Citation
Assessment of Tree-Rings as a Record of Pre-Historic Stream Flow in a Subtropical Environment

Material Information

Title:
Assessment of Tree-Rings as a Record of Pre-Historic Stream Flow in a Subtropical Environment
Creator:
Crockett, Kris Daniel
Place of Publication:
[Gainesville, Fla.]
Florida
Publisher:
University of Florida
Publication Date:
Language:
english
Physical Description:
1 online resource (67 p.)

Thesis/Dissertation Information

Degree:
Master's ( M.S.)
Degree Grantor:
University of Florida
Degree Disciplines:
Geology
Geological Sciences
Committee Chair:
Martin, Jonathan B.
Committee Members:
Screaton, Elizabeth J.
Brenner, Mark
Graduation Date:
8/11/2007

Subjects

Subjects / Keywords:
Climate models ( jstor )
Correlation coefficients ( jstor )
Correlations ( jstor )
Dendroclimatology ( jstor )
Growth rings ( jstor )
Modeling ( jstor )
Precipitation ( jstor )
Rivers ( jstor )
Statistical models ( jstor )
Statistics ( jstor )
Geological Sciences -- Dissertations, Academic -- UF
dendohydrology, dendrochronology, tree
Waccasassa River ( local )
Genre:
bibliography ( marcgt )
theses ( marcgt )
government publication (state, provincial, terriorial, dependent) ( marcgt )
born-digital ( sobekcm )
Electronic Thesis or Dissertation
Geology thesis, M.S.

Notes

Abstract:
As trees age, they produce annual layers inside their trunks called tree-rings. How wide these rings are depends on how much a tree grows in one year. Wide rings will correspond to favorable growing seasons while thin ones will correspond to poor growing seasons. By analyzing cores taken from trees we can obtain a record of past growing seasons. By understanding what factors of climate make a growing season either favorable or unfavorable for trees, we can extend these climate factors back through time. The climate factor of interest in this study is streamflow. By extending streamflow records back in time we can more effectively plan for the wise use of our water resources here in the state of Florida. ( en )
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.
Thesis:
Thesis (M.S.)--University of Florida, 2007.
Local:
Adviser: Martin, Jonathan B.
Statement of Responsibility:
by Kris Daniel Crockett.

Record Information

Source Institution:
UFRGP
Rights Management:
Copyright Crockett, Kris Daniel. Permission granted to the University of Florida to digitize, archive and distribute this item for non-profit research and educational purposes. Any reuse of this item in excess of fair use or other copyright exemptions requires permission of the copyright holder.
Resource Identifier:
660256280 ( OCLC )
Classification:
LD1780 2007 ( lcc )

Downloads

This item has the following downloads:

crockett_k.pdf

crockett_k_Page_30.txt

crockett_k_Page_46.txt

crockett_k_Page_49.txt

crockett_k_Page_05.txt

crockett_k_Page_33.txt

crockett_k_Page_61.txt

crockett_k_Page_01.txt

crockett_k_Page_10.txt

crockett_k_pdf.txt

crockett_k_Page_56.txt

crockett_k_Page_12.txt

crockett_k_Page_11.txt

crockett_k_Page_22.txt

crockett_k_Page_40.txt

crockett_k_Page_50.txt

crockett_k_Page_04.txt

crockett_k_Page_53.txt

crockett_k_Page_44.txt

crockett_k_Page_25.txt

crockett_k_Page_41.txt

crockett_k_Page_15.txt

crockett_k_Page_59.txt

crockett_k_Page_07.txt

crockett_k_Page_39.txt

crockett_k_Page_20.txt

crockett_k_Page_21.txt

crockett_k_Page_64.txt

crockett_k_Page_26.txt

crockett_k_Page_28.txt

crockett_k_Page_03.txt

crockett_k_Page_17.txt

crockett_k_Page_19.txt

crockett_k_Page_47.txt

crockett_k_Page_48.txt

crockett_k_Page_57.txt

crockett_k_Page_32.txt

crockett_k_Page_27.txt

crockett_k_Page_02.txt

crockett_k_Page_23.txt

crockett_k_Page_45.txt

crockett_k_Page_58.txt

crockett_k_Page_66.txt

crockett_k_Page_63.txt

crockett_k_Page_16.txt

crockett_k_Page_36.txt

crockett_k_Page_38.txt

crockett_k_Page_06.txt

crockett_k_Page_43.txt

crockett_k_Page_14.txt

crockett_k_Page_09.txt

crockett_k_Page_55.txt

crockett_k_Page_67.txt

crockett_k_Page_52.txt

crockett_k_Page_42.txt

crockett_k_Page_24.txt

crockett_k_Page_18.txt

crockett_k_Page_37.txt

crockett_k_Page_60.txt

crockett_k_Page_31.txt

crockett_k_Page_51.txt

crockett_k_Page_29.txt

crockett_k_Page_54.txt

crockett_k_Page_65.txt

crockett_k_Page_08.txt

crockett_k_Page_34.txt

crockett_k_Page_13.txt

crockett_k_Page_62.txt

crockett_k_Page_35.txt


Full Text





CHAPTER 5
DISCUSSION

The results of this study provide insights into the link between hydrologic cycle and

longleaf pine growth in Florida and how records of this growth could be used to estimate pre-

historical flow in the region. Climate, geology, topography, vegetation, and other landscape

characteristics that influence runoff are highly variable in Florida. These characteristics will

influence river discharge. In order to record effectively this average value of runoff in a drainage

basin a TRI thus needs to come from a sample site or sites that reflect the same average climate

and landscape characteristics as the drainage basin of interest.

Tree-Ring Indices

The reported EPS values of 97% indicate that the sample size of the Goethe and Camp

Branch forest were sufficiently large for this study. These high EPS values also show that a

strong common variable or variables control the growth of longleaf pine trees in the forest

sample sites. Both TRls have a mean of 1.0 after approximately 1910 that indicates growth in

the two forests was steady during this period (Figures 4-1 and 4-2). Prior to this period the

Goethe TRI had a mean of 0.75 which indicates that the trees were going through a time of slow

growth that was not adequately compensated for during the detrending process.

Tree-rings produced during the first 10 to 20 years in the life of a tree often provide the

least reliable climatic information, and are often disregarded in dendrohydrology studies (Fritz,

1976). A major drought in 1904 (Henry, 1931) may have had severe impacts on the young trees.

Impacts could have included a period of declined growth from which they didn't fully recover

until around 1910 (Figures 4-2.). The extreme low in the Camp Branch TRI that occurred in

1907 (Figure 4-4) also corresponds with the driest year on record for Lake City (Figure 1-1;









CHAPTER 4
RESULTS

Tree-ring Data

COFECHA

From the 55 cores collected at the Goethe site, 31 cores from 23 trees could be dated by

the COFECHA program and were included in the Goethe TRI. From the 47 cores collected at

the Camp Branch site, 42 cores from 26 trees and 2 stumps could be dated by the COFECHA

program and were included in the Camp Branch TRI. These TRIs extend from 1875 to 2003 for

the Goethe State forest site, and extend from 1806 to 2002 for the Camp Branch site. Series

intercorrelation was calculated to be 0.513 for the Goethe site and 0.522 for the Suwannee site.

The EPS was calculated to be 97% for both sites. This value is well above the value of 85%

suggested by Wigley et al. (1984) for acceptable statistical quality.

ARSTAN

The TRls produced from the ARSTAN program (STNDRD and ARSTAN versions) are

displayed in Figure 4-1 for the Goethe TRI and in Figure 4-2 Camp Branch TRI. In the

following discussion only the STNDRD versions of the TRls were used because:

* The STNDRD and the ARSTAN versions for the Goethe TRI and Camp Branch TRI are
similar (Figures 4-1 and 4-2).

* The ARSTAN version did not provide a significant improvement on the correlation results
(r<0.01) and in most cases lowered the Pearson correlation coefficient.

* The STNDRD version produces a simpler prediction model, thereby preserving additional
degrees of freedom that would be lost through the use of the autocorrelation modeling in
the ARSTAN versions (Fritz, 1976).

The means, standard deviations, minimum and maximum values, and ranges for the Goethe TRI

and Camp Branch TRI are displayed in table 4-1. The correlation between the Goethe TRI and

Camp Branch TRI had a Pearson correlation coefficient of 0.58.










LIST OF REFERENCES


Bamnes, J., 1998, Florida's hurricane history: Chapel Hill, North Carolina, The University of
North Carolina Press, 330 p.

Case, R.A., and MacDonald, G.M., 2003, Tree ring reconstructions of streamflow for three
Canadian Prairie rivers: Journal of the American Water Resources Association, v. 39, p.
703-716.

Cleaveland, M.K., and Stahle D.W., 1989, Tree-ring analysis of surplus and deficit runoff in the
White River, Arkansas: Water Resources Research, v. 25, p. 1391-1401.

Cook, E.R., and Holmes, R.L., 1997, ARSTAN: chronology development. in Grissino-Mayer,
H.D., Holmes, R.L., and Fritts, H.C., eds., The international tree-ring data bank program
library, version 2. 1. User' s manual: Laboratory of tree-ring research, The University of
Arizona, Tucson, Arizona, p. 75-87.

Cook, E.R., and Kaririukstis, L.A., 1990, Methods of Dendrochronology: Dordrecht, The
Netherlands, Kluwer Academic Publishers, 394 p.

Coile, T.S., 1936, The effects of rainfall and temperature on the annual radial growth of pine in
the Southemn United States: Ecological Monographs, v. 6, p. 533-562.

Dingman, S.L., 1994, Physical hydrology: New Jersey, Prentice Hall, Inc., p.575.

Ford, C.R and Brooks, J.R., 2002, Detecting forest stress and decline in response to increasing
river flow in southwest Florida, USA: Forest Ecology and Management, v. 160 p. 45-64.

Foster, T.E., and Brooks, J.R., 2001, Long-term trends in growth of Pinus palustris and Pinus
elliottii along a hydrological gradient in central Florida: Canadian Journal of Forest
Research, v. 31, p.1661-1670.

Fritts, H.C., 1976, Tree rings and climate: New York, Academic Press, 567 p.

Gedalof, Z., Peterson, D.L., and Mantua, N.J., 2004, Columbia River flow and drought since
1750: Journal of the American Water Resources Association, v. 40, p. 1579-1592.

Gray, S.T., Jackson, S.T., and Betancourt, J.L., 2004, Tree-ring based reconstructionss of
interannual to decadal-scale precipitation variability for northeastern Utah since 1226
A.D: Joumnal of the American Water Resources Association, v. 40, p. 947- 960.

Grissino-Mayer, H.D., 2001, Evaluating crossdating accuracy: A manual and tutorial for the
computer program COFECHA. Tree-Ring Research, v. 57, p. 205-221.

Hand, J., Tauxe, V., and Friedemann, M., 1988, Water quality assessment for the State of










4-14 Discharge of the Steinhatchee River in m3/S minus mean flow plotted against the
residuals of the TRIGt-Steinhatchee discharge linear regression. ................ ................ .46

4-15 Usher Tower raw precipitation and precipitation minus potential evapotranspiration
through tim e. .............. ...............48....

4.16 Precipitation plotted against potential recharge. .............. ...............49....

4-17 The correlation coefficients of all 14 stream gauging stations plotted to the Usher
Tower raw precipitation data vs. correlation coefficients of the same stream gauging
stations to the Goethe tree ring TRIGt and TRICB .............. ...............50....

4-18 Actual and predicted Waccasassa River discharge in m3/S from 1875 -2003 ................... .51

4-19 Actual and predicted Steinhatchee River discharge in m3/S from 1875 -2003. .................5 1









correlation coefficients between the two forest sample sites also improved by 37% after removal

of the warm phase (r = 0.41 for the warm phase and r = 0.65 for the cool phase).

Final model

Despite the influence of the PDO, the Goethe TRI and Camp Branch TRI show positive

correlations with the precipitation data (both raw and adjusted for evapotranspiration) as well as

for flow measurement at all gauging stations used in this study. The reported Pearson correlation

coefficients for the Waccasassa and Steinhatchee discharge data to the Goethe TRI (r = 0.86 and

r =0.64, respectively) are the highest out of every stream gauging station used in this study and

are in the range of values found in the reviewed literature from predominantly arid sites. The RE

statistics calculated for both these rivers also demonstrate some predictive qualities to the

Waccasassa and Steinhatchee models (Table 4-3).

The Waccasassa and Steinhatchee drainage basins share characteristics of small size and

close proximity to the Goethe State Forest compared to the other 12 gauging stations. Both

rivers also receive no input from first order magnitude springs and thus are linked to surface

water runoff and precipitation in the Goethe State Forest better than any other gauging station.

This link is further illustrated by the fact that the Waccasassa and Steinhatchee discharge record

have better correlation coefficients with the Usher Tower rain gauge station than the other

gauging stations used in this study. Since the Waccasassa River basin encompasses part of the

Goethe State Forest the discharge record for this site should be well correlated with the Goethe

TRI during both the PDO warm and cool phase. Because most of the short discharge record from

the Waccasassa River (89%) occurs during a PDO cool phase; this record cannot provide support

for or evidence against a control by the PDO in the region.












Table 4-3. Validation Statistics

Rivers r MSE

Waccasassa 0.864 5.47

Steinhatchee 0.64 15.82

Odd years 0.43 17.59

Even years 0.77 14.35

Early years 0.79 13.08

Late years 0.45 17.56


RE

0.714

0.39

0.12

0.56

0.59

0.14


CE

N\A

N\A

0.48

0.1

-0.09

0.48


Model

(-3.71) +( 12.53) x (TRIMt

(-5.29) +( 14.54) x (TRIG,

(-1.82) +( 10.52) x (TRIMt

(-6.91) +( 16.74) x (TRIMt

(-10.45) +( 18.89) x (TRIMt

(-0.94) +( 10.45) x (TRIMt









CHAPTER 2
STUDY AREA

Climate, geology, hydrology, and human interactions have a strong effect on both tree

growth and stream flow. An understanding of how these characteristics influence the growth-

flow relationship is important to dendrohydrologic studies. These characteristics are described

for north-central Florida, with emphasis on the Waccasassa and Steinhatchee drainage basins,

which provided data for the most in-depth analysis. Limited analyses are included for an

additional 12 watersheds (Figure 1-1).

Florida is one of the fastest growing states, with population increasing by 4.7 %percent

between 2003 and 2004 (2004 U.S. Census Bureau). This rapid population growth places large

demands on Florida' s ground water supply and other natural resources. Florida' s five water

management districts are responsible for managing ground and surface water supplies as well as

evaluating current and future water needs. To date, water policy and management decisions

employed by these districts have been based primarily on instrumental climatic and hydrologic

records with maximum lengths of around 75 years. Based on available records, the Suwannee

River Water Management District (SRWMD), which encompasses the study area, has concluded

water supplies are sufficient for demands over the next 20 years. All regulatory agencies

continually reassess their demands and sources of water as new techniques and data become

available. Long-term climate records from tree-rings could provide valuable information for this

assessment.

North Florida has a humid subtropical climate with long, hot, rainy summers, short and

usually mild winters (Henry et al., 1994). The seasons in Florida are determined more by

precipitation than by temperature. The winters and falls are cool and dry and the summers and

springs are rainy and warm (Winsberg, 1990). Due to the mild winter temperatures, snow and









Limited analysis of tree-rings has been completed to estimate hydrologic variations

through time in subtropical regions and this previous work has been erratic and contradictory. At

least two possible reasons for this lack of research exist: there are limited sites with sufficiently

old trees (Varner and Pederson, 2002), and dendrohydrologic research is usually conducted

where severe water stress strengthens the relationship between tree-ring growth and climate

(Fritts, 1976; Loaiciga et al., 1993; and North, 2006). Lodewick (1930) was one of the first to

investigate the relationship of climate to tree growth in subtropical regions. He found a positive

correlation between tree-ring width and mid-June to mid-October precipitation, but no

correlation between tree-ring width and temperature. Coile (1936) found that February to April

precipitation has a positive correlation on young longleaf pine growth in the subtropics and that

June to August temperature had a negative correlation. Schumacher and Day (1939) performed

similar studies in several areas of North Florida but with varied results. More recently Ford and

Brooks (2002) and Foster and Brooks (2001) found a positive correlation between tree-ring

widths, precipitation, and the discharge of the Myakka River in several species of trees from

south Florida.

In order to assess the use of tree-rings as a proxy for stream flow, data has been compiled

from a network of longleaf pine trees located in two old growth forests in northern Florida,

several USGS stream gauging stations and National Weather Service rain gauging stations

(Figure 1-1). My study uses these data to test the validity of dendrohydrologic analysis in a

subtropical environment. If successful, the technique of utilizing tree-ring analysis could be used

to provide information about stream flow prior to instrumental data. Long-term records

developed from tree-ring proxies may offer a useful tool in the management of water resource

systems in the area.










































































-



-


Waccasassa


1964 1969 1974 1979 1984 1989

Year Feb.-Jan.

-Predicted Q -Measured Q


Figure 4-12. Waccasassa River predicted and actual discharge.


1994 1999


Steinhatchee


30



25



20





S10





5


V


0
1950 1960 1970 1980 1990

Year Feb.-Jan
-Measured Q -Predicted Q


Figure 4-13. Steinhatchee River predicted and actual discharge.


2000


V V










Table 5-1. The growth-discharge correlation for the cool phase PDO cycle.
Gauging stations r % improvement*
1 Waccasassa River 0.8746 2%
2 Steinhatchee River 0.8247 22%
3 Fenholloway River 0.7635 23%
4 Jumper Creek Canal 0.6922 28%
5 Rainbow Springs 0.5825 11%
6 Withlacoochee River 0.4791 5%
7 Santa Fe River 0.6108 26%
8 Suwannee River 0.5679 21%
9 Santa Fe River 0.6081 29%
10 Suwannee River 0.4439 15%
11 Suwannee River 0.4832 25%
12 Suwannee River 0.5763 40%
13 Withlacoochee River 0.4562 34%
14 Alapaha River 0.5463 47%









These simplified models imply that overfitting is not the cause of the low CE calculated for the

early and even, split sample data sets.

The differences in the descriptive statistics of discharge between the even and odd data

sets and the early and late data sets are mostly attributed to extreme events that occurred during

these years, in particular the high rainfall in 1964. Removal of this year from the analyses cuts

the difference between the means of the Steinhatchee even and odd periods by 73% and the

difference between the means of the Steinhatchee late and early periods by 97%. So it is

unlikely that the higher means, standard deviations, and ranges of discharge for the even and

early data sets cause their favorable r, RE, and MSE validation statistics (Table 4-2 and 4-3).

The difference in the validation statistics of the Steinhatchee even and odd calibration

models indicates that there is a breakdown of the linear relationship during odd years. During

even years, the Steinhatchee experiences considerably greater range of precipitation and

discharge than during odd years. Since there is no known biological or climatological

phenomenon occurring on biennial timescales that could cause differences in a prediction model

based on even years compared to odd, the difference in these validation statistics is attributed to

sample size. Three of the biggest outliers fall on even years (1956, 1964, and 1998). The

highest year of flow (1964, 25.92 m3/S) and lowest year of flow (2002, 1.45 m3/S) during the

Steinhatchee period of record both occurred during even numbered years. The discharge of the

Steinhatchee during 1964 (25.92 m3/S) is 179% larger than the highest year of discharge during

the Steinhatchee odd period (14.49 m3/S during 1985).

The differences in range of flow and precipitation lead to large disparity in the correlation

models between the subsets and resulted in significant prediction errors. The influence that a

single outlier can have on a 25 year long calibration model is demonstrated by the 32% reduction





























To trees. They are such useful little buggers.











Final Graphs

The final validated actual and predicted Waccasassa and Steinhatchee River discharge


from 1875 -2003 are plotted in figures 4-18 and 4-19 respectively. Prior to 1885 there is a very


large amplitude increase in discharge, which is probably the result of low number of core


measurements. A period of low flow is suggested by the graphs extending from 1884 to


approximately 1910. During this period of time, 44% of the discharge predictions for the


Waccasassa River fall outside the range of flow that occurred during the Waccasassa calibration


period and only three of the years during this time have predicted flow above the average


measured discharge. For the Steinhatchee River five years of predicted flow during the 1884 to


1910 period fall below the range of flow that occurred during the Steinhatchee calibration period


and only one year during this time has predicted flow above the average measured discharge.


TRIG




1.8

1.6

1.4 -

1.2-



0.8-

0.6 -

0.4

0.2


1875 1885 1895 1905 1915 1925 1935 1945 1955 1965 1975 1985 1995 2005
Year
|STNDRD -ARSTAN


Figure 4-1. STNDRD TRIG and ARSTAN TRIG plOtted through time.









period starts in 1900). This dry year may have also had a negative effect on the trees in this

region that lasted for several years.

Overall the Goethe TRI correlated with most gauging stations analyzed in this study better

than the Camp Branch TRI (Figure 4-8). The better correlation of the Goethe TRI than the

Camp Branch TRI could reflect the proximity of the Camp Branch site to the Suwannee River.

Trees in close proximity to a river can be more susceptible to floods or may be receiving a more

consistent supply of water, thereby limiting the broad climatic effects on tree-ring growth.

Biological indications of the results

The predictive capabilities of the two reconstruction models indicate that water availability

in Florida is a predominant factor in the growth rates for longleaf pine species in the humid

subtropical regions of the Southeastern United States. This supports the findings Meldahl et al.

(1999), who determined that climate plays a significant role in growth of longleaf pine and that it

is a good candidate for southeastern tree-ring analyses. Longleaf pine' s inability to effectively

compete in wetter environments that are more conducive to growth, and their common

occurrence on poorly drained soils, may explain why they are moisture limited (Foster and

Brooks, 2001) and thus useful for tree-ring analyses. An example of the link between tree

growth and precipitation is shown in Figures 4-12 and 4-13 during the 1964 high flow event,

when predictions of flow based on the Goethe TRI reflect a large percent flow for Steinhatchee

and Waccasassa Rivers as shown by the instrumental record.

The growth-discharge correlation was the strongest with annual discharge as opposed to

seasonal or monthly, indicating that water availability throughout the year positively affects

growth of longleaf pines in both forests that were studied. Longleaf pine growth is continuous

throughout the year and does not go dormant during the colder or drier parts of the year, as is

common in some species of trees in more temperate sites (Fritz 1976). The continuous growth









ASSESSMENT OF TREE-RINGS AS A RECORD OF PRE-HISTORIC STREAM FLOW IN
A SUBTROPICAL ENVIRONMENT



















By

KRIS CROCKETT


A THESIS PRESENTED TO THE GRADUATE SCHOOL
OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OF
MASTER OF SCIENCE

UNIVERSITY OF FLORIDA

2007










Precipitation and Evapotranspiration Correlation

The correlation shown in figure 4-17 shows a strong relationship (r = 0.69) between the

correlation coefficients of precipitation to flow compared with correlation coefficients of flow to

the Goethe TRI. This relationship demonstrates a useful way for checking how strongly a

potential forest sample site may correlate with discharge stations in an area. If there is a rain

gauging station in close proximity to a forest sample site then the degree of correlation between

the station's data and a nearby stream gauging station will provide a meaningful assessment of

how well the forest might predict streamflow before even collecting and analyzing any tree

cores.

The results shown in table 4-3 indicate that there is little to no improvement in the

correlation coefficients of the precipitation after being adjusted for the effects of

evapotranspiration data. If the annual average of evapotranspiration rates is constant through

time (as indicated by Figure 4-16) then the effects of evapotranspiration on tree growth need not

be accounted for when using tree-ring proxies for climate in the state of Florida. The

Thornthwaite-Mather evapotranspiration model, however, does not account for the duration of a

rain event or the amount of time between rain events, which may have a large effect on the

amount of evapotranspiration affecting the soil. For instance, a moderate rainfall of one inch on

a dry soil may only wet the upper 10 inches of soil, while the lower areas remain dry, so unless

more moisture is added, water will not move to deeper levels by gravity flow (Fritz, 1976).

Flow Proxies

The reconstruction presented here produces a proxy for flow to 1875 for the Waccasassa

and Steinhatchee Rivers. These proxies show a period of low flow in 1932 and 1916-1917 that

corresponds with the dust bowl and a period of low precipitation (Figures 4-18 and 4-19; Henry,

193 1). The period of predominant high flow stretching from 1919 to 1924 also corresponds well









A residual of a linear regression model is the observed data minus the predicted data

(Equation 3-2).

residual = vt -ft (3 -2)

Residual analyses were carried out on the reconstructions of discharge based on a TRI. To test

the assumption that the discharge-TRI relationship is linear, a scatter plot of the residuals of

discharge versus TRI was analyzed. To test the assumption that errors have a constant residual

variance with varying tree-ring widths, scatter plots of the absolute values of the residuals of

discharge versus TRI, residuals of discharge over time, and the residuals of discharge versus

discharge minus the mean discharge were analyzed. Any functional relationship between these

variables would be a violation of the assumption of linearity. Patterns or trends apparent in these

plots would violate the assumptions of constant residual variance.

Reconstructionn models were verified using a data splitting method (Fritts 1976), which

involves dividing data into a calibration period and a validation period to assess the accuracy of

the reconstructionn model. Data were split in two ways: late and early years (i.e. 1951-1976 and

1977-2003) and odd and even years. Correlation models were constructed for all subsets of data

and model estimates were compared to data not used to create the models. Long term variations

could introduce bias into the data split by early and late years, but would not occur when data are

split into odd and even years.

The Pearson correlation coefficient (Eq. 3-1), mean squared error (MSE), reduction of

error (RE), and coefficient of efficiency (CE) verification statistics were used to assess the

accuracy of streamflow predictions based on the difference between the validation and

calibration periods (North, 2006). The MSE of predicted discharge (9) (equation 3-3)

MSE (9) = (1/N) 1 (yt- 9t)2 (3-3)










LIST OF FIGURES


FiMr page

1-1 Map of study area with the l ocati on of US GS Stream gauging stati ons ................... .........1 4

2-1 Steinhatchee River and gauging station (adapted from http ://tiger.census.gov/. ...............20

2-2 Map of Waccasassa River and gauging station along with the Goethe sample site. .........21

2-3 Map of the Presettlement range of longleaf pines (light gray) and their current extent
(dark gray) ................. ...............22......_._. .....

4-1 STNDRD TRIG and ARSTAN TRIG plOtted through time.................. ...............3

4-2 STNDRD TRICB and ARSTAN TRICB plOtted through time. ............. .....................3

4-3 Histogram of number of tree ring width measurements per year for the Goethe site
with STNDRD Goethe TRI (TRIG) OVerlay .................... ...............3

4-4 Histogram of number of tree ring width measurements per year for the Camp Branch
site with STNDRD Camp Branch (TRICB) OVerlay............... ...............39

4-5 How the EPS fluctuates with sample size for the TRIG and TRICB. The red line
represents the 0.85 critical threshold. ............. ...............40.....

4-6 Average daily discharge of the Steinhatchee River per month for the entire period of
record. ............. ...............41.....

4-7 Average daily discharge of the Waccasassa River per month for the entire period of
record. ............. ...............41.....

4-8 Correlation coefficients of the 14 gauging stations TRIGt and TRICB. All reported
correlation coefficients for this histogram are for the entire yearly record for each
year adjusted to the starting month of best fit ................. ...............43........... .

4-9 Histogram of 12 month correlation coefficients for the TRIG taken for a 12 month
period starting at different month forward and backwards from a calendar year. .............43

4-10 Waccasassa discharge plotted against TRIG. The Waccasassa prediction model and r
value are given above the best fit line............... ...............44..

4-11 Steinhatchee discharge plotted against TRIG. The Steinhatchee prediction model and
r value are given above the best fit line. ............. ...............44.....

4-12 Waccasassa River predicted and actual discharge. .............. ...............45....

4-13 Steinhatchee River predicted and actual discharge ................. .............................45









have dramatic shifts in the dominant tree species in an area (Foster and Brooks, 2001). Of the

numerous tree species predominant in Florida' s forested ecosystems, longleaf pine (Pinus

palustris) was chosen for this analysis because of its characteristics of longevity (>400 years),

wide geographic range, and resistance to fire, disease, drought, and insects (Meldahl et al.,

1999). Also the relationship of longleaf pines to water availability does not depend on depth to

the aquifer. This factor has been shown to be important to the growth of slash pine (Pinus

elliottii), which is another dominant pine in north Florida (Foster and Brooks, 2001). Although

longleaf pine has been shown to prefer mesic sites to xeric sites in terms of growth rate, it is out-

competed by slash pine in mesic areas (Foster and Brooks, 2001). Consequently longleaf pines

are common in xeric sites where frequent fires limit the growth of slash pine and other native

Florida tree species (Myers and Ewel, 1990; Foster and Brooks, 2001). Water availability alone

is thus not responsible for where longleaf pine will dominate (Foster and Brooks, 2001).

Longleaf pine forests used to blanket the southeastern United States (Figure 2-3).

Logging, development, and fire suppression have depleted this ecosystem by 97% of its

presettlement extent (Loudermilk, 2005; Van Lear et al., 2005) and consequently there are few

sites with a significant number of old trees. Two promising old-growth forests were sampled for

this study. One is the Goethe State Forest located partly within the Waccasassa drainage basin

and the other is an old turpetined flatwoods ecosystem growing along the Suwannee River,

which is referred to in this study as the Camp Branch Forest (Figure 1-1).










North, G.R. (chr), 2006, Committee on Surface Temperature Reconstructions for the Last 2,000
Years, National Research Council, Surface temperature reconstructionss for the last 2,000
years: The National Academies Press, 160 p.

Obeysekera, J., Trimble, P., Neidrauer, C., Pathak, C., VanArman, J., Strowd, T., and Hall, C.,
2006, Consideration of long-term climatic variability in regional modeling for SFWMD
planning & operations: South Florida Water Management Distict, 45 p.

Pederson, N., Jacoby, G.C., D'Arrigo, R.D., Cook, E.R., Buckley, B.M., Dugarjay, C., and
Mij iddorj, R., 2001, Hydrometeorological reconstructionss for Northeastern Mongolia
derived from tree rings: AD 1651-1995: Journal of Climatology, v. 14, p. 872-881.

Taylor, G.F., and Snell, L.J., 1978, Water resources of the Waccasassa River basin an
adjacent areas, Florida: U.S. Geological Survey Water-Resources Investigations 77-
101, 1 sheet.

Scott, T.M., Means, G.H., Meegan, R.P., Means, R.C., Upchurch, S.B., Copeland, R.E. Jones, J.,
Roberts, T., and Willet A., 2004, Springs of Florida: Tallahassee, Florida, Florida
Geological Survey Bulletin No. 66, 377 p.

Schmidt, N., Lipp, E.K., Rose, J.B., and Luther, M.E., 2001, ENSO influences on seasonal
rainfall and river discharge in Florida: Journal of Climate, v. 14, p. 615-628.

Schumacher, F.X., and Day, B.B., 1939, The influences of precipitation upon the width of annual
rings of certain timber trees: Ecological Monographs, v. 9, p.387-429.

SRWMD, 1995, Waccasassa River watershed management plan: Live Oak, Florida, Suwannee
River Water Management District Surface Water Improvement and Management
Program, 65 p.

SRWMD, 1989, Steinhatchee River basin assessment: Live Oak, Florida, Suwannee River
Water Management District Interim Report, 86 p.

Thornthwaite, C.W., and Mather, J.R., 1957, Instructions and tables for computing potential
evapotranspiration and the water balance. Drexel Institute of Technology, Publications in
Climatology, vol X, 311 p.

Tootle, G.A., Piechota, T.C., and Singh, A.K., 2005, Coupled oceanic-atmospheric variability
and U.S. streamflow: Water Resources Research, v. 41, W12408,
doi: 10. 1029/2005WROO43 81 .

Watson, I., and Burnett, A., 1995, Hydrology, an environmental approach: New York, Lewis
Publishers/Crc Press, Times Mirror Book, 702 p.

Wendland, W.M., and Watson-Stegner, D., 1983, A technique to reconstructt river discharge
history from tree-rings: Water Resources Bulletin, v. 19, p. 175-181.












Waccasassa Q vs TRIG


0.8 1 1 .2 1 .4 1 .6 1 .8
TRIG


Figure 4-10. Waccasassa discharge plotted against TRIG. The Waccasassa prediction model and
r value are given above the best fit line.


Steinhatchee Q vs. TRIG


20









5


TRIG


Figure 4-11. Steinhatchee discharge plotted against TRIG.
and r value are given above the best fit line.


The Steinhatchee prediction model












TABLE OF CONTENTS


page

ACKNOWLEDGMENT S ................. ...............4.......... ......


LIST OF TABLES ................ ...............7............ ....


LI ST OF FIGURE S .............. ...............8.....


AB S TRAC T ............._. .......... ..............._ 10...


CHAPTER


1 INTRODUCTION ................. ...............12.......... ......


2 STUDY AREA ................ ...............15........... ....


3 METHODS .............. ...............23....


Tree-Ring Data .............. ...............23....
COFECHA .............. ...............23....
ARSTAN ................. ...............24.......... .....

Dischar ge Data ................. ...............25.................
Statistical Method s............... ...............26


4 RE SULT S ................. ...............3.. 1......... ...


T ree-ring Data ................. ...............3.. 1..............
COFECHA .............. ...............3 1....
ARS TAN ................. ...............3.. 1..............

Dischar ge Statistics............... ...............3
Steinhatchee River ................. ...............32.................
W accasassa River ................. ....... ........ .......... .............3

Statistical Analysis of the Discharge-TRI Correlations. ......____ .......___ ...............33
Residual Analysis .............. ...............34....
Validation statistics ............... .. ........... ...........3

Precipitation and Evapotranspiration Correlation .............. ...............36....
Final Graphs............... ...............37.


5 DI SCUS SSION ............. ..... ._ ............... 2....


Tree-Ring Indices .............. ..... ...............52
Biological indications of the results .............. ...............53....
Validation Statistics ................ ....... .. ..__....... ..._ .. ................5

Influence of Pacific Decadal Oscillation (PDO) on Tree-Ring Indices ................ ...............56
Final m odel .............. ......... .. .. ............5

Precipitation and Evapotranspiration Correlation .............. ...............59....















1.


0.8

0.7


0.6"
0.5
8 0.4

0.3

0.2

0.1











I Goethe Chronology I Camp Branch



Figure 4-8. Correlation coefficients of the 14 gauging stations TRIGt and TRICB. All reported
correlation coefficients for this histogram are for the entire yearly record for each

year adjusted to the starting month of best fit.



Adjusted Year Correlations




0.9

0.8 --

0.7 --



s 0.6



U 0.4-

0.3-

0.2-

0.1-



Jul-Jun Aug-Jul Sep-Aug Oct-Sep Nov-Oct Dec-Nov Jan-Dec Feb-Jan Mar-Feb Apr-Mar May-Apr Jun-May
Month

SWaccasassa H Steinhatchee


Figure 4-9. Histogram of 12 month correlation coefficients for the TRIG taken for a 12 month
period starting at different month forward and backwards from a calendar year.











































































1875 1895 1915 1935 1955 1975 1995
Year Feb.-Jan

-Measured Q -Predicted Q

Figure 4-19. Actual and predicted Steinhatchee River discharge in m3/S from 1875 -2003.


- -11


-


--
I g I


Waccasassa Final Graph


1875


0


1935
Year Feb.-Jan.


'VMeasured IPredicted

Actual and predicted Waccasassa River discharge in m3/S from 1875 -2003.


Figure 4-18.


Steinhatchee Final Graph


30


25


20


E 15


10


5









to increased climatic heterogeneity in the study area. This heterogeneity implies that climate

signals recorded in tree-rings are more likely to differ from climate signals that are recorded in

the discharge and rain gauging stations that are some distance apart. Since 81% of the

Steinhatchee late calibration period falls during a PDO warm event this is a likely mechanism for

the loss of correlation calculated for the late period.

The effects of the PDO on the study area can be further seen in the Madison and Ocala rain

gauging stations. As shown in Figure 1-1 these stations are located on opposite edges of the

study area. The combined yearly precipitation averages of these two stations for the warm phase

is 130.5 cm/year and for the cool phase is 132.2 cm/year representing a difference of only 1.3 %.

Correlation of these stations' precipitation data over the 1977 to 1998 period for the warm phase

of the PDO cycle had a Pearson correlation coefficient of 0. 14. In contrast, these stations had a

Pearson correlation coefficient of 0.55 from 1951 to 1976 and 1999 to 2003 during the cool

phase. This further illustrates the climate heterogeneity occurring during the PDO warm phase

years of 1977 to 1998.

The effect of the PDO cycle on the Goethe TRI to discharge correlation can be further seen

in Figure 4-13 during the 1977 to 1998 warm phase. During this warm phase there is a low

correlation between the actual and predicted discharge record (r = 0.20 for the warm phase and

0.81 for the cool phase). The RE statistic calculated for the Steinhatchee discharge-Goethe TRI

correlation during the warm phase is -0.04, which indicates that the Goethe TRI and the

Steinhatchee River discharge show little to no common variability during this time. By

disregarding the warm phase of the PDO, there is an average of 23% improvement in the Pearson

correlation coefficients of all 14 gauging stations, suggesting that the warm phase of the PDO

introduces a large degree of error in every gauging station used in this study (Table 5-1). The










Florida: Tallahassee, Florida Bureau of Surface Water Management, Division of Water
Management, Department of Environmental Regulation, Standards and Monitoring
Section, 289 p.

Henry, A.J., 1931, The calendar year as a time unit for drought statistics: Monthly Weather
Review, v. 59, p. 151-154.

Henry, J.A., Porter, K.M., and Coyne, J., 1994, The climate and weather of Florida: Sarasota,
Florida, Pineapple Press, Inc., 279 p.

Jacobs and Ripo 2002, Minimum flows and levels for the Lower Suwannee River
implementation methodology: Live Oak, Florida, Suwannee River Water Management
District Final Report, p. 129.

LeBlanc, D., and Terrell, M., 2001, Dendroclimatic analyses using Thornthwaite-Mather-Type
evapotranspiration models: A bridge between dendroecology and forest simulation
models: Tree-Ring Research, v. 57, p. 55-56.

Lodewick, E.J., 1930, Effect of certain climatic factors on the diameter growth of longleaf pines
in Western Florida: Journal of Agricultural Research, v. 41, p. 349-363.

Loaiciga, H.A., Haston, L., and Michaelsen, J. 1993, Dendrohydrology and long-term hydrologic
phenomena: Reviews of Geophysics, v. 31, p. 151-171.

Loudermilk, E.L., 2005, Spatial and demographic modeling techniques applied to the longleaf
pine (Pinus Palustris) ecosystem of North Central Florida [M.S. Thesis]: Gainesville,
University of Florida, 80 p.

Mantua, N.J., and Hare, S.R., 2002, The Pacific Decadal Oscillation: Journal of Oceanography,
v. 58, p.35-44.

Meko, D., Stockton, C.W., and Boggess, W.R., 1995, The tree-ring record of severe sustained
drought: Water Resources Bulletin, v. 31, p. 789-813.

Meko, D., and Graybill, D.A., 1995, Tree-ring reconstruction of Upper Gila River discharge:
Water Resources Bulletin, v. 31, p. 605-616.

Myers, R.L., and Ewel, J.J., 1990, Ecosystems of Florida: Orlando, University of Central
Florida Press, 765 p.

Meldahl, R.S., Pederson, N., Kush, J.S., and Varner, J.M., 1999, Dendrochronological
investigations of climate and competitive effects on longleaf pine growth. in Wimmer, R.
and Vetter, R.E. eds., Tree Ring Analysis: Biological, Methodological and Environmental
Aspects: Wallingford, UK., CABI Publishing, p. 265-285.









correlation changes with different water year is shown for Waccasassa and Steinhatchee gauging

stations (Figure 4-9). By using a February water year for the Steinhatchee and Waccasassa

gauging stations, both stations show a 1.5% improvement in correlation from using a water year

beginning in January and a 3% improvement from using a water year beginning in March.

To illustrate the correlations between the Goethe TRI and discharge, the Goethe TRI was

plotted against the recorded discharge for both the Waccasassa and Steinhatchee rivers (Figures

4-10 and 4-11). The correlation equations were used to reconstruct discharge for the Waccasassa

and Steinhatchee. Model discharge for the Waccasassa River is derived by equation 4-1

Q = 12.53 x Goethe TRIGt 3.71 (4-1)

and derived for the Steinhatchee River in equation 4-2,

Q = 14.54 x TRIGt 5.29 (4-2)

where Q is discharge in m3/S and TRIGt is the value given by the Goethe tree-ring index at year t.

Predictions of discharge using equations 4-1 and 4-2 are plotted with the observed

discharge over both the Waccasassa (Figures 4-12) and Steinhatchee (Figures 4-13) gauging

histories. Both stations correlate well with the Goethe TRI during the period of time between

1964 and 1977. Outside of this time period the correlation is not as strong. This period of time

includes the highest flow event and one of the most severe droughts in both periods of record.

This period of time also makes up 72% of the total Waccasassa gauging record, and thus it is not

possible to determine how well the model predicts actual flow outside of this period.

Residual Analysis

Residual analysis was conducted only for the Steinhatchee-Goethe TRI correlation,

because of the shortness of the gauging record for the Waccasassa River. Both the Waccasassa

and Steinhatchee gauging records correlate well with each other (r = 0.74) and the rivers are only






































Figure 2-3. Map of the Presettlement range of longleaf pines (light gray) and their current extent
(dark gray) (adapted from http://biology .usgs.gov/s+t/SNT/noframe/sel130.htm).














1 13. --l ..
-' 1 1. Waccasessa River
a 2 Steinha~chee Rver
-t Qi i Fenlh iloway River
~4, Jumpe Creek Canal
5. T.am~bwrr Sprngs
i~ bWethlaco~hee Raver
1, I 7 Santa Fe River
8 1 :: L~ ~k, Suw~amneeRiver
-/i 9r Santa Fe BRiver
10.t ~r Suwaames River
~-~" _1 11. Suwannee River
,'12. Suwairnee River
1 ~'13. P.Frlat.a. >: hc h River
14 Alspadu P;ivr
~:- ~ : ~.JNOAA Precipitation Stationsa
1, Ushler ToweFr
It L- I 1( --2 MadBisOn
LEGEND 3. Ocala
A sheamatrnosliosrlon ----- Coutyri 4h Lake Cllr
MDA Pratnoap ~pUn Simores
SGoelhe StateFntest
ICelpEntanchrd

Scale 1:23811181 lo, o7~$67ionlp 14 1,


Station Identification #i
2313700
2324000
2324400
231264~0
123313100
23 1 *%100
2321500
2',?5;00
2322500
T23 41.500
2315500
2320500
2;319000
".'37500

89120
85275
86414
$4771


Figure 1-1. Map of study area with the location of USGS Stream gauging stations and the NOAA
Precipitation Station where flow and precipitation data were compiled. Areas of
regions where tree cores were collected are shown in dark green for Goethe and light
green for Camp Branch (adapted from http://tiger.census .gov/).










LIST OF TABLES

Table page

4-1 Summary of TRI Statistics............... ...............3

4-2 Annual (Feb. to Jan.) Descriptive Statistics of Discharge ........... _..... ..._ ............42

4-3 Validation Statistics .............. ...............47....

4-4 Precipitation and Recharge Correlations .............. ...............49....

5-1 The growth-discharge correlation for the cool phase PDO cycle. .................. ...............61









CHAPTER 3
METHOD S

Tree-Ring Data

Standard core sampling techniques (Fritts, 1976) were used by Tom Mirti to collect tree

cores from the two forests. Large diameter trees with minimal defects (e.g., lack of storm

damage and/or rot) were selected by assuming them to be among the oldest specimens. From

each tree at least one core was extracted with a 5.15 mm incremental borer 0.5 to 1.5 meters

above ground level. Two cores are preferable to aid in the identification of missing rings which

can occur when a tree is under stress (e.g. drought or severe competitive suppression).

From the Goethe site, 55 cores were collected from 29 longleaf pines. From the Camp

Branch site, 47 cores were collected from 30 living longleaf pines and 2 longleaf pine stumps.

After extraction, the cores were air dried and affixed to wooden mounts, and labeled. The ring

widths were measured at the University of Tennessee Tree-Ring Laboratory to the nearest 0.01

mm using a Velmex measuring system interfaced with Measure J2X software. Each core was

measured from the first complete innermost ring to the last complete ring before reaching bark.

COFECHA

The raw tree-ring width data were processed at the University of Tennessee Tree-Ring

Laboratory with COFECHA, which is a computer program that assesses the quality of dating and

measurement accuracy of a tree-ring series (Grissino-Mayer, 2001). COFECHA checks the

accuracy of assigned tree-ring dates by correlating individual core measurements with a master

tree-ring record constructed from the remaining series (Holmes, 1983; Grissino-Mayer, 2001).

Individual cores that correlate poorly with the master record are flagged by COFECHA and then

visually inspected for missing or false rings. If the source of the error can be identified, the core

is redated and reanalyzed with the COFECHA program. Cores that cannot be dated accurately









CHAPTER 6
CONCLUSION

The forest sample sites chosen for this study were picked for their ability to provide

suitable cores, and not for their propensity to reflect runoff of any particular drainage basin.

However, despite this limitation, results of this study indicate some value in the use of

dendrohydrologic reconstructions of historic discharge in Florida. The 128 year Waccasassa and

Steinhatchee river reconstructions are examples of how dendrohydrologic reconstructions could

provide insight to the range of short hydrologic variability that has occurred in Florida' s past.

These reconstructions demonstrate the ability to hind-cast several droughts outside the

calibration period that correspond with instrumental records. Longleaf pine growth in Florida

responds well to variable water availability and seems to accurately reflect runoff in the area

where they reside.

The maj or inaccuracies in the reconstructions appear to come largely from the

underestimation of maxima and overestimation of minima of hydrologic predictions and from

Florida's variable regional climate. This climate heterogeneity increases during the warm phase

of the PDO cycles which are time periods of poor correlations between gauging stations and

forest sites. These dendrohydrologic reconstruction techniques may be of use, however, because

most large scale climate anomalies (droughts or floods) occur during PDO cool phases and

climate anomalies are the periods of greatest interest to water resource planners. As demands for

water resources in the State continue to grow, our susceptibility to drastic hydrologic changes

will also increase. Insights into the magnitude of departure from current perception of "average

hydrologic conditions" could be critical in water resource planning. Results of this study should

aid in how dendrohydrologic studies are employed in this area and give some perspective on

what climatic information may be garnered from the rings of trees.









against the master record due to the presence of too many false or missing rings are not included

in subsequent analyses. Each core used in the master record is correlated with the others to

create the series intercorrelation, which is a statistical quantity representing the variability

between all tree cores in a set. The series intercorrelation is expressed as a Pearson correlation

coefficient (r) that measures the strength of a linear relationship (Equation 3-1).

r= [{L(yt-9) 2- f (t- ~t 2/ (Yt- )2 (3-1)

Where yr is the measured value at some year t (in this study the measured value is either

standardized tree-ring widths, discharge, or precipitation), it is equal to the predicted value at

year t, and y is equal to the average of all measured values. The significance of the series

intercorrelation can be shown by the expressed population signal (EPS). The EPS is a measure

of how well a tree-ring record made from a sample set is likely to portray a hypothetical tree-ring

record comprised from an entire population and is derived from the series intercorrelation and

the number of trees in the sample (Cook and Kairiukstis, 1990). A value above 85% is set as the

threshold for acceptable statistical quality (Wigley et al., 1984).

ARSTAN

The confidently dated tree-ring measurements were then analyzed at the University of

Tennessee Tree-Ring Laboratory with the ARSTAN program. The ARSTAN program fits a

cubic spline to each set of measurements for every core. The program then creates a

dimensionless tree-ring index (TRI) by taking the ratio between the predicted value of the cubic

spline and the measured ring width value (Cook and Holmes, 1997). This procedure adjusts for a

tree' s natural tendency to produce smaller ring widths with age and also allows values of the

tree-ring widths to be averaged without artifacts from variable growth rates (Cook and Holmes,

1997). The samples were averaged into two initial TRIs (labeled STNDRD version by the









Table 4-2. Annual (Feb. to Jan.) Descriptive Statistics of Discharge
Length Standard
reof Mean Deviation Minimum Maximum Range
Years m3/S
Waccasassa 18 8.21 4.50 3.45 19.45 16.00

Steinhatchee 53 8.85 5.17 1.45 25.92 24.47
Odd years 27 8.5 4.56 1.52 18.66 17.14
Even years 26 9.2 5.80 1.45 25.92 24.47
Early years 26 9.11 5.77 1.64 25.92 24.28
Late years 27 8.59 4.60 1.45 14.49 13.04









calibrated for the even and early data sets had the highest r and RE statistics of the Steinhatchee

data subsets, but did not verify well over their corresponding validation periods (odd and late).

The odd and late models showed a stronger relationship with their validation periods than they

do with themselves as evident by their low REs and high CEs.

Precipitation and Evapotranspiration Correlation

Precipitation data taken from the Usher Tower rain gauge station and potential recharge

(precipitation minus evapotranspiration calculated using the Thornthwaite-Mather

evapotranspiration model) are shown in figure 4-15. The potential recharge calculated using the

Thornthwaite-Mather model shows that 69% of the rainfall in the area returns to the atmosphere

via evapotranspiration. The slope of the line (m=1.0) and the Pearson correlation coeffcient

calculated between the Usher tower precipitation and potential recharge data reveal only minor

differences in trend (r=0.98; Figure 4-16). This result suggests that the average annual amount

of precipitation lost through evapotranspiration is a constant 1059 mm/year through the entire

period of record.

Table 4-4 shows the Pearson correlation coeffcients of the Usher Tower rain gauge

station precipitation and potential recharge data compared with the Goethe TRI, the Waccasassa

gauging data, and the Steinhatchee gauging data including the even and odd subset. The early

and late subset was not included in this comparison, because the Usher Tower rain gauging

record is shorter than the Steinhatchee gauging record and does not provide an equal comparison

among the subsets. The Pearson correlation coeffcients between all 14 stream gauging stations

and the Usher Tower precipitation data is compared to the Pearson correlation coeffcients

between these stream gauging stations and the Goethe TRI (Figure 4-17). This graph shows a

positive correlation (r = 0.69), indicating that as correlation improves between discharge and

precipitation, the correlation also improves between discharge and the TRI.












Precipitation/Potential Recharge


3000



2500




Precipitation- = .0*Potential recharge + 1059
~ r =0.98












0 200 400 600 800 1000 1200 1400 1600
Potential recharge

Figure 4.16. Precipitation plotted against potential recharge.



Table 4-4. Precipitation and Recharge Correlations
Recharge
Precipitation m3/S m3/S
Correlation Coefficients
TRIG 0.68 0.71
Waccasassa 0.9 0.91
Steinhatchee 0.77 0.76
Even years 0.87 0.87

Odd years 0.58 0.56









in the difference between the CE statistics for the even and odd calibration periods after the

removal of the 1964 year (Table 4-3). Despite the differences in the even and odd validation

statistics all values indicate some predictive capacity of the full Steinhatchee calibration model.

Nonetheless care should be taken when assessing the error of the model in its predictive

capability for discharge that falls outside the range of flow that occurs during a calibration period

(Meko et al., 1995).

Influence of Pacific Decadal Oscillation (PDO) on Tree-Ring Indices

There is even more pronounced difference between the validation statistics of the early and

late Steinhatchee sets compared to the even and odd validation statistics. For example the early

calibration model fails to predict the late Steinhatchee validation period (CE = -0.09). These

differences in the predictive capabilities may be attributed to the Pacific Decadal Oscillation

(PD O).

The PDO is a long-lived ENSO-like pattern of Pacific climate variability (Zhang et al.,

1997). The PDO is detected as a shift in the sea surface temperature of North Pacific Ocean to

either a warm phase or a cool phase (Mantua and Hare, 2002). During a warm phase, the

western Pacifie becomes cooler and part of the eastern ocean becomes warmer; during a cool

phase, the opposite pattern occurs (Mantua and Hare, 2002). Cool phases occurred from 1890-

1924 and from 1947-1976, while the warm phases occurred from 1925-1946 and from 1977 to a

possible shift back after 1998 (Mantua and Hare, 2002). These changes in the sea surface

temperatures in the north Pacific alter the path of the j et stream, the conveyor belt for storms

across the continent. During a warm phase there is a shift of the j et stream towards the equator

(Tootle et al., 2005). This equatorial shift brings increased frontal precipitation to the

southeastern United States and as a result summer and fall precipitation and stream flow are

more likely to be the result of highly localized convective storms (Schmidt, 2001) and may lead










program), one each for the Goethe and Camp Branch forests. Due to the effects of the

standardizing and detrending, both TRls have dimensionless means of approximately 1.

The ARSTAN program also produces an additional TRI (labeled ARSTAN version by the

program) from the STNDRD version that removes the effects of autocorrelation in TRI. Tree

growth during an individual year may be affected by the previous year' s favorable or

unfavorable growing season (Fritts, 1976). For instance, a favorable growing season can allow a

tree to build up an excess of sugars that will cause it to grow more the following year than it

would have otherwise given the hydrologic conditions for that year. The ARSTAN program

attempts to remove this trend through autocorrelation modeling. In the following discussion both

final ARSTAN and STNDRD version TRls are used to estimate correlations between tree-ring

records and streamflow records.

Discharge Data

Fourteen gauging stations located in northern Florida were chosen for the initial correlation

analysis between TRI and discharge. These stations were chosen on the basis of their significant

length of record and proximity to the Goethe State and Camp Branch forest sample sites (Figure

1-1). Streamflow records for all gauging stations were taken from the United Sates Geological

Survey (USGS) (http://waterdata.usgs .gov/fl/nwi s/rt). The streamflow records for the

Waccasassa River USGS gauging station 02313700 were sporadic and were only available for

the periods of March 1963 to September 1978, November 1980 to September 1984

(fragmentary), October 1984 to September 1992, and October 1998 to current year. All other

gauging records are continuous over the periods analyzed. Discharge was computed from a

continuous velocity record obtained from a water-current meter.









CHAPTER 1
INTTRODUCTION

Understanding long-term patterns of hydrologic variability is important in planning for

the wise use of water resources. The longest climate and streamflow records in the United States

rarely extend past the 1890s, which may not be long enough to capture the frequency of severe

droughts or floods. This lack of direct observation of climate and flow requires developing

proxy records of these variables. Records of long-term hydro-climatic behavior may be found in

the sediments of streams, lakes, and other depositional basins, but these records are rarely

sensitive enough to capture annual variability. In contrast, trees have a natural and consistent

response to seasonal hydrologic forcing that can record hydrologic variability and record high-

resolution hydrologic phenomena. The linear relationships between tree-ring and climate records

are at least equal to, and often exceed, those found for other proxies (North, 2006).

Tree-ring width has been shown to correlate with historic precipitation and temperature

data in many places around the world (Coile, 1936; Schumacher and Day, 1939; Fritts, 1976;

Pederson, et al. 2001). Using tree-rings to hind cast climate records can be complicated,

however, if they are linked to two or more variables (i.e. precipitation, temperature,

evapotranspiration, and duration of rainfall). Streamflow, which is a function of precipitation,

temperature, and evapotranspiration, is a single variable that has been shown to correlate well

with tree-ring variations (Fritts, 1976). This correlation has already been shown to be valid in

many areas across North America such as the southeastern United States (Cleaveland and Stahle,

1989), the southwestern United States (Meko and Graybill, 1995), the northwestern United

States (Gedalof et al., 2004), and the Canadian prairie region (Case and MacDonald, 2003). The

correlation was strong (r-0.60 to 0.87) between river discharge (Q) and tree-rings in these

studies.









with published climate tables (Henry, 1931). Even prior to 1910, which depicts a period of

extended low flow that is most likely inaccurate based on cursory climate records, some of the

peaks and valleys correspond to state wide highs and lows for precipitation, such as the higher

than normal precipitation year of 1900 and the lower than normal year of 1904 (Figures 4-18 and

4-19; Henry, 1931).









cycle of longleaf pines most likely relates to Florida' s mild winters and their lack of seasonal

defoliation. This characteristic indicates that the earlywood and latewood growth patterns seen

in the longleaf pines relate to seasonal trends in water availability (Figures 4-6 and 4-7). A

similar result has been found by Lodewick (1930) and Foster and Brooks (2001), who saw no

strong correlation between temperature and growth in Florida longleaf pines.

The water year for tree growth extends from February to January as estimated by the best

correlation between tree-rings and discharge (Figure 4-9). This February to January water year

reflects the findings of Jacobs and Ripo (2002) who concluded the ideal statistical water year for

SRWMD area is the period from February 3rd to the following February 2nd. These dates were

chosen by the authors because of the small number of extreme events that had occurred on

preceding and subsequent days. The commonly applied USGS water year of October 1 st to

September 30th divides a period of high flow seen in the Waccasassa and Steinhatchee average

daily discharge records (Figures 4-6 and 4-7) and is not applicable for dendrohydrologic

reconstructions in the state of Florida.

Validation Statistics

Split sample validation is not commonly used in simple linear regression due to the

straightforwardness of a model that is based on only two variables. Data transformation and

autocorrelation modeling was also not performed on the data sets in order to create the simplest

model possible. As additional variables are added into a prediction model and as data

transformation is done, a prediction model will begin to fit the data. Model overfitting that can

occur is minimized in this study by keeping the prediction models used as simple as possible.

Calibration models developed from the early and even years of the Steinhatchee discharge data

did not predict the corresponding validation periods as well as the late and odd years (Table 4-2).
























r of Q-precipitation = 1.068 x r of Q-TRIG 0.1534
r = 6


T VS T


0.8



0.7






S0.4




0.3





0.2


r of Q-TRIG


Figure 4-17. The correlation coefficients of all 14 stream gauging stations plotted to the Usher
Tower raw precipitation data vs. correlation coefficients of the same stream gauging
stations to the Goethe tree ring TRIGt and TRICB -







































LEGEND

A Steinhatchee Gauging station County line
0 2 4 6 8 10 mi
Sca le 1: 335650 a eI 2 0 115 12 k



Figure 2-1. Steinhatchee River and gauging station (adapted from http:.//tiger.census.gov/









Statistical Analysis of the Discharge-TRI Correlations

Annual descriptive statistics for the Waccasassa and Steinhatchee gauging stations and the

split data sets for the Steinhatchee flow for the adjusted year of February through January are

given in table 4.2. Subsets for the Waccasassa gauging record were not made due to the short

period of recorded discharge for this station (Table 4-2).

All stations showed a positive correlation between TRIs and discharge, with Pearson

correlation coeffieients ranging from 0.12-0.86 (Figure 4-8). At most gauging stations the

Goethe TRI correlates with flow better than the Camp Branch TRI. The only gauging stations

that correlated better with the Camp Branch TRI were stations 10 and 11 along the Suwannee

River, both of which do not have reported discharge for at least the last 25 years (1928 to 1978

for station 10 and 1928 to 1957 for station 11). The Pearson correlation coefficient (Eq.3-1)

calculated for the Steinhatchee River gauging data and the Goethe TRI (r = 0.64) is within the

range of values from reviewed literature (Cleaveland and Stahle, 1989; Meko and Graybill,

1995; Gedalof et al., 2004; and Case and MacDonald, 2003) and on this basis further regression

analysis and model development was carried out between these two data sets. The Pearson

correlation coefficients (Eq. 3-1) calculated between the Waccasassa River gauging data and

both the Goethe TRI and Camp Branch TRI (r = 0.86 for both) are also well within the range of

the reviewed literature. The Goethe TRI was used in the regression analysis and model

development for the Waccasassa River historic flow reconstruction because of its better

correlation to most of the other gauging stations with longer records and proximity to the

Waccasassa River (approximately 15.8 km apart).

Seven of the gauging stations show a maximum correlation for a water year beginning in

February and ending in January, Hyve show a maximum correlation from January to December,

and two show a maximum correlation from March to February. An example of how the




































O 2007 Kris Crockett
















Goethe EPS


1.2







0.8





S0.4


0.2



0.


0 5 10 15 20 25 30 35 40 45 50

Number of trees

-EPS -0.85 Threshold

Figure 4-5. How the EPS fluctuates with sample size for the TRIG and TRICB. The red line
represents the 0.85 critical threshold.









ACKNOWLEDGMENTS

I thank my advisor Dr. Martin and my committee members Dr. Brenner and Dr. Screaton

for sticking by me through to the end. I thank Tom Mirti for introducing me to this field of study

and providing me with the necessary data. I thank my family for all their support. Finally I

thank all the professors and my friends here at the University of Florida Department of Geologic

Sciences for making my experience here at UF the best time of my life.









correlated with all 14 streamflow records to see how discharge in these stations reflected

precipitation events experienced by the forest. Streamflow that has been changed through human

influences (i.e. dams or weirs) or that is receiving runoff from a greatly differing climatic area

should correlate poorly.

Both raw precipitation data and precipitation data adjusted for the effects of monthly

potential evapotranspiration (PEm) were used in the TRI to precipitation correlations. PEm was

calculated using the Thornthwaite-Mather (1957) evapotranspiration model is derived by

equation 3-9.

PEm = 16 DLm [10 Tm /HI]a mm (3-9)

Where Tm is the mean monthly temperature in degrees C, HI is the heat index and equaled to the

sum of the 12 HIm values (equation 3-10).

HIm = [Tm/5]1. (3-10)

The a is a coefficient derived by equation 3-11.

a = (6.7e-7) (HI3) (7.7e-5) (I2) + (0.018) (HI) + 0.49 (3-11)

DLm in equation 3-9 is the monthly day length multiplier calculated by equation 3-12.

DLm ={a +b (Latitude)1.35})/a (3-12)

The DLm relates hours of sunlight at certain latitudes and is only valid for a range of latitudes

from 0 to 50o (LeBlanc and Terrell, 2001). Values for "a" and "b-" ineuto -2aegvnb

LeBlanc and Terrell (2001).

The Thornthwaite-Mather evapotranspiration model assumes that the only effects on

evapotranspiration are meteorological conditions and ignores the density of vegetation. Despite

these simplifications, this method gives a reasonable approximation of PET, and is especially

suited for humid regions such as Florida (Watson and Burnett, 1995). Mean monthly air













FI ow Proxies .............. ...............59....


6 CONCLU SION................ ..............6


LIST OF REFERENCES ................. ...............63................


BIOGRAPHICAL SKETCH .............. ...............67....









Abstract of Thesis Presented to the Graduate School
of the University of Florida in Partial Fulfillment of the
Requirements for the Master of Science Degree

ASSESSMENT OF TREE-RINGS AS A RECORD OF PRE-HISTORIC STREAM FLOW IN
A SUBTROPICAL ENVIRONMENT

By

Kris Crockett

August 2007

Chair: Jonathan Martin
Major: Geology

Understanding long-term patterns of hydrologic variability is important in planning for the

wise use of water resources. One method that could extend the historical record of hydrologic

variability is by analysis of tree-ring widths. Such use of tree-rings has been rare in subtropical

environments because dendrohydrologic research is usually conducted where severe water stress

strengthens the relationship between tree-ring growth and climate. My study assesses the utility

of tree rings in subtropical regions of north Florida as records of hydrologic variability. Tree-

ring indices from two forests were compared with annual discharge from 14 river gauging

stations in North Florida with linear correlation coeffieients ranging from 0. 12 to 0.86. Annual

flows from two gauging stations, one each on the Steinhatchee and Waccasassa rivers were

found to correlate well with tree-ring indices (r = 0.64 and 0.86 respectively) and the correlation

models were used to extend river flow records to 1875. Accuracy of these linear flow models

was tested by statistical comparison with historical climate records. A maximum reduction in

error of 71% over just using average discharge for the prediction model alone was achieved by

the tree-ring model. Correlations between tree rings and flow were diminished during the warm

phase of the Pacific Decadal Oscillation (PDO), but were greatly improved during the PDO cool

phase. The warm phase of the PDO brings an equatorial shift of the j et stream, which increases





Year
-SNDRD -ARSTA


Figure 4-2. STNDRD TRICB and ARSTAN TRICB plOtted through time.












Table 4-1. Summary of TRI Statistics
TRIG TRICB
Count 197 129
Mean 0.98 0.93
Standard
deviation 0.26 0.3
Minimum 0.26 0.31
Maximum 1.67 1.83

Range 1.41 1.52


TRICB


1800


02


1815 1830 1845 1860 1875 1890 1905 1920 1935 1950 1965 1980 1995















Residuals vs. Q Minus Mean Q


Q m /s mean Q m /s


Figure 4-14. Discharge of the Steinhatchee River in m3/S minus mean flow plotted against the
residuals of the TRIGt-Steinhatchee discharge linear regression.









As is generally the case in dendroclimatic studies, fewer cores are available to create the

TRls back in time, thereby reducing their EPS (Figure 4-3 and 4-4) (Gray et al., 2004). An

example of how the EPS fluctuates with sample size for the Goethe TRI is given in Figure 4-5.

Since the series intercorrelations calculated for the Goethe TRI and Camp Branch TRI are close,

Figure 4-5 is used to represent the Camp Branch TRI as well. The subsample signal strength for

the Goethe TRI and Camp Branch TRI does not drop below 85% probability of representing the

hypothetical population TRI until 1889 for the Goethe TRI and 1860 for Camp Branch TRI when

the number of trees that make up the TRIs drops below 5 (Figures 4-3 and 4-4).

Discharge Statistics

Steinhatchee River

The annual mean daily discharge at this station over the calendar year is 8.87 cubic

meters per second (m3/S); with a maximum of 27.27 m3/S in 1964 to a minimum of 1.44 m3/S in

2000. Fifty-two years of daily observations were used to create an average daily flow

hydrograph for one calendar year (Figure 4-6). Flow peaks for the year in March at an average

discharge of 14.34 m3/S and then again around August and September at an average discharge of

14.05 m3/S.

Waccasassa River

The annual mean daily discharge at this station over the calendar year is 7.75 m3/S; with a

maximum of 18.75 m3/S in 1964 to a minimum of 2.00 m3/S in 2000. During periods of drought

reported monthly flow averages can be negative indicating reverse flow because of the incoming

tide. Twenty two years of daily observations were used to create an average daily flow

hydrograph for one calendar year (Figure 4-7). Flows peaks for the year in March at an average

discharge of 10. 12 m3/S and then again in September at an average discharge of 13.42 m3/S.














TRIG


1875 1885 1895 1905 1915 1925 1935 1945 1955 1965 1975 1985 1995
Year

INumber ofCores -STNDRD

Figure 4-3. Histogram of number of tree ring width measurements per year for the
with STNDRD Goethe TRI (TRIG) OVerlay.


Goethe site


TRIn


10

15




1805 1815 1825 1835 1845 1855 1865 1875 1885 1895 1905 1915 1925 1935 1945 1955 1965 1975 1985 1995
Year

ENumber of Cores -STNDRD

Figure 4-4. Histogram of number of tree ring width measurements per year for the Camp Branch
site with STNDRD Camp Branch (TRICB) OVerlay.










Wigley, T.M.L., Briffa, K.R., and Jones, P.D., 1984, On the average value of correlated time
series, with applications in dendroclimatology and hydrometeorology: Journal of
Climate and Applied Meterology, v. 23, p. 201-213.

Winsberg M. D., 1990, Florida weather: Orlando, Florida, University of Central Florida Press,
171 p.

Van Lear, D.H., Carroll, W.D., Kapeluck, P.R., and Johnson, R., 2005, History and restoration of
the longleaf pine-grassland ecosystem implications for species at risk: Forest Ecology
and Mangement, v. 211, p. 150-165.

Varner, J.M., and Pederson, N., 2002, Potential for dendrochronological analysis of the
Suwannee River basin tree species: Suwannee River Water Management District, Live
Oak, F.L., 20 p.

Zhang, Y., Wallace, J.M., and Battisti, D.S., 1997, ENSO-like interdecadal variability:
Journal of Climate, v. 10, p. 1004-1020.









in the Steinhatchee River basin (SRWMD, 1989). Relatively thin layers of alluvial deposits and

peat accumulations occur on top of these formations (SRWMD, 1989). Within the Steinhatchee

River basin, most groundwater occurs in the unconfined Floridan Aquifer with minor amounts

occurring in the Surficial Aquifers (SRWMD, 1989). Rainwater percolating through the

overburden recharges the Floridan and Surficial aquifers directly, and during high stages the

Steinhatchee River recharges the aquifers through the streambed and at the Steinhatchee sink

located downstream of the gauging station (SRWMD, 1989). In the northern and northeastern

portions of the basin, in Lafayette County, a transitional zone exists where the Floridan Aquifer

is semi-confined by the Hawthorn Formation and is periodically under artesian conditions

(SRWMD, 1989).

Waccasassa River Basin

The Waccasassa River, along with its principal tributaries Wekiva River, Cow Creek,

Otter Creek, and Magee Branch, drains approximately 2424 km2 (SRWMD, 1995). The

Waccasassa River basin lies in North Central Florida along the Big Bend Coast in Levy,

Alachua, and Gilchrist counties (SRWMD, 1995) (Figures 1-1 and 2-2). The river is about 47

km long and originates west of Bronson from diffuse swamp waters of southern Waccasassa

Flats. The Waccasassa River doesn't develop a defined channel until receiving flow from Levy

Blue Spring, which has an average discharge of 0.251 m3/S (Scott et al., 2004). Elevations

within the area range from sea level at the coast to a maximum of about 43 m above ms1 in

Alachua County.

Similar to the Steinhatchee, the Waccasassa drainage basin is defined by karst

topography. The Floridan Aquifer system is largely unconfined in the Waccasassa River basin.

The uppermost geologic unit of the aquifer in the area is Ocala Limestone (SRWMD, 1995).









BIOGRAPHICAL SKETCH

Kris Crockett was born on March 31, 1978 in Daytona Beach, Florida. The middle of

three children, he grew up in Palm Coast, Florida and graduated from Flagler Palm Coast High

School in 1997. He received his B.S. in geology from University of Florida in 2004. Kris

received his M. S. in the geological sciences from the University of Florida in 2007. He plans to

pursue a career in hydrology here in Florida.





















20




15




10










Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
Month


Figure 4-6. Average daily discharge of the Steinhatchee River per month for the entire period of
record.



Waccasassa Mean Daily Flow

25




20




15








10





Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
Month


Figure 4-7. Average daily discharge of the Waccasassa River per month for the entire period of
record.


Steinhatchee Mean Daily Flow









frontal precipitation to the southeastern United States and as a result summer and fall

precipitation and stream flow are more likely to be the result of highly localized convective

storms. These different patterns in precipitation may lead to an increase in climatic

heterogeneity in the study area that is reflected in the tree-ring indices. Dendrohydrologic

reconstructions of streamflow in subtropical Florida appear to be as useful as in temperate

regions where they are more commonly utilized. Care needs to be taken for successful

utilization of this technique due to Florida's complex hydrology and climate.










temperature and precipitation values were recorded at the Usher Tower rain gauge station

(Figure 1-1).









84 km apart, so the analysis and verification results of the Steinhatchee discharge-Goethe TRI

correlation are assumed to approximate the Waccasassa discharge-Goethe TRI correlation.

Scatter plots show no discernable trends or patterns between Steinhatchee model residuals

versus Goethe TRI, absolute values of the residuals versus Goethe TRI, and residuals over time.

A scatter plot show a positive trend between residuals versus discharge minus the mean

discharge (r-0.77; Figures 4-14). This trend demonstrates that on average, the residuals of the

predictive model are negative during years when flow was less than average and positive when

flow is greater than average. Underestimation of maxima and minima of hydrologic data is a

common feature of tree-ring based reconstructions of flow because there is a biological limit to

the response of tree growth to an extreme high and low precipitation event (Fritz, 1976; Loaiciga

et al., 1992).

Validation statistics

The results of the data splitting and validations statistics are given in table 4-3. The

corresponding r, MSE, RE, and the prediction model are shown for both rivers, and for the

Steinhatchee split data sets. The CE values are given for the calibration periods of the

Steinhatchee data subsets. Any positive RE or CE value is considered to be significant

(Cleaveland and Stahle, 1989).

All r values displayed in Table 4-3 indicate a positive growth-discharge relationship. The r

values reflect better correlation for the even and early years than the corresponding odd and late

year subsets. All RE statistics displayed in Table 4-3 are positive, indicating some predictive

capability within the respective calibrations sets. The odd, even, and late CE statistics are

positive, indicating some predictive capability of those calibration models to their corresponding

validation periods. The model developed with the early calibration sets performed worse than

just using the sample mean of its respective validation set for the prediction model. Models









freezing temperatures are rare in North Florida. Florida is one of the wettest states, with

afternoon thunderstorms being frequent from spring until early autumn (Winsberg, 1990).

Florida is also vulnerable to the effects of hurricanes; in the 20th century alone approximately 60

hurricanes made landfall on the state (Barnes, 1998). These storms mostly occur during the late

summer months, causing severe flood and wind damage. For example, Hurricane Dora in 1964

dumped almost 24 inches of rain on the Steinhatchee River basin.

Long-term variations in sea surface temperatures (SST) in the Pacific and Atlantic Oceans

also may impact Florida' s climate. At least three long-term climate oscillations, the El

Nifio/Southern Oscillation (ENSO), the Atlantic Multidecadal Oscillation (AMO) and the Pacific

Decadal Oscillation (PDO), seem to influence hydrologic conditions in Florida (Obeysekera et

al., 2006). The influences of these climate oscillations in Florida are still not very well

understood. Because the AMO and PDO cycles are on decadal scales, information over long

time scales would be required to develop climate prediction models.

Steinhatchee River Basin

The Steinhatchee River basin is located within portions of Dixie, Taylor, and Lafayette

counties in the Florida panhandle (Figures 1-1 and 2-1). The Steinhatchee River, along with its

two tributaries, Eightmile Creek and Kettle Creek, drain an area of approximately 1518 km2

(Hand et al., 1988). The basin lies entirely within the coastal lowlands of the Florida

physiographic province and is characterized by flat topography (SRWMD, 1989). Elevations

typically range from 0 to 15 meters above mean sea level (msl) in most of the basin, with

elevations 30+ meters above ms1 in the upper basin (SRWMD, 1989).

The basin is characterized by karst topography. Crystal River and Suwannee Limestone

from the Eocene series are the dominant limestone formations that comprise the Floridan Aquifer



































LEGEND
A Waccasassa Gauging station County line
ajBGoethe State Forest
0 2 4 6 1.8 10 mi
Scale 1:316601 9 I 10 1k

Figure 2-2. Map of Waccasassa River and gauging station along with the Goethe sample site
(adapted from http://tiger.census. gov/).









Statistical Methods

The statistical relationships between annual tree growth and natural streamflow were

assessed by linear regression, in which the dependent variable (Y) is streamflow and the

independent variable (X) is the TRI. The correlation between the dependent and independent

variables was calculated using the NCSS 2004 Deluxe Statistical Analysis and Graphics

Software package. Both the STNDRD version and the ARSTAN version of the two TRIs were

correlated with fourteen gauging stations located in North Florida (Figure 1-1).

Regression analyses were carried out over the period when TRI correspond to observed

flow data. Monthly, seasonal, and calendar year averages of discharge (in units of m3/SOC) WeTO

correlated with the annual TRI to determine what fraction of the year or if the whole year

provided the best correlation to annual tree growth. Annual averages of discharge produced the

highest correlation with the tree-ring data and are used in subsequent analysis.

For initial evaluations, annual averages of discharge were taken over the calendar year for

each of the 14 gauging stations and then correlated with the two TRls. Subsequently, the time

window for annual averages of discharge for each station were shifted one month both forward

and backward for six months in each direction. These twelve yearly averages, produced from

each gauging station, were sequentially correlated with the Goethe TRI and Camp Branch TRI to

observe which interval provided the best correlation. The period of time when annual tree-ring

width and discharge correlate the best represents the water year for the study area. The water

year is generally accepted as an appropriate period to separate data and has previously been

defined to be a period when maximum flow is unlikely to carry over into the next consecutive

water year and the year-to-year variation in storage levels is minimized (Dingman 1994).

Commonly applied water year delineations include the calendar year and the classic USGS water

year (October 1st to September 30th).









Similar to the Steinhatchee, the limestone in the Waccasassa River basin is porous and

permeable, overlain by a thin layer of alluvial sediments (SRWMD, 1995). Because of the

relatively thin overburden of sand between the land surface and the Floridan aquifer, there is a

high degree of exchange between the surface waters and the aquifer (Taylor, 1978).

Climate in the Steinhatchee and Waccasassa River Basins

The climate in the two main river basins used in this study (climate data compiled from

National Weather Service Usher Tower weather gauge 1957-2005 (Figure 1-1)) is humid and

subtropical, with an average annual temperature of about 20.50 C. June through August is the

warmest period with an average maximum air temperature of 33.30 C. January through February

is usually the coldest period with an average minimum temperature of 50 C and with at least

several nights of freezing temperatures.

Basin rainfall has ranged over the period of record from a maximum of 253 cm/yr in

1964 to a minimum of 106 cm/year in 2000, with a mean rainfall over the basin of 154 cm/year.

Most of the precipitation falls in the summer (June through September). The highest rainfall

usually occurs in August, with a mean of 26 cm, and the lowest usually occurs in November,

with a mean of 6 cm. Between 105 and 1 15 cm/year of precipitation returns to the atmosphere

by evaporation and transpiration (Thornthwaite, 1948). Winter frontal activity causes another

smaller peak in the amount of precipitation in late winter (Winsberg, 1990). Flooding in the

rivers is more likely during the winter than the summer, due to the longer winter rain events

coupled with lower evapotranspiration rates (SRWMD, 1995).


Forest Sample Collection Sites

Forested lands in Florida are particularly sensitive to changes in hydrology, because of

the low water-holding capacity of sandy soils. A slight change in the depth to ground water can















3000


2500


2000



1 00 m


1000




50




Year


Figure 4-15. Usher Tower raw precipitation and precipitation minus potential evapotranspiration
through time.


Rainfall









represents the average of the residuals squared, where N equals the number of time points in the

data set. This statistic demonstrates how close a set of predictions are to the measured values.

The RE (equation 3-4)

RE=1-(MSE (9)/MSE (9) (3 -4)

compares the MSE (9) of the reconstruction to the MSE (9) (equation 3-5),

MSE (9) = (1/N) 1 (yt- 9c)2 (3-5)

where ye is the average discharge of the calibration period. Predictive models should perform

better than just using an unvarying sample mean and thus RE > 0. The CE is derived by

equation 3-6

CE = 1- MSE(f e-v)/ MSE(yv) (3-6)

and compares the MSE of the performance of a model (equation 3-7)

MSE(f e-v)= (1/N)1(yty- 90-v)2 (3-7)

to the MSE of the validation period which is derived by equation 3-8.

MSE(yv)= (1/N)1(yt- Yv)2 (3-8)

Where f e-v is the predicted discharge of the calibration period without the validation period and

yv is the validation period average. The CE will always be less than the RE. The difference

between the two values increases as the differences between the sample means for the validation

and calibration periods increases (North, 2006).

After each subset was evaluated with the verification tests, a final model was developed

from the entire TRI series and gauging record. A 95% prediction limit was used to quantify the

uncertainty in the reconstructed values. The TRI or TRIs used in the final reconstruction and the

closest rain gauging station to the TRI sample site were used in a linear regression analysis (Eq.

3-1) to determine their regression coefficients. This same rain gauging station was then




Full Text

PAGE 1

ASSESSMENT OF TREE-RINGS AS A RECORD OF PRE-HISTORIC STREAM FLOW IN A SUBTROPICAL ENVIRONMENT By KRIS CROCKETT 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 2007 1

PAGE 2

2007 Kris Crockett 2

PAGE 3

To trees. They are such useful little buggers. 3

PAGE 4

ACKNOWLEDGMENTS I thank my advisor Dr. Martin and my co mmittee members Dr. Brenner and Dr. Screaton for sticking by me through to the end. I thank To m Mirti for introducing me to this field of study and providing me with the necessary data. I th ank my family for all their support. Finally I thank all the professors and my friends here at the University of Florid a Department of Geologic Sciences for making my experience here at UF the best time of my life. 4

PAGE 5

TABLE OF CONTENTS page ACKNOWLEDGMENTS ...............................................................................................................4 LIST OF TABLES ...........................................................................................................................7 LIST OF FIGURES .........................................................................................................................8 ABSTRACT ...................................................................................................................................10 CHAPTER 1 INTRODUCTION................................................................................................................. .12 2 STUDY AREA................................................................................................................... ....15 3 METHODS...................................................................................................................... .......23 Tree-Ring Data .......................................................................................................................23 COFECHA ......................................................................................................................23 ARSTAN .........................................................................................................................24 Discharge Data ........................................................................................................................25 Statistical Methods ..................................................................................................................26 4 RESULTS...................................................................................................................... .........31 Tree-ring Data .........................................................................................................................31 COFECHA ......................................................................................................................31 ARSTAN .........................................................................................................................31 Discharge Statistics .................................................................................................................32 Steinhatchee River ...........................................................................................................32 Waccasassa River ............................................................................................................32 Statistical Analysis of the Discharge-TRI Correlations ..........................................................33 Residual Analysis ............................................................................................................34 Validation statistics .........................................................................................................35 Precipitation and Evapot ranspiration Correlation ..................................................................36 Final Graphs ............................................................................................................................37 5 DISCUSSION................................................................................................................... ......52 Tree-Ring Indices ...................................................................................................................52 Biological indications of the results .......................................................................................53 Validation Statistics ................................................................................................................54 Influence of Pacific Decadal Os cillation (PDO) on Tree-Ring Indices .................................56 Final model .............................................................................................................................58 Precipitation and Evapot ranspiration Correlation ..................................................................59 5

PAGE 6

Flow Proxies ...........................................................................................................................59 6 CONCLUSION................................................................................................................... ....62 LIST OF REFERENCES ...............................................................................................................63 BIOGRAPHICAL SKETCH .........................................................................................................67 6

PAGE 7

LIST OF TABLES Table page 4-1 Summary of TRI Statistics .................................................................................................38 4-2 Annual (Feb. to Jan.) Descri ptive Statistics of Discharge .................................................42 4-3 Validation Statistics ...........................................................................................................47 4-4 Precipitation and Recharge Correlations ...........................................................................49 5-1 The growth-discharge correlation for the cool phase PDO cycle. .....................................61 7

PAGE 8

LIST OF FIGURES Figure page 1-1 Map of study area with the locati on of USGS Stream gauging stations. ...........................14 2-1 Steinhatchee River and gauging station (adapted from http://tiger.census.gov/ ................20 2-2 Map of Waccasassa River and gauging stat ion along with the Goethe sample site. .........21 2-3 Map of the Presettlement range of longleaf pines (light gray) and their current extent (dark gray). .........................................................................................................................22 4-1 STNDRD TRI and ARSTAN TRI plotted through time.G G..............................................37 4-2 STNDRD TRI and ARSTAN TRI plotted through time.CB CB..........................................38 4-3 Histogram of number of tree ring width measurements per year for the Goethe site with STNDRD Goethe TRI (TRI ) overlay.G.....................................................................39 4-4 Histogram of number of tree ring width measurements per year for the Camp Branch site with STNDRD Camp Branch (TRI ) overlay.CB..........................................................39 4-5 How the EPS fluctuates with sample size for the TRI and TRI The red line represents the 0.85 cr itical threshold.G CB................................................................................40 4-6 Average daily discharge of the Steinhatch ee River per month for the entire period of record. ................................................................................................................................41 4-7 Average daily discharge of the Waccasassa River per month for the entire period of record. ................................................................................................................................41 4-8 Correlation coefficients of the 14 gauging stations TRI and TRI All reported correlation coefficients for this histogram are for the entire yearly record for each year adjusted to the starting month of best fit.Gt CB...................................................................43 4-9 Histogram of 12 month correla tion coefficients for the TRI taken for a 12 month period starting at different month forward and backwards from a calendar year.G.............43 4-10 Waccasassa discharge plotted against TRI The Waccasassa prediction model and r value are given above the best fit line.G...............................................................................44 4-11 Steinhatchee discharge pl otted against TRIG. The Stei nhatchee prediction model and r value are given above the best fit line. ............................................................................44 4-12 Waccasassa River predicted and actual discharge. ............................................................45 4-13 Steinhatchee River predic ted and actual discharge. ...........................................................45 8

PAGE 9

4-14 Discharge of the Steinhatchee River in m /s minus mean flow plotted against the residuals of the TRI Steinhatchee discharge linear regression.3 Gt.....................................46 4-15 Usher Tower raw precipitation and preci pitation minus potentia l evapotranspiration through time. ......................................................................................................................48 4.16 Precipitation plotted against potential recharge. ................................................................49 4-17 The correlation coefficients of all 14 st ream gauging stations plotted to the Usher Tower raw precipitation data vs correlation coefficients of the same stream gauging stations to the Goethe tree ring TRI and TRI .Gt CB............................................................50 4-18 Actual and predicted Waccasassa River discharge in m /s from 1875 -2003.3...................51 4-19 Actual and predicted Steinhatchee River discharge in m /s from 1875 -2003.3.................51 9

PAGE 10

Abstract of Thesis Presen ted to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Master of Science Degree ASSESSMENT OF TREE-RINGS AS A RECORD OF PRE-HISTORIC STREAM FLOW IN A SUBTROPICAL ENVIRONMENT By Kris Crockett August 2007 Chair: Jonathan Martin Major: Geology Understanding long-term patterns of hydrologic va riability is important in planning for the wise use of water resources. One method that could extend the historic al record of hydrologic variability is by analysis of tree-ring widths. Such use of tree -rings has been rare in subtropical environments because dendrohydrologic research is usually conducted where severe water stress strengthens the relationship betw een tree-ring growth and climat e. My study assesses the utility of tree rings in subtropical regi ons of north Florida as record s of hydrologic variability. Treering indices from two forests were compared with annual discharge from 14 river gauging stations in North Florida with linear correlation coefficients ranging from 0.12 to 0.86. Annual flows from two gauging stations, one each on th e Steinhatchee and Waccas assa rivers were found to correlate well with tree-ring indices (r = 0.64 and 0.86 re spectively) and the correlation models were used to extend river flow records to 1875. Accuracy of these linear flow models was tested by statistical comparison with historical climate records. A maximum reduction in error of 71% over just using av erage discharge for the predic tion model alone was achieved by the tree-ring model. Correlations between tree rings and flow we re diminished during the warm phase of the Pacific Decadal Os cillation (PDO), but were greatly improved during the PDO cool phase. The warm phase of the PDO brings an equa torial shift of the jet stream, which increases 10

PAGE 11

frontal precipitation to the s outheastern United States and as a result summer and fall precipitation and stream flow are more likely to be the result of highly localized convective storms. These different patterns in precipita tion may lead to an increase in climatic heterogeneity in the study area that is reflect ed in the tree-ring i ndices. Dendrohydrologic reconstructions of streamflow in subtropical Florida appear to be as useful as in temperate regions where they are more commonly utilized Care needs to be taken for successful utilization of this technique due to Floridas complex hydrology and climate. 11

PAGE 12

CHAPTER 1 INTRODUCTION Understanding long-term patterns of hydrologi c variability is important in planning for the wise use of water resources. The longest climate a nd streamflow records in the United States rarely extend past the 1890s, which may not be long enough to capture the frequency of severe droughts or floods. This lack of direct obser vation of climate and flow requires developing proxy records of these variables. Records of long-term hydro-climatic behavior may be found in the sediments of streams, lakes, and other de positional basins, but th ese records are rarely sensitive enough to capture annual variability. In contrast, trees have a natural and consistent response to seasonal hydrologic fo rcing that can record hydrologic variability and record highresolution hydrologic phenomena. The linear rela tionships between tree-ring and climate records are at least equal to, and often exceed, t hose found for other proxies (North, 2006). Tree-ring width has been shown to correlate wi th historic precipitation and temperature data in many places around the world (Coile, 1936; Schumacher and Day, 1939; Fritts, 1976; Pederson, et al. 2001). Using tree-rings to hind cast climate records can be complicated, however, if they are linked to two or more variables (i.e. precipitation, temperature, evapotranspiration, and duration of rainfall). St reamflow, which is a f unction of precipitation, temperature, and evapotranspirati on, is a single variable that has been shown to correlate well with tree-ring variations (Fritts, 1976). This correlation has alr eady been shown to be valid in many areas across North America such as the sout heastern United States (Cleaveland and Stahle, 1989), the southwestern United States (Meko an d Graybill, 1995), the northwestern United States (Gedalof et al., 2004), and the Canadian prairie region (Case and MacDonald, 2003). The correlation was strong (r=0.60 to 0.87) between river discharge (Q) and tree-rings in these studies. 12

PAGE 13

Limited analysis of tree-rings has been completed to estimate hydrologic variations through time in subtropical regions and this previous work has been erratic and contradictory. At least two possible reasons for this lack of resear ch exist: there are limited sites with sufficiently old trees (Varner and Pederson, 2002), and dend rohydrologic research is usually conducted where severe water stress strengthens the rela tionship between tree-ri ng growth and climate (Fritts, 1976; Loaiciga et al., 1993; and North, 2006). L odewick (1930) was one of the first to investigate the relationship of climate to tree grow th in subtropical regions. He found a positive correlation between tree-ring width and mid-J une to mid-October precipitation, but no correlation between tree-ring width and temperature. Coile (1936) found that February to April precipitation has a positive correl ation on young longleaf pine growth in the subtropics and that June to August temperature had a negative corr elation. Schumacher and Day (1939) performed similar studies in several areas of North Florida bu t with varied results. More recently Ford and Brooks (2002) and Foster and Brooks (2001) found a positive correlation between tree-ring widths, precipitation, and the disc harge of the Myakka River in several species of trees from south Florida. In order to assess the use of tree-rings as a proxy for stream flow, data has been compiled from a network of longleaf pine trees located in two old growth forests in northern Florida, several USGS stream gauging stations and Na tional Weather Service rain gauging stations (Figure 1-1). My study uses these data to te st the validity of dendr ohydrologic analysis in a subtropical environment. If succe ssful, the technique of utilizing tr ee-ring analysis could be used to provide information about stream flow prio r to instrumental data. Long-term records developed from tree-ring proxies may offer a useful tool in the management of water resource systems in the area. 13

PAGE 14

Figure 1-1. Map of study area with the location of USGS Stream gauging stations and the NOAA Precipitation Station where flow and preci pitation data were compiled. Areas of regions where tree cores were collected are shown in dark green for Goethe and light green for Camp Branch (adapted from http://tiger.census.gov/). 14

PAGE 15

CHAPTER 2 STUDY AREA Climate, geology, hydrology, and human interact ions have a strong effect on both tree growth and stream flow. An understanding of how these characteristics influence the growthflow relationship is important to dendrohydrolog ic studies. These characteristics are described for north-central Florida, with emphasis on th e Waccasassa and Steinhatchee drainage basins, which provided data for the most in-depth anal ysis. Limited analyses are included for an additional 12 watersheds (Figure 1-1). Florida is one of the fastes t growing states, with populat ion increasing by 4.7 %percent between 2003 and 2004 (2004 U.S. Census Bureau). This rapid population growth places large demands on Floridas ground water supply and othe r natural resources. Floridas five water management districts are responsible for mana ging ground and surface water supplies as well as evaluating current and future water needs. To date, water policy and management decisions employed by these districts have been based pr imarily on instrumental climatic and hydrologic records with maximum lengths of around 75 years. Based on available records, the Suwannee River Water Management District (SRWMD), wh ich encompasses the study area, has concluded water supplies are sufficient for demands over th e next 20 years. All regulatory agencies continually reassess their demands and sources of water as new techniques and data become available. Long-term climate records from tree-ri ngs could provide valuable information for this assessment. North Florida has a humid subtropical clim ate with long, hot, rainy summers, short and usually mild winters (Henry et al., 1994). Th e seasons in Florida are determined more by precipitation than by temperature. The winters and falls are cool and dry and the summers and springs are rainy and warm (Winsberg, 1990). Due to the mild winter temperatures, snow and 15

PAGE 16

freezing temperatures are rare in North Florida. Florida is one of th e wettest states, with afternoon thunderstorms being frequent from sp ring until early autumn (Winsberg, 1990). Florida is also vulnerable to the effects of hurricanes; in the 20th century alone approximately 60 hurricanes made landfall on the stat e (Barnes, 1998). These storms mostly occur during the late summer months, causing severe flood and wind damage. For example, Hurricane Dora in 1964 dumped almost 24 inches of rain on the Steinhatchee River basin. Long-term variations in sea su rface temperatures (SST) in the Pacific and Atlantic Oceans also may impact Floridas climate. At leas t three long-term climate oscillations, the El Nio/Southern Oscillation (ENSO), the Atlantic Multidecadal Oscillation (AMO) and the Pacific Decadal Oscillation (PDO), seem to influence hy drologic conditions in Florida (Obeysekera et al., 2006). The influences of these climate osci llations in Florida are still not very well understood. Because the AMO and PDO cycles are on decadal scales, information over long time scales would be required to develop climate prediction models. Steinhatchee River Basin The Steinhatchee River basin is located within portions of Dixie, Taylor, and Lafayette counties in the Florida panhandle (F igures 1-1 and 2-1). The Stei nhatchee River, along with its two tributaries, Eightmile Creek and Kettle Creek, drain an area of approximately 1518 km2 (Hand et al., 1988). The basin lies entirely within the coastal lowlands of the Florida physiographic province and is characterized by flat topography (SRWMD, 1989). Elevations typically range from 0 to 15 meters above mean sea level (msl) in most of the basin, with elevations 30+ meters above msl in the upper basin (SRWMD, 1989). The basin is characterized by karst topogr aphy. Crystal River and Suwannee Limestone from the Eocene series are the dominant limestone formations that comprise the Floridan Aquifer 16

PAGE 17

in the Steinhatchee River basin (SRWMD, 1989). Rela tively thin layers of alluvial deposits and peat accumulations occur on top of these form ations (SRWMD, 1989). Within the Steinhatchee River basin, most groundwater occurs in the unc onfined Floridan Aquifer with minor amounts occurring in the Surficial Aquifers (SRWMD 1989). Rainwater percolating through the overburden recharges the Floridan and Surficial aquifers direc tly, and during high stages the Steinhatchee River recharges the aquifers thr ough the streambed and at the Steinhatchee sink located downstream of the gauging station (SRW MD, 1989). In the northern and northeastern portions of the basin, in Lafayette County, a trans itional zone exists where the Floridan Aquifer is semi-confined by the Hawthorn Formation a nd is periodically under artesian conditions (SRWMD, 1989). Waccasassa River Basin The Waccasassa River, along with its princi pal tributaries Wekiva River, Cow Creek, Otter Creek, and Magee Branc h, drains approximately 2424 km2 (SRWMD, 1995). The Waccasassa River basin lies in North Central Florida along the Big Bend Coast in Levy, Alachua, and Gilchrist counties (SRWMD, 1995) (Figures 1-1 and 2-2). The river is about 47 km long and originates west of Bronson from diffuse swamp waters of southern Waccasassa Flats. The Waccasassa River doesnt develop a defined channel until receiving flow from Levy Blue Spring, which has an average discharge of 0.251 m3/s (Scott et al., 2004). Elevations within the area range from sea level at the coast to a maximum of about 43 m above msl in Alachua County. Similar to the Steinhatchee, the Waccasassa drainage basin is defined by karst topography. The Floridan Aquifer system is larg ely unconfined in the Waccasassa River basin. The uppermost geologic unit of the aquifer in the area is O cala Limestone (SRWMD, 1995). 17

PAGE 18

Similar to the Steinhatchee, the limestone in the Waccasassa River basin is porous and permeable, overlain by a thin layer of alluvial sediments (SRWMD, 1995). Because of the relatively thin overburden of sa nd between the land surface and the Floridan aquifer, there is a high degree of exchange between the surface waters and the aquifer (Taylor, 1978). Climate in the Steinhatchee and Waccasassa River Basins The climate in the two main river basins us ed in this study (clima te data compiled from National Weather Service Usher Tower weathe r gauge 1957-2005 (Figure 1-1)) is humid and subtropical, with an average annual temperature of about 20.5 C. June through August is the warmest period with an average maximum air temp erature of 33.3 C. January through February is usually the coldest period with an average mi nimum temperature of 5 C and with at least several nights of freezing temperatures. Basin rainfall has ranged over the period of record from a maximum of 253 cm/yr in 1964 to a minimum of 106 cm/year in 2000, with a mean rainfall over the basin of 154 cm/year. Most of the precipitation falls in the summer (June through Sept ember). The highest rainfall usually occurs in August, with a mean of 26 cm and the lowest usually occurs in November, with a mean of 6 cm. Between 105 and 115 cm/year of precipitation return s to the atmosphere by evaporation and transpiration (Thornthwaite, 1948). Winter frontal activity causes another smaller peak in the amount of precipitation in late winter (Winsberg, 1990). Flooding in the rivers is more likely during the winter than the summer, due to the longer winter rain events coupled with lower evapotra nspiration rates (SRWMD, 1995). Forest Sample Collection Sites Forested lands in Florida are particularly sensitive to changes in hydrology, because of the low water-holding capacity of sandy soils. A slight change in the depth to ground water can 18

PAGE 19

have dramatic shifts in the dominant tree speci es in an area (Foster and Brooks, 2001). Of the numerous tree species predominant in Florid as forested ecosystems, longleaf pine ( Pinus palustris ) was chosen for this analysis because of its characteristics of longevity (>400 years), wide geographic range, and resi stance to fire, disease, drought and insects (Meldahl et al., 1999). Also the relationship of longleaf pines to water availabili ty does not depend on depth to the aquifer. This factor has been shown to be important to the growth of slash pine ( Pinus elliottii ), which is another dominant pine in nort h Florida (Foster and Brooks, 2001). Although longleaf pine has been shown to prefer mesic sites to xeric sites in terms of growth rate, it is outcompeted by slash pine in mesic areas (Foster and Brooks, 2001). Consequently longleaf pines are common in xeric sites where frequent fires limit the growth of slash pine and other native Florida tree species (Myers a nd Ewel, 1990; Foster and Brooks, 2001). Water availability alone is thus not responsible for where longleaf pine will dominate (Foster and Brooks, 2001). Longleaf pine forests used to blanket the southeastern United St ates (Figure 2-3). Logging, development, and fire suppression have depleted this ecosystem by 97% of its presettlement extent (Loudermilk, 2005; Van Lear et al., 2005) and consequently there are few sites with a significant number of old trees. Tw o promising old-growth forests were sampled for this study. One is the Goethe State Forest locat ed partly within the Waccasassa drainage basin and the other is an old turpetined flatwoods ecosystem growing along the Suwannee River, which is referred to in this study as the Camp Branch Forest (Figure 1-1). 19

PAGE 20

Figure 2-1. Steinhatchee River and gauging station (adapted from http://tiger.census.gov/ 20

PAGE 21

Figure 2-2. Map of Waccasassa River and gauging station along with the Goethe sample site (adapted from http://tiger.census.gov/). 21

PAGE 22

Figure 2-3. Map of the Presettlement range of longl eaf pines (light gray) a nd their current extent (dark gray) (adapted from http://biology.usgs.gov/s+t/SNT/noframe/se130.htm ). 22

PAGE 23

CHAPTER 3 METHODS Tree-Ring Data Standard core sampling techniques (Fritts, 1976) were used by Tom Mirti to collect tree cores from the two forests. Large diameter tr ees with minimal defect s (e.g., lack of storm damage and/or rot) were select ed by assuming them to be among the oldest specimens. From each tree at least one core was extracted with a 5.15 mm incremental borer 0.5 to 1.5 meters above ground level. Two cores are preferable to aid in the identificati on of missing rings which can occur when a tree is under stress (e.g. dr ought or severe competitive suppression). From the Goethe site, 55 cores were collected from 29 longl eaf pines. From the Camp Branch site, 47 cores were collected from 30 liv ing longleaf pines and 2 longleaf pine stumps. After extraction, the cores were air dried and a ffixed to wooden mounts, and labeled. The ring widths were measured at the University of Tennessee Tree-Ring Laboratory to the nearest 0.01 mm using a Velmex measuring system interfaced with Measure J2X software. Each core was measured from the first complete innermost ring to the last complete ring before reaching bark. COFECHA The raw tree-ring width data were processe d at the University of Tennessee Tree-Ring Laboratory with COFECHA, which is a computer program that assesses the quality of dating and measurement accuracy of a tree-ring series (G rissino-Mayer, 2001). COFECHA checks the accuracy of assigned tree-ring date s by correlating individual core measurements with a master tree-ring record constructed from the remaini ng series (Holmes, 1983; Grissino-Mayer, 2001). Individual cores that correlate poorly with the master record ar e flagged by COFECHA and then visually inspected for missing or fa lse rings. If the source of the er ror can be identified, the core is redated and reanalyzed with the COFECHA pr ogram. Cores that cannot be dated accurately 23

PAGE 24

against the master record due to the presence of too many false or missing rings are not included in subsequent analyses. Each core used in the master record is correlated with the others to create the series inte rcorrelation, which is a statistical quantity representing the variability between all tree cores in a set. The series inte rcorrelation is expressed as a Pearson correlation coefficient (r) that measures the strengt h of a linear relationship (Equation 3-1). r = [{ (yt) 2(ytt) 2}/ (yt) 2 ] (3-1) Where yt is the measured value at some year t (i n this study the measured value is either standardized tree-ri ng widths, discharge, or precipitation), t is equal to the predicted value at year t, and is equal to the average of all measured values. The significance of the series intercorrelation can be shown by the expressed population signal (EPS). The EPS is a measure of how well a tree-ring record made from a samp le set is likely to portr ay a hypothetical tree-ring record comprised from an entire population and is derived from the seri es intercorrelation and the number of trees in the sample (Cook and Kairi ukstis, 1990). A value above 85% is set as the threshold for acceptable statistical quality (Wigley et al., 1984). ARSTAN The confidently dated tree-ring measurements we re then analyzed at the University of Tennessee Tree-Ring Laboratory with the ARST AN program. The ARSTAN program fits a cubic spline to each set of measurements fo r every core. The program then creates a dimensionless tree-ring index (T RI) by taking the ratio between th e predicted value of the cubic spline and the measured ring width value (Cook a nd Holmes, 1997). This procedure adjusts for a trees natural tendency to produce smaller ring widt hs with age and also allows values of the tree-ring widths to be averaged without artifacts from variable growth rates (Cook and Holmes, 1997). The samples were averaged into two initial TRIs (labeled STNDRD version by the 24

PAGE 25

program), one each for the Goethe and Camp Br anch forests. Due to the effects of the standardizing and detrending, both TRIs have dimensionl ess means of approximately 1. The ARSTAN program also produces an additional TRI (labeled ARSTAN version by the program) from the STNDRD version that removes the effects of autocorrelation in TRI. Tree growth during an individual year may be aff ected by the previous years favorable or unfavorable growing season (Fritts, 1976). For in stance, a favorable growing season can allow a tree to build up an excess of sugars that will cau se it to grow more the following year than it would have otherwise given the hydrologic cond itions for that year. The ARSTAN program attempts to remove this trend through autocorr elation modeling. In the following discussion both final ARSTAN and STNDRD version TRIs are used to estimate correlations betw een tree-ring records and streamflow records. Discharge Data Fourteen gauging stations located in northern Florida were chos en for the initial correlation analysis between TRI and discharge. These stati ons were chosen on the ba sis of their significant length of record and proximity to the Goethe Stat e and Camp Branch forest sample sites (Figure 1-1). Streamflow records for all gauging statio ns were taken from the United Sates Geological Survey (USGS) (http://waterdata.usgs.gov/fl/nwi s/rt). The streamflow records for the Waccasassa River USGS gauging station 02313700 were sporadic and were only available for the periods of March 1963 to Septem ber 1978, November 1980 to September 1984 (fragmentary), October 1984 to September 1992, and October 1998 to current year. All other gauging records are continuous ove r the periods analyzed. Di scharge was computed from a continuous velocity record obtained from a water-current meter. 25

PAGE 26

Statistical Methods The statistical relationships between annua l tree growth and natu ral streamflow were assessed by linear regression, in which the depe ndent variable (Y) is streamflow and the independent variable (X) is the TRI. The correlation between the dependent and independent variables was calculated using the NCSS 2004 De luxe Statistical Analysis and Graphics Software package. Both the STNDRD version and the ARSTAN version of the two TRIs were correlated with fourteen ga uging stations located in North Florida (Figure 1-1). Regression analyses were ca rried out over the period when TRI correspond to observed flow data. Monthly, seasonal, and calendar ye ar averages of discharge (in units of m3/sec) were correlated with the annual TRI to determine what fraction of the year or if the whole year provided the best correlation to annual tree growth. Annual averag es of discharge produced the highest correlation with the tree-ring data and are used in subsequent analysis. For initial evaluations, annual averages of di scharge were taken over the calendar year for each of the 14 gauging stations and then correlated with the two TRIs. Subsequently, the time window for annual averages of discharge for eac h station were shifted one month both forward and backward for six months in each direction. These twelve yearly averages, produced from each gauging station, were sequentially correlated with the Goethe TRI and Camp Branch TRI to observe which interval provided the best correlation. The period of time when annual tree-ring width and discharge correlate the best represents the water year for the study area. The water year is generally accepted as an appropriate peri od to separate data and has previously been defined to be a period when maximum flow is unlikely to carry over in to the next consecutive water year and the year-to-year variation in storage levels is minimized (Dingman 1994). Commonly applied water year delin eations include the calendar ye ar and the classic USGS water year (October 1st to September 30th). 26

PAGE 27

A residual of a linear regression model is th e observed data minus the predicted data (Equation 3-2). residual = yt t (3-2) Residual analyses were carried out on the reconstructions of discharge based on a TRI. To test the assumption that the discharge-TRI relationship is linear, a scatter plot of the residuals of discharge versus TRI was analyze d. To test the assumption that errors have a constant residual variance with varying tree-ring widt hs, scatter plots of the absolute values of the residuals of discharge versus TRI, residuals of discharge over time, and the residuals of discharge versus discharge minus the mean discharge were analyz ed. Any functional rela tionship between these variables would be a violation of the assumption of linearity. Pattern s or trends apparent in these plots would violate the assumptions of constant residual variance. Reconstruction models were verified usi ng a data splitting method (Fritts 1976), which involves dividing data into a calibration period an d a validation period to assess the accuracy of the reconstruction model. Data were split in two ways: late and early years (i.e. 1951-1976 and 1977-2003) and odd and even years. Correlation models were constructed for all subsets of data and model estimates were compared to data not us ed to create the models. Long term variations could introduce bias into the data split by early and late years, but would not occur when data are split into odd and even years. The Pearson correlation coefficient (Eq. 3-1) mean squared error (MSE), reduction of error (RE), and coefficient of efficiency (CE) verification statistics we re used to assess the accuracy of streamflow pred ictions based on the difference between the validation and calibration periods (North, 2006). Th e MSE of predicted discharge ( ) (equation 3-3) MSE ( ) = (1/N) (ytt)2 (3-3) 27

PAGE 28

represents the average of the residuals squared, where N equals the number of time points in the data set. This statistic demons trates how close a set of predicti ons are to the measured values. The RE (equation 3-4) RE=1-(MSE ( )/MSE ( ) (3-4) compares the MSE ( ) of the reconstruction to the MSE ( ) (equation 3-5), MSE ( ) = (1/N) (ytc)2 (3-5) where c is the average discharge of the calibration period. Predic tive models should perform better than just using an unvarying sample m ean and thus RE > 0. The CE is derived by equation 3-6 CE = 1MSE( c-v)/ MSE( v) (3-6) and compares the MSE of the perf ormance of a model (equation 3-7) MSE( c-v)= (1/N)(ytvc-v)2 (3-7) to the MSE of the validation peri od which is derived by equation 3-8. MSE( v)= (1/N)(ytv)2 (3-8) Where c-v is the predicted discharge of the calibration period without the validation period and v is the validation period average. The CE w ill always be less than the RE. The difference between the two values increases as the differenc es between the sample means for the validation and calibration periods in creases (North, 2006). After each subset was evaluated with the verification tests, a final model was developed from the entire TRI series and gauging record. A 95% prediction limit was used to quantify the uncertainty in the reconstructed values. The TRI or TRIs used in the final reconstruction and the closest rain gauging stati on to the TRI sample site were used in a linear regression analysis (Eq. 3-1) to determine their regre ssion coefficients. This same rain gauging station was then 28

PAGE 29

correlated with all 14 streamflow records to se e how discharge in these stations reflected precipitation events experienced by the forest. Streamflow that has be en changed through human influences (i.e. dams or weirs) or that is receiving runoff from a greatly differing climatic area should correlate poorly. Both raw precipitation data and precipitation data adjusted for th e effects of monthly potential evapotranspiration (PEm) were used in the TRI to precipitation correlations. PEm was calculated using the Thornthw aite-Mather (1957) evapotrans piration model is derived by equation 3-9. PEm = 16 DLm [10 Tm /HI]a mm (3-9) Where Tm is the mean monthly temperature in degrees C, HI is the heat index and equaled to the sum of the 12 HIm values (equation 3-10). HIm = [Tm/5]1.5 (3-10) The a is a coefficient derived by equation 3-11. a = (6.7e-7) (HI3) (7.7e-5) (I2) + (0.018) (HI) + 0.49 (3-11) DLm in equation 3-9 is the monthly day length multiplier calculated by equation 3-12. DLm = {a + b (Latitude)1.35}/a (3-12) The DLm relates hours of sunlight at certain latitudes and is only valid for a range of latitudes from 0 to 50 (LeBlanc and Terrell, 2001). Va lues for a and b in equation 3-12 are given by LeBlanc and Terrell (2001). The Thornthwaite-Mather ev apotranspiration model assumes that the only effects on evapotranspiration are meteorologi cal conditions and ignores the de nsity of vegetation. Despite these simplifications, this method gives a reason able approximation of PET, and is especially suited for humid regions such as Florida (W atson and Burnett, 1995). Mean monthly air 29

PAGE 30

temperature and precipitation values were reco rded at the Usher Tower rain gauge station (Figure 1-1). 30

PAGE 31

CHAPTER 4 RESULTS Tree-ring Data COFECHA From the 55 cores collected at the Goethe site, 31 cores from 23 trees could be dated by the COFECHA program and were included in the Go ethe TRI. From the 47 cores collected at the Camp Branch site, 42 cores from 26 trees and 2 stumps could be dated by the COFECHA program and were included in the Camp Branch TRI. These TRIs extend from 1875 to 2003 for the Goethe State forest site, and extend from 1806 to 2002 for the Camp Branch site. Series intercorrelation was calculated to be 0.513 for the Goethe site and 0.522 for the Suwannee site. The EPS was calculated to be 97% for both sites. This value is well above the value of 85% suggested by Wigley et al. (1984) for acceptable statistical quality. ARSTAN The TRIs produced from the ARSTAN program (STNDRD and ARSTAN versions) are displayed in Figure 4-1 for the Goethe TRI a nd in Figure 4-2 Camp Branch TRI. In the following discussion only the STNDRD versions of the TRIs were used because: The STNDRD and the ARSTAN versions for the Goethe TRI and Camp Branch TRI are similar (Figures 4-1 and 4-2). The ARSTAN version did not provide a signif icant improvement on the correlation results (r<0.01) and in most cases lowered th e Pearson correlation coefficient. The STNDRD version produces a simpler prediction model, thereby preserving additional degrees of freedom that would be lost thr ough the use of the autocorrelation modeling in the ARSTAN versions (Fritz, 1976). The means, standard deviations minimum and maximum values, a nd ranges for the Goethe TRI and Camp Branch TRI are displayed in table 4-1. The correlation between the Goethe TRI and Camp Branch TRI had a Pearson correlation coefficient of 0.58. 31

PAGE 32

As is generally the case in dendroclimatic studi es, fewer cores are avai lable to create the TRIs back in time, thereby reducing their EPS (Figure 4-3 and 4-4) (Gray et al., 2004). An example of how the EPS fluctuates with sample size for the Goethe TRI is given in Figure 4-5. Since the series intercorrelations calculated for the Goethe TRI and Camp Branch TRI are close, Figure 4-5 is used to represent the Camp Branch TRI as well. The subsam ple signal strength for the Goethe TRI and Camp Branch TRI does not drop below 85% probability of representing the hypothetical population TRI until 1889 for the Goet he TRI and 1860 for Camp Branch TRI when the number of trees that make up the TRIs drops below 5 (Figures 4-3 and 4-4). Discharge Statistics Steinhatchee River The annual mean daily discharge at this st ation over the calendar year is 8.87 cubic meters per second (m3/s); with a maximum of 27.27 m3/s in 1964 to a minimum of 1.44 m3/s in 2000. Fifty-two years of daily observations were used to create an average daily flow hydrograph for one calendar year (Fi gure 4-6). Flow peaks for the year in March at an average discharge of 14.34 m3/s and then again around August and September at an average discharge of 14.05 m3/s. Waccasassa River The annual mean daily discharge at this station over the calendar year is 7.75 m3/s; with a maximum of 18.75 m3/s in 1964 to a minimum of 2.00 m3/s in 2000. During periods of drought reported monthly flow averages can be negative indicating revers e flow because of the incoming tide. Twenty two years of daily observations were used to create an average daily flow hydrograph for one calendar year (Fi gure 4-7). Flows peaks for the year in March at an average discharge of 10.12 m3/s and then again in September at an average discharge of 13.42 m3/s. 32

PAGE 33

Statistical Analysis of th e Discharge-TRI Correlations Annual descriptive statistics fo r the Waccasassa and Steinhatch ee gauging stations and the split data sets for the Steinhatchee flow for the adjusted year of February through January are given in table 4.2. Subsets for the Waccasassa ga uging record were not made due to the short period of recorded discharge for this station (Table 4-2). All stations showed a positive correlation between TRIs and discharge, with Pearson correlation coefficients ranging from 0.12-0.86 (Fi gure 4-8). At most gauging stations the Goethe TRI correlates with flow better than th e Camp Branch TRI. The only gauging stations that correlated better with the Camp Branch TRI were stations 10 and 11 along the Suwannee River, both of which do not have reported discharge for at least the last 25 years (1928 to 1978 for station 10 and 1928 to 1957 for station 11). The Pearson correlation coefficient (Eq.3-1) calculated for the Steinhatchee River gauging data and the Goethe TRI (r = 0.64) is within the range of values from reviewed literature (Cleaveland and Stahle, 1989; Meko and Graybill, 1995; Gedalof et al., 2004; and Case and MacDonald, 2003) and on this basis further regression analysis and model development was carried out between these two data sets. The Pearson correlation coefficients (Eq. 3-1) calculated between the Waccasassa River gauging data and both the Goethe TRI and Camp Branch TRI (r = 0.86 for both) are also well within the range of the reviewed literature. The Goethe TRI was used in the regression analysis and model development for the Waccasassa River historic flow reconstruction be cause of its better correlation to most of the othe r gauging stations with longer records and proximity to the Waccasassa River (approximately 15.8 km apart). Seven of the gauging stations show a maximu m correlation for a water year beginning in February and ending in January, five show a maximum correlation from January to December, and two show a maximum correlation from Marc h to February. An example of how the 33

PAGE 34

correlation changes with different water year is shown for Waccasassa and Steinhatchee gauging stations (Figure 4-9). By using a February water year for the Stei nhatchee and Waccasassa gauging stations, both stations show a 1.5% improve ment in correlation from using a water year beginning in January and a 3% improvement fr om using a water year beginning in March. To illustrate the correlations between the Go ethe TRI and discharge, the Goethe TRI was plotted against the recorded discharge for both th e Waccasassa and Steinhatchee rivers (Figures 4-10 and 4-11). The correlation equations were used to reconstruct discha rge for the Waccasassa and Steinhatchee. Model discharge for the Waccasassa River is derived by equation 4-1 Q = 12.53 x Goethe TRIGt 3.71 (4-1) and derived for the Steinhatchee River in equation 4-2, Q = 14.54 x TRIGt 5.29 (4-2) where Q is discharge in m3/s and TRIGt is the value given by the Goethe tree-ring index at year t. Predictions of discharge using equations 4-1 and 4-2 are plotted with the observed discharge over both the Waccasassa (Figures 4-12) and Steinhatchee (Figures 4-13) gauging histories. Both stations correlate well with the Goethe TRI during the period of time between 1964 and 1977. Outside of this time period the correla tion is not as strong. This period of time includes the highest flow event and one of the most severe droughts in bo th periods of record. This period of time also makes up 72% of the to tal Waccasassa gauging record, and thus it is not possible to determine how well the model predic ts actual flow outside of this period. Residual Analysis Residual analysis was conducted only for the Steinhatchee-Goethe TRI correlation, because of the shortness of the gauging record for the Waccasassa River. Both the Waccasassa and Steinhatchee gauging records correlate well w ith each other (r = 0.74) and the rivers are only 34

PAGE 35

84 km apart, so the analysis and verification re sults of the Steinhatchee discharge-Goethe TRI correlation are assumed to approximate the Waccasassa discharge-Goethe TRI correlation. Scatter plots show no discernable trends or patterns between Steinhatchee model residuals versus Goethe TRI, absolute values of the residu als versus Goethe TRI, and residuals over time. A scatter plot show a positive trend between residuals versus discharge minus the mean discharge (r=0.77; Figures 4-14). This trend demonstrates that on average, the residuals of the predictive model are negative during years when fl ow was less than average and positive when flow is greater than average. Underestimation of maxima and minima of hydrologic data is a common feature of tree-ring based reconstructions of flow because there is a biological limit to the response of tree growth to an extreme high and low precipitati on event (Fritz, 1976; Loaiciga et al., 1992). Validation statistics The results of the data splitti ng and validations statistics are given in table 4-3. The corresponding r, MSE, RE, and the prediction model are shown for both rivers, and for the Steinhatchee split data sets. The CE values are given for the calibration periods of the Steinhatchee data subsets. Any positive RE or CE value is considered to be significant (Cleaveland and Stahle, 1989). All r values displayed in Table 4-3 indicate a positive growth-d ischarge relationship. The r values reflect better correlation for the even and early years than th e corresponding odd and late year subsets. All RE statistics displayed in Table 4-3 are posi tive, indicating some predictive capability within the respective calibrations se ts. The odd, even, and late CE statistics are positive, indicating some predictive capability of those calibration models to their corresponding validation periods. The model developed with th e early calibration sets performed worse than just using the sample mean of its respective validation set for the prediction model. Models 35

PAGE 36

calibrated for the even and early da ta sets had the highest r and RE statistics of the Steinhatchee data subsets, but did not verify well over their corresponding validation periods (odd and late). The odd and late models showed a stronger relation ship with their validation periods than they do with themselves as evident by their low REs and high CEs. Precipitation and Evapotranspiration Correlation Precipitation data taken from the Usher Towe r rain gauge station and potential recharge (precipitation minus evapotranspiration calculated using the Thornthwaite-Mather evapotranspiration model) are shown in figure 415. The potential rechar ge calculated using the Thornthwaite-Mather model shows that 69% of the rainfall in the area returns to the atmosphere via evapotranspiration. The slope of the line (m=1.0) and the Pearson correlation coefficient calculated between the Usher tower precipitation a nd potential recharge da ta reveal only minor differences in trend (r=0.98; Figure 4-16). This result suggests that th e average annual amount of precipitation lost through ev apotranspiration is a constant 1059 mm/year through the entire period of record. Table 4-4 shows the Pearson correlation co efficients of the Usher Tower rain gauge station precipitation and potential recharge data compared with the Goethe TRI, the Waccasassa gauging data, and the Steinhatchee gauging data in cluding the even and odd subset. The early and late subset was not included in this co mparison, because the Usher Tower rain gauging record is shorter than the Steinhatchee gauging record and does not provi de an equal comparison among the subsets. The Pearson correlation coeffi cients between all 14 st ream gauging stations and the Usher Tower precipitation data is compar ed to the Pearson correlation coefficients between these stream gauging stations and the Goethe TRI (Figure 4-17). This graph shows a positive correlation (r = 0.69), indicating that as correlation improves between discharge and precipitation, the correlation also im proves between discharge and the TRI. 36

PAGE 37

Final Graphs The final validated actual and predicted W accasassa and Steinhatchee River discharge from 1875 -2003 are plotted in figur es 4-18 and 4-19 respectively. Prior to 1885 there is a very large amplitude increase in discharge, which is probably the result of low number of core measurements. A period of low flow is s uggested by the graphs extending from 1884 to approximately 1910. During this period of time, 44% of the discharge predictions for the Waccasassa River fall outside the range of flow that occurred during the Waccasassa calibration period and only three of the years during this time have predicted flow above the average measured discharge. For the Steinhatchee River five years of predicted flow during the 1884 to 1910 period fall below the range of flow that occurred during the Steinhatchee calibration period and only one year during this time has predicte d flow above the average measured discharge. TRIG0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 18751885189519051915192519351945195519651975198519952005YearIndex STNDRD ARSTAN Figure 4-1. STNDRD TRIG and ARSTAN TRIG plotted through time. 37

PAGE 38

TRICB0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.818001815183018451860187518901905192019351950196519801995YearIndex STNDRD ARSTAN Figure 4-2. STNDRD TRICB and ARSTAN TRICB plotted through time. Table 4-1. Summary of TRI Statistics TRIGTRICB Count 197 129 Mean 0.98 0.93 Standard deviation 0.26 0.3 Minimum 0.26 0.31 Maximum 1.67 1.83 Range 1.41 1.52 38

PAGE 39

TRIG0 5 10 15 20 25 30 35 1875188518951905191519251935194519551965197519851995YearCores Number of Cores STNDRD Figure 4-3. Histogram of number of tree ring width measurements per year for the Goethe site with STNDRD Goethe TRI (TRIG) overlay. TRICB0 5 10 15 20 25 30 35 40 45 18051815182518351845185518651875188518951905191519251935194519551965197519851995YearCores Number of Cores STNDRD Figure 4-4. Histogram of number of tree ring width measurements per year for the Camp Branch site with STNDRD Camp Branch (TRICB) overlay. 39

PAGE 40

Goethe EPS 0 0.2 0.4 0.6 0.8 1 1.2 05101520253035404550Number of treesEPS EPS 0.85 Threshold Figure 4-5. How the EPS fluctuates with sample size for the TRIG and TRICB. The red line represents the 0.85 cr itical threshold. 40

PAGE 41

Steinhatchee Me an Daily Flow0 5 10 15 20 25 JanFebMarAprMayJunJulAugSepOctNovDecMonthQ m3/s Figure 4-6. Average daily discharg e of the Steinhatchee River per month for the entire period of record. Waccasassa Mean Daily Flow0 5 10 15 20 25 JanFebMarAprMayJunJulAugSepOctNovDecMonthQ m3/s Figure 4-7. Average daily discharge of the Waccasa ssa River per month for the entire period of record. 41

PAGE 42

Table 4-2. Annual (Feb. to Jan.) Descriptive Stat istics of Discharge Mean Standard Deviation Minimum Maximum Range Length of record Years m3/s Waccasassa 18 8.21 4.50 3.45 19.45 16.00 Steinhatchee 53 8.85 5.17 1.45 25.92 24.47 Odd years 27 8.5 4.56 1.52 18.66 17.14 Even years 26 9.2 5.80 1.45 25.92 24.47 Early years 26 9.11 5.77 1.64 25.92 24.28 Late years 27 8.59 4.60 1.45 14.49 13.04 42

PAGE 43

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 11 Wac c a s a ss a R iver 2 Steinhatchee River 3 F e n h o l lo w a y R i ver 4 J u m p e r Cre e k Canal 5 Rain b o w S p rings 6 Wit h lacooc h e e Ri v e r 7 Santa Fe River 8 S u wan n e e R iv e r 9 Santa Fe River 10 S u w annee River 11 S uwannee River 12 Suwannee River 1 3 Wi th l a c o o c he e Ri ve r 14 Alapaha R i ver Correlation Goethe Chronology Camp Branch Figure 4-8. Correlation coefficients of the 14 gauging stations TRIGt and TRICB. All reported correlation coefficients for this histogram are for the entire yearly record for each year adjusted to the starting month of best fit. Adjusted Year Correlations0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Jul-JunAug-JulSep-AugOct-SepNov-OctDec-NovJan-DecFeb-JanMar-FebApr-MarMay-AprJun-MayMonthCorrelation Waccasassa Steinhatchee Figure 4-9. Histogram of 12 month co rrelation coefficients for the TRIG taken for a 12 month period starting at different month forward and backwards from a calendar year. 43

PAGE 44

Waccasassa Q vs TRIGQ = 12.53 x TRIG 3.71 r = 0.86 0 5 10 15 20 25 0.6 0.8 1 1.2 1.4 1.6 1.8TRIGQ m3/s Figure 4-10. Waccasassa discharge plotted against TRIG. The Waccasassa prediction model and r value are given above the best fit line. Steinhatchee Q vs. TRIGQ = 14.54 x TRIG 5.29 r = 0.64 0 5 10 15 20 25 30 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8TRIGQ m3/s Figure 4-11. Steinhatchee discha rge plotted against TRIG. The Steinhatchee prediction model and r value are given ab ove the best fit line. 44

PAGE 45

Waccasassa0 5 10 15 20 25 19641969197419791984198919941999Year Feb.-Jan.Q m3/s Predicted Q Measured Q Figure 4-12. Waccasassa River predicted and actual discharge. Steinhatchee0 5 10 15 20 25 30 1950 1960 1970 1980 1990 2000Year Feb.-JanQ m3/s Measured Q Predicted Q Figure 4-13. Steinhatchee River pr edicted and actual discharge. 45

PAGE 46

Residuals vs. Q Minus Mean Q-15 -10 -5 0 5 10 15 -10 -5 0 5 10 15 20Q m3/s mean Q m3/sResiduals Figure 4-14. Discharge of th e Steinhatchee River in m3/s minus mean flow plotted against the residuals of the TRIGtSteinhatchee discharge linear regression. 46

PAGE 47

Table 4-3. Validation Statistics Rivers r MSE RE CE Model Waccasassa 0.864 5.47 0.714 N\A (-3.71) + ( 12.53) x (TRIGt) Steinhatchee 0.64 15.82 0.39 N\A (-5.29) + ( 14.54) x (TRIGt) Odd years 0.43 17.59 0.12 0.48 (-1.82) + ( 10.52) x (TRIGt) Even years 0.77 14.35 0.56 0.1 (-6.91) + ( 16.74) x (TRIGt) Early years 0.79 13.08 0.59 -0.09 (-10.45) + ( 18.89) x (TRIGt) Late years 0.45 17.56 0.14 0.48 (-0.94) + ( 10.45) x (TRIGt) 47

PAGE 48

Rainfall0 500 1000 1500 2000 2500 30001 9 5 7 1 9 5 9 1 9 6 1 1963 1 965 1967 1969 1971 1973 1975 1 9 7 7 1979 1 9 8 1 1 9 8 3 1 9 8 5 1 9 8 7 1 9 8 9 1 991 1993 1 995 1997 1999 2001 200 3 2005Yearmm Figure 4-15. Usher Tower raw precipitation and precipitation minus poten tial evapotranspiration through time. 48

PAGE 49

Precipitation/Pot ential RechargePrecipitation = 1.0*Potential recharge + 1059 r = 0.980 500 1000 1500 2000 2500 3000 0 200 400 600 800 1000 1200 1400 1600Potential rechargePrecipitation Figure 4.16. Precipitation plotted against potential recharge. Table 4-4. Precipitati on and Recharge Correlations Precipitation m3/s Recharge m3/s Correlation Coefficients TRIG0.68 0.71 Waccasassa 0.9 0.91 Steinhatchee 0.77 0.76 Even years 0.87 0.87 Odd years 0.58 0.56 49

PAGE 50

r vs rr of Q-precipitation = 1.068 x r of Q-TRIG 0.1534 r = 0.69 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9r of Q-TRIGr of Q-precipitation Figure 4-17. The correlation coefficients of all 14 stream gauging stations plotted to the Usher Tower raw precipitation data vs. correlation coefficients of the same stream gauging stations to the Goethe tree ring TRIGt and TRICB 50

PAGE 51

Waccasassa Final Graph0 5 10 15 20 25 1875 1895 1915 1935 1955 1975 1995Year Feb.-Jan.Q m3/s Measured Q Predicted Q Figure 4-18. Actual and predicted Waccasassa River discharge in m3/s from 1875 -2003. Steinhatchee Final Graph0 5 10 15 20 25 30 1875 1895 1915 1935 1955 1975 1995Year Feb.-JanQ m3/s Measured Q Predicted Q Figure 4-19. Actual and predicted Steinhatchee River discharge in m3/s from 1875 -2003. 51

PAGE 52

CHAPTER 5 DISCUSSION The results of this study provide insight s into the link between hydrologic cycle and longleaf pine growth in Florida and how records of this growth could be used to estimate prehistorical flow in the region. Climate, geology, topography, vegetati on, and other landscape characteristics that influence runoff are highly variable in Florida. These characteristics will influence river discharge. In orde r to record effectively this aver age value of runoff in a drainage basin a TRI thus needs to come from a sample site or sites that reflect the same average climate and landscape characteristics as th e drainage basin of interest. Tree-Ring Indices The reported EPS values of 97% indicate that the sample size of the Goethe and Camp Branch forest were sufficiently large for this study. These high EPS values also show that a strong common variable or variables control the growth of longleaf pine trees in the forest sample sites. Both TRIs have a mean of 1.0 after approximately 1910 that indicates growth in the two forests was steady during this period (Fi gures 4-1 and 4-2). Pr ior to this period the Goethe TRI had a mean of 0.75 which indicates th at the trees were going through a time of slow growth that was not adequately compensa ted for during the detrending process. Tree-rings produced during the first 10 to 20 y ears in the life of a tree often provide the least reliable climatic information, and are ofte n disregarded in dendrohyd rology studies (Fritz, 1976). A major drought in 1904 (Henry, 1931) may have had severe impacts on the young trees. Impacts could have included a period of declined growth from which they didnt fully recover until around 1910 (Figures 4-2.). The extreme low in the Camp Branch TRI that occurred in 1907 (Figure 4-4) also corresponds with the driest year on record for Lake City (Figure 1-1; 52

PAGE 53

period starts in 1900). This dry year may have also had a negative effect on the trees in this region that lasted for several years. Overall the Goethe TRI correlated with most gauging stations analyzed in this study better than the Camp Branch TRI (Figure 4-8). The better correlat ion of the Goethe TRI than the Camp Branch TRI could reflect the proximity of the Camp Branch site to the Suwannee River. Trees in close proximity to a river can be more susceptible to floods or may be receiving consistent supply of water, thereby limiting th e broad climatic effects on tree-ring growth. a more Biological indications of the results The predictive capabilities of the two reconstruc tion models indicate that water availability in Florida is a predominant fact or in the growth rates for long leaf pine species in the humid subtropical regions of the Sout heastern United States. This supports the findings Meldahl et al. (1999), who determined that climate plays a significan t role in growth of l ongleaf pine and that it is a good candidate for southeastern tree-ring analyses. Longleaf pines inability to effectively compete in wetter environments that are mo re conducive to growth, and their common occurrence on poorly drained soils may explain why they are mo isture limited (Foster and Brooks, 2001) and thus useful fo r tree-ring analyses. An exam ple of the link between tree growth and precipitation is show n in Figures 4-12 and 4-13 du ring the 1964 high flow event, when predictions of flow based on the Goethe TR I reflect a large percent flow for Steinhatchee and Waccasassa Rivers as shown by the instrumental record. The growth-discharge correlation was the str ongest with annual discharge as opposed to seasonal or monthly, indicating th at water availability throughout the year positively affects growth of longleaf pines in both forests that were studied. Longleaf pine growth is continuous throughout the year and does not go dormant during the colder or dr ier parts of the year, as is common in some species of trees in more temper ate sites (Fritz 1976). The continuous growth 53

PAGE 54

cycle of longleaf pines most likely relates to Fl oridas mild winters and their lack of seasonal defoliation. This characteristic indicates that the earlywood and latewood growth patterns seen in the longleaf pines relate to seasonal trends in water availabi lity (Figures 4-6 and 4-7). A similar result has been found by Lodewick ( 1930) and Foster and Brooks (2001), who saw no strong correlation between te mperature and growth in Florida longleaf pines. The water year for tree growth extends from February to January as estimated by the best correlation between tree-rings and discharge (Figure 4-9). This February to January water year reflects the findings of Jacobs and Ripo (2002) who concluded the ideal statistical water year for SRWMD area is the period from February 3rd to the following February 2nd. These dates were chosen by the authors because of the small numb er of extreme events that had occurred on preceding and subsequent days. The commonly a pplied USGS water year of October 1st to September 30th divides a period of high flow s een in the Waccasassa and Steinhatchee average daily discharge records (Figures 4-6 and 4-7) and is not applicable for dendrohydrologic reconstructions in the state of Florida. Validation Statistics Split sample validation is not commonly used in simple linear regression due to the straightforwardness of a model th at is based on only two variab les. Data transformation and autocorrelation modeling was also not performed on th e data sets in order to create the simplest model possible. As additiona l variables are added into a prediction model and as data transformation is done, a predicti on model will begin to fit the da ta. Model overfitting that can occur is minimized in this study by keeping the pr ediction models used as simple as possible. Calibration models developed from the early and ev en years of the Steinhatchee discharge data did not predict the corresponding validation periods as well as the late and odd years (Table 4-2). 54

PAGE 55

These simplified models imply that overfitting is not the cause of the low CE calculated for the early and even, split sample data sets. The differences in the descri ptive statistics of discharge between the even and odd data sets and the early and late data sets are mostly attributed to extreme events that occurred during these years, in particular the hi gh rainfall in 1964. Removal of this year from the analyses cuts the difference between the means of the Stei nhatchee even and odd pe riods by 73% and the difference between the means of the Steinhatchee late and early periods by 97%. So it is unlikely that the higher means, standard deviati ons, and ranges of discharge for the even and early data sets cause their favorable r, RE, a nd MSE validation statistics (Table 4-2 and 4-3). The difference in the validation statistics of the Steinhatchee even and odd calibration models indicates that there is a breakdown of the linear relati onship during odd years. During even years, the Steinhatchee experiences cons iderably greater range of precipitation and discharge than during odd year s. Since there is no known biological or climatological phenomenon occurring on biennial timescales that could cause differences in a prediction model based on even years compared to o dd, the difference in these validati on statistics is attributed to sample size. Three of the biggest outlie rs fall on even years (1956, 1964, and 1998). The highest year of flow (1964, 25.92 m3/s) and lowest year of flow (2002, 1.45 m3/s) during the Steinhatchee period of record both occurred during even numbered years. The discharge of the Steinhatchee during 1964 (25.92 m3/s) is 179% larger than the hi ghest year of discharge during the Steinhatchee odd period (14.49 m3/s during 1985). The differences in range of flow and precipitatio n lead to large dispar ity in the correlation models between the subsets and resulted in significant prediction errors. The influence that a single outlier can have on a 25 year long calibration model is de monstrated by the 32% reduction 55

PAGE 56

in the difference between the CE statistics for the even and odd calibration periods after the removal of the 1964 year (Table 4-3). Despite the differences in the even and odd validation statistics all values indicate so me predictive capacity of the fu ll Steinhatchee calibration model. Nonetheless care should be taken when assess ing the error of the model in its predictive capability for discharge that falls outside the rang e of flow that occurs during a calibration period (Meko et al., 1995). Influence of Pacific Decadal Oscilla tion (PDO) on Tree-Ring Indices There is even more pronounced difference betwee n the validation statistics of the early and late Steinhatchee sets compared to the even and odd validation statistics. For example the early calibration model fails to predict the late Steinhatchee validation period (CE = -0.09). These differences in the predictive capabilities may be attributed to the Pacific Decadal Oscillation (PDO). The PDO is a long-lived ENSO-like pattern of Pacific climate variab ility (Zhang et al., 1997). The PDO is detected as a shift in the se a surface temperture of North Pacific Ocean to either a warm phase or a cool phase (Man tua and Hare, 2002). During a warm phase, the western Pacific becomes cooler and part of the eastern ocean becomes warmer; during a cool phase, the opposite pattern occurs (Mantua and Hare, 2002). Cool phases occurred from 18901924 and from 1947-1976, while the warm phases occurred from 1925-1946 and from 1977 to a possible shift back after 1998 (Mantua and Hare, 2002). These changes in the sea surface temperatures in the north Pacific alter the path of the jet stream, the conveyor belt for storms across the continent. During a warm phase there is a shift of the jet stream towards the equator (Tootle et al., 2005). This equatorial shift brings increase d frontal precipitation to the southeastern United States and as a result summ er and fall precipitation and stream flow are more likely to be the result of highly localized convective storms (Schmidt, 2001) and may lead 56

PAGE 57

to increased climatic heterogeneity in the study area. This heterogeneity implies that climate signals recorded in tree-rings are more likely to differ from climate signals that are recorded in the discharge and rain gauging stations that are some distance apart. Since 81% of the Steinhatchee late calibration peri od falls during a PDO warm event this is a likely mechanism for the loss of correlation calculated for the late period. The effects of the PDO on the study area can be further seen in the Madison and Ocala rain gauging stations. As shown in Figure 1-1 these stations are located on opposite edges of the study area. The combined yearly precipitation aver ages of these two stati ons for the warm phase is 130.5 cm/year and for the cool phase is 132.2 cm /year representing a difference of only 1.3 %. Correlation of these stations pr ecipitation data over the 1977 to 1998 period for the warm phase of the PDO cycle had a Pearson co rrelation coefficient of 0.14. In contrast, these stations had a Pearson correlation coefficient of 0.55 from 1951 to 1976 and 1999 to 2003 during the cool phase. This further illustrates the climate he terogeneity occu rring during the PDO warm phase years of 1977 to 1998. The effect of the PDO cycle on the Goethe TRI to discharge correlation can be further seen in Figure 4-13 during the 1977 to 1998 warm phase. During this warm phase there is a low correlation between the actual and predicted di scharge record (r = 0.20 for the warm phase and 0.81 for the cool phase). The RE statistic calculated for the Steinhatchee discharge-Goethe TRI correlation during the warm phase is -0.04, whic h indicates that the Goethe TRI and the Steinhatchee River discharge show little to no common variability during this time. By disregarding the warm phase of the PDO, there is an average of 23% improvement in the Pearson correlation coefficients of all 14 gauging stati ons, suggesting that the warm phase of the PDO introduces a large degree of error in every gauging station used in this study (Table 5-1). The 57

PAGE 58

correlation coefficients between the two forest sample sites also improved by 37% after removal of the warm phase (r = 0.41 for the warm phase and r = 0.65 for the cool phase). Final model Despite the influence of the PDO, the Goethe TRI and Camp Branch TRI show positive correlations with the precipitation data (both raw and adjusted for evapotranspiration) as well as for flow measurement at all gaugi ng stations used in this study. The reported Pearson correlation coefficients for the Waccasassa and Steinhatche e discharge data to the Goethe TRI (r = 0.86 and r =0.64, respectively) are the highe st out of every stream gauging station used in this study and are in the range of values found in the reviewed l iterature from predominantly arid sites. The RE statistics calculated for both these rivers al so demonstrate some predictive qualities to the Waccasassa and Steinhatchee models (Table 4-3). The Waccasassa and Steinhatchee drainage basi ns share characteristics of small size and close proximity to the Goethe State Forest co mpared to the other 12 gauging stations. Both rivers also receive no input from first order magnitude springs and thus are linked to surface water runoff and precipitation in the Goethe State Forest better than any other gauging station. This link is further illustrated by the fact that the Waccasassa a nd Steinhatchee discharge record have better correlation coefficients with the Usher Tower rain gauge station than the other gauging stations used in this study. Since the Waccasassa River basin encompasses part of the Goethe State Forest the discharge record for this site should be well correlated with the Goethe TRI during both the PDO warm and cool phase. Becau se most of the short discharge record from the Waccasassa River (89%) occurs during a PDO cool phase; this record cannot provide support for or evidence against a control by the PDO in the region. 58

PAGE 59

Precipitation and Evapotranspiration Correlation The correlation shown in figure 4-17 shows a strong relationship (r = 0.69) between the correlation coefficients of precipita tion to flow compared with correla tion coefficients of flow to the Goethe TRI. This relationship demonstr ates a useful way for checking how strongly a potential forest sample site may correlate with disc harge stations in an area If there is a rain gauging station in close proximity to a forest sa mple site then the degree of correlation between the stations data and a nearby stream gauging st ation will provide a meaningful assessment of how well the forest might predict streamflow before even co llecting and analyzing any tree cores. The results shown in table 4-3 indicate that there is little to no improvement in the correlation coefficients of the precipitation after being adjusted for the effects of evapotranspiration data. If the annual average of evapotranspiration ra tes is constant through time (as indicated by Figure 4-16) then the effect s of evapotranspiration on tree growth need not be accounted for when using tree-ring proxies for climate in the state of Florida. The Thornthwaite-Mather evapotranspiration model, however, does not account for the duration of a rain event or the amount of tim e between rain events, which may have a large effect on the amount of evapotranspiration affect ing the soil. For instance, a m oderate rainfall of one inch on a dry soil may only wet the upper 10 inches of so il, while the lower areas remain dry, so unless more moisture is added, water will not move to deeper levels by gravity flow (Fritz, 1976). Flow Proxies The reconstruction presented here produces a proxy for flow to 1875 for the Waccasassa and Steinhatchee Rivers. These proxies show a period of low flow in 1932 and 1916-1917 that corresponds with the dust bowl and a period of low precipitation (Figures 4-18 and 4-19; Henry, 1931). The period of predominant high flow st retching from 1919 to 1924 also corresponds well 59

PAGE 60

with published climate tables (Henry, 1931). Even prior to 1910, which depicts a period of extended low flow that is most likely inaccurate based on cursory climate records, some of the peaks and valleys correspond to state wide highs and lows for precipitation, such as the higher than normal precipitation year of 1900 and the lo wer than normal year of 1904 (Figures 4-18 and 4-19; Henry, 1931). 60

PAGE 61

Table 5-1. The growth-discharge correla tion for the cool phase PDO cycle. Gauging stations r % improvement* 1 Waccasassa River 0.8746 2% 2 Steinhatchee River 0.8247 22% 3 Fenholloway River 0.7635 23% 4 Jumper Creek Canal 0.6922 28% 5 Rainbow Springs 0.5825 11% 6 Withlacoochee River 0.4791 5% 7 Santa Fe River 0.6108 26% 8 Suwannee River 0.5679 21% 9 Santa Fe River 0.6081 29% 10 Suwannee River 0.4439 15% 11 Suwannee River 0.4832 25% 12 Suwannee River 0.5763 40% 13 Withlacoochee River 0.4562 34% 14 Alapaha River 0.5463 47% 61

PAGE 62

CHAPTER 6 CONCLUSION The forest sample sites chosen for this st udy were picked for their ability to provide suitable cores, and not for their propensity to reflect runoff of any particular drainage basin. However, despite this limitation, results of this study indicate some value in the use of dendrohydrologic reconstructions of historic discharge in Florida. The 128 year Waccasassa and Steinhatchee river reconstructions are examples of how dendr ohydrologic reconstructions could provide insight to the range of short hydrologic variabil ity that has occurred in Floridas past. These reconstructions demons trate the ability to hind-cas t several droughts outside the calibration period that correspond w ith instrumental records. L ongleaf pine growth in Florida responds well to variable water availability and seems to accura tely reflect runoff in the area where they reside. The major inaccuracies in th e reconstructions appear to come largely from the underestimation of maxima and overestimation of minima of hydrologic predictions and from Floridas variable regional climate. This clim ate heterogeneity increases during the warm phase of the PDO cycles which are time periods of poor correlations betw een gauging stations and forest sites. These dendrohydrol ogic reconstruction techniques ma y be of use, however, because most large scale climate anomalies (droughts or floods) occur during PDO cool phases and climate anomalies are the periods of greatest interest to water res ource planners. As demands for water resources in the State continue to grow our susceptibility to drastic hydrologic changes will also increase. Insights into the magnitude of departure from current perception of average hydrologic conditions could be cri tical in water resource planni ng. Results of this study should aid in how dendrohydrologic studies are employed in this area and give some perspective on what climatic information may be garnered from the rings of trees. 62

PAGE 63

LIST OF REFERENCES Barnes, J., 1998, Floridas hurrican e history: Chapel Hi ll, North Carolina, The University of North Carolina Press, 330 p. Case, R.A., and MacDonald, G.M., 2003, Tree ring r econstructions of streamflow for three Canadian Prairie rivers: Jour nal of the American Water Resources Association, v. 39, p. 703-716. Cleaveland, M.K., and Stahle D.W., 1989, Treering anal ysis of surplus and deficit runoff in the White River, Arkansas: Water Re sources Research, v. 25, p. 1391 1401. Cook, E.R., and Holmes, R.L., 1997, ARSTAN: chronology development. in Grissino-Mayer, H.D., Holmes, R.L., and Fritts, H.C., eds., The international treering data bank program library, version 2.1. Users manual: Laboratory of tree-ring research, The University of Arizona, Tucson, Arizona, p. 75. Cook, E.R., and Kaririukstis, L.A., 1990, Met hods of Dendrochronology: Dordrecht, The Netherlands, Kluwer Academic Publishers, 394 p. Coile, T.S., 1936, The effects of rainfall and temperat ure on the annual radial growth of pine in the Southern United States: Ecological Monographs, v. 6, p. 533-562. Dingman, S.L., 1994, Physical hydrology: New Jers ey, Prentice Hall, Inc., p.575. Ford, C.R and Brooks, J.R., 2002, Detecting forest stress and decline in re sponse to increasing river flow in southwest Florida, USA: Forest Ecology and Management, v. 160 p. 45-64. Foster, T.E., and Brooks, J.R., 2001, Long-term trends in growth of Pinus palustris and Pinus elliottii along a hydrological gradient in central Florida: Canadian Journal of Forest Research, v. 31, p.1661-1670. Fritts, H.C., 1976, Tree rings and climat e: New York, Academic Press, 567 p. Gedalof, Z., Peterson, D.L., and Mantua, N.J ., 2004, Columbia River flow and drought since 1750: Journal of the American Water Resources Association, v. 40, p. 1579-1592. Gray, S.T., Jackson, S.T., and Betancourt, J. L., 2004, Tree-ring base d reconstructions of interannual to decadal-scale precipitation variability for northeastern Utah since 1226 A.D: Journal of the American Water Resources Association, v. 40, p. 947960. Grissino-Mayer, H.D., 2001, Evaluating crossdati ng accuracy: A manual and tutorial for the computer program COFECHA. Tree-Ring Research, v. 57, p. 205. Hand, J., Tauxe, V., and Friedemann, M., 1988, Water quality assessment for the State of 63

PAGE 64

Florida: Tallahassee, Florida Bureau of Surface Water Management, Division of Water Management, Department of Environmen tal Regulation, Standards and Monitoring Section, 289 p. Henry, A.J., 1931, The calendar year as a time unit for drought statistic s: Monthly Weather Review, v. 59, p. 151-154. Henry, J.A., Porter, K.M., and Coyne, J., 1994, The climate and weather of Florida: Sarasota, Florida, Pineapple Press, Inc., 279 p. Jacobs and Ripo 2002, Minimum flows and le vels for the Lower Suwannee River implementation methodology: Live Oak, Florida, Suwannee River Water Management District Final Report, p. 129. LeBlanc, D., and Terrell, M., 2001, Dendroclimatic analyses using Thornthwaite-Mather-Type evapotranspiration models: A bridge between dendroecology and forest simulation models: Tree-Ring Research, v. 57, p. 55-56. Lodewick, E.J., 1930, Effect of certain climatic fa ctors on the diameter growth of longleaf pines in Western Florida: Journal of Ag ricultural Research, v. 41, p. 349363. Loaiciga, H.A., Haston, L., and Michaelsen, J. 1993, Dendrohydrology and long-term hydrologic phenomena: Reviews of Geophysics, v. 31, p. 151-171. Loudermilk, E.L., 2005, Spatial and demographic m odeling techniques applie d to the longleaf pine ( Pinus Palustris) ecosystem of North Central Flor ida [M.S. Thesis]: Gainesville, University of Florida, 80 p. Mantua, N.J., and Hare, S.R., 2002, The Pacific Decadal Oscillation: Journal of Oceanography, v. 58, p.35-44. Meko, D., Stockton, C.W., and Boggess, W.R., 1995, The tree-ring record of severe sustained drought: Water Resources Bulletin, v. 31, p. 789-813. Meko, D., and Graybill, D.A., 1995, Tree-ring reconstruction of Uppe r Gila River discharge: Water Resources Bulletin, v. 31, p. 605-616. Myers, R.L., and Ewel, J.J., 1990, Ecosystems of Florida: Orlando, University of Central Florida Press, 765 p. Meldahl, R.S., Pederson, N., Kush, J.S., and Varner, J.M., 1999, Dendrochronological investigations of climate and compe titive effects on longleaf pine growth. in Wimmer, R. and Vetter, R.E. eds., Tree Ring Analysis: Bi ological, Methodological and Environmental Aspects: Wallingford, UK., CABI Publishing, p. 265-285. 64

PAGE 65

North, G.R. (chr), 2006, Committee on Surface Temp erature Reconstructions for the Last 2,000 Years, National Research Council, Surface temp erature reconstructions for the last 2,000 years: The National Academies Press, 160 p. Obeysekera, J., Trimble, P., Neidrauer, C., Pathak, C., VanArman, J., Strowd, T., and Hall, C., 2006, Consideration of long-term climatic variability in regional modeling for SFWMD planning & operations: South Florida Water Management Distict, 45 p. Pederson, N., Jacoby, G.C., D'Arrigo, R.D., Cook E.R., Buckley, B.M., Dugarjav, C., and Mijiddorj, R., 2001, Hydrometeorological rec onstructions for Northeastern Mongolia derived from tree rings: AD 1651-1995: Journal of Climatology, v. 14, p. 872-881. Taylor, G.F., and Snell, L.J., 1978, Water re sources of the Waccasassa River basin an adjacent areas, Florida: U.S. Geological Survey Water-Resources Investigations 77101, 1 sheet. Scott, T.M., Means, G.H., Meegan, R.P., Means, R.C., Upchurch, S.B., Copeland, R.E. Jones, J., Roberts, T., and Willet A., 2004, Springs of Florida: Tallahassee, Florida, Florida Geological Survey Bulletin No. 66, 377 p. Schmidt, N., Lipp, E.K., Rose, J.B., and Luther, M.E., 2001, ENSO influences on seasonal rainfall and river discharge in Flor ida: Journal of Climate, v. 14, p. 615-628. Schumacher, F.X., and Day, B.B., 1939, The influences of precipitation upon the width of annual rings of certain timber trees: Ecological Monographs, v. 9, p.387-429. SRWMD, 1995, Waccasassa River watershed manage ment plan: Live Oak, Florida, Suwannee River Water Management District Surf ace Water Improvement and Management Program, 65 p. SRWMD, 1989, Steinhatchee Rive r basin assessment: Live Oak, Florida, Suwannee River Water Management District Interim Report, 86 p. Thornthwaite, C.W., and Mather, J.R ., 1957, Instructions and tables for computing potential evapotranspiration and the water balance. Drex el Institute of Technol ogy, Publications in Climatology, vol X, 311 p. Tootle, G.A., Piechota, T.C., and Singh, A.K., 2005, Coupled oceanic-atmospheric variability and U.S. streamflow: Water Re sources Research, v. 41, W12408, doi:10.1029/2005WR004381. Watson, I., and Burnett, A., 1995, Hydrology, an environmental approach: New York, Lewis Publishers/Crc Press, Times Mirror Book, 702 p. Wendland, W.M., and Watson-Stegner, D., 1983, A t echnique to reconstruct river discharge history from tree-rings: Water Resources Bulletin, v. 19, p. 175-181. 65

PAGE 66

Wigley, T.M.L., Briffa, K.R., and Jones, P.D ., 1984, On the average value of correlated time series, with applications in dendrocli matology and hydrometeorology: Journal of Climate and Applied Meterology, v. 23, p. 201-213. Winsberg M. D., 1990, Florida weather: Orlando, Flor ida, University of Central Florida Press, 171 p. Van Lear, D.H., Carroll, W.D., Kapeluck, P.R., and Johnson, R., 2005, History and restoration of the longleaf pine-grassland ecosystem impli cations for species at risk: Forest Ecology and Mangement, v. 211, p. 150. Varner, J.M., and Pederson, N., 2002, Potential for dendrochronological analysis of the Suwannee River basin tree species: Suwannee River Water Management District, Live Oak, F.L., 20 p. Zhang, Y., Wallace, J.M., and Battisti, D.S., 1997, ENSO-like interdecadal variability: Journal of Climate, v. 10, p. 1004-1020. 66

PAGE 67

BIOGRAPHICAL SKETCH Kris Crockett was born on March 31, 1978 in Daytona Beach, Florida. The middle of three children, he grew up in Palm Coast, Flor ida and graduated from Flagler Palm Coast High School in 1997. He received his B.S. in geol ogy from University of Florida in 2004. Kris received his M.S. in the geological sciences from the University of Florida in 2007. He plans to pursue a career in hydrology here in Florida. 67