Field and basin scale water quality models for evaluating agricultural nonpoint pollution abatement programs in a South Florida flatwoods watershed

Material Information

Field and basin scale water quality models for evaluating agricultural nonpoint pollution abatement programs in a South Florida flatwoods watershed
Heatwole, Conrad Dean, 1956-
Publication Date:
Physical Description:
xi, 185 leaves : ill., maps ; 28 cm.


Subjects / Keywords:
Hydrological modeling ( jstor )
Land use ( jstor )
Modeling ( jstor )
Nutrients ( jstor )
Pastures ( jstor )
Phosphorus ( jstor )
Simulations ( jstor )
Water tables ( jstor )
Watersheds ( jstor )
Wetlands ( jstor )
Water -- Pollution -- Mathematical models ( lcsh )
Water quality -- Florida ( lcsh )
Water quality -- Mathematical models ( lcsh )
Watersheds -- Florida ( lcsh )
bibliography ( marcgt )
theses ( marcgt )
non-fiction ( marcgt )


Thesis (Ph. D.)--University of Florida, 1986.
Includes bibliographical references (leaves 141-152).
General Note:
General Note:
Statement of Responsibility:
by Conrad Dean Heatwole.

Record Information

Source Institution:
University of Florida
Holding Location:
University of Florida
Rights Management:
Copyright [name of dissertation author]. 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:
000989384 ( ALEPH )
AEW6246 ( NOTIS )
17689525 ( OCLC )

Full Text








Kendra and Cheryl


Many people contributed, directly and indirectly, to the successful

completion of my PhD program. To them, I express my gratitude.

I greatly appreciate the help of Dr. A.B. (Del) Bottcher, who ably

guided my graduate program and research, provided the necessary finan-

cial support, and in general, provided a positive environment that help-

ed make this endeavor an enjoyable one. Dr. Kenneth Campbell served as

chairman of the supervisory committee and also provided invaluable

assistance and support along the way. Special thanks are extended to

the other members of the supervisory committee: Associate Professor

L.B. Baldwin either had, or knew where to find, important information

dealing with the project area; Dr. L.H. Allen, Jr., introduced me to the

complexity of soil phosphorus chemistry; Dr. Wayne Huber served as

advisor for my minor and was instructor for several challenging courses.

Data and additional assistance were provided by Alan Goldstein and

Gary Ritter of the South Florida Water Management District, and by the

staff of the USDA-ARS Southeast Watershed Research Program, Tifton,

Georgia. Financial support was provided by grants from the U.S. Army

Corps of Engineers and the South Florida Water Management District.

Behind the scenes, providing much more than technical support,

were my loving wife and girls, who rejoiced with me in the good times,

and cheered me up in the bad. To Janie, Kendra and Cheryl, I express my

deep appreciation and love.



LIST OF TABLES . . . .vii

LIST OF FIGURES. . . .. ix




The "Best Management Practice" Concept
Evaluating BMP Effectiveness .
Mathematical Models .
Research Problem and Objective .


Description of the Taylor Creek--Nubbin Slough
General Characteristics . .
Hydrologic Processes . .

Data Sources . . .
Taylor Creek--Nubbin Slough .
Upland Detention/Retention Demonstration
Ash Slough (Bass East and Bass West)
SEZ dairy . .
Lower Kissimmee River Basin .

Phosphorus Transformations in Sandy Soils .
Phosphorus Retention in Soil .
Modeling Soil/Phosphorus Reactions .
Phosphorus Relationships in Sandy Soils .

. . x

. . 7

* .
* .

. .
. .
. .
. .
. .

Nutrient Uptake in Wetland Systems .
Component Studies . .
Nitrogen transformations .
Phosphorus transformations .
Nutrient uptake/release by aquatic
System Studies . .
Natural wetlands .
Artificial wetlands .
Summary . .


Mathematical Models for Hydrology and Water Quality 40
The Curve Number Method for Predicting Runoff 40
Water Quality Models . . 43


Development . . . 48
Description of the CREAMS Model . 48
Background . . 48
Hydrology algorithm . .. 49
Limitations for flatwoods watersheds 50
Concepts of CREAMS-WT . . 51
Description of Algorithms . 52
Water table simulation . 52
Water table in the root zone . 53
Water table in the lower zone . 54
Runoff prediction . 56
Calibration and Testing . 57
Calibration . . 57
Bass West simulation . 58
Evaluation of the CREAMS-WT Runoff Equation 61

Verification . . 65
Armstrong Slough . . 65
SEZ Dairy . . 67
Long Term Simulation . . 67

Sensitivity Analysis . . 74
Parameter Estimation . . 79


Development . . 81
Description of the CREAMS Nutrient Model. ... 81
Soluble nitrogen and phosphorus. ... 81
Parameter Estimation . 83
Applicability of P Model to Flatwoods Watersheds. 85
CREAMS-WT Enhancements for Flatwoods Watersheds 87
Denitrification. . 87
Phosphorus in rainfall . 88
Soluble phosphorus model . 88
Testing and Calibration . . 89
Issues in calibration and verification 89
Simulation of Bass West pasture. ... 92

Verification . . . 94
SEZ Dairy . . 96
Bass East Pasture . . 97
Discussion. .. 99
Sensitivity Analysis . . .100


Concepts . . . .104
Spatial Discretization. . .105
Time Scale. . . .105

Components . . . .106
Nutrient Attenuation. . . .106
Algorithm. .... . .106
Background concentrations.. . 108
Rate coefficients. . .109
Overland flow . .109
Attenuation to sub-basin outlet .110
Main channel. . .112
Impoundments. . .112
Distance estimates . ... .112
Runoff Detention by Impoundments. . .113
Direct Loads to Wetlands and Streams. . .114
Background Loading Rates. . .116

Implementation . . .116
Cell Database . . .116
Cell BMP File . . .118
BASIN Parameter File. . .... .119
CREAMS-WT Cell Loads. . . .119
Access code. . .. .120
Hydrology simulations. . .120
Nutrient simulations . .120

Verification . . .121
Taylor Creek--Nubbin Slough . .121
Otter Creek--Annual Response. . .125
BMP Simulation. . . .129

Sensitivity Analysis . . .129

Summary . . .133
Conclusions . . .138
Recommendations for Additional Research .139








. .104



2-1 Characteristics of the Upland Detention/Retention
Demonstration Project watersheds . 16

3-1 Annual water balance for Armstrong Slough. ... 68

3-2 Annual water balance for SEZ Dairy . 68

3-3 Annual water balance for 20-year CREAMS-WT simulation. 72

3-4 Sensitivity analysis of selected CREAMS-WT parameters. 76

3-5 Sensitivity analysis of selected CREAMS-WT parameters
with deep seepage simulated. . .. 77

4-1 Bass West pasture simulated and observed average annual
flow weighted concentrations . 93

4-2 SEZ Dairy simulated and observed average annual
flow weighted concentrations . 98

4-3 Bass East pasture simulated and observed average annual
flow weighted concentrations . 98

4-4 Sensitivity analysis of selected CREAMS-WT nutrient
model parameters . . .101

4-5 Model response to time and number of applications of a
constant annual fertilizer amount. . .103

5-1 Rate coefficients for nutrient uptake. . .111

5-2 Rate coefficients for nutrient uptake used in BASIN. .111

5-3 Annual nutrient content of cattle waste, by land use .115

5-4 Base annual nutrient loads used in BASIN simulation. .122

5-5 Characteristics of watersheds in the Taylor Creek--Nubbin
Slough basin . . .122

5-6 Verification of BASIN model with TCNS basin data .123

5-7 BASIN predictions and mean time series concentrations
for several watersheds . . .127

5-8 BASIN predictions and mean annual flow weighted
concentrations for Otter Creek . .127

5-9 Comparison of base and BMP simulations . .128

5-10 Sensitivity of predicted nitrogen loads
to selected parameters . . .131

5-11 Sensitivity of predicted phosphorus loads
to selected parameters . . .132

A-I Curve number and equivalent storage. . .154

B-I Hydrology parameter files for Bass West pasture. .155

B-2 Hydrology parameter files for Armstrong Slough .156

B-3 Hydrology parameter files for SEZ Dairy. . .157

B-4 Hydrology parameter files for 20-year simulation .158

C-I Nutrient parameter file for Bass West pasture. .159

C-2 Nutrient parameter file for Bass East pasture. .160

C-3 Nutrient parameter file for SEZ Dairy. . .161

D-I BASIN parameter file for the TCNS simulation .163

D-2 CREAMS-WT cell loads for the TCNS simulation .164

CREAMS-WT hydrology parameter files for TCNS simulation..

CREAMS-WT nutrient parameter files for TCNS simulation .

BASIN model output for TCNS simulation . .

BASIN model output from BMP simulation for the TCNS basin.








2-1 Florida Flatwoods physiographic regions. . 8

2-2 Taylor Creek--Nubbin Slough basin. . .. 13

2-3 Location of the Upland Detention/Retention
Demonstration Project watersheds . 18

2-4 Ash Slough (Bass East and Bass West pasture) watershed 19

2-5 SEZ Dairy watershed. . . .. 20

2-6 Available storage versus depth to water table (ARS). .. 42

2-7 Available storage versus depth to water table (SFWMD). 42

3-1 Bass West pasture water table. . .. 60

3-2 Graphical representation of the CREAMS-WT runoff equation. 64

3-3 Armstrong Slough water table . . 69

3-4 SEZ Dairy water table. . . .. 70

4-1 Bass West pasture simulated and observed monthly
flow weighted phosphorus concentrations. ... 95

5-1 Flow chart of BASIN model. . . .107

5-2 Data paths in the implementation of the BASIN model. .117

Abstract of Dissertation Presented to the Graduate School
of the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Doctor of Philosophy



Conrad Dean Heatwole

August 1986

Chairman: Kenneth L. Campbell
Co-chairman: A. B. Bottcher
Major Department: Agricultural Engineering

The CREAMS-WT and BASIN water quality models are designed to

reflect the hydrologic, nutrient (nitrogen and phosphorus), and

management characteristics typical of flat, sandy, high-water-table

South Florida flatwoods watersheds. Characteristics to be considered

include 1) the storage based hydrologic system; 2) low chemical

reactivity with the sand soils; 3) nutrient uptake in wetlands;

4) management practices appropriate for pasture and dairy land use.

CREAMS-WT, a modified version of CREAMS, simulates the dynamic

water table, limits deep seepage, and uses a modified curve number

method for predicting runoff. Predicted annual water balance com-

ponents compare favorably with data from three watersheds in South

Florida and are dramatically better than CREAMS predictions. A

20-year simulation shows that the model is stable and performs

predictably over the range of natural rainfall variability.

The nutrient component of CREAMS was changed to represent the

lower phosphorus buffering capacity of flatwoods soils. Predicted

annual nutrient loads compare well with data from three watersheds that

cover the range of expected areal nutrient loading rates.

The BASIN model predicts the delivery of field/cell nutrient loads

to sub-basin and basin outlets. A first-order rate equation is used to

attenuate nutrient loads in streamflow and flow through wetlands. Rate

coefficients were determined from nutrient uptake studies reported in

the literature. BASIN also computes nutrient loads from cattle in

streams and wetlands, simulates effects of fencing and impoundments,

and provides background loads for non-agricultural land use.

The 464 km2 Taylor Creek--Nubbin Slough (TCNS) basin was simulated

without calibration using 5 ha grid cells as "fields". Parameters were

selected from the literature and from knowledge of the area. Predicted

long term average annual nutrient yields compare closely with mean

observed values at the basin outlet and at gaged sub-basin outlets.

The CREAMS-WT and BASIN models can be used for evaluating the

relative effectiveness of alternative management practices at both the

field and basin level in the TCNS basin, and should be applicable to

other watersheds in Coastal Plain flatwoods regions.


Protecting and managing our water resources is an issue of

national interest. Of the leading water resource problems facing the

nation, nonpoint source stream pollution has been identified as one of

continuing concern (DeCoursey, 1985). The State of Florida is experi-

encing increasing stress on its water resources due to the growing

population and the accompanying land development. In the 1960s,

development in the Miami area resulted in dairy farmers moving from

that vicinity to the open range lands north of Lake Okeechobee, locat-

ing in the Taylor Creek and Nubbin Slough watersheds. These large

dairies grow very little of their own feed, choosing rather to buy it

from outside sources. The result is that large amounts of nitrogen and

phosphorus are imported to these watersheds. Not surprisingly, some of

this nitrogen and phosphorus reaches the watershed outlet. In calcu-

lating the nutrient budget of Lake Okeechobee, Federico et al. (1981)

found that on an annual basis, the Taylor Creek--Nubbin Slough (TCNS)

basin contributes approximately 30% of the phosphorus and 5% of the

nitrogen to the lake while accounting for only 4% of the water inflow.

Concern over the eutrophication status of Lake Okeechobee has

focused attention on reducing the nutrient load flowing into the lake.

Accordingly, both state and federal funds have been allocated to the

TCNS basin with the goal of reducing nutrient loading to the lake. The

state funded Taylor Creek Headwaters Program (TCHP) and the federally

funded Rural Clean Waters Program (RCWP) have both provided cost

sharing to land owners for implementing management practices which have

been identified as having potential for reducing the nonpoint source

nutrient load coming from this basin.

The "Best Management Practice" Concept

Effort has gone into identifying, often conceptually, the manage-

ment practices which might be expected to reduce nonpoint sources of

pollution when compared with the standard management. These "best

management practices" (BMPs) focus on ways to minimize the negative

water quality impact of a particular land use while keeping it profit-

able for the land user. While reduction of point source pollution

loads has focused on collection and treatment immediately before

discharge, the emphasis for reducing nonpoint source loads has been on

reducing the pollution potential at the source through the use of BMPs.

With the emphasis on utilizing BMPs for controlling nonpoint

pollution, an obvious need has been to have some measure of the effec-

tiveness of different BMPs. Proper management of any type of system

depends on having reasonable estimates of the expected outcomes of the

different alternatives being considered. So too with water resources,

the land user and the governing water quality agency must resolve the

often conflicting management objectives of maximizing return and

minimizing pollution. A logical decision can be reached only with

adequate data. Estimates of the effectiveness of BMPs can be used to

a) select the most appropriate BMP for a particular site; b) estimate

the overall impact of BMPs applied in a watershed; c) rank the

cost-effectiveness of BMP alternatives; d) determine an "optimum" BMP

program based on a specified objective.

The objectives) used in determining what is "most appropriate" or

for an optimization scheme will depend on the individual situation but

may include meeting some desired water quality goal, or achieving the

best possible reduction in pollutant load. The possible financial

objectives are numerous. Some examples would be maximizing the utility

of cost-sharing monies, minimizing cost (initial investment or mainte-

nance costs or both) to the farmer, or maximizing return while meeting

some minimum water quality standard.

Evaluating BMP Effectiveness

Two approaches have been used for evaluating the effectiveness of

BMPs: watershed data collection and mathematical modeling. While

measuring the actual effects may be the most desirable way to evaluate

the response of a system, there are significant drawbacks to using data

collection for general BMP evaluation. Early data collection programs

comparing paired watersheds found, in many cases, that differences

between the plots masked the treatment effects being evaluated

(DeCoursey, 1985). There is tremendous variability in the watershed

characteristics that affect the rainfall-runoff process and chemical

transformations. A large number of sampling sites (watersheds) would

be required to get a statistically significant result, and this is

prohibitively expensive. Time is a second limiting factor. Due to the

variability in rainfall and other meteorological data, it is desirable

to have a long term record. Not only is it expensive to maintain a

sampling program, but we cannot wait years to get these data before

taking action on the water resource issues that face us today.

The second approach to BMP evaluation is the use of mathematical

models. Physical process models attempt to describe the physical,

chemical and biological processes that define the natural system being

modeled. Models developed around our understanding of physical

processes enable us to examine in greater detail the various inter-

relationships between those processes. The other main class of models,

empirical models, are less applicable for use in BMP evaluation.

Empirical models are mathematical formulations based on data analysis

and do not necessarily represent the actual processes controlling the

system. They should not be used beyond the range of data used in their

development, and in addition, empirical models often use fitted

parameters thus are not well suited for simulating BMPs.

Mathematical Models

Although long term watershed data are needed for model development

and testing, it is now recognized that the primary means of evaluating

the effectiveness of BMPs is through the use of detailed water quality

models designed for that purpose (Beasley et al., 1982; DeCoursey,

1985). The impact of a BMP on offsite water quality is determined

through modeling by simulating the pre-BMP and post-BMP conditions,

and comparing the results. BMPs are simulated by making changes in the

model parameter values that are representative of the actual changes

that would be expected were the BMP actually implemented.

However, models also have their limitations. Extensive amounts of

data are often required, some of which will not be readily available.

In some cases, more is known about the governing processes than can be

implemented in a model due to computational (cost) and data limita-

tions. We are also limited in our understanding of the details of the

processes that govern the natural system. There is a tremendous amount

of variability in the response of any natural system, and while current


physically-based models may be able to predict mean response with some

certainty, much of the variability is not addressed as the causal

processes are not fully understood (McDowell et al., 1980).

In practice, BMPs are implemented on a field by field basis; thus,

the most accurate assessment of the effectiveness of a BMP can be ob-

tained with field-scale simulation. However, it may also be desirable

to evaluate the effects of a BMP implementation program over a water-

shed or basin to provide information for evaluating broader planning,

funding and water quality objectives.

Research Problem and Objective

Management practices with potential for reducing the nonpoint

nutrient load discharged from the TCNS basin have been identified

(Ritter and Allen, 1982). Under the TCHP and RCWP programs, the BMPs

deemed most effective have been implemented (RCWP, 1982; 1984), and

already, the question of the effectiveness of this basin-wide BMP

program is being debated. The water quality data collection effort

faces the problems discussed above in attempting to categorize the

effectiveness of the BMP program. A water quality model could provide

some indication of the long term water quality improvement that might

be expected from various BMP programs.

Bottcher and Baldwin (1981) developed a BMP application model

which uses a digitized database of the TCNS basin and matches BMPs with

appropriate land use. Selection of BMPs is based on land use, cattle

density, and distance from waterways. With the inclusion of cost data,

the model can be used to determine the total amounts of BMPs applied

and the associated costs for different basin-wide BMP scenarios. For a

more complete evaluation of BMP alternatives, Bottcher and Baldwin

(1981) recognized the need for a water quality model which could pro-

vide estimates of the water quality improvement associated with the BMP


Numerous water quality models have been designed for general

application. However, their use in evaluating BMP effectiveness in the

TCNS basin (and in similar South Florida flatwoods watersheds) is sig-

nificantly limited by several somewhat unique features of this region.

1) The hydrologic response is dominated by the flat terrain, sandy

soils, and a high water table.

2) Phosphorus transport and transformations in the sandy soils do

not follow conventional concepts.

3) Wetlands comprise a significant portion of the area and often

receive (by design and default) nutrient-ladened runoff.

4) This region is predominantly pasture with very little tilled

land, and appropriate BMPs may not be readily reflected in the

parameters of the existing models which tend to be oriented

towards row crop land use.

The objective of this research is to develop a water quality model

that can be used to evaluate the effectiveness of BMPs in South Florida

flatwoods watersheds. The model should encompass both field and basin

scale simulation and must address the unique hydrologic, chemical, and

management conditions characteristic of these watersheds.


Description of the Taylor Creek--Nubbin Slough Basin

General Characteristics

The Taylor Creek--Nubbin Slough (TCNS) basin lies in the Southern

Florida Flatwoods physiographic region (Figure 2-1). The region is

very flat with sandy soils. The term "flatwoods" is still used even

though in most areas the pine forests have been cleared for range and

pasture. The TCNS basin is very similar to the larger adjoining Lower

Kissimmee River (LKR) basin which has similar physiographic and land

use characteristics.

The soils of this area are generally classified as fine sand,

often with less than 1% clay content, and typically have very high

hydraulic conductivities (>16 cm/hr). Internal drainage may be very

rapid to slow, so a dual hydrologic classification is used, A/D or B/D,

with the appropriate classification dependent on what drainage improve-

ments have been made. Slopes range from 0 to 2 percent but are

typically less than 0.5 percent. As a result of the low slopes, there

is little or no erosion.

Water tables vary with topography and soil classification but

generally range from the surface to 2 meters deep. The water table in

the phreatic aquifer is underlaid by the impermeable Hawthorne

Formation, and beneath that is the artesian Floridan aquifer. Thus,

deep seepage is generally assumed to be negligible (Stewart, 1980;







Figure 2-1. Florida Flatwoods physiographic regions and site location.

Yates et al., 1982). During the wet summer months, the water table is

near the surface and is the controlling factor in the hydrologic

response of the area.

Wetlands are an important component of the basins, impacting both

hydrology and water quality. There are many isolated ponded areas,

some seasonal and many year-round. Due to the lack of sufficient

gradient, these areas often remain wet even when ditched. The primary

drainage ways are poorly defined channels through marshes and wooded

wetlands. Total wetland area in the TCNS basin accounted for approx-

imately 8% of the total area in 1980.

Land use in the area has moved towards more intensive management,

with dairy farming and ranching now dominating. For the Upper Taylor

Creek watershed, Allen et al. (1982b) report 1959 land use as 24%

improved pasture, 44% unimproved pasture, 17% forest and range, and 15%

miscellaneous (including roads, buildings, and ponded wetlands). By

1980 it was 3% citrus, 69% improved pasture, 16% forest and range, 4%

urban, and 8% miscellaneous. The Nubbin Slough basin has similar land

use, while the larger Lower Kissimmee River basin is less developed,

with higher percentages of unimproved pasture and range land.

The basins are ditched in varying degrees depending on the land

use. Ditching densities increase from unimproved pasture, to improved

pasture, with intensively managed pasture along with citrus and row

crops having the greatest ditching densities (18 km/sq km reported by

Huber et al., 1976). Ditches often help blur watershed boundaries as

they inter-connect over wide areas. Watershed boundaries are generally

poorly defined, and direction of flow in a ditch may be dependent on

prevailing wind, type of storm, and the depth of water in the ditch.

Several factors uniquely impact the water quality of this region.

There is evidence that phosphorus will leach through sandy flatwoods

soils (Neller et al., 1951; Blue, 1970; Chaiwanakupt and Robertson,

1976). This is a rare phenomenon for phosphorus in soils. Very

reactive in solution, phosphorus normally is readily adsorbed by soil

minerals with negligible concentrations remaining in the soil solution

(Buckman and Brady, 1969; Larsen, 1967). Because of the very low clay

content, the adsorptive capacity of these soils is greatly reduced

resulting in significant phosphorus concentrations in the soil water

and in runoff.

Wetlands play an important role in water quality. They are most

commonly considered as sinks for nutrients although under certain con-

ditions they may be a source of nutrients. A number of studies have

been done in the LKR basin (Davis, 1982; Federico et al., 1978;

Goldstein, 1986), but pertinent data can also be obtained from studies

along the Eastern Coastal Plain (Simpson et al., 1983; Craig and

Kuenzler, 1983) and Louisiana (Conner and Day, 1982).

Hydrologic Processes

Runoff from these flat, sandy soils is not produced by the common

concept of an infiltration limiting 'rainfall excess'. Runoff is more

likely a result of rainfall on saturated areas and subsurface flow to

ditches and drainage ways both during and following a storm. This

concept of the rainfall/runoff process is very similar to the 'variable

source area' concept (Hewlett and Nutter, 1970; Hewlett and Troendle,

1975; Betson and Marius, 1969). These researchers were studying forest

hydrology where infiltration rates were also rarely, if ever, a limiting


The very flat terrain also challenges some of the concepts of

classical hydrology. The standard terms of overland flow, interflow,

and base flow are difficult to apply to flatwoods watersheds. Accepted

interpretations associate these terms with particular flow processes

and specific time intervals of discharge hydrographs. Although the

same flow processes occur on flatwoods watersheds, they are often

difficult to separate and not associated with the same hydrograph time

interpretations. Direct runoff is typically defined as including over-

land flow and interflow. Base flow is defined as being nearly constant

and originating from groundwater. More appropriate terms for describ-

ing runoff from flatwoods watersheds are rapid, intermediate, and slow

flow. These terms do not attempt to explain process but refer only to

flow rate (Speir et al., 1969).

A water balance for this area shows that the rainfall and soil

moisture storage leave the system primarily by evapotranspiration and

secondarily by runoff with percolation or deep seepage being negligible

(Stewart, 1980; Yates et al., 1982). Thus, the level of the water

table (i.e. change in soil moisture storage) directly reflects input

from rainfall, and losses by evapotranspiration and runoff.

Data Sources

There are numerous sources of data describing the hydrology and

water quality of Southern Florida Flatwoods. The Upper Taylor Creek

watershed has been studied in greatest detail, with initial data

collection beginning in 1955 (Speir et al., 1969). In the past 20

years, concern over the nutrient status of Lake Okeechobee has spawned

a spate of special studies specifically structured to survey nutrient

dynamics in the Lower Kissimmee River flatwoods. Additional data can

be gathered from studies of other watersheds in the flatwoods of

Florida (Riekerk et al., 1979) and the Eastern Coastal Plain (Craig and

Kuenzler, 1983).

Taylor Creek--Nubbin Slough

Initial research interests in the Upper Taylor Creek watershed

were to determine rainfall-runoff-evapotranspiration relationships for

natural watersheds in Central Florida (Knisel et al., 1985). This was

broadened to include analysis of the effects of channel improvements

and water control structures on storm runoff, water yield and ground

water in the 1960's. Beginning in 1972, water quality sampling was

begun at 15 sites. Additional flow gaging and water quality sampling

sites have been added in the intervening years. Three agencies have

cooperated in collecting data: the USDA-ARS, USGS, and the South

Florida Water Management District (SFWMD).

Most of the data collected in this basin are from the Upper Taylor

Creek watershed (also identified as watershed W-2), and from sub-

watersheds within this watershed, Northwest Taylor Creek (W-3),

Williamson Ditch (W-5) and Otter Creek (W-13) (See Figure 2-2). In

June 1973, Upper Taylor Creek was diverted into the Nubbin Slough canal

to be discharged into Lake Okeechobee through structure S-191. The

drainage area contributing to the flow at S-191 has since been referred

to collectively as the Taylor Creek--Nubbin Slough basin.

Comprehensive reports on the hydrology and hydrogeology of the

Upper Taylor Creek watershed (W-2 and W-3) have been published by the

USDA-ARS (Speir et al., 1969; Knisel et al., 1985). Additional reports

JJJJiJ.!iiJ NW Taylor Creek

... f|?;||.|. Otter Creek

^ill~illllJJJ~lllllj^Hli~l I Willilamson Ditch
4 ~jfiJJIiilljl JIi ilUr SJIi
:: ... ~ ~.. ........i~ii!!ii~

.. .. .... ........ .... .......... ... .. ..

....... ..... ..... WIIi IIa m o n ut h
............ ..........

ubbin Slough

Upper Taylor Creek .-

0 5 S-191
/ Lake
/ Okeechobee


Figure 2-2. Taylor Creek--Nubbin Slough basin and sub-watersheds.

to Creek

have focused on specific aspects of the data: hydrology (Allen et al.,

1976; Allen et al., 1982a), effects of channel modifications (Yates et

al., 1982), and water quality (Stewart et al., 1978; Allen et al.,


With the initiation of the TCHP and RCWP cost sharing programs,

the focus has turned to water quality, and additional sampling was

started in the lower portion of the basin. Water quality sampling and

analysis has been done by the SFWMD, with samples routinely collected

on a biweekly basis. Samples have been collected at all flow gaging

stations and at numerous other locations in the watershed in an attempt

to categorize the nutrient concentrations that result from different

land use practices. Ritter and Allen (1982) provide a detailed report

and analysis of the water quality data from Upper Taylor Creek. Data

collected since publication of that report (and including data from

Nubbin Slough and at S-191) are summarized in the RCWP Annual Progress

Reports (RCWP, 1982; 1983; 1984).

There are two problems that plague collection and analysis of flow

data from these watersheds. First, due to the very low slopes in some

channels accurate flow measurement is difficult. Particular uncer-

tainty exists in the streamflow measurement at the W-2 outlet, and flow

conditions may also affect the chemical concentrations in the channel

(Knisel et al., 1985). Thus, calculated nutrient loads from this

watershed have been noted as being "estimated" (RCWP, 1984). Knisel et

al. (1985) consider "observed" concentrations and calculated loads to

be in the right order of magnitude, but caution that uncertainty in the

"observed" values must be kept in mind.

A second problem in hydrologic analysis of these watersheds is

what appears to be the dynamic nature of the watershed boundaries.

This is due to the low slopes and the widespread use of ditches for

land drainage. The areas of all the gaged watersheds in the TCNS basin

(with few exceptions), have been changed during the course of the data

collection program as documented by Knisel et al. (1985) and in the

RCWP reports. While most of the major changes in ditching and drainage

patterns have surely been identified, the very fact that this occurs is

indicative of the uncertainty that exists in defining the watershed


Upland Detention/Retention Demonstration Project

Data have been collected from several small watersheds in the LKR

and TCNS basins under the Upland Detention/Retention Demonstration

Project (UD/RDP) implemented by the SFWMD (Goldstein, 1986). The

objectives of this project were to obtain information on the water

quality problems associated with typical agricultural practices in

these basins, and to evaluate the potential of using wetland areas for

pollution abatement. Data from this project have been valuable in

helping to characterize hydrologic processes on the field scale, and in

identifying the water quality response of individual land uses.

Some characteristics of these watersheds are listed in Table 2-1,

with their locations shown in Figure 2-3. This data collection program

was maintained for approximately three years (Sept. 1979 through 1982)

with measurement of precipitation, runoff, water table elevation, and

biweekly samples taken for water quality analysis. Unfortunately,

1980-1981 was one of the driest periods on record. Not only were there

Table 2-1. Characteristics of the Upland Detention/Retention
Demonstration Project watersheds (Capece, 1984).

Parameter Armstrong Peavine SEZ Ash W. Ash E.
Slough Pasture Dairy Pasture Pasture

Drainage area (ha) 1457 314 291 65 8
Channel slope (%) 0.03 0.02 0.02 0.10 0.02
Overland slope (%) 2 0.15 0.17 0.08 0.12 0.10
Drainage dens. (km/km2) 0.66 1.00 4.26 23.6 12.8
Ponds and marsh (%) 13 21 7 0 0
Length/width ratio 3.5 2.3 4.8 1.1 0.3

fewer data, but how well any of the data represent "typical" conditions

must be questioned.

Analysis of the runoff data from several of these sites has raised

questions as to their integrity. In some cases runoff exceeds rainfall

even over a period of several months. In addition to the typical prob-

lems of defining the watershed boundary in this flat region, there were

problems with maintenance of the sites (Capece, 1984). The estimated

range of expected errors in flow measurement were 5 to 15 percent, with

the flumes designed for optimum measurement of low to normal flow

events (Goldstein, 1986). Thus, as runoff volumes increase, measure-

ment errors will also increase considering the backwater conditions

that tend to occur. This may partially explain the unreasonably high

runoffs "measured" in 1982.

Data from the three smaller sites were used in the development and

testing of the field scale model and will be described briefly here.

Additional description of the sites can be found in the report by

Goldstein (1986) and from the hydrologic analysis of Capece (1984).

Ash Slough (Bass West and Bass East). These two pastures on

Myakka fine sand are on either side of the Ash Slough wetland (Figure

2-4). Both pastures are used for grazing beef cattle. The East pas-

ture drainage boundary is poorly defined, with the contributing area

estimated to be between 8 and 20 ha. The West pasture has a deep pe-

rimeter ditch with an internal network of shallow (2 ft) ditches.

Cattle movements resulted in numerous breaches in the low levee around

the pasture and no estimate of possible flow errors could be made. In

addition, there was no adjustment for possible subsurface flow into the

perimeter ditch from "outside" the watershed.

SEZ dairy. The SEZ dairy watershed is a 291 ha area surrounded by

a deep ditch (Figure 2-5). The soils are predominantly fine sands

(Immokalee, Myakka), with approximately 7% of the watershed being wet-

lands. As a dairy, part of the area (56 ha) is holding/staging pasture

for the dairy herd, with 217 ha used for heifer and dry cow pasture and

hayland. The remaining 18 ha is the farmstead area.

Lower Kissimmee River Basin

There have been several studies in the Lower Kissimmee River basin

that have related primarily to water quality. Federico (1982) charac-

terized the water quality of the Lower Kissimmee River canal (C-38) and

its tributaries with monthly sampling over a five year period (1973-78).

Federico et al. (1978) report on environmental studies in the Chandler

Slough marsh. A study in the Boney Marsh was designed to provide data

on the potential use of wetland areas for removing nutrients from

runoff (Davis, 1981). Pertinent information from these studies will be

included in following sections.








5 10 1
1"C 10.8 MILES

Figure 2-3.

5 20


Location of the Upland Detention/Retention
Demonstration Project watersheds.




--------= O

_-.--__- 2

< I i

<. ..I > -
.. -,

I'..... .... ... .. ...--
c- z

| \ i0 I C- )

S__L __ .__ .1

---- -, -
I ~-.- -. m

,4 crI y/Lh^'*"*>, i LUf n

4U IJ -- >L

CD < cr< E

LJ < 0 0- L 4)
o r a coa

< c co o
10 -Ss^ urr~j Ln
SS .1 <4B
Ira 0

Phosphorus Transformations in Sandy Soils

Phosphorus Retention in Soil

It has long been recognized that when soluble phosphate is added

to the soil it rapidly disappears from the soil solution. In his

review article, Wild (1949) noted that this phenomenon had been demon-

strated 100 years earlier. Volumes of reports and related phosphorus

(P) research have followed in the past 130 years. It has been sug-

gested that the subject of P retention in soils has prompted more

research than any other aspect of soil-fertilizer-plant interactions

(Sample et al., 1980).

Along with the hundreds of studies demonstrating rapid P retention

by the soil, there have also been a few that did not observe this

phenomenon but found that P would actually pass through the soil

profile in significant amounts. Phosphorus leaching has been reported

in sandy soils in Florida (Neller, 1946; Blue, 1970; and others) and in

Australia (Hingston, 1959; Russell, 1960; Ozanne et al., 1961) and is

somewhat unusual in soil/phosphorus interactions. This response of

some sandy soils can lead to problems with water quality as a result of

excessive P in ground and surface waters, concerns that do not apply to

most soils.

Soluble phosphorus added to the soil as fertilizer is often ren-

dered unavailable even under the most ideal field conditions. In

general, the overall phosphorus problem for agriculture is 1) the

small total amount present in soils, 2) the unavailability of this

native P, and 3) the rapid sorption of added soluble P (Buckman and

Brady, 1969). Thus, the fixation of P in soils is most commonly viewed

as a problem because it decreases the efficiency of fertilization,

rapidly rendering the applied P unavailable to plants (Sanchez and

Uehara, 1980). This is, in general, directly opposite the concerns of

P management in the TCNS basin.

The availability of inorganic P is primarily controlled by: 1)

soil pH, 2) soluble Fe, Al, and Mn, 3) minerals containing Fe, Al, and

Mn, 4) available calcium and Ca containing minerals, 5) amount and

decomposition of organic matter, and 6) microorganism activity (Buckman

and Brady, 1969). The first four are related, largely through their

dependence on pH. Under low pH conditions, P is fixed by Fe and Al

compounds, and in high pH calcareous soils, Ca is the primary fixing

agent. In a more recent review, Sanchez and Uehara (1980) define

factors affecting the amounts of P fixed in terms of soil charac-

teristics: clay mineralogy, clay content, x-ray amorphous colloid

content, exchangeable Al, and soil organic matter. They state that

these factors are not strictly additive in their effects but interact

with each other.

Modeling Soil/Phosphorus Reactions

The phosphorus present in the soil has been categorized as being

in three states: phosphate in the soil solution, adsorbed phosphate

which is in some state of equilibrium with the solution, and nonlabile

P which is more firmly held (Larsen, 1967). A variety of reactions

have been proposed as those maintaining the state of equilibrium

between the soil P pools, with precipitation and adsorption receiving

the most attention. The chemistry of these soil phosphorus reactions

is described by Larsen (1967), Olsen and Khasawneh (1980) and others.

Barrow (1980) presents evidence that on the addition of P to a

soil, there is a two step reaction. The first step, the adsorption of

P, is rapid, with over 90% of the P commonly fixed within the first

hour of contact (Sanchez and Uehara, 1980). The second step is a slow

one in which some of the rapidly adsorbed P is converted to more firmly

held forms. These reactions may last for days or months and are

suggested to be the diffusion of P into inner surfaces of hydroxides or

a rearrangement of the crystal structure (Barrow, 1974). Sanchez and

Uehara (1980) also note that precipitates formed with Al and Fe, though

highly insoluble, will gradually release phosphate ions into the soil

solution over several years.

Hemwall (1957), after summarizing precipitation and adsorption

studies concluded that regardless of the type of the reaction, the

compounds formed and the mechanism of reaction seem to be essentially

the same. Now there is widespread acceptance of the concept of a

continuum of reaction mechanisms in a media as diverse as soil. There

is little concern for distinguishing between precipitation and adsorp-

tion reactions, and both phenomena are generally considered together as

sorptionn" (Berkheiser et al., 1980).

The phosphorus isotherm defines the equilibrium relationship

between the P content of the soil and the concentration in the soil

solution. Phosphorus sorption in soils has most frequently been char-

acterized using the Freundlich or the Langmuir adsorption isotherms.

However, many additional equations have been proposed (Barrow, 1978;

Berkheiser et al., 1980), some being modifications of the Langmuir or

Freundlich equations. There has been a great deal of controversy in

discussing the particular advantages of these two models and in

arguing for one as being 'best'. However, theoretical distinctions

between the two have been narrowed, and Kurtz (1981) concludes that it

makes little difference which equation is used as both have empirically

fitted parameters.

The adsorption/desorption process is not completely reversible,

with desorption occurring at a slower rate than adsorption (Selim et

al., 1975; Enfield and Ellis, 1983). The Langmuir and Freundlich

isotherms assume complete reversibility thus do not adequately describe

the adsorption/desorption process. The rate of desorption of P has

been found to decrease proportional to the cube root of elapsed time

(Barrow, 1980). Research on desorption is largely from the viewpoint

of plant availability of residual P and is and not concerned with the

source of the P. Thus, it is not clear if this release of P to the

soil solution comes from adsorbed P or from more tightly held P. No

distinct discontinuity has been observed in the rate of release; thus

no clear differentiation on the source of the P can be made (Barrow,


With the growing need for predicting nonpoint sources of phos-

phorus, algorithms describing P desorption are being developed (Chien

and Clayton, 1980; Sharpley et al., 1981; Sharpley et al., 1985).

Sharpley (1983) evaluated several models, relating physical and

chemical properties of 60 soils to the models parameters. Sharpley

concludes that these equations may have practical application in

predicting soluble P release in modeling water quality, particularly

the model by Sharpley et al. (1981) whose parameters were found to be

proportional to a soil's percent clay/organic C content ratio, thus

readily determinable.

Phosphorus Relationships in Sandy Soils

The factors conducive to leaching are essentially the absence of

the soil constituents that are responsible for P fixation. Thus

leaching of P has been observed primarily in sandy soils with low clay

content (typically less than 5%) and is primarily correlated to low

content of extractable Al and Fe (Ballard and Fiskell, 1974; Humphreys

and Pritchett, 1971; Yuan and Lucas, 1982; Chaiwanakupt and Robertson,

1976). Barrow (1980) also attributes the low water retention capacity

of sands as a contributory factor to leaching of P, presumably because

water may move through the profile faster, and there is less reaction

time for P in solution.

Despite the ready leaching of P from flatwoods soils as described

above, it has also been shown that P can accumulate in the surface soil

in significant amounts provided a higher pH is maintained with regular

liming. Blue (1970) found that with an average soil pH of 6.0, 70% of

the P added to a Leon fine sand over a period of 18 years was retained

in the surface soil of the white clover-perennial grass pasture. In

similar pastures, phosphorus contents of 156 and 327 ppm have been

found (Rodulfo and Blue, 1970). Spencer (1957), working with Lakeland

fine sand planted to citrus, found that over a 15 year period where the

pH had been maintained at 5.6 or above, 26.4% of the added P had

accumulated in the surface 6 inches, but where the pH had been allowed

to drop to 4.0, no P was retained. This build-up of P has also been

observed over a five year period in the top three inches of an

Immokalee fine sand with pH greater than 5.5 (Neller et al., 1951).

Even though the relationship between liming and P retention in

sandy soils has been observed repeatedly, the mechanism involved in

this P retention has not been clearly identified. Under the conditions

of high Ca content, pH approaching neutrality (6.0) and low Fe and Al

in the profile, it is expected that the P would be held as Ca compounds

(Blue, 1970). However, this has not been observed. Yuan et al. (1960)

analyzed for water-soluble, aluminum, iron and calcium phosphates

fractions in three Florida soils. In a Leon fine sand (pH 5.4) to

which 100 ppm P was added, 90% of the P was found associated with Al

and 7% with Fe, with the water-soluble and Ca forms negligible.

Thus, it is uncertain how much P may accumulate in a soil under

these conditions. In a normal P fixing soil, it has been observed that

a decrease in P sorption was related to previous P applications (Smyth

and Sanchez, 1980), which is consistent with the concept of adsorption

sites becoming saturated. For sandy soils with lime added, such

assumptions are less sure because of the uncertainty of the mechanism

of P retention. The soil P content may increase with continued liming,

but this P is poorly buffered and may be released readily.

Release of P held in limed sandy soils is most often associated

with a decrease in the pH when liming is halted (Neller et al., 1951;

Spencer, 1957; Forbes et al., 1974). Rodulfo and Blue (1970) have also

shown that a Leon fine sand with high P content (327 ppm) had suffi-

cient P available to plants so that no supplemental fertilizer P was

needed. The logical conclusion is that when the soil-P pools are in a

relatively stable state, changing one aspect of the balance (such as

decreasing the pH) will affect the others.

Nutrient Uptake in Wetland Systems

Among the management practices suggested for reducing nonpoint

source pollution loading to receiving waters is the possibility of

using wetlands as a water purifier. Using wetlands as nutrient sinks

has been proposed as a BMP for South Florida range and dairy land

(Allen et al., 1982b). Wetlands have been used for municipal waste-

water polishing for several years (Boyt et al., 1977; Tilton and

Kadlec, 1979). However, this use of wetlands has not gone without

debate. The use of wetlands for filtering of agricultural runoff must

be re-evaluated in light of recent reports. Can wetlands effectively

filter nutrients (in particular, phosphorus) from runoff? What type of

wetland systems might be appropriate for such use, and what uptake

rates should be considered?

To understand nutrient movement and storage in wetlands, Valiela

and Teal (1978) suggest study on two scales: first, the whole eco-

system, looking at inputs, outputs, and internal processes, and second,

analyzing in depth the component processes of the system. Component

processes that have been studied are nitrogen and phosphorus transfor-

mations, interactions with sediment, and nutrient uptake and release by

macrophytes. System studies have been done with natural wetlands and

with artificial wetland reservoirs.

Component Studies

Nitrogen transformations in the water column and sediment

Soluble nitrogen in floodwater occurring as ammonia and/or nitrate

can undergo several transformations. Ammonia in floodwater can undergo

volatilization, assimilation by plants, nitrification, and can be

adsorbed and retained in the soil (Chen and Patrick, 1981; Reddy and

Graetz, 1981). In an aerobic water column, ammonia is readily oxidized

to nitrate completing one side of the nitrification/denitrification

process (Patrick and Reddy, 1976). In a water column with high pH (8.5

up), ammonia volatilization can take place and nitrification is slowed

(Reddy, 1983).

Nitrate transformations include immobilization by incorporation

into organic forms, plant assimilation, reduction of nitrate to ammon-

ia, and denitrification. Denitrification has been shown to be the

primary pathway of nitrate removal from soil samples incubated with

floodwater (Bartlett et al., 1979; Reddy et al., 1980; Reddy and

Graetz, 1981; Sompongse, 1982). This is true even at lower tempera-

tures where there is an increase in nitrate reduction to ammonia

(Bartlett et al., 1979; Reddy et al., 1980), and immobilization to

organic N (Reddy et al., 1980). Denitrification is effective for both

mineral and organic soils (Krottje, 1980; Reddy and Graetz, 1981), and

in laboratory studies has accounted for over 95% of the nitrate lost

(Reddy et al., 1980; Sompongse, 1982).

The rate of denitrification is a function of floodwater depth,

nitrate concentration, presence of organic carbon as an energy source,

pH, and temperature (Reddy et al., 1980; Krottje, 1980). Since deni-

trification of wastewater occurs when nitrate moves down into the

anaerobic soil profile, the rate of diffusion of nitrate into the soil

is often the limiting factor (Engler et al., 1976; Reddy and Graetz,

1981). Adequate residence time of the wastewater is crucial to nitrate

removal, especially if ammonia is also present in the wastewater.

Rapid nitrification of the ammonia to nitrate may initially increase

the concentrations of nitrate over several days (Reddy and Graetz,

1981). Reddy and Graetz (1981) found a 12 to 24 day residence time

necessary for effective nitrate removal using agricultural drainage

water incubated on undisturbed cores from wetland soils. However,

actual nitrate removal rates and the required residence time will vary

greatly, depending in part on the factors mentioned above.

Phosphorus transformations in the water column and sediment

Various processes contribute to the equilibrium concentration of

soluble phosphorus. Phosphorus is removed from wetland water through

plant uptake and by adsorption and precipitation. Desorption can

release sediment bound P into solution. There is no process that

removes P from the system comparable to the nitrogen processes of

ammonia volatilization and denitrification, a fact having important

consequences in considering the ultimate fate of P added to a wetland.

There are many factors that control adsorption-desorption (and

precipitation). Three that recur in the literature are redox poten-

tial, pH, and availability of reactive compounds of calcium, aluminum

and iron. Redox potential affects the release of P to the water. The

general characterization is that P is released into solution under

anaerobic conditions (low redox potential) (Ponnamperuma, 1972;

Berkheiser et al., 1980; Sloey et al., 1978). In reality, the inter-

action of other factors is very important. Reddy and Graetz (1981)

found that an organic soil with an aerobic water column increased

soluble P concentration, while the anaerobic reservoir sorbed

phosphorus. Sompongse (1982) found that P was removed from nutrient

enriched water by sorption under both aerobic and anaerobic conditions.

The effect of pH on phosphate adsorption appears to show no uni-

fied trend. However, it is recognized as an important factor: directly

through effects on chemical reactions, and indirectly through effects

on plant decomposition (Berkheiser et al., 1980; Sloey et al., 1978).

A third dominating factor in P sorption is the presence of Ca, Fe,

and Al in the water and sediment. The availability of Ca compounds is

strongly related to precipitation of P (Sloey et al., 1978; Reddy and

Graetz, 1981), and Al and Fe oxides in the sediment are the primary

sorption agents for P (Berkheiser et al., 1980; Sompongse, 1982).

Richardson (1985) found that the phosphorus adsorption potential in a

variety of wetland soils could be predicted solely by the extractable

Al content of the soil. The generally low Al and Fe content of organic

soils means they have a limited capacity for removing P from the over-

lying water. Thus, while there may be good uptake initially, removal

rates decline quickly as the system becomes saturated, and P export may

eventually occur (Kadlec and Hammer, 1982; Richardson, 1985).

The variability in wetland response is demonstrated in the uptake

of P in wetlands with organic soils. Several studies have found

organic soils to be a source of soluble P under anaerobic conditions

(Peverly, 1982; Reddy and Graetz, 1981; Reddy et al., 1982b). Reddy

and Graetz (1981) characterized flooded organic soils as having poor

adsorptive capacity. They found equilibrium phosphorus concentration

(EPC)--the concentration at which P is neither adsorbed or desorbed--to

be 2.25 mg/L for a flooded organic soil, and 0.05 mg/L for a calcareous

marl soil used to line a reservoir. Other researchers have noted good

potential for adsorption in organic soils (Tilton and Kadlec, 1979;

Sloey et al., 1978). It appears difficult to characterize in simple

terms the sorption of P in a wetland system. The exact physico-

chemical nature of the water and sediment control the response, and

changes in one or more factors could dramatically affect that response.

Nutrient uptake/release by aquatic macrophytes

Following the pioneering work of Seidel (1976) in Germany in the

mid-1950s, there has been growing interest in the ability of aquatic

macrophytes to purify wastewater. Some of the species that have been

studied are waterhyacinth, bulrush, cattail, reeds, rush, bamboo,

pennywort and elodea (Lakshman, 1979; Wolverton et al., 1983; Reddy,

1983; Ogwada, 1983). Macrophytes affect the nutrient dynamics of the

wetland system through nutrient assimilation, nutrient release by

decomposition, altering the physico-chemical environment of the water

column and through sediment/water nutrient cycling.

Macrophytes can alter the physico-chemical environment of the

water affecting nutrient transformations. Denitrification, ammonia

volatilization and precipitation of ortho-phosphate can be increased

significantly as a result of plant presence creating anaerobic

conditions (Reddy, 1983).

Prentki et al., (1978) studied nutrient movement in a Wisconsin

marsh that did not have external nutrient inputs. Macrophytes rooted

in the sediment withdrew nutrients during the growing season, returning

those nutrients to the water and sediment surface upon decomposition.

In this habitat, nutrient cycling by aquatic plants was found to be a

significant source of nutrients in the water.

However, most of the interest in studying macrophytes is for

nutrient uptake. Nutrient removal by macrophytes depends on the

climate, type of plant, and nutrient availability. The ability of

macrophytes to remove nutrients from municipal and agricultural waste-

water has been well documented in micro-environments, reservoirs, and

natural wetlands (Boyd, 1970; Spangler et al., 1976; Lakshman, 1979;

Reddy, 1983; DeBusk et al., 1983).

The waterhyacinth has attracted most attention because of its

prodigious growth rate and accompanying large uptake of nutrients. The

waterhyacinth takes up nutrients in proportion to the concentrations in

the water. Tissue concentrations of 37 g N/kg and 9.4 g P/kg have been

observed for plants grown in sewage effluent. Average tissue concen-

trations for plants taken from 19 eutrophic water bodies in Florida had

1.61 + 5.0 g N/kg and 3.1 + 1.8 g P/kg (Reddy and Sutton, 1984).

Nutrient uptake by waterhyacinth under ideal conditions in a

nutrient rich solution has been documented up to 5350 kg N/ha/yr and

1260 kg P/ha/yr (Reddy and Tucker, 1983). Removal rates for natural

stands have been estimated as 480 kg N/ha/yr and 93 kg P/ha/yr (Reddy

and Sutton, 1984). In organic soil drainage water at Zellwood,

Florida, 730 kg N/ha/yr and 159 kg P/ha/yr were removed (Reddy et al.,

1982a). DeBusk et al. (1983) measured waterhyacinth nutrient uptake

from secondary sewage effluent (9.96 mg/l total N, and 4.68 mg/l total

P) between May and August. This period corresponds to 50% of the

annual biomass yield in Florida (Reddy and Sutton, 1984) giving

projected annual uptakes of 648 kg N/ha/yr and 207 kg P/ha/yr.

Plant assimilation of nutrients only occurs during the growing

season, and plant decay releases many of these nutrients back to the

water. Decay rates and nutrient fate have been studied under a variety

of conditions for different plants, and frequently, a rapid release of

a major fraction of N and P has been observed (Simpson et al., 1978;

Hill, 1979; Ogwada, 1983). Regeneration of N and P varies greatly

between different species and is dependent on temperature, pH and

water oxygen concentration (Jewell, 1971; Davis and van der Valk, 1978;

Reddy and Sacco, 1981; Sompongse, 1982). Depending on these factors

(but primarily on the species), a small fraction of plant material may

resist decomposition and retain N and P (Jewell, 1971; DeBusk et al.,


Rates of detritus accumulation have not been well documented.

Estimates of annual accumulation range from 3.5 70 kg/ha for N, and

0.05 2.4 kg/ha P (Richardson, 1985). After 17 weeks of decomposition

of waterhyacinth, DeBusk et al. (1983) observed a fairly constant rate

of detritus accumulation. Assuming a six month growing/senescence

season, those rates would project to accumulation of 25 kg/ha N and 1.8

kg/ha P per year. These values are clearly very low in terms of a

annual budget and are only slightly higher than the average loadings

from rainfall (19.5 kg/ha N and 1.2 kg/ha P) in the LKR and TCNS basins

(Goldstein, 1986).

Many researchers have concluded that plant nutrient uptake is pri-

marily a temporary storage factor in the nutrient dynamics of wetlands

(Prentki et al., 1978; Simpson et al., 1983; Hill, 1979). However,

this time lag can be very beneficial in reducing summer algal blooms in

lakes receiving wetland water (Lee et al., 1975). Harvesting is the

best approach for permanent removal of plant nutrients and is important

for maintaining optimum growth and nutrient uptake in artificial

wetland systems (Boyd, 1970; DeBusk et al., 1983; Ogwada, 1983).

In considering the role of nutrient removal by macrophytes, only

two conditions result in permanent removal: physical removal of plant

material through harvesting and detritus accumulation. In considering

the annual nutrient budget, the only process with a significant impact

is harvesting.

System Studies

Natural wetlands

The fate of nutrients in natural wetlands on an ecosystem basis,

looking at inputs, outputs and sometimes internal processes, has been

the subject of numerous reports in the past ten years. Investigators

have examined the use of wetlands as a final treatment of municipal

wastewater (Fetter et al., 1978; Boyt et al., 1977; Sloey et al.,

1978; Tilton and Kadlec, 1979). Agricultural nonpoint source runoff

flowing through wetlands has been studied in Minnesota (Hill, 1979),

New York (Peverly, 1982), Delaware (Simpson et al., 1983), Louisiana

(Conner and Day, 1982), and Florida (Kangas, 1977; Davis, 1982;

Goldstein, 1986).

High nutrient removal efficiencies characterize the majority of

municipal wastewater applications. Tilton and Kadlec (1979) tabulate

nine such studies with most showing nutrient reductions of above 90%,

and only one case each of reductions of ortho-P and ammonia being below

10%. Sloey et al. (1978) present a similar summary but also include a

few cases showing less promising results. In general, results of

municipal wastewater "treatment" using wetlands has been positive as

indicated in the literature.

Studies on the fate of nutrients from agricultural runoff in

wetlands have been more variable in outcome. Peverly (1982) studied a

3000 ha muckland in New York over two years. The upper region is

drained and cultivated, and the drainage water and runoff drains

through the lower region which is maintained as a natural wetland. The

wetland acted as an N and P sink only during the second year, when flow

was about half that of the first year. Otherwise, N, P, K, C, and Ca

were exported from the wetland.

Conner and Day (1982) found that a Louisiana swamp was effective

in the removal of nitrate-N from agricultural runoff, but that on the

average, ortho-P, organic-N and organic-P were added to the water.

During the winter the oxygen concentration in the water increased, and

the increase in redox potential led to phosphorus removal by the


There have been several investigations in the Kissimmee River

basin in southern Florida. Davis (1982) reports that the 142 ha Boney

Marsh functioned as a year-round phosphorus sink. Phosphorus storage

in both living and dead plant material continued to increase over the

three year study. An average P retention of 4200 kg/ha/yr was calcu-

lated, a figure Davis (1982) says compares closely with the uptake of

Chandler Slough as calculated by Federico et al., (1978). Since the P

input to Chandler Slough was six times the Boney Marsh input, Davis

(1982) suggests this uptake rate may be the maximum assimilative

capacity for this region.

Davis (1982) noted that nutrient uptake took place over a short

distance after which the concentrations were reduced to background

levels. This nutrient uptake front has moved downstream, with a perma-

nent change in plant community noted behind the front (S.M. Davis per-

sonal communication, 1985). A similar progression of a nutrient removal

front has been documented by Kadlec and Hammer (1982). The implication

is that any significant long term storage must be reflected by a change

in the flora to a community which contains higher nutrient content per

unit area. The desirability of such a change in the plant community is

a factor that must be considered in any management decision.

Goldstein (1986) calculated nutrient budgets for two wetland sys-

tems, Ash Slough and Armstrong Slough, and concluded that these systems

appeared to be in equilibrium with nutrient loading from the surround-

ing pastures. Reduction in exported nutrient load was attributed to

the reduction in outflow volume over inflow, and one year with high

outflows had a net export of total N and total P.

Allen et al. (1982b) report on a 12 ha wetland used as a second

stage impoundment for dairy barn wash water which first enters an

anaerobic lagoon. Outflow from the second impoundment had high total P

(9.55 mg/1) and ortho-P (5.12 mg/1) concentrations. They concluded

that the receiving area was too small and the high water and nutrient

inflow overtaxed the wetland.

For all natural wetland systems, the indications are that if

regular inputs are continued over a long term, a steady state response

will be reached. ("Steady state" is used here as being descriptive of

the average long term response. Variability due to meteorologic inputs

will still certainly be evident.) This apparent equilibrium has been

observed in the wetlands that have been receiving runoff from agri-

cultural areas over many years (Peverly, 1982; Conner and Day, 1982;

Goldstein, 1986). Wetlands where significant uptake has been observed

after initiating wastewater application have with time demonstrated an

apparent nutrient saturation where active uptake no longer occurs

(Kadlec and Hammer, 1982; Davis, 1982).

Artificial wetlands

Flooded reservoirs stocked with aquatic plants have been used for

waste treatment in Europe for many years (Seidel, 1976; DeJong, 1976),

and have been evaluated for purification of municipal (DeBusk et al.,

1983) and agricultural water (Reddy et al., 1982a,b). These artificial

wetlands utilize the processes of natural wetlands without endangering

or disturbing those resources, and in addition, enable better control

and management of the system.

Research has focused on quantifying the response of the system,

identifying the important nutrient pathways, and determining optimum

management techniques. Again, results of studies reflect findings for

a particular set of conditions, and caution must be used in general

application of the results. Many of the studies looking at nutrient

uptake by macrophytes as discussed previously, were done using arti-

ficial wetlands, and are directly applicable to this topic.

Permanent removal via plant uptake in these systems is dependent

on frequent harvesting of the plants. This removes nutrients and also

enables optimum growth to continue. Plant uptake has been found to be

the most significant pathway of P removal (DeBusk et al., 1983; Reddy

et al., 1982a), although Reddy (1983) demonstrated that under the right

physico-chemical conditions, P could also be removed by sorption to the

sediment. Denitrification can equal or exceed plant uptake in nitrogen

removal (Reddy, 1983; DeBusk et al., 1983).

Artificial wetland reservoirs have been found to effectively

remove N and P from wastewater. Regular harvesting is necessary, and

proper management is crucial. The system must be maintained at desired

operating conditions to be effective (Reddy, 1983).


Two methods of presenting experimental results have been encoun-

tered which need some comment in that they tend to obscure the "bottom

line." A nutrient budget, with total average annual input and output

loadings, is the only way to determine if a wetland is acting as a sink

or a source (Richardson et al., 1978). Data presented as concentra-

tions without flow information say nothing about net nutrient flux, as

lower output concentration may simply be a result of dilution.

A second problem arises in presenting only the percentage of

reduction. This is less than adequate since it says nothing about out-

fall concentrations and total load exported, which are more important

criteria (Richardson et al., 1978). The results are doubly obscure

when only the percent reduction in concentration is reported--an all

too frequent occurrence.

In contrast to the enthusiasm shown by some early proponents, more

caution is now being expressed regarding the use of natural wetlands as

wastewater filters. Caution is due for the following reasons.

1. Wetland ecosystems are vastly different depending on type of

wetland, climate, soil, plant population, hydrologic regime.

Tilton and Kadlec (1979) appropriately warn that their results

apply only to the specific wetland being studied. Each site

must be considered separately with careful testing and eval-

uation (Whigham, 1982; Stearns, 1978; Sloey et al., 1978).

Loading rate is a key. A small load in a very large area may

be a fraction of the total natural nutrient assimilation and

ultimately be negligible even on a long term basis.

2. Most research has been short-term studies, and long-range

effects are largely unknown (Stearns, 1978; Sloey et al.,

1978). To determine the extended impact on all flora and fauna

will require very detailed monitoring, and some current

information suggests that the risk of damage may not be worth

the gain (Stearns, 1978).

In a system view of a natural wetland, inflowing nutrients can be

removed from the system through volatilization and plant removal, can

be stored in the sediment, organic matter, and plants, or can be

exported from the system. Wetlands can unequivocally function as a

nitrogen sink as a result of denitrification. The fate of phosphorus is

less certain. If it is stored, for what length of time can the wetland

continue to assimilate phosphorus? Because of the dynamics of

adsorption/desorption and the natural fluctuations in the hydrologic

regime of a wetland, phosphorus export will most likely occur, whether

annually, intermittently or continuously, and will then have a greater

storage supply from which to draw.

Thus it is difficult to try to characterize a given wetland as a

source or a sink. Because of the higher natural nutrient levels,

wetlands often have higher background nutrient output. Nutrient output

is very sensitive to the hydrology of the system, to draining (Lee et

al., 1975; Klopatek, 1978) and variations in rainfall. A compounding

problem is that the agricultural runoff of greatest concern is the

large storms, for which the treatment capacity of a wetland is greatly

reduced because of the decreased residence time. Management of

artificial systems is possible, for control of inputs, maintenance of

desirable conditions, harvesting, and control of the hydrologic regime.

The use of artificial wetlands has the potential of capitalizing on the

benefits of marshes without the prospect of damaging natural wetlands.

Mathematical Models for Hydrology and Water Quality

The Curve Number Method for Predicting Runoff

The SCS runoff equation was developed to provide a simple means of

estimating runoff, while being able to reflect differences due to soils,

landform, land use, and management practices. Due to its simplicity

(it is a single parameter model) and its in-house use by the USDA-SCS,

the method has become widely used for estimating runoff. The history

of the development of the equation has recently been presented

(Rallison, 1980).

The SCS runoff equation is written as

(P 0.2S)2
Q = ----------- [2-1]
(P + 0.8S)

where Q is runoff, P is rainfall, and S is a watershed storage param-

eter, with all having units of depth (ins). The equation was intended

to be used for individual storms but is often used with daily rainfall

amounts since those are readily available. The single parameter, S, is

usually determined by way of a "curve number," a transformation

introduced to linearize the value of the parameter. The curve number,

CN, is related to S by the relationship:

CN = ---- [2-2]

The SCS has published tables of values of curve numbers based on

different land use, management practices and other conditions.

While enjoying widespread use, some researchers have questioned the

reliability of the curve number method. It has been found that the

method does not respond to rainfall intensity (Hawkins, 1978) or storm

duration (Rallison, 1980), and that the relationship between initial

abstraction and S used in the derivation of the equation may be

dependent on location (Hjelmfelt, 1980) and storm characteristics

(Rallison, 1980).

Capece (1984) evaluated several methods of predicting storm runoff

volume to determine their applicability for use on flatwoods water-

sheds. Data for his analysis were from the Upland Detention/Retention

Demonstration Project (Goldstein, 1986). The runoff prediction methods

analyzed were all forms of the SCS runoff equation (equation 2-1). The

difference between the various methods lies in the approach used to

determine S.

The three most accurate methods related S to an estimate of the

actual physical storage available in the profile, using curves that

relate available profile storage as a function of the water table

depth. The "ARS" curve (Figure 2-6) was developed from Taylor Creek

watershed data (Speir et al., 1969). The South Florida Water

Management District (SFWMD, 1983) developed a curve for storage on

natural sites, along with an alternate curve for developed sites

(Figure 2-7). The ARS curve predicts less available storage than does

the SFWMD curve, with the differences becoming very significant for

water table depths greater than two feet. Capece (1984) found that



-) 2

Figure 2-6.

2 -

4 -


Sz *

0 i


Water storage characteristics of sandy soils from the
Upper Taylor Creek watershed (Speir et al., 1969).


Figure 2-7.

Available soil profile storage as a function of depth
to the water table for watersheds in South Florida
(SFWMD, 1983).

directly predicting available storage using the ARS curve gave the most

accurate runoff prediction.

Water Quality Models

Numerous water quality models have been developed in the past ten

years. Many of these have been designed for specific situations, but

there are several models which have been designed for general use, and

have been adequately documented, distributed and supported. There are

several compilations of hydrologic/water quality models which include

both specialized and general application models (Huber and Heaney,

1982; Renard et al., 1982).

There are several models which are applicable for use in eval-

uating the impact of various management practices on offsite water

quality. Models developed for this purpose have generally been

physically based, distributed parameter models. These models have

parameters that are based on physical characteristics of the system so

parameter values can be obtained by measurement rather than through

calibration (Beasley et al., 1982). Several of the currently available

models, and some just recently developed will be described briefly.

A field scale model for Chemicals, Runoff, and Erosion from

Agricultural Management Systems (CREAMS), developed by a group of USDA

scientists, is designed specifically for evaluating the relative

effects of different management practices (Knisel, 1980). Conceptually

it is a distributed parameter, physically based model, and requires no

calibration. Hydrology, sediment, nitrogen and phosphorus transfor-

mations and pesticides are modeled continuously for up to twenty years

using a daily time step. A "field" is defined as an area of uniform

hydrologic and chemical characteristics under a single land use.

CREAMS has been widely distributed and used (Bengtson and Carter, 1983;

Foster and Ferreira, 1981; DelVecchio and Knisel, 1982; Crowder et al.,


Several models have been developed which rely heavily on the

CREAMS algorithms. The Simulator for Water Resources in Rural Basins

(SWRRB) is a basin scale model which builds on the daily rainfall

hydrology component of CREAMS (Williams et al., 1985). The model adds:

a return flow component, spatial linking of watersheds within a basin,

reservoir storage, simple flood routing, and a weather simulation

model. In addition to water, SWRRB simulates sediment yield, but no

nutrients are considered.

The Erosion-Productivity Impact Calculator (EPIC) simulates

erosion and its effects on the long term productivity of soils

(Williams et al., 1983, 1984). This is a field scale model, based on

the CREAMS hydrology component. It has a much more complex nutrient

model than does CREAMS. Where CREAMS has a very simple phosphorus

model, EPIC contains a comprehensive soil phosphorus cycle, with

phosphorus pools and reactions calculated separately in up to ten soil

layers (Jones et al., 1984a,b; Sharpley et al., 1984). However, this

model has not been released for public use.

The AGNPS model (Young et al., 1985) uses CREAMS runoff, sediment

and nutrient algorithms, with some simplifications. There are two

major changes in model design: AGNPS is an event model, and it

simulates yield from multiple source areas. A grid is used with each

cell assumed to have uniform conditions. Flow and sediment are routed

in overland and channel flow to the watershed outlet for a storm event.

The ANSWERS model (Areal, Nonpoint Source Watershed Environmental

Response Simulation) predicts runoff, along with erosion, sediment

transport and deposition, from field or basin scale watersheds on a

storm event basis (Beasley and Huggins, 1982). The watershed is

overlayed with a square grid (typically 1-4 hectares per cell) with

physically-based parameter values specified for each cell. The ability

to define this level of spatial variability in the watershed results in

a more accurate representation of the watershed and enables evaluation

of changes in land use and management practices at the cell level. In-

cluded is the ability to simulate various structural practices such as

diversions and field detention basins. More recent versions of the

model include simple nutrient yield predictions, and a version includ-

ing a detailed phosphorus model has been planned (Beasley et al., 1982).

Several water quality models have been developed around the hydro-

logic framework of the Stanford Watershed Model (SWM). The history of

the development of this "family" of models is detailed by Barnwell and

Johanson (1981). Since an adaptation of the SWM serves as the hydro-

logic base, these models all follow the lumped parameter concept and

must be calibrated for each site.

The first use of the SWM for water quality was the Pesticide

Transport and Runoff (PTR) model (Crawford and Donigian, 1973). This

was expanded into the Agricultural Runoff Management (ARM) model

(Donigian and Crawford, 1976a), which was further improved by Donigian

et al. (1977). ARM predicts runoff, sediment loss, nutrient uptake by

plants, and soil/water interactions of nutrients and pesticides. The

Non-point Source (NPS) model (Donigian and Crawford, 1976b) was devel-

oped as a simpler version of ARM and included algorithms to make it

applicable to urban modeling. Both ARM and NPS are continuous simu-

lation models; neither have channel routing routines, thus provide only

edge-of-field loadings.

The Hydrologic Simulation Program--Fortran (HSPF) (Johanson et

al., 1980) is a culmination of these earlier modeling efforts, combin-

ing components of ARM and NPS in a model framework that includes a data

manager (Time Series Management System), channel routing and instream

chemical processes. Improved instream processes and statistical anal-

ysis for risk assessment have been included in recent versions of the

model (Johanson et al., 1984; Donigian et al., 1983). Although it is a

lumped parameter model in design, HSPF can account for some watershed

variability through its ability to route flow from different sub-

watersheds delineated within the watershed of interest. HSPF has been

proposed for use in agricultural nonpoint source modeling (Woodruff et

al., 1982; Donigian et al., 1983), and BMP evaluation (Bicknell et al.,

1984), however, as a lumped parameter model, calibration is required,

and the availability of suitable data will often be a limiting factor.

Of the current water quality models, none adequately assess the

effectiveness of BMPs on both the field and watershed level. CREAMS is

specifically designed for evaluating alternative practices, but is

strictly a field scale model. HSPF predicts runoff and water quality

constituents at the basin outlet, but provides no information at inter-

ior points in the basin. Networking field sized areas together would

be possible, but this approach is neither practical nor is it consis-

tent with the model's design and intended application. ANSWERS and the

AGNPS model do provide the desired modeling framework, predicting both

field (cell) and basin scale results. A major limitation of both is


that they are event models, and for water quality modeling, long term

simulation is desirable (DelVecchio and Knisel, 1982; Frere et al.,

1980; DeCoursey, 1985).



Description of the CREAMS Model

Background. CREAMS was developed as a tool for evaluating the

relative effectiveness of alternative management practices in reducing

nonpoint pollution (Knisel, 1980). Objectives in the design of CREAMS

were that the model should 1) simulate major physical processes that

control water balance, erosion, sediment yield, and movement of plant

nutrients and pesticides, 2) use physically based parameters that can

reflect changes in management systems, 3) operate on a daily time step

for computational efficiency, and 4) be field scale, since this is the

common base for selection of alternative management practices

(DelVecchio and Knisel, 1982). Model documentation includes an

extensive users manual (Knisel, 1980) and there are several good

summaries of the model in the literature (Knisel et al., 1983; Knisel

and Foster, 1980).

CREAMS is composed of hydrology, erosion and chemistry sub-models.

The three sub-models are run independently (in sequence), with "pass"

files used to link the models providing the necessary data for the

next sub-model. The model was designed this way so that the sub-models

could be used separately if desired and so the model could be used on

smaller computers. The obvious disadvantage is that there is no feed-

back between the modules. This structure also made the model easier to

modify, as each module can be handled separately.

Hydrology algorithms. The hydrology sub-model uses rainfall,

temperature, average monthly radiation, soils and land use parameters

to predict runoff, evapotranspiration, percolation and soil water

content. Using daily rainfall records, a modified SCS curve number

method is used to predict runoff. A second hydrology option is

available which uses hourly or breakpoint rainfall with an infiltration

equation. The daily rainfall option was selected as the base for

CREAMS-WT because 1) breakpoint rainfall data are generally not

available, 2) an infiltration model does not conceptually represent the

rainfall/runoff process on flatwoods watersheds, and 3) the analysis by

Capece (1984) indicated that the curve number method has good potential

for predicting runoff from flatwoods watersheds.

The soil profile is divided into seven layers with accounting of

the soil moisture within each layer. Percolation is computed by moving

soil water down through the soil layers, to be lost out the bottom as

deep seepage. Evapotranspiration (ET) is calculated using an adapta-

tion of the Penman equation. Leaf-area-index, radiation and available

soil moisture determine actual ET from potential ET. A simple plant

root growth model is used to partition the ET extraction of soil water

among the different soil layers.

The daily rainfall option of CREAMS uses the SCS runoff equation

(equation 2-1) for predicting runoff. The storage parameter, S, is

calculated as,

S = Smx ----- [3-1]

where ULE is the upper limit of soil water storage in the profile, SM

is the soil water content, and Smx is the maximum value of S. This

maximum value is determined from the SCS equation relating curve number

and S as,

Smx = ---- 10 [3-2]

where CNI is the curve number for antecedent moisture condition I. The

curve number (condition II) is an input parameter and is converted by

the model to the equivalent CNI value. The relationship between CNII

and Smx is described in greater detail in Appendix A.

As calculated in equation 3-1, all layers of the profile are

equally important in determining the soil moisture, hence storage.

Runoff is normally expected to be more sensitive to the moisture

content in the surface layers of the soil. To better represent this,

CREAMS uses a weighted form of equation 3-1. The surface layer is

weighted heaviest with weighting factors decreasing with depth.

Limitations for flatwoods watersheds. The modified SCS Curve

Number method used by CREAMS for predicting runoff volumes was one of

the methods evaluated by Capece (1984). This algorithm was compared

with several other predictive methods and found to be one of the best

ones tested. However, good estimates of runoff volume were found to be

dependent on knowing the available storage in the soil. The CREAMS

model uses an internal moisture accounting procedure to determine S,

whereas Capece (1984) used the ARS storage curve.

For use in flatwoods watersheds, a significant problem with the

CREAMS model is its inability to account for a fluctuating water table

in the soil profile. There are two associated effects. First, the

high saturated conductivity of the sandy soil results in the model

predicting a significant amount of deep seepage, while in reality, deep

percolation in these basins is negligible. Second, available storage

in the soil is overestimated when the water table is in the profile,

resulting in underprediction of runoff. Thus, the effect on runoff

prediction is especially significant since runoff occurs only when the

water table is near the surface, and this is the condition where the

model's estimate of available storage is least accurate.

Bengtson and Carter (1983) applied CREAMS to a Mississippi water-

shed that is also very flat and has a seasonally high water table.

They found that CREAMS underestimated runoff during wet periods and

overestimated runoff during dry periods. A similar result is reported

by Pathak et al. (1982), who simulated native rangeland runoff in


Concepts of CREAMS-WT

For application to flat, high water table watersheds, CREAMS must

be modified to limit the amount of deep seepage that is predicted.

Without modification, the CREAMS estimate of the available storage in

the profile is not accurate. This error is then reflected in the

runoff volume predicted by the curve number method.

The CREAMS-WT model corrects these limitations. A high water

table is modeled with limited or no deep seepage and a modified method

is used to calculate the available storage. The result is more

accurate estimation of the actual storage available, thus better

prediction of runoff volumes.

The following objectives guided the development of the CREAMS-WT


1) The model should represent the physical processes occurring in

flatwoods watersheds: a high water table, and limited deep


2) The method of estimating available storage, a crucial

parameter in the curve number method, must be improved.

3) The general modeling objectives for CREAMS should be followed:

use physically-based parameters, and

be computationally efficient.

4) The current operation of CREAMS should not be disrupted. This

modification can then be incorporated as a third hydrology

option if desired. It will also ease incorporation of this

modification in future versions of CREAMS.

Description of Algorithms

Water table simulation

CREAMS considers only the defined root zone depth (input parameter

RD) in its soil water accounting. Additional routines were thus needed

in two areas. First, to follow the water table in the root zone, and

second, to follow the water table when it falls below the root zone

into the "lower zone".

Water content above field capacity is "routed" by CREAMS, layer by

layer, down through the root zone. For sandy soil, the high saturated

hydraulic conductivity can easily result in no effective detention of

downward water movement. Thus, on a daily time step and given adequate

rainfall, the water content of the profile is simply restored to field

capacity and any excess becomes deep seepage. The way this deep seep-

age is handled in CREAMS-WT depends on which zone the water table is in.

Water table in the root zone. CREAMS-WT retains all the CREAMS

routines for calculating the moisture balance in the root zone. Added

routines take the "deep seepage" predicted by the CREAMS, and beginning

at the bottom and working upwards, fill to saturation successive layers

in the root zone. An estimate of the position of the water table is

then made by interpolating to find the depth of saturation in the layer

just above the topmost layer that is fully saturated.

A new parameter, DSP, specifies a rate (in inches/day) of subsur-

face water movement out of the modeled watershed system. This may be

used to simulate a limited degree of deep percolation, or lateral

seepage. Deep percolation occurs at the rate specified by DSP while

the water table is in the root zone. When the water table falls below

the root zone, no deep percolation occurs. One possibility is to use

this component to model baseflow on larger watersheds. In this case,

the predicted "percolation" should be added to the volume of runoff


When the water content in the root zone is at or below field

capacity the water table is assumed to be at the boundary of the two

zones. On the following day (next time step), provided there is no

rainfall to raise the water content in the root zone above field

capacity, the water table is considered to be in the lower zone.

Water table in the lower zone. Water movement into and out of the

lower zone is tracked with the aid of a water table recession curve.

Speir et al. (1969) used well data from the Taylor Creek watershed to

develop a recession curve showing water table depth versus number of

rainless days. With an estimate of the drainable porosity of the lower

zone, a storage versus time relationship can be derived from the

recession curve.

The function used to represent the recession curve is,

D = Tk [3-3]

where D is depth of the water table from the surface, in feet, T is

time in days, and k is a coefficient. A value of k = 0.33 closely

represents the curve presented by Speir et al. (1969). The recession

curve begins at the surface, but the only portion of the curve used in

the model is that which lies below the root zone depth (RD).

Introducing a parameter for the drainable porosity, PORS (in/ft),

depth in the lower zone can be converted to storage, LZS (in inches),

LZS = PORS 0 [3-4]

which combined with equation [3-3] becomes,



This storage versus time relationship is used to track the water table

and provide a water balance in the lower zone. In the following

discussion, the subscripts 'i-l' and 'i' refer, respectively, to the

previous and current time step (day).

When the water table is identified as being in the lower zone,

rainfall first fills the root zone to field capacity before any excess

moves into the lower zone (variable SEP). For days without water re-

charging the lower zone, the time location on the curve is increased by

one day, with one exception. For days with rainfall greater than 0.1

inch, no change is made in the water table since ET is assumed to be

satisfied from the root zone. For days with seepage into the lower

zone, the position on the curve is moved up representative of that

volume of seepage. These relationships can be expressed in the

following way.

When SEP = 0 and precipitation < 0.1 inch,

Ti = Ti + 1 [3-6]

DPET = PORS ( Tk Ti I ) [3-7]

where DPET is the volume of water lost from the lower zone in that time

step. These withdrawals from the lower zone are summed and reported as

"Lower Zone ET".

When SEP > 0, the new storage is found as,



and the corresponding reference time on the recession curve is

Ti = ( LZSi/PORS )**(l/k) [3-9]

The total volume of seepage into the lower zone is reported as "Lower

Zone Recharge". For any period of time where the water table starts in

the root zone, moves to the lower zone, and ends up back in the root

zone, the Lower Zone Recharge will equal the Lower Zone ET.

When the volume of seepage (SEP) from the root zone is greater

than the available storage in the lower zone (between the current water

table and the bottom of the root zone), the excess is transferred back

to the root zone. The soil water is then modeled with the root zone

component in the following time step.

Runoff prediction

The CREAMS curve number method requires an estimate of soil

moisture (SM in equation 3-1) which is translated to an effective curve

number. CREAMS uses a weighting of the actual storage in the seven

layers of the root zone. The top layer is weighted most heavily and

the weighting decreases moving down the profile. This is an attempt to

more closely represent the effects of soil moisture on infiltration

but is not desirable for the storage based system being modeled. This

weighting was removed so that for CREAMS-WT the actual available

storage in the root zone, rather than a weighted value, is used in

predicting runoff.

Runoff is the first of the various components of the daily water

balance that is computed. The difference between precipitation and

runoff is then assumed to infiltrate. To ensure that this volume does

not exceed the available storage in the profile, the "strict storage"

prediction of runoff as precipitation minus available storage is cal-

culated, and the larger of these two runoff predictions is used. The

curve number prediction will generally be higher but comparing against

the strict storage prediction ensures that the available storage in the

profile will not be exceeded.

Calibration and Testing

Calibration. Few parameters were adjusted during the calibration

of the model. A minor change was made in the exponent of the water

table recession curve ('k' in equation 3-3, equivalent to model

parameter RXP). This was changed from 0.33 to a value of 0.36 which

better matched recession curves from Bass West well data. This also

matches closely with a value of 0.355 determined by Capece (1984).

The most important calibration was of the annual predicted ET. As

will be discussed later, the ET model component is very sensitive to

parameter changes. This is especially critical since ET is a large

component in the annual water balance. Leaf-area-index (LAI) is the

primary parameter affecting ET that can be reasonably varied. (For

instance, it would not be reasonable to vary temperature from recorded

values.) While LAI is a physically based parameter, it is difficult to

determine appropriate values. Data from Jones et al. (1984c) were used

to determine initial estimates, and these values were then adjusted so

the model output matched annual ET estimates better. Additional com-

ments on estimating LAI appear in the following sections on sensitivity

analysis and parameter estimation.

With additional calibration it would be possible to match the

observed watershed response more closely. While this would improve the

perception of the model's performance, additional calibration was not

done for the following reasons. An objective for the model is that it

be applicable for predicting the hydrologic and water quality response

of ungaged watersheds. Thus, the ability to achieve reasonable simula-

tions without extensive calibration demonstrates the model's capacity

for use in ungaged, uncalibrated application. In addition, uncertain-

ties in parameter values as well as in the observed watershed response

imply some expected variability between predicted and observed values.

Bass West simulation. Data from the Bass West site of the Upland

Detention/Retention Demonstration Project (Goldstein, 1986) were used

during model development, for initial evaluation of the model and for

calibration of some parameters. Comparison with runoff volumes was not

possible for this site, as precipitation and water table elevation were

the only data judged to be suitable for use in model evaluation.

Parameters for CREAMS-WT are physically based and were estimated

from various sources: Goldstein (1986), the CREAMS manual, Soil Survey

reports, Capece (1984) and Speir et al. (1969). The simulation of the

Bass West pasture was run from June 1979 to December 1982, the length

of the rainfall record, although water table observations did not begin

until 1980. This allowed a period for model equilibration which mini-

mized the effects of the initial parameter values chosen. The para-

meter file is listed in Appendix B.

Accurate runoff prediction is the goal of the model. However, the

importance of the soil moisture content in the runoff model would sug-

gest that the accuracy of the predicted soil water content is crucial


to the accurate representation of the hydrologic system and to runoff

prediction. The predicted water table was thus used as one measure of

model evaluation. Several comments on the observed water table data

are appropriate. First, recording the water table elevation is a

simple and reliable procedure, thus it provides good data for use in

model evaluation. However, one well in a 20 to 3600 acre watershed is

clearly limited in how well it represents the average storage over the

entire watershed. This must be considered in comparing water tables

since the CREAMS-WT predicted water table is a value representative of

the watershed and thus may at times reasonably differ from a water

table elevation measured at one particular location.

The simulated and observed water tables are compared in Figure

3-1. The general trends in the water table are followed fairly well.

The high frequency 'spikes' in the observed water table are not

followed by the model due to two factors: 1) the model operates on a

daily time step, and 2) the runoff volume predicted by the curve number

method is subtracted from the water balance the same day, whereas in

reality, the runoff hydrograph will extend over several days. One

consistent discrepancy is that the simulated water table is too high in

the winter months. This is due to inaccuracies in estimating ET,

resulting from the lack of specific site radiation, temperature, and

leaf area index data. The low water table predicted in 1983 may in

part be due to uncertainties in estimating ET parameters, and may also

indicate that values for the parameters PORSUB and FUL are too high.

However, the primary cause is attributed to rainfall events missing in

the recorded data.




____-* ..J CJ
IM 0
I*? - 17) C

_, ________--_ "-
-- ------ ---------------------

-- -------- -


-0 C -

I- E


F cc

5~C w
--- I-u LO C ai4-


Evaluation of the CREAMS-WT Runoff Equation

One aspect of model evaluation is to determine if the algorithms

respond in a way that is reasonable, or expected, of the system being

modeled. The following discussion will explore the characteristics of

the rainfall/runoff relationship used in CREAMS-WT. As mentioned in

the literature review section, the appropriateness of the curve number

method has come into question for a wide range of reasons. Since the

method was developed from data on soils and watersheds that are more

typical of the rest of the country, it is even more important to

address the issue of the theoretical and conceptual appropriateness of

the method for simulating flatwoods watersheds. (Note that Capece

[1984] addresses the pragmatic question of "does it work"?)

Many of the current hydrologic models are based on the concept of

limited infiltration producing runoff. The curve number method of

estimating runoff is based on a storage concept rather than on infil-

tration. It does not consider rainfall intensity and duration--crucial

factors in the generation of runoff in most of the country. However,

for modeling a physical system that is storage based, as are the

flatwoods watersheds, the curve number method is a good model.

The SCS runoff equation (equation 2-1) can be arranged in the


Q [ 1 0.2S/P ]2
= --------------- [3-10]
P [ 1 + 0.8S/P ]

By defining available storage in the profile as

Sav = ULE SM [3-11]

equation 3-1 can be written

S = Say Smx/ULE [3-12]

Substituting this in equation 3-10 gives the effective relationship which

is used to calculate runoff in CREAMS-WT,

Q [ 1 0.2 (Smx/ULE) (Sav/P) ]2
------------------------------- [3-13]
P [ 1 + 0.8 (Smx/ULE) (Sav/P) ]

For ease in graphical presentation, consider unit precipitation

which simplifies the expression to,

[ 1 0.2 (Smx/ULE) Sav ]2
Q --------------------------- [3-14]
[ 1 + 0.8 (Smx/ULE) Sav ]

A graph of Q versus Sav is shown in Figure 3-2 for several values of

Smx/ULE. The factor Smx/ULE is fixed for a given simulation and

represents the ratio of the maximum profile storage as determined by

the curve number to the maximum storage as specified by the input

parameter UL. When this ratio is one, equation 3-13 reduces to a

formula analogous to the SCS equation (3-11). Also shown in Figure 3-2

is the straight line represented by the strict storage relationship,

Q = P Sav [3-15]

which for unit precipitation becomes,

Q = 1 Sav [3-16]

While the flatwoods watersheds respond as a storage based system,

the strict storage relationship is less than ideal. Lower areas of a

watershed (such as sloughs) begin contributing to runoff before the

entire watershed is saturated. The curve number method (the Smx/ULE=1

curve in Figure 3-2) is more representative of this process. Runoff

begins before available storage is completely filled, and predicted

runoff is always greater than the strict storage relationship would

indicate. For the original curve number method (Smx/ULE = 1), runoff

is predicted out to the point where storage is five times the amount of


The CREAMS-WT runoff algorithm has an additional variable, the

Smx/ULE ratio. With a higher curve number (inversely related to Smx)

this ratio is less than unity and increased runoff is predicted for a

given value of Sav. When the ratio is greater than unity, decreased

runoff is predicted. When the Smx/ULE ratio is greater than one, the

curve crosses the straight line representing the strict storage

relationship. As discussed previously in the "Runoff Prediction"

section, the model forces the runoff prediction to satisfy the strict

storage relationship. Thus when the Smx/ULE ratio is greater than one,

the runoff prediction follows the curve (moving from right to left)

" 0.6--





Figure 3-2.


Runoff predicted by: the CREAMS-WT runoff equation
graphedd for several values of Smx/ULE), and the linear,
strict storage relationship of Q=P-Sav. These functions
are graphed considering unit precipitation (P=I).


until it intersects the straight line, then it follows the straight

line relationship to zero storage.

While the runoff prediction must conform, at minimum, to the

strict storage relationship, the length of the "tail" to the right,

which represents increased runoff, can be controlled through the

Smx/ULE ratio. This feature may be useful in representing morphologic

variability between watersheds as well as the effects of some manage-

ment practices. Thus, the implementation of the curve number method as

used in CREAMS-WT appears to be particularly appropriate for predicting

runoff from flat, high water table watersheds. In a simple manner, it

represents the variable source area concept and is descriptive of the

rainfall/runoff process on flatwoods watersheds. Further analysis is

needed to determine the relationship between the various parameters and

watershed characteristics. Unfortunately, analysis is limited at this

point by a lack of applicable data.


Two sources of data were used in verification of the hydrology

model. Small watershed data from the UD/RDP were used for specific

comparison of model predictions with observed data, and long term water

balance studies on the Taylor Creek watershed were used to test how

reasonable and representative the model predictions are over a long


Armstrong Slough

The Armstrong Slough site was simulated without calibration. Both

CREAMS-WT and the original CREAMS models were run using this same

parameter set (included in Appendix B). Parameters were taken largely

from the Bass West pasture simulation. A small amount of percolation

was modeled, in part as a demonstration of this aspect of the model.

Since there is little specific data related to this pathway, its

inclusion here is merely conjecture. In a larger watershed there may

be some losses due to deep seepage or lateral outflow, or this could be

used to represent baseflow.

The simulated and observed water tables are shown in Figure 3-3.

The inconsistencies in the water table prediction are due in part to

inaccuracies in simulating ET as mentioned previously. However, it

also appears that most of the discrepancy may be due to errors in the

rainfall data. In June and July of 1980, there are two significant

rises in the observed water table which are not echoed in the simulated

water table. This is typical of the type of discrepancy that would

occur when a rainfall event on the watershed does not appear in the

recorded rainfall data. Errors in the simulated water table are

perpetuated until the water table is "reset" by reaching the surface,

which in this case is more than a year later. It is important to

recognize these limitations in the observed data when making

comparisons with model results.

The components of the annual water balance as observed and simula-

ted by the two models are compared in Table 3-1. The CREAMS simulation

was included to demonstrate the significance of the simulated water

table. The excessive percolation predicted by the unmodified CREAMS

version clearly biases the other components of the water balance.

Specifically, runoff and ET are both underpredicted.

In evaluating the performance of CREAMS-WT, the following points

must be kept in mind. The error in the simulated water table due to

missing rainfall data resulted in a low water table through most of

1981. This would produce high amounts of "lower zone ET" thus

inflating the ET estimate, and would also under-predict runoff because

of the excess storage in the profile. Thus, the effects of this error

carry through to the water balance. Water extracted from profile

storage in 1980 inflates ET for that year. This storage is refilled in

1981, with both ET and runoff underpredicted because of the very low

water table. For 1982, the accuracy of the observed runoff data must

be questioned, as "observed" ET is unreasonably low for a year with

above average rainfall.

Considering the above comments, CREAMS-WT models the site very

well. Runoff and other components of the water balance are all pre-

dicted reasonably well. And again, this simulation was not calibrated.

SEZ Dairy

The SEZ Dairy watershed was simulated without calibration. The

water table and annual water balance are compared with observed data in

Figure 3-4, and Table 3-2. The accuracy of the observed 1982 runoff

must again be questioned based on the unreasonably low ET values. The

above comments on the accuracy of the water table simulation also apply


Long Term Simulation

Two to three years of data are minimal for good testing of a

model. Rainfall varies greatly from year to year, and a short period

of record can include only a limited sampling of the range of

Table 3-1.

Annual water balance for Armstrong Slough. Comparison of
observed data and predictions by CREAMS and CREAMS-WT.

------ 1980 ----- --- 1981 ---- -- 1982 ---
-----------------------------(ins)--------- --------------

PRECIP 37.80 37.80 37.80 36.50 36.48 36.48 55.00 54.96 54.96
RUNOFF 1.65 1.45 1.66 5.58 1.14 4.08 30.20 3.50 15.25
PERC 0.00 5.63 0.26 0.00 12.59 0.27 0.00 17.34 0.73
ET 35.30 30.85 40.00 32.70 22.84 27.60 24.80 34.06 40.47
STORAGE 0.86 -0.16 -1.73 -1.73 -0.12 2.09 0.00 0.00 -1.56

a Rainfall, runoff and storage from recorded data, percolation assumed
zero, and ET estimated as the difference.

Table 3-2.

Annual water balance for SEZ Dairy.
data and predictions by CREAMS-WT.

Comparison of observed

---- 1980 --- -- 1981 --- ---- 1982 ---

PRECIPITATION 36.20 35.88 33.20 33.12 51.00 51.00
RUNOFF 4.19 1.21 5.09 4.68 24.40 14.14
PERCOLATION 0.00 0.00 0.00 0.00 0.00 0.00
ET 31.70 34.48 29.00 29.72 26.60 36.38
STORAGE CHANGE 0.29 0.19 -0.86 -1.29 0.00 0.47

LOWER ZONE RECHARG 2.889 3.785 4.158
LOWER ZONE ET 2.889 3.785 4.158

a Rainfall, runoff and storage from recorded data, percolation assumed
zero, and ET estimated as the difference.


r-= co

0 <
1-- -4





-- ----


-4 0




0i~~- m O -) S





r L-"


C -

-* ^ -e
*" _^'- 0 w** C-0
--*''"C s

"c ^ x '^ -o

-- i .r ~r ^ i___i (

o csjn 3 ir)^ow
1I -o 11Q

conditions that may occur. It is desirable to test a model with the

full range of input conditions under which it may be used. The Upland

Detention/Retention Demonstration Project unfortunately occurred during

a period of record drought. While this may check the response of the

model to low rainfall conditions, this is generally the condition of

least concern. Runoff prediction for average to high flow conditions

are usually of greater interest.

Hydrologic data have been collected on the Upper Taylor Creek

watershed for over fifteen years. The CREAMS model was designed to

simulate field sized areas, thus cannot be used directly to simulate

this watershed. However, long term studies of this watershed have

yielded some consistent relationships between rainfall, runoff, and

evapotranspiration, the major components of the water balance for this

area (Speir et al., 1969; Allen et al., 1982a).

As a means of testing the long term response of the model, the

parameters for the Bass West site were used with twenty years of

rainfall records from this area (Okeechobee Hurricane Gate-6, 1960-

1979). Actual rainfall records from the region were used so that

natural variability in precipitation patterns and amounts would be

represented. Annual rainfall amounts over the twenty year period of

record used for simulation have a mean of 46.8 inches and standard

deviation of 8.20. Over the more than 50 year period of record at that

station, similar statistics of 47.59 for the mean, and 8.94 for the

standard deviation (Knisel et al., 1985) show that the twenty year

period used is representative of the long term record in both amount

and annual distribution. Thus, the range of dry to wet years were

represented in this time span.

Table 3-3.

Annual water balance summary for 20-year CREAMS-WT


60 55.840 16.737 40.787 -1.685
61 28.320 0.026 32.995 -0.595
62 51.610 10.909 34.513 2.082
63 42.020 1.303 39.687 1.029
64 48.220 8.182 41.954 -1.916

65 34.420 0.150 37.428 -0.956
66 48.410 6.501 38.378 1.330
67 37.870 3.131 32.990 1.749
68 45.960 10.768 36.045 -0.853
69 56.970 14.529 41.889 0.552

70 47.710 11.395 38.648 -2.333
71 50.810 10.781 38.238 1.790
72 44.990 6.936 39.304 -1.250
73 49.250 4.492 43.531 1.227
74 47.710 10.328 38.428 -1.046

75 41.000 5.634 34.281 1.085
76 46.900 8.875 37.994 0.031
77 40.700 4.180 35.186 1.334
78 52.200 9.623 42.933 -0.357
79 64.900 26.426 40.769 -2.296

Average 46.790 8.545 38.299 2.370

The predicted water balance for the 20-year period is shown in

Table 3-3. The results of this long term simulation show that the

CREAMS-WT model is stable and that long term predictions are


Keeping in mind the possible differences between the hydrologic

response of the field and watershed, limited comparison can be made

with observed data. On a long term basis, the water balance for the

Taylor Creek watershed is represented by,

Rainfall = Runoff + Evapotranspiration

Of the other components usually found in a water balance, change in

storage is considered negligible over a long period. Recharge to deep

aquifers as deep percolation is negligible in this area and can be

ignored (Stewart, 1980). For an annual rainfall of 47.9 inches,

estimates of ET for Upper Taylor Creek (W-2) and Northwest Taylor Creek

(W-3) are 35.4 and 37.1 inches respectively (Speir et al., 1969). This

compares closely with the predicted annual ET of 38.3 inches, which was

averaged over the 20 years simulated. A slightly higher ET from the

Bass Pasture simulation is to be expected for two reasons. First,

runoff would be expected to increase (and ET decrease) in moving from a

pasture, to a headwaters watershed (W-3), to a larger watershed (W-2).

Second, higher ET amounts may result from the conversion of unimproved

pasture and range to improved pasture (Yates et al., 1982).

This twenty year simulation provides some important verification

of the operation, appropriateness, and applicability of CREAMS-WT. In

summary: the model is stable, it performs predictably over the range

of natural rainfall variability, and long term average annual predic-

tions compare favorably with observed values.

Sensitivity Analysis

The twenty year simulation described above was used as the base

simulation for the parameter sensitivity analysis. Sensitivity was

computed using the twenty year average annual value of the desired

output variables. The effect of a parameter on model response may vary

greatly between a wet and a dry year. Simulation using this period of

historical rainfall averages the parameter effects over the range of

rainfall conditions, minimizing the effects of individual years.

Parameters chosen for analysis were those known to be sensitive,

those that may be more difficult to determine, and the new parameters

introduced in CREAMS-WT. Because the magnitudes and the range of

variation of various parameters may be greatly different, a standard

percentage change in parameters was not used. Rather, parameters were

varied from the base value according to the reasonable range of values

for that parameter.

Relative sensitivity is calculated as,

Onew Ob Pb
Sens = --------- ---- [3-17]
Ob Pnew Pb

where Onew and Ob are the values of the output variable for the new and

the base simulations, and Pnew and Pb are the corresponding parameter

values. This method of calculating sensitivity provides a consistent

measure for comparison between parameters. A value of zero means that

a change in the parameter had no effect on the output. Higher values

of the relative sensitivity indicate increased sensitivity of the model

output to the parameter in question. A value of one means the output

changes by the same percentage the parameter is changed.

The results of the sensitivity analysis are shown in Table 3-4.

Note that sensitivities cannot be compared between different output

variables, for instance between runoff and ET, since the relative

sensitivity is inversely related to the base magnitude of the variable.

The actual values of the variables have also been included in the table

for a reference. The parameters evaluated fall into three categories:

soil characterization, runoff, and ET. The parameters, RC, FUL, POROS,

PORSUB are generally insensitive and will be discussed individually in

the following section on parameter estimation.

The curve number (CN) and the total free water storage, plant

available and drainable, in the profile (UL) are parameters used in the

runoff algorithms. The following observations can be made.

1. The curve number is much more sensitive for higher values

(lower storage) than it is for lower values.

2. UL is sensitive for both positive and negative changes,

although there is some increase in sensitivity with decreasing

storage as is the case with CN.

3. UL is more sensitive than CN. If the curve number is trans-

lated to its equivalent storage value, the difference in

sensitivity is magnified. This is reasonable as CN appears

only in the Smx/ULE ratio, whereas UL is a part of that ratio

and is also a key component in the soil moisture accounting.

Table 3-4. Sensitivity analysis of selected CREAMS-WT parameters.

Parameter Param. % of --- Runoff -- ----- ET ---- Avg Soil Water
(Base) Value Base Value Sens Value Sens Value Sens


































RC (in/hr)








-20 12.80 -2.492 34.01 0.561 1.726 1.360
20 5.95 -1.518 40.93 0.343 2.816 0.938

-20 11.18 -1.543 35.64 0.348 1.940 0.909
20 7.12 0.834 39.75 1.038 2.600 1.097

-50 8.54 0.000 38.30 0.000 2.371 0.000
100 8.54 0.000 38.30 0.000 2.371 0.000




9.15 -0.177 37.70 0.039 2.246 0.132
8.12 -0.124 38.73 0.028 2.412 0.043

8.54 0.000 38.30 0.000 2.371 0.000
8.54 0.000 38.30 0.000 2.371 0.000








C /)
0 0

I- -




10 0NNM r-

c q3- n (1) r- M

000 -4


I I _: ( -i

C 0M -4o

M r o r-r-0

S00 C

-- -4 .-t 0 0


- N- Nr_ c c

C; CM c; CO

0 00 0

00 00
%i tT rt iO

I-i co '-0

C L 0n o ')
L( OCjn -<


N- CM CM c;

as M 0

CN4 00 0-

-n -4 m P

cn ; *<<

mn CMi


0 0

m a


m o
* M
.n i

* *


i i


lqr Ln -:I-
* D C
0 0 0

1-4 -|-4 -4


M 0 N -


- -4 t.0 ifl

CO ) '=0 CDM

C\ CM -


Il l i

0I I 00
m "'.

C) r~ :*,
t.0 CD r~-

CMJ r-- al
- 00



0 t.0c r

in c-

cII-4. enrV cM" d" In C\jC LInLn)

oU)m Cno in 0o Co U-
r-O r-00 C l 0 r

nNO C' Jn -:
Nen cr

*r- CA
4-) L/o

I eA



4- ,


C c



r -A

CNM n :d 00 o
O w3.0 L) Ln LO
C-4 CO ro 3 .-4 ro
00 0


~0 -



0 0 -4 -4

00 0
0 *



-:)C= C)

= 0D CD

0 -
03- CD0
0 **-
Im f


o ----

Leaf area index (LAI), and temperature (TEMP) and radiation (RADI)

data are all used in predicting ET. As can be seen, the model's

response is very sensitive to each of these parameters.

Temperature and radiation are input as monthly values. This sen-

sitivity is a detriment to comparison of model results with historical

data as in the testing and validation of a model. CREAMS-WT uses aver-

age monthly temperature and radiation thus daily or weekly fluctuations

in the historical period being simulated cannot be represented. This

will produce inaccuracies in the simulated ET, and result in discrepan-

cies between observed data and the simulated values. However, for

general predictive use, this sensitivity presents no real drawback to

using the model. Average temperature and radiation data are known with

reasonable accuracy and will not change from one simulation to the

next. Temperature and radiation are physical values that are generally

well defined, thus leaf area index becomes one of the most important

parameters to be estimated.

The base simulation used in Table 3-4 had no deep seepage (DSP=0).

To evaluate the sensitivity of this parameter, a second set of simula-

tions were run with a new base parameter set having DSP = 0.004. The

results of this sensitivity analysis are given in Table 3-5. Several

of the parameters are evaluated again along with the new parameter,

DSP. There are only small changes in the sensitivity of the basic

parameters, with all the previous observations still valid. The

parameter DSP is significant in the prediction of deep percolation.

Runoff appears to be only slightly sensitive to DSP, but comparison of

the actual values shows there is some effect.

Parameter Estimation

The CREAMS manual provides extensive help for coming up with the

necessary parameter values for the model, and is the primary source for

information for the CREAMS-WT model. However, there are several new

parameters, and the emphasis on others has changed. Some comments on

estimating these parameters for CREAMS-WT follows.

CN curve number. The effect of this comes in the Smx/ULE ratio. A

table of curve numbers and corresponding storage are included in

Appendix A. The standard curve number may be used, or alternately,

when ULE has been determined, a curve number can be chosen to give

a desired Smx/ULE ratio with the corresponding runoff curve as in

Figure 3-2.

UL(I) profile storage by layer (ins). Summing the seven layers gives

ULE, the storage in the root zone. This should be a realistic

measure of the actual storage in the profile to the depth of RD.

Storages calculated using 0.1 bar water content appear to be

consistently high. Better estimates may be found in the curves of

depth to water table versus storage from the ARS (Figure 2-6) and

the SFWMD (Figure 2-7). Capece (1984) found these to be the best

storage estimates for predicting storm runoff. These values may be

multiplied by the factor 1/(1-FUL) (FUL is defined below) as depth

to water table may represent only the drainable water and not the

water held at "field capacity". An additional estimate is from

Allen et al. (1982a), who found that 2 to 2.5 inches of stored

water corresponded to a one-foot change in the water table.

This is an important parameter, and the best estimate of the

active storage in the profile should be used.

RD root depth. Generally in the range of 18 to 36 inches. This

should be to the top of the spodic horizon if one exists.

LAI leaf area index. Since the land use is predominantly pasture,

this will not have to be changed much. Caution must be used in

making changes because the model is sensitive to this parameter.

Calibration may be needed to achieve a reasonable annual water bal-

ance. When the total leaf-area-days (printed in the model output)

that gives an appropriate water balance has been determined, the

distribution of that total within the year can be adjusted

depending on the crop.

RC saturated hydraulic conductivity (in/hr). Use a representative

value. However, values above 1.0 in/hr generally have no effect as

this corresponds to moving 24 inches of water per day through the

profile which exceeds any rainfall capacity.

FUL fraction of ULE held at field capacity, that is, the ratio of

field capacity to field capacity plus gravitational water.

POROS Has no effect. Not needed.

PORSUB porosity of the lower zone. This is an estimate of drainable

or fillable porosity. Can be estimated from soil characterization

data. Data from the Bass West site indicate a value in the range

of 0.06 to 0.1. This parameter primarily affects the recovery of

the water table after a long dry period. If the value is too high,

recovery would be slower and less runoff would be predicted. A low

value would result in greater runoff, thus generally would be the

conservative choice. The parameter PORS in equation 3-4 is equal

to PORSUB 12.

DSP deep seepage rate (in/day). User specified.



Description of the CREAMS Nutrient Model

The CREAMS nutrient model simulates soluble and sediment-attached

nitrogen and phosphorus in runoff, considering the top one centimeter

of the soil as the active zone. Nitrogen is modeled in more detail

than is P. A nitrogen budget considers plant uptake, denitrification,

mineralization of organic nitrogen and leaching of nitrate. Fertilizer

nitrogen and phosphorus can be added to the surface or incorporated in

the profile in single or multiple applications during a year. Since

the erosion component is being bypassed for simulation of flatwoods

sites, evaluation and added enhancements were limited to the portion of

the model describing movement of soluble nutrients.

Soluble nitrogen and phosphorus. The soluble nutrients in the top

centimeter of the soil are assumed to be available for leaching and

movement into runoff. Phosphorus is assumed to be adsorbed by the

soil, thus no leaching of phosphorus is considered. Similar relation-

ships are used for modeling both nitrogen and phosphorus. The basic

assumption is that the rate of change in concentration of soluble

nutrient is proportional to the difference between the existing concen-

tration and a base concentration. This can be written,

dC/dt = KI f(t) (Cb-C) [4-1]

where C is the concentration in the soil, Cb is the base concentration,

K1 is a rate constant for downward movement, and f(t) describes infil-

tration as a function of time. For nitrogen, the base concentration is

taken as the concentration of nitrogen in the rainfall and for phospho-

rus, the base concentration is assumed to be the equilibrium phosphorus

concentration of the soil. Both values for base concentration are

input parameters.

For simplification, it is assumed that infiltration occurs at a

uniform rate, after which runoff occurs. Nutrient interaction with

runoff is assumed to be similar to that for leaching (as in equation

4-1), with a rate constant K2 used for movement into runoff. These

relationships are integrated to give the concentration at the end of a


Ct+1 = Cb + (Ct Cb) exp(-K1*F -K2*Q) [4-2]

where subscripts t and t+1 indicate the beginning of the current and

the next time period respectively, Q is the total runoff (mm), and

concentrations are in units of mg/L. F is the effective infiltration

(mm), calculated as the difference between rainfall and runoff and

additionally subtracting the pore volume of the surface layer. Since

the model uses a daily time step, this equation is used to calculate

the concentration for the next day, given the predicted depths of

infiltration and runoff from the hydrology model. Similar equations

are derived for the mean concentration during infiltration and mean

concentration during runoff.

The soluble N and P in runoff (kg/ha) are then determined by

-RON = Cr EXKN Q 0.01 [4-3]

ROP = Cr EXKP Q 0.01 [4-4]

where Cr is the mean concentration of the nutrient during runoff. The

runoff "extraction coefficients" are defined as

EXKN = K2N POR d [4-5]

EXKP = K2p POR d [4-6]

with similar expressions for the vertical extraction coefficients EXKN1

and EXKP1. POR is the soil porosity, and d = 10 mm is the depth of the

surface layer.

The phosphorus buffering capacity of the soil is assumed to keep

the soluble P concentration from dropping below a level characteristic

of the soil and adsorb the soluble P that is "leached" from the surface

layer. Since this is the nutrient of greatest concern in the TCNS

basin and since the assumptions used in developing the CREAMS P compo-

nent are less appropriate for flatwoods watersheds, the discussion here

will focus on the P component of the model.

Parameter estimation. In predicting the movement of N and P into

runoff, there are two primary input parameters to be specified, the

runoff extraction coefficient and the base concentration. For nitro-

gen, the additional parameters that describe the other transformations

also come into play, but are less significant. However, the phosphorus

model is almost completely described by the extraction coefficients,

EXKP1 and EXKP, the base buffered soil P concentration, SOLP, and any

fertilizer inputs.

In the CREAMS manual, Frere et al. (1980) comment that data

suggest extraction coefficients for pesticides and nutrients range from

0.01 to 0.4. Since data are not readily available for defining these

parameters, the vertical extraction coefficients, EXKP1 and EXKN1, have

been fixed in the model with values of 0.25. This was done to simplify

the input requirements, letting the variability of the extraction

process be represented by the one remaining coefficient. The value of

0.25 was a somewhat arbitrarily chosen value which gave reasonable

results with available data. Frere et al. (1980) assume that leaching

efficiency is greater than extraction into runoff, so the constraint on

EXKP and EXKN is that they be less than 0.25.

The CREAMS manual discusses methods of estimating the parameters

SOLP and EXKP. The best method is to determine the equilibrium P

concentration in soil samples. The next best method indicated is to

use measured data from several storms to fit the equation,

PPMP = EXKP CP [4-7]

where PPMP is the concentration in runoff (ppm), and CP is the concen-

tration in the pore water of the surface soil (ppm). Unfortunately,

these data are not readily available, and very few model users will

make the effort to acquire them. The above equation is simply a form

of the basic relationship in equation 4-4. If some estimate of base or

background concentrations from a watershed are available, the following

expression may be used to help estimate the correct range of values for


RP = SOLP EXKP 25 [4-8]

where RP (ppm) is a base concentration in runoff, SOLP has units of

kg/ha, and the 25 is for units conversion.

Applicability of P Model to Flatwoods Watersheds

The performance of the phosphorus model is clearly a function of

making good estimates of EXKP and SOLP. The nutrient load to the

watershed from fertilizer and animal waste is extremely important but

it is assumed that these inputs can be adequately defined. The soluble

N and P model in CREAMS as described above, especially with the neces-

sity of estimating the runoff extraction coefficient, is acknowledged

by its developers as being the weakest part of the nutrient model

(Frere et al., 1980). This is clearly a concern for use in flatwoods

watersheds where the P nutrient model is defined solely in terms of the

soluble component since there is no significant erosion and sediment

transported P. Several questions must be raised in regards to the

appropriateness of the P model and the reasonableness of the assump-

tions for flatwoods watersheds. Is the basic model appropriate? Is

the assumption of a highly buffered soil reasonable? Is the assumption

of EXKP1 = 0.25 reasonable?

The concept of a thin surface layer being the active zone for

movement of soluble P into runoff is a reasonable one, and has been

demonstrated (Ahuja et al., 1981; Sharpley et al., 1981; Sharpley,

1985). There has also been evidence of a relationship between this

active depth and the extraction coefficient, demonstrating that there

may be some physical basis to the simple algorithm used in CREAMS and

that it can provide accurate predictions of mean annual P concentra-

tions (Sharpley et al., 1982). However, this work has been done with

silt loam to clay soils and the general applicability and procedures

for determining EXKP have not been established. Thus, while the

soluble P model has some physical basis, its use generally involves

empirical fitting of some form of the basic relationship shown in

equation 4-4.

The assumption of a highly buffered soil which adsorbs soluble P

and maintains a constant equilibrium phosphorus content (the value of

SOLP) is questionable for the sandy flatwoods soils. As discussed in

the literature review section, data from many Florida sands and from

flatwood spodosols in particular, indicate that P readily moves down

through the profile. The P adsorption isotherms for these soils have

very low partition coefficients. Thus, the addition or removal of P

from the soil readily changes the EPC (equilibrium phosphorus concen-

tration), whereas for highly buffered soils, large changes in P will

have only a minor effect on the EPC. The assumption of a highly buf-

fered soil is clearly a weakness of the model for flatwoods watersheds.

The extraction coefficients are parameters that describe the

efficiency of nutrient extraction into percolating water and surface

runoff. The phosphorus model further assumes that there is no actual

movement of P down through the soil profile, as all phosphorus extrac-

ted by vertical flow is rapidly bound up by the soil. Phosphorus

removed from the soluble pool by this process is not considered any

further by the model. Thus, while the derivation of the above algo-

rithms considers the extraction coefficients to be descriptive of the

flow process, in the case of P, what is being modeled is actually the

processes of adsorption and precipitation.

In flatwoods soils, the removal of soluble P from the active run-

off extraction layer in the profile is a combination of physical move-

ment and chemical adsorption. Due to the low buffering capacity of the

soils, it would be expected that the set value of EXKP1 = 0.25 may be

too high. Initial simulations showed that soil P was rapidly reduced

to the base P level (specified by SOLP) after a fertilizer application,

while data indicate that fertilizer applications may contribute to

runoff loads for several months (Goldstein, 1986).

CREAMS-WT Enhancements for Flatwoods Watersheds

Denitrification. The CREAMS denitrification algorithm is a

function of the number of days of drainage from the bottom layer of the

profile. This follows the assumption that while drainage is occurring

and the soil profile is above field capacity, anaerobic conditions will

be present in the soil thus enabling denitrification. Under the modi-

fied hydrologic model of CREAMS-WT, days when saturated conditions

exist in the upper profile are determined explicitly as a result of

following the level of the water table. The number of days the water

table is in the upper profile is now passed for use by the denitrifica-

tion algorithm in place of the original CREAMS value. This modifica-

tion was actually made in the hydrology model but its effect is seen in

the nutrient model.

Phosphorus in rainfall. The ability to specify the concentration

of phosphorus in rainfall has been added to the model. This input is

handled the same as the nitrogen loading in rainfall and is the same in

concept to a fertilizer application. The load in kg/ha is calculated

knowing the amount of rainfall. This input load is then added to the

nutrient content in the active layer before calculating extraction by

leaching and runoff.

Soluble phosphorus model. Phosphorus removal from the active

layer is controlled by two factors, the value of the extraction coeffi-

cient, EXKP1, and the volume of effective infiltration. While directly

changing the value of EXKP1 is possible, an indirect way of reducing P

removal was implemented. The depth of the active layer was set in the

CREAMS model at 1 cm. Increasing this depth affects the model's re-

sponse in two ways. First, it results in a larger active P pool, and

extraction by "vertical" removal and by runoff has a proportionally

smaller effect on the P pool. The result is that the effective extrac-

tion coefficients are inversely proportional to the increased thickness

of the active layer. Second, an increase in the depth of the active

layer reduces the amount of "effective infiltration" (F in equation

4-2) that is used to calculate P removal from the active pool.

There are several advantages in using this approach to modify the

response of the P model. 1) It has the equivalent effect of reducing

EXKP1. 2) Extraction coefficient values remain in the same range as

those discussed in the CREAMS manual. 3) A reduction in the volume of

effective infiltration is reasonable because of the effects of a high

water table. An increase in depth of the active layer is also

physically reasonable for the flatwoods system, where the majority of

runoff passes through the upper soil profile on its way to a ditch or


Testing and Calibration

Issues in calibration and verification. Measured data all have

some degree of error or uncertainty associated with them. There are

several sources of uncertainty in water quality data from the UD/RDP

sites that need to be recognized, especially in regard to model cali-

bration and verification.

1. Dates and amounts of fertilizer applications are not known

exactly. The timing of fertilizer application and a rainfall

event can significantly affect the amount of nutrients in


2. The activity of cattle cannot be predicted by the model, but

can have a dramatic impact on the nutrient concentrations in

waterways (Goldstein, 1986; Ritter and Allen, 1982). Thus,

measured in-stream concentrations may at times not strictly

represent the quality of runoff from the watershed, but may be

due in part to the activity of cattle in streams and ditches.

3. Water quality samples were collected biweekly. The adequacy

of representing the true time series concentrations with this

sampling frequency must be questioned. Indeed, on the SEZ

dairy watershed, Goldstein (1986) noted that: "The sensiti-

vity of P concentrations to events on the watershed began to

show up vividly when the resolution of the sampling interval

was increased from weekly to daily." The low sampling fre-

quency introduces uncertainty in the calculated nutrient loads

Full Text
xml version 1.0 encoding UTF-8
REPORT xmlns http:www.fcla.edudlsmddaitss xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.fcla.edudlsmddaitssdaitssReport.xsd
INGEST IEID E10W4OWVJ_EORQ6I INGEST_TIME 2017-07-13T15:44:31Z PACKAGE AA00003388_00001