO3-SRP (mg/kg) 0.43 115 0.31 40 1.38 7.92 -0.74 Total P (mg/kg) OL* 0.72 124 0.40 46 1.26 341 -34.9 Outlier(s) removed from set Table 3-3. Sediment attribute spectral results using scans of wet samples.. Wet scan Sediment attribute Calibration efficiency Sample Validation efficiency SampleRPD RMSE ME (bias) Mg (mg/kg) 0.91 116 0.70 58 2.39 6311 -1381 Ca (mg/kg) 0.97 116 0.80 58 2.28 28266 -2286 Total N (g/kg) 0.88 116 0.83 58 2.09 4.23 -0.35 KCl-NH4-N (mg/kg) 0.67 105 0.61 52 2.08 20.2 -2.35 Moisture content (%) 0.89 108 0.73 54 2.08 13.1 -1.31 Loss-on-ignition (%) 0.84 116 0.80 58 2.07 11.7 -1.74 Total C (g/kg) 0.87 116 0.79 58 2.05 62.6 -1.62 Bulk density (g/cm3)* 0.86 107 0.66 53 1.86 0.30 -0.05 Al (mg/kg) 0.83 116 0.69 58 1.74 338 -18.60 Fe (mg/kg) 0.84 116 0.57 58 1.55 1894 -317.2 Total P (mg/kg)* 0.85 115 0.57 57 1.54 280 -37.3 HCl-P (mg/kg)* 0.71 115 0.56 57 1.51 214 -28.3 KCl-P (mg/kg) 0.89 105 0.42 52 1.51 0.82 0.01 NaHCO3-SRP (mg/kg) 0.28 105 0.31 52 1.26 8.69 -0.76 NaHCO3-TP (mg/kg) 0.59 105 0.29 52 1.13 10.6 -1.15 Outlier(s) removed from set

PAGE 78

78 Table 1-4. Porewater attribute spectral results using scans of dry samples. Dry scan Porewater attribute Calibration efficiency Sample Validation efficiency SampleRPD RMSE ME (bias) Fe (mg/L)* 0.41 112 0.27 38 1.26 1.11 -0.35 Total P (mg/L)* 0.33 114 0.22 40 1.38 0.70 -0.19 Soluble reactive P (mg/L)* 0.54 114 0.24 40 1.17 1.35 -0.32 NH4-N (mg/L)* 0.28 114 0.27 40 1.14 1.09 -0.28 KCl-P (mg/kg) 0.75 115 0.05 40 1.11 0.74 0.00 TKN (mg/L) 0.64 115 0.46 40 1.93 2.03 -1.20 pH 0.23 114 0.13 40 1.59 0.25 0.05 Ca (mg/L) 0.48 115 0.14 40 1.01 29.8 4.47 Mg (mg/L) 0.40 115 0.10 40 1.01 6.52 0.32 Conductivity (uS) 0.30 114 0.07 40 0.89 236 33.9 Outlier(s) removed from set Natural log transformation performed on variable Table 3-5. Porewater attribute spectral results using scans of wet samples. Wet scan Porewater attribute Calibration efficiency Sample Validation efficieny SampleRPD RMSE ME (bias) Soluble reactive P (mg/L)* 0.02 105 0.00 51 1.07 0.22 0.02 TKN (mg/L) 0.55 105 0.68 52 1.05 3.77 0.53 NH4-N (mg/L) 0.64 105 0.66 52 1.00 3.52 0.34 Total P (mg/L)* 0.42 105 0.27 51 0.91 0.27 0.00 pH 0.29 104 0.22 52 0.90 0.44 -0.03 Al (mg/L)* 0.14 104 0.03 52 0.82 0.08 0.00 Mg (mg/L) 0.12 105 0.12 52 0.74 9.58 1.57 Conductivity (uS) 0.01 104 0.01 52 0.73 303 59.4 Ca (mg/L) 0.34 105 0.01 52 0.72 42.6 4.73 Fe (mg/L)* 0.10 105 0.02 51 0.61 1.26 0.12 Outlier(s) removed from set

PAGE 79

79 Table 3-6. Spectral results for 5 analytes base d on spatial selection of the validation set. Sediment attribute Calibration efficiency Sample Validation efficieny Sample RPD RMSE ME (bias) TP (mg/kg) 0.69 134 0.74 40 2.02 239 -79.6 Mg (mg/kg) 0.98 132 0.72 42 1.54 9785 -1521 TN (g/kg) 0.94 134 0.74 40 1.52 5.81 -1.80 LOI (%) 0.92 132 0.84 42 1.98 12.2 -6.62 Ca (mg/kg) 0.93 135 0.83 39 2.45 26324 3285 Figure 3-1. Map of sampling sites fro m 2006 sampling effort used in study

PAGE 80

80 y = 0.8654x + 49.189 R2 = 0.3959 0 200 400 600 800 1000 1200 1400 0200400600800100012001400160 0 Predicted TP (mg/kg)Observed TP (mg/kg) Figure 3-2. Spectrally predicted vs. observed to tal P for hold out validation samples (outliers removed) based on spectra from dry samples. All units expressed as mg kg-1.

PAGE 81

81 y = 0.9418x 2.4142 R2 = 0.5692 -200 0 200 400 600 800 1000 1200 1400 0200400600800100012001400 Predicted TP (mg/kg)Observed TP (mg/kg) Figure 3-3. Spectrally predicted vs. observed to tal P for hold out validation samples (outliers removed) based on spectra from wet samples. All units expressed as mg kg-1.

PAGE 82

82 y = 1.1429x 107.28 R2 = 0.3398 -500 0 500 1000 1500 2000 2500 3000 3500 0200400600800100012001400 Predicted TP (mg/kg)Observed TP (mg/kg) Figure 3-4. Spectral TP scatterp lot of predicted vs. observed values from wet scans, including outliers. All units expressed as mg kg-1. y = 0.8349x + 47.718 R2 = 0.5045 0 200 400 600 800 1000 1200 1400 02004006008001000120014001600 Predicted TP (mg/kg)Observed TP (mg/kg) Figure 3-5. Spectral TP scatterp lot of predicted vs observed va lues from dry scans, including outliers. All units expressed as mg kg-1.

PAGE 83

83 y = 1.4956x 3.8118 R2 = 0.8557 0 5 10 15 20 25 30 35 40 45 010203040 Predicted TN (mg/kg)Observed TN (mg/kg) Figure 3-6. Spectral TN scatterplo t of predicted vs observed values from dry scans. All units expressed as g kg-1. y = 1.3206x 3.4167 R2 = 0.8276 0 5 10 15 20 25 30 35 0510152025 Predicted TN (g/kg)Observed TN (g/kg) Figure 3-7. Spectral TN scatterplo t of predicted vs observed values from wet scans. All units expressed as g kg-1.

PAGE 84

84 Figure 3-13. Comparison of dry vs wet scan RPD fo r sediment spectral results. All units in %; anything above 100% indicates that dry sca nning performed better than wet scanning. Anything below 100% indicates that wet scanning performed better. 0 20 40 60 80 100 120 140 160Ca N a H C O3-TP Bu lk D e ns LOI KCl-NH4-N F e TC N a H CO 3-S RP Moist HCl-P TP A l T N Mg KCl-PDry/Wet RPD x 100 %

PAGE 85

85 Collect samples Scan and discard samples in field Identify and remove outliers Process for wet chemistry Process for dry spectral scan Re-wet and scan Scan dry Use model and collected spectra to estimate values Use chemometricsto create model Obtain laboratory values Use spectra to estimate values Compare: if good (Optional ) Mathematical transforms Collect scan to add data to model Figure 3-14. Flowchart of methodology for chemometrics.

PAGE 86

86 Cost Per Sample by Method0 50 100 150 200 250 Wet ChemDry ScanWet ScanCost per sample ($) Analytical Processing Sampling $ 172 $ 162 $219 Figure 3-15. Comparison of cost for different analysis methods. All units in dollars.

PAGE 87

87 Figure 3-8. Maps of observed Ca, predicted Ca, error between predicted and observed, a nd the standard error of prediction for the same spatial region based on the lakewide wet chemistry-derived values. All units in mg kg-1.

PAGE 88

88 Figure 3-9. Maps of observed LOI, predicte d LOI, error between predicted and observe d, and the standard error of prediction ma p for the same spatial region based on the lakewide traditionally-derived values. All units in %.

PAGE 89

89 Figure 3-10. Maps of observed TP, predicte d TP, error between predicted and observed, a nd the standard error of prediction map for the same spatial region based on the lakewide we t chemistry-derived values. All units in mg kg-1.

PAGE 90

90 Figure 3-11. Maps of observed TN, predicted TN, error between predicted and observed, and the standard error of prediction map for the same spatial region based on the lakewide we t chemistry-derived values. All units in g kg-1.

PAGE 91

91 Figure 3-12. Maps of observed Mg, predicted Mg, error between predicted and observed, and the standard error of prediction map for the same spatial region based on the lakewide we t chemistry-derived values. All units in mg kg-1.

PAGE 92

92 CHAPTER 4 PROJECT REVIEW AND SYNTHESIS Lake Okeechobee has b een subjected to elevated P loads in its recent history, impairing its water quality and presenting management challe nges. It is a system whose economic and ecological services merit protect ion, whose degradation is a ca use for concern and action, and whose size makes implementing plans and moni toring their results labor intensive and expensive. Lake sediments act as a sink for P, and the biogeochemical conditions in them control the forms in which P is stored and the potential under which it can be released to the water column. The purpose of this study was to: i) quantify spatial and temporal changes in sediment distribution and sediment chemistry parameters, and ii) investigate the application of spectrosc opy to monitoring and mapping efforts in the lake The results of this study suggest that mud substrate in the la ke is decreasing in depth and increasing in area, while losing total volume. Th is finding suggests that hurricane activity could be responsible for mass export of entrained mud sediments. While sampling methodologies were different between the study years, which could introduce significant error in the estimation of mud sediment volume, it is unlikely that this di fference in sampling could explain the magnitude of difference in mud sediments determined in this study. Spatial structure was found to be low to moderate for most sediment properties, and this did not change appreciably with time. Spatial structure of temporal change was generally weak These results suggest that for most analytes there is no strong spatial distribut ion pattern in the lake sediment s, in spite of physicallyand chemically-distinct sediment types occurring in different areas of th e lake. Continuity of substrate type proved to have an effect on the ability to statistically detect significant change in

PAGE 93

93 chemical concentrations between sampling years. Differences were clearer when considering only those sites whose substrate remained uncha nged between samplings. This suggests that basic differences in substrate characteristic s are large enough to mask systematic changes occurring in the lake. Sediment chemistry change s were variable in behavior for example, P increased between 88/98 and then decreased be tween 98/06, resulting in no net change over the entire 18 year period while Fe decreased between 88/98 and then increased enough between 98/06 to be significantly higher than in both and Other an alytes experienced no change (TC, TN) and some only changed over one or th e other periods. The current temporal sampling regime seems sufficient to pick up at least so me changes occurring in the sediments of Lake Okeechobee. It is unclear whether the more ambi guous results concerning spatial coherence of change and distribution are a resu lt of spatial sampling deficien cies or natural noise in the system. Perhaps the sediment zones described in the past are just more mottled and heterogeneous than might have been expected. Many sediment properties were predictabl e using chemometric methods; porewater properties, however, were not. Dry spectroscopy RPDs for sediment properties outperformed wet ones by 7% on average. These result s indicate that while moisture does have a deleterious effect on spectral modeling, at least for Okeechobee sedi ments the magnitude of this effect is reasonably small and does not drive the quality of the models to an unusable level. Interpolation maps created using spectral results were shown to be comparable to maps derived from traditional wet chemistry methods. The results suggest that spectral methods performed on wet samples in the field could be used as a cost-effective way to increase the density of spatial and temporal sampling of sediments, potentially leading to more accurate mapping of chemical gradients and hotspots in the lake.

PAGE 94

94 Finally, the data collected for this study will be used for comparison against future data to see how the lake continues to evolve in the face of ongoing anthropogenic pressures and restoration efforts, contributing to ongoing efforts to understand change in this large system. The use of spectroscopy may improve our ability to det ect spatial and temporal changes in the lake sediments by providing a lower-cost and lower-effo rt means by which to sample and collect data.

PAGE 95

95 APPENDIX Figure A-1. 1988 Ca distribution. Units in mg kg-1.

PAGE 96

96 Figure A-2. 1998 Ca distribution. Units in mg kg-1.

PAGE 97

97 Figure A-3. 2006 Ca distribution. Units in mg kg-1.

PAGE 98

98 Figure A-4. Change in Ca distribution between 1988 and 1998. Units in mg kg-1.

PAGE 99

99 Figure A-5. Change in Ca distribution between 1998 and 2006. Units in mg kg-1.

PAGE 100

100 Figure A-6. 1988 Fe distribution. Units in mg kg-1.

PAGE 101

101 Figure A-7. 1998 Fe distribution. Units in mg kg-1.

PAGE 102

102 Figure A-8. 2006 Fe distribution. Units in mg kg-1.

PAGE 103

103 Figure A-9. Change in Fe distribution between 1988 and 1998. Units in mg kg-1.

PAGE 104

104 Figure A-10. Change in Fe distribu tion between 1998 and 2006. Units in mg kg-1.

PAGE 105

105 Figure A-11. 1998 HCl-P distribution. Units in mg kg-1.

PAGE 106

106 Figure A-12. 2006 HCl-P distribution. Units in mg kg-1.

PAGE 107

107 Figure A-13. Change in HCl-P distribution between 1998 and 2006. Units in mg kg-1.

PAGE 108

108 Figure A-14. 1988 Mg distribution. Units in mg kg-1.

PAGE 109

109 Figure A-15. 1998 Mg distribution. Units in mg kg-1.

PAGE 110

110 Figure A-16. 2006 Mg distribution. Units in mg kg-1.

PAGE 111

111 Figure A-17. Change in Mg distribu tion between 1998 and 2006. Units in mg kg-1.

PAGE 112

112 Figure A-18. Change in Mg distribu tion between 1998 and 2006. Units in mg kg-1.

PAGE 113

113 Figure A-19. 1988 Mud depth. Units in cm.

PAGE 114

114 Figure A-20. 1998 Mud depth. Units in cm.

PAGE 115

115 Figure A-21. 2006 Mud depth. Units in cm.

PAGE 116

116 Figure A-22. Change in mud dept h between 1988 and 2006. Units in cm.

PAGE 117

117 Figure A-23. Change in mud dept h between 1998 and 2006. Units in cm.

PAGE 118

118 Figure A-24. 1988 porewater SRP distribution. Units in mg L-1.

PAGE 119

119 Figure A-25. 1998 porewater SRP distribution. Units in mg L-1.

PAGE 120

120 Figure A-26. 2006 porewater SRP distribution. Units in mg L-1.

PAGE 121

121 Figure A-27. Change in porewater SRP di stribution between 1988 and 1998. Units in mg L-1.

PAGE 122

122 Figure A-28. Change in porewater SRP di stribution between 1998 and 2006. Units in mg L-1.

PAGE 123

123 Figure A-29. 1998 TC distribution. Units in g kg-1.

PAGE 124

124 Figure A-30. 2006 TC distribution. Units in g kg-1.

PAGE 125

125 Figure A-31. Change in TC distri bution between 1998 and 2006. Units in g kg-1.

PAGE 126

126 Figure A-32. 1988 TKN distribution. Units in g kg-1.

PAGE 127

127 Figure A-33. 1998 TN distribution. Units in g kg-1.

PAGE 128

128 Figure A-34. 2006 TN distribution. Units in g kg-1.

PAGE 129

129 Figure A-35. Change in TKN/TN dist ribution between 1988 and 1998. Units in g kg-1.

PAGE 130

130 Figure A-36. Change in TN distri bution between 1998 and 2006. Units in g kg-1.

PAGE 131

131 Figure A-37. 1988 TP distribution. Units in mg kg-1.

PAGE 132

132 Figure A-38. 1998 TP distribution. Units in mg kg-1.

PAGE 133

133 Figure A-39. 2006 TP distribution. Units in mg kg-1.

PAGE 134

134 Figure A-40. Change in TP distribution between 1988 and 1998. Units in mg kg-1.

PAGE 135

135 Figure A-41. Change in TP distribution between 1998 and 2006. Units in mg kg-1.

PAGE 136

136 Figure A-42. Location of spectral/kriging mapping area for TP.

PAGE 137

137 Figure A-43. Location of spectral/kriging mapping area for Ca.

PAGE 138

138 Figure A-44. Location of spectral/kriging mapping area for LOI.

PAGE 139

139 Figure A-45. Location of spectral/kriging mapping area for Mg.

PAGE 140

140 Figure A-46. Location of spectral/kriging mapping area for TN.

PAGE 141

141 LIST OF REFERENCES BOGREKCI, I. AND W. S. LEE. 2005. Improving Phosphorus Sensing by Eliminating Soil Particle Size Effect in Spectral Measurement. Transactions of the ASAE. 48: 1971-1978. BOGREKCI, I. AND W. S. LEE. 2005. Spectral Measurement of Common Soil Phosphates. Transactions of the ASAE. 48: 2371-2378. BREZONIK, P. L. AND D. R. ENGSTROM. 1998. Modern and historic accumulation rates of phosphorus in Lake Okeechobee, Fl orida. J. Paleolimnol. 20: 31-46. BROWN, D. J., K. D. SHEPHERD, M. G. WALSH, M. DEWAYNE MAYS AND T. G. REINSCH. 2006. Global soil characterization with VNIR di ffuse reflectance spectroscopy. 132: 273-290. CAMBARDELLA, C. A., T. B. MOORMAN, J. M. NOVAK, T. B. PARKIN, D. L. KARLEN, R. F. TURCO AND A. E. KONOPKA. 1994. Field-scale variabil ity of soil properties in central Iowa soils. Soil Sci. Soc. Am. J. 58: 1501-1511. CAMBARDELLA, C. A., T. B. MOORMAN, J. M. NOVAK, T. B. PARKIN, D. L. KARLEN, R. F. TURCO AND A. E. KONOPKA. 1994. Field-scale variabil ity of soil properties in central Iowa soils. Soil Sci. Soc. Am. J. 58: 1501-1511. CHANG, C., D. A. LAIRD, M. J. MAUSBACH AND C. R. HURBURGH JR. 2001. Near-Infrared Reflectance Spectroscopy-Principal Components Regression Analyses of Soil Properties. Soil Sci Soc Am J. 65: 480-490. CHEMNEY, M. J. 2005. Surface Seiche and Wind Setup on Lake OKeechobee (Florida, USA) During Hurricanes Frances and Jeanne. 21: 465-473. COHEN, M. J., J. P. PRENGER AND W. F. DEBUSK. 2005. Visible-Near Infrared Reflectance Spectroscopy for Rapid, Nondestructive Asse ssment of Wetland Soil Quality. J. Environ. Qual. 34: 1422-1434. DUNN, B. W., G. D. BATTEN, H. G. BEECHER AND S. CIAVARELLA. 2002. The potential of nearinfrared reflectance spectros copy for soil analysis a case study from the Riverine Plain of south-eastern Australia. Aust. J. Exp. Agric. 42: 607-614. FEARN, T. 2000. Savitzky-Golay filters. 6: 14-15. FISHER, M. M., K. R. REDDY AND R. T. JAMES. 2005. Internal Nutrient Loads from Sediments in a Shallow Subtropical Lake. 21: 338-349. FLORIDA DEPARTMENT OF ENVIRONMENTAL PROTECTION. 2001. Total Maximum Daily Load for Total Phosphorus Lake Okeechobee, Florida. HARRIS, W. G., M. M. FISHER, X. CAO, T. OSBORNE AND L. ELLIS. 2007. Magnesium-Rich Minerals in Sediment and Suspended Part iculates of South Florida Water Bodies: Implications for Turbidity. J Environ Qual. 36: 1670-1677.

PAGE 142

142 HAVENS, K. E., K. JIN, A. J. RODUSKY, B. SHARFSTEIN, M. A. BRADY, T. L. EAST, N. IRICANIN, R. T. JAMES, M. C. HARWELL AND A. D. STEINMAN. 2001. Hurricane Effects on a Shallow Lake Ecosystem and Its Response to a Controll ed Manipulation of Wa ter Level. 1: 44-70. HAVENS, K. E. AND D. E. GAWLIK. 2005. Lake Okeechobee Conceptual Model. Wetlands. 25: 908-925. HELLAWELL, J. M. AND J. D. HOLLOWAY. 1977. Change in Natural and Managed Ecosystems: Detection, Measurement and Assess ment [and Discussion]. 197: 31-57. JAMES, R. T., J. ZHANG, S. GORNAK, S. GRAY, G. RITTER, AND B. SHARFSTEIN. 2006. Chapter 10: Lake Okeechobee Protection Program State of the Lake a nd Watershed. 10-1-10-102. JANIK, L. J., J. O. SKJEMSTAD AND R. H. MERRY. 1998. Can mid infrared diffuse reflectance analysis replace soil extractions? Aust. J. Exp. Agric. 38: 681-696. KEMPER, T. AND S. SOMMER. 2002. Estimate of Heavy Metal C ontamination in Soils after a Mining Accident Using Refl ectance Spectroscopy. Environ. Sci. Technol. 36: 2742-2747. KOOISTRA, L., J. WANDERS, G. F. EPEMA, R. S. E. W. LEUVEN, R. WEHRENS AND L. M. C. BUYDENS. 2003. The potential of field spectrosco py for the assessment of sediment properties in river fl oodplains. 484: 189-200. LEE, W. S., J. F. SANCHEZ, R. S. MYLAVARAPU AND J. S. CHOE. 2003. Estimating Chemical Properties of Florida Soils Using Spectral Reflectance. Transactions of the ASAE. 46: 1443-1453. LOBELL, D. B. AND G. P. ASNER. 2002. Moisture Effects on Soil Re flectance. Soil Sci Soc Am J. 66: 722-727. LYONS, J. B., J. H. GORRES AND J. A. AMADOR. 1998. Spatial temporal variability of phosphorus retention in a riparian forest soil. J. E nviron. Qual. 27: 895-903. MALLEY, D. F. AND P. C. WILLIAMS. 1997. Use of Near-Infrared Reflectance Spectroscopy in Prediction of Heavy Metals in Freshwater Sediment by Their Association with Organic Matter. Environ. Sci. Technol. 31: 3461-3467. MOORE, A., JR. AND K. R. REDDY. 1994. Role of Eh and pH on Phosphorus Geochemistry in Sediments of Lake Okeechobee, Fl orida. J Environ Qual. 23: 955-964. MOORE, P. A., JR., K. R. REDDY AND M. M. FISHER. 1998. Phosphorus Flux between Sediment and Overlying Water in Lake Okeechobee, Flor ida: Spatial and Temporal Variations. J Environ Qual. 27: 1428-1439. NILSSON, M. B., E. DABAKK, T. KORSMAN AND I. RENBERG. 1996. Quantifying Relationships between Near-Infrared Reflectance Spectra of Lake Sediments and Water Chemistry. Environ. Sci. Technol. 30: 2586-2590.

PAGE 143

143 PARR, T. W., A. R. J. SIER, R. W. BATTARBEE, A. MACKAY AND J. BURGESS. 2003. Detecting environmental change: science and societ yperspectives on long-term research and monitoring in the 21st century. 310: 1-8. REDDY, K. R., J. R. WHITE, M. M. FISHER, H. K. PANT, Y. WANG, K. GRACE, AND W. G. HARRIS. 2002. Potential impacts of sediment dredgi ng on internal phosphorus load in Lake Okeechobee. SHEPHERD, K. D., C. A. PALM, C. N. GACHENGO AND B. VANLAUWE. 2003. Rapid Characterization of Organic Resource Quality for Soil and Livestock Management in Tropical Agroecosystems Using Near-Infrared Spectroscopy. Agron J. 95: 1314-1322. SHEPHERD, K. D. AND M. G. WALSH. 2002. Development of Reflectance Spectral Libraries for Characterization of Soil Propertie s. Soil Sci Soc Am J. 66: 988-998. SOUTH FLORIDA WATER MANAGEMENT DISTRICT. 2002. Lake Okeechobee Surface Water Improvement and Management (SWIM) Plan. U.S. ENVIRONMENTAL PROTECTION AGENCY. 1993a. Methods for the determination of inorganic substances in environmental samples. Environmental Monitoring Systems Lab, Cincinnati, OH.

PAGE 144

144 BIOGRAPHICAL SKETCH Justin Vogel earned his bachelors degree in P hysics from the University of South Florida in 2005. Unable to find a job blowing things up or hunting ghosts, he returned to school to pursue a masters degree in the environmental sc iences, combining his interests in intellectual pursuits with his love of the outdoors. Justins interests are varied and include su ch things as paranorm al phenomena, music, Lovecraftian myth, and effects of excessive alcohol consumption on the human mind and body. He hopes someday to reach escape velocity of the world of academia and blast off forever to the stars, living happily ever after.