Top Navigation Bar

ASAE Journal Article

ADAPT: Model Use, Calibration, and Validation

P. H. Gowda, D. J. Mulla, E. D. Desmond, A. D. Ward, D. N. Moriasi


Published in Transactions of the ASABE Vol. 55(4): 1345-1352 ( 2012 American Society of Agricultural and Biological Engineers ).

Submitted for review in September 2011 as manuscript number SW 9426; approved for publication by the Soil & Water Division of ASABE in June 2012.

USDA is an equal opportunity provider and employer.

The authors are Prasanna H. Gowda, ASABE Member, Research Agricultural Engineer, USDA-ARS Conservation and Production Research Laboratory, Bushland, Texas; David J. Mulla, Professor, Department of Soil, Water, and Climate, University of Minnesota, St. Paul, Minnesota; Eric D. Desmond, ASABE Member, Research Associate, and Andrew D. Ward, ASABE Member, Professor, Department of Food, Biological and Agricultural Engineering, The Ohio State University, Columbus, Ohio; and Daniel N. Moriasi, ASABE Member, Research Hydrologist, USDA-ARS Great Plains Agroclimate and Natural Resources Research Unit, El Reno, Oklahoma. Corresponding author: Prasanna H. Gowda, USDA-ARS Conservation and Production Research Laboratory, P.O. Drawer 10, Bushland, TX 79012; phone: 806-356-5730; e-mail: Prasanna.Gowda@ars.usda.gov.


Ab stract. This article presents an overview of the Agricultural Drainage and Pesticide Transport (ADAPT) model and a case study to illustrate the calibration and validation steps for predicting subsurface drainage and nitrate-N losses from an agricultural system. The ADAPT model is a daily time step, field-scale water table management model that was developed as an extension of the GLEAMS model. The GLEAMS algorithms were augmented with algorithms for subsurface drainage, subsurface irrigation, deep seepage, and related water quality processes. Recently, a frost depth algorithm was incorporated to enhance the model’s capability to predict flow during spring and fall months. In addition to the normal GLEAMS output, ADAPT gives estimates of pesticides and nutrients in drainage. The model has four components: hydrology, erosion, nutrient transport, and pesticide transport. Predictions of surface runoff and subsurface drainage by ADAPT are very sensitive to hydrology input parameters, such as NRCS curve number, hydraulic conductivity, depth of the impeding layer, and hydraulic conductivity of the impeding layer. In the erosion component, slope, hydraulic length, and crop management are the most sensitive factors. Nutrients generally follow the trends in surface runoff and subsurface drainage. In addition, nitrogen and phosphorus concentrations in soil horizons are sensitive to nutrient losses. Recently, the ADAPT model was further calibrated and validated in southern Minnesota to evaluate impacts of subsurface drain spacing and depth, rate and timing of nitrogen application, and precipitation changes on water quality. ADAPT is written in FORTRAN, and the source code is available to interested model users. Considering the limited technical support and text editor-based input files, development of a user-friendly interface to create input files would greatly enhance ADAPT’s acceptability by users involved in modeling agricultural systems equipped with subsurface drains.

Keywords. Nonpoint-source pollution, Subsurface drainage system, Upper Midwest, Water quality, Water table management.

The Agricultural Drainage and Pesticide Transport (ADAPT) model is a daily time step, field-scale simulation model that was developed in the Department of Agricultural Engineering of Ohio State University to simulate the quantity and quality of flows associated with water table management systems. ADAPT was developed by integrating Groundwater Loading Effects of Agricultural Management Systems (GLEAMS; Leonard et al., 1987) with DRAINMOD (Skaggs, 1982), a subsurface drainage model. The two models were combined to create a model that would predict both the quantity and quality of surface and subsurface effluents. ADAPT was written in FORTRAN with modular programming techniques.

The first publications describing ADAPT in the refereed literature were Chung et al. (1992) and Ward et al. (1993). In addition to describing ADAPT, Chung et al. (1992) evaluated the hydrologic component of the model, while Ward et al. (1993) described development studies with the ADAPT model. A key initial application of ADAPT was performed by Desmond et al. (1996), who compared daily water table depth prediction by four simulation models. Gowda et al. (1999a) developed a spatial process model that uses ADAPT for predicting flow and nutrient discharges in the Rock Creek watershed of northern Ohio and successfully used this model in numerous water quality studies in Minnesota at plot, field, and watershed levels (Davis et al., 2000; Dalzell et al., 2004). In 2000, a frost depth algorithm developed by Benoit and Mostaghimi (1985) was incorporated to enhance ADAPT’s capability to predict flow during spring and fall months and was tested with flow data from the lower Minnesota River basin (Dalzell et al., 2004). Numerous studies with ADAPT were successfully completed in Ohio (Gowda et al., 1999b), Iowa (Gowda et al., 2008), Illinois (Sogbedji and McIsaac, 2006), Minnesota (Sands et al., 2003; Johansson et al., 2004; Dalzell et al., 2004; Updegraff et al., 2004; Petrolia and Gowda, 2006a, 2006b; Gowda et al., 2007), and Kansas (Anand et al., 2007) for evaluating agricultural alternative management practices and development of total maximum daily loads. Recently, the ADAPT model was calibrated and validated in southern Minnesota to evaluate impacts of subsurface drain spacing and depth (Nangia et al., 2010a), rate and timing of N application (Nangia et al., 2008; Nangia et al., 2010b), and changes in precipitation (Nangia et al., 2010c) on N losses in southern Minnesota.

The ADAPT model theoretical documentation and users’ manuals can be obtained from the developers at Ohio State University, but technical support for users is limited. There is a small group of developers and users, primarily associated with the USDA-ARS Conservation and Production Research Laboratory in Bushland, Texas, and the University of Minnesota. ADAPT is an open-source model. Currently, there is no GIS user interface available for ADAPT; instead, the model uses a text editor for creating input files. The objectives of this study were to: (1) provide a brief description of ADAPT including the components simulated, (2) present ADAPT calibration and validation procedures, (3) present a case study to illustrate the calibration and validation process, and (4) discuss ADAPT model weaknesses and future research needs.

ADAPT Description

The ADAPT model has four components: hydrology, erosion, nutrient transport, and pesticide transport. The hydrology component includes snowmelt, surface runoff, macropore flow, evapotranspiration (ET), infiltration, subsurface drainage, sub-irrigation, and deep seepage. The snowmelt component in ADAPT is based on a theory proposed by Anderson and Crawford (1964) and Viessman et al. (1989). Snowmelt water depth is computed as the summation of snowmelt due to radiation, rainfall, conduction, convection, and condensation (Chung et al., 1992). Potential ET is computed using either the Ritchie method used in GLEAMS (Knisel, 1980) or the Dorenbos-Pruitt method (James, 1988). A modified Green-Ampt equation (Mein and Larson, 1971) is used to determine infiltration time. Subsurface drainage and sub-irrigation algorithms are based on DRAINMOD (Skaggs, 1980). Subsurface drainage flow rate is calculated using either Kirkham’s or Hooghoudt’s equation depending on the water table condition. Deep seepage through the impermeable layer is determined using Darcy’s equation. A detailed description of the equations used to compute the different hydrologic processes in ADAPT can be obtained from Chung et al. (1992).

Soil erosion is estimated using the Universal Soil Loss Equation (Foster et al., 1980). Nutrient transformation processes in the ADAPT model were adopted from the GLEAMS nutrient model (Knisel et al., 1993). Two nutrients, nitrogen and phosphorus, are included in the model. The nitrogen cycle used in ADAPT includes routines for mineralization from crop residue, soil organic matter, and animal waste, immobilization to crop residue, plant uptake, partitioning between soil and solution phases, nitrogen fixation by legumes, denitrification, and fertilization. Nitrogen in the soil is divided into active and stable pools and changes daily as a function of their relative size and the C:N ratio of organic materials, such as crop residue, roots, and animal waste. The phosphorus cycle used in the ADAPT model includes routines for mineralization, immobilization, fertilization, animal waste application, and crop uptake. Phosphorus losses are simulated at a daily time step based on rates of sediment loss and soil concentrations of phosphorus, as well as rates of runoff and concentrations of soluble phosphorus (Knisel et al., 1993). A detailed description of how these components are computed is given by Chung et al. (1992).

The pesticide component in the ADAPT model uses parameter values from the hydrology and erosion components to simulate partitioning and degrading of pesticides each day. Pesticide input parameters include pesticide application date, amount, method, pesticide water solubility, foliar and soil half-lives, and the partition coefficient. Up to ten pesticides can be simulated at a time for determining their fate and transport in the agricultural system. Pesticide output includes concentration and mass in surface runoff, sediment, subsurface drainage, and deep seepage. ADAPT also provides pesticide concentrations in the soil layers, and masses decomposed and taken up by plants. Decomposition of the pesticide accounts for biodegradation and hydrolysis and is assumed to follow first-order kinetics both in the soil and on leaves (Knisel, 1980). The vertical movement of pesticides includes macropore flow, infiltration, and deep seepage. Pesticide movements with evaporation and plant uptake are the same as in GLEAMS. Evaporation moves pesticide in solution upward in the soil profile, while plants uptake the pesticide in each soil layer through transpiration depending on the root distribution in the soil. Pesticide mass in subsurface drainage is calculated using the solution concentration in the water table layer. ADAPT allows the user to inject pesticides through sub-irrigation. It is assumed that pesticides move up along with the sub-irrigated water. Based on the principle of mass conservation, the pesticide mass balance is checked daily.

Model Input and Output

Input data for the ADAPT model include weather, soil, crop, sub-irrigation and subsurface drainage systems, and nutrient and pesticide data. The weather data required to drive the ADAPT model are daily rainfall, average air temperature, wind speed, relative humidity, and solar radiation for the duration of the simulation. Soil data include the texture, thickness, organic matter content, soil water characteristics, and saturated hydraulic conductivity for each horizon. In addition, surface storage depth, NRCS curve number at antecedent soil water condition II, and relationships between water table depth, upward fluxes, and drained-to-equilibrium soil water contents are required. Crop data include effective rooting depth, leaf area index as a function of growth stage, and planting and harvesting dates. Sub-irrigation and subsurface drainage system input parameters consist of drain depth, spacing, diameter, outlet weir height, and depth to impermeable layer.

Nutrient parameters in the ADAPT model include fertilizer and manure application date, amount, method, crop name, leguminous or not, potential yield, dry matter ratio, C:N ratio, and nitrogen concentration in rainfall. Initial conditions are defined by parameters such as crop residue on the soil surface, base saturation, pH, calcium carbonate content, and various nitrogen and phosphorus concentrations in each soil horizon. Pesticide parameters include application date, amount, method, water solubility, foliar and soil half-lives, and the adsorption constant.

Output of the ADAPT model consists of daily, monthly, and annual estimates of surface runoff, subsurface drainage, monthly rainfall, evapotranspiration, deep seepage, water table depth, and sub-irrigation volume. Nutrient output includes concentration and mass in surface runoff, sediment, subsurface drainage, and deep seepage. ADAPT also provides nutrient concentrations in the soil layers and masses taken up by plants. Pesticide output includes concentration and mass in surface runoff, sediment, subsurface drainage, and deep seepage. ADAPT also provides pesticide concentrations in each horizon and pesticide masses decomposed and taken up by plants.

ADAPT Calibration and Validation

Ideal calibration and validation of the ADAPT model should begin with the hydrology component. If the model is predicting the water budget well, it is more likely that the erosion, nutrient, and pesticide components will follow the trend. However, further calibration may be required to obtain more accurate predictions of nitrate-N, phosphorus, and pesticide losses. For an agricultural system with subsurface drainage, the most important aspect is to make sure that the model is partitioning the flow between surface runoff and subsurface drainage. Subsurface drainage is the major carrier of nitrate-N from agricultural systems to streams and rivers. Therefore, errors in partitioning of the flow will affect prediction of nutrient and pesticide losses, and errors in surface runoff will affect prediction of soil erosion losses. The calibration parameter values used in the simulation should be reasonable and supported by the literature. Further, model calibration parameters can be determined by sensitivity analysis. Model parameters are calibrated manually by comparing predictions against measured data for the calibration period only (Mulla and Addiscott, 1999), and the performance is determined using graphical and statistical performance measures. Once the calibration is completed, the predictions for the validation period should be compared against measured data for deriving performance statistics.

For the ADAPT model, it is recommended that simulations for calibration and validation be done simultaneously in order to measure the model performance effectively. The two most commonly reported methods for selecting calibration and validation periods are to: (1) divide the simulation period into two equal parts and use one part for calibration and the other for validation, or (2) use alternate years of simulation as the calibration and validation periods. It is important for ADAPT users to know that each of these methods can give different model performance statistics depending on the years selected for calibration and validation. Irrespective of the method used for establishing the calibration and validation periods, the rationale for choosing a particular method should be given. It is also necessary to allow a model spin-up period at the beginning of the simulation period to initialize the model and allow soil pools to reach a quasi-state of equilibrium.

There are no absolute criteria in the literature for judging the performance of ADAPT or any other model. Numerous graphical and statistical model calibration and validation performance measures have been used with respect to ADAPT applications. Davis et al. (2000) compared observed and predicted means and their standard deviations, coefficient of determination, slope and intercept of a least square regression between predicted and observed values, root mean square error (RMSE), and index of agreement (Willmott, 1981) to assess the level of agreement between predicted and observed values. Other model performance statistics that can be used are presented in detail by Moriasi et al. (2007). Graphical measures include time series (Gowda et al., 2007) and 1:1 graphs (Davis et al., 2000). Different modelers use different criteria depending on the application, availability of measured data, and data quality.

Calibration Parameters

Prediction of total flow (surface runoff and subsurface drainage) by the ADAPT model is very sensitive to hydrology input parameters, such as NRCS curve number, hydraulic conductivity, depth of the impeding layer, and hydraulic conductivity of the impeding layer. Total flow and partitioning of the flow between surface runoff and subsurface drainage during the winter and spring periods are sensitive to cutoff average daily temperature (TCUT), which is required for triggering infiltration. Evapotranspiration is one of the important parameter in the water budget that affects the water available for runoff and subsurface drainage. It is sensitive to leaf area index (LAI) and effective rooting depth (RD). In the absence of observed data on surface runoff and subsurface drainage, model predictions should be compared with values available in the literature for similar agricultural systems. A sensitivity analysis by Chung et al. (1992) showed that surface runoff estimates are sensitive to changes in the curve number, and subsurface drainage estimates are sensitive to deep seepage predictions (Chung et al., 1992; Desmond et al., 1996).

Prediction of soil erosion losses with ADAPT are highly sensitive to the field size (DAOVR), average slope (AVGSLP), slope length of the overland flow profile (SLNGTH), soil erodibility factor (KSOIL), and soil loss ratio (CFACT) parameters (Gowda et al., 1999a). In the nutrient component, total nitrogen and phosphorus, nitrate-N and labile-N concentrations, and potentially mineralizable nitrogen in each soil horizon are important parameters in predicting nitrate-N and phosphorus losses. In the absence of measured data, users can start the calibration of these parameters with values reported in the literature. Nutrient losses increase with increase in nitrate-N and labile-P concentrations in the soil horizons. Losses of N by denitrification are expected to be significant during high water table and low soil temperature conditions. Any errors in the prediction of denitrification require calibration of hydrology parameters, such as the hydraulic conductivity of each soil horizon, depth of the impeding layer and its hydraulic conductivity, and frost depth related parameters. As expected, prediction of pesticide losses is sensitive to pesticide properties, such as soil half-life (SOLLIF) and partitioning coefficient (KOC).

Case Study

Objective and Study Area

The main objective of this case study is to demonstrate how ADAPT can be calibrated and validated for subsurface drainage and associated nitrate-N losses in a typical Upper Midwest U.S. row crop agricultural system with continuous corn under conventional tillage. Although calibration and validation of ADAPT for pesticide and P losses are not presented in this case study, they follow similar calibration procedures. Calibration and validation of ADAPT was done using flow and nitrate-N measurements made in three experimental plots for the period 1983-1996. These field measurements were part of a subsurface drainage study (Randall et al., 1997; Randall and Iragavarapu, 1995; Buhler et al., 1993) at the University of Minnesota’s Southern Experiment Station near Waseca, Minnesota. The size of the each plot was 13.5 m × 15.0 m on poorly drained Webster clay loam soil. These plots were designed to simulate a subsurface drain spacing of 27 m. Subsurface drains were installed at a depth of 1.2 m with a slope of 0.1%. The diameter of the subsurface drain was 100 mm. Since 1982, these plots were planted with continuous corn under moldboard plow tillage. Nitrogen fertilizer was broadcast in the spring before planting at an annual rate of 202 kg N ha -1 from 1983 to 1993, 154 kg N ha -1 in 1994, 204 kg N ha -1 in 1995, and 149 kg N ha -1 in 1996. Subsurface drain flows were measured daily, and water samples from the subsurface drain outlet were taken every Monday, Wednesday, and Friday for quantifying nitrate-N in subsurface drainage.

Measured Data

Weather data, including daily values of precipitation, average air temperature, solar radiation, wind speed, and average relative humidity recorded at a weather station located about 0.5 km from the experimental plots, were used in the simulation. Tables 1 and 2 present soil properties and subsurface drainage system calibration parameters for the hydrology and nitrate-N components. These parameters were held constant for all simulations unless otherwise stated. Soil properties such as depth of each horizon, particle size distribution, and organic matter content reported by Cully (1986) were used in the simulations. The moisture retention curve for each soil horizon was estimated using the DMSOIL program (NRCS, 1994). The NRCS runoff curve number was estimated for the hydrologic group rating of the Webster soil (B/D rating, poor drainage improved by tiling) under a straight row cropping system. The values of hydraulic conductivity and depth to impermeable layer were determined by calibration, since no experimental measurements of these parameters were available. The initial N content of the soil was obtained from Randall (1983).

Table 1. Values used for representative depth profile of soil and subsurface drainage systems in the study plot.

Input Variable

Unit

Value

NRCS curve number (AMC II)

-

78

Evaporation constant

mm d -0.5

4.0

Effective rooting depth

cm

115

Surface sealing threshold

cm

4.4

Surface storage depth

cm

2.5

Depth to impermeable layer

m

4.57

Drain spacing

m

27

Depth of tile drain

m

1.2

Table 2. Values for layer-specific soil input variables. [a]

Horizon

Thickness

(cm)

Porosity

(cm 3 cm -3 )

BR15

(cm 3 cm -3 )

OM

(%)

Clay

(%)

Silt

(%)

Vert. K

(cm h -1 )

Horiz. K

(cm h -1 )

Initial N

(Mg kg -1 )

1

31

0.45

0.23

6.1

33

38

4.8

4.8

6.7

2

31

0.41

0.21

2.1

31

33

4.8

4.8

6.4

3

31

0.39

0.19

1.0

30

31

4.8

4.8

5.5

4

87

0.39

0.19

1.0

29

32

4.8

4.8

4.6

[a] BR15 = moisture content at wilting point, OM = organic matter, Vert. K = vertical saturated hydraulic conductivity, and Horiz. K = horizontal hydraulic conductivity. Values for saturated hydraulic conductivity were obtained by model calibration.

Calibration and Validation Procedure

Although ADAPT is a daily time step model, model outputs can be obtained at daily, monthly, or annual time steps. Because of the lack of complete daily nitrate-N datasets, monthly nitrate-N values were estimated and used in the model calibration and validation. Therefore, in this case study, calibration and validation of the ADAPT model consisted of predicting and comparing monthly subsurface drainage and associated nitrate-N losses with measured data between April and August. For calibration, measured data for the years 1983, 1985, 1987, 1989, 1991, 1993, and 1995 were used. The selection of odd years for calibration was made in order to improve model performance across a wide range of climatic conditions. For example, the years 1989 and 1993 were the driest and wettest years, respectively, in the last 40 years. If we had simply used the first seven years for model calibration, the model would have been biased toward the driest year. If we had used the last six years for calibration, it would have been biased toward the wettest year.

In this study, ADAPT was calibrated by varying the hydraulic conductivity, rooting depth, leaf area index, drainage coefficient, and soil moisture retention curves to achieve the closest agreement between observed and predicted values of monthly subsurface drainage and nitrate-N losses. Five statistical procedures were used to assess the level of agreement between the predicted and observed values, including: (1) observed and predicted means and their standard deviations, (2) coefficient of determination (r 2 ), (3) slope and intercept of a least square regression between the predicted and observed values, (4) root mean square error (RMSE), and (5) index of agreement (d) (Willmott, 1981; Moriasi et al., 2007). For perfect model performance, the RMSE should be zero, and the index of agreement should be one. Efforts were made to minimize the RMSE to zero and index of agreement close to one. In practice, model performance is never perfect, and RMSE values below 75% or an index of agreement greater than 0.75 indicate satisfactory model performance (Moriasi et al., 2007). In addition, simulated water and nitrogen budgets were compared with available measured data to evaluate ADAPT’s ability to predict individual components of the budgets. This was done by calculating the average annual water and nitrogen budgets for the period 1983-1996 and comparing them with available field measurements and data reported in the literature.

Calibration and Validation Results

Table 3. Comparison of predicted and observed cumulative, mean, and standard deviation values for subsurface drainage and nitrate-N losses in subsurface drainage during April-August in calibration and validation years.

Statistic

Calibration Period

Validation Period

Subsurface

Drainage

(cm)

Nitrate-N in

Subsurface

Drainage

(kg h -1 )

Subsurface

Drainage

(cm)

Nitrate-N in

Subsurface

Drainage

(kg h -1 )

Cumulative

Observed

159.8

240.5

129.9

228.9

Predicted

159.9

234.3

141.4

214.0

Monthly mean

Observed

4.6

6.9

3.7

6.5

Predicted

4.6

6.7

4.0

6.1

Standard deviation

Observed

6.1

11.2

4.3

9.5

Predicted

4.3

6.5

3.8

5.7

Table 3 shows good agreement between the monthly predicted and observed subsurface drainage and associated nitrate-N losses during the calibration and validation periods (Moriasi et al., 2007). During calibration, we attempted to minimize the RMSE and obtain d values close to one (table 4). Figure 1 illustrates the 1:1 relationship between predicted and measured monthly subsurface drainage for the calibration years. The ideal values for slope and intercept are one and zero. The closer the slope is to one, the better the model performance. Values below the 1:1 line indicate model underprediction, while values above the 1:1 line indicate overprediction. The results show that the model underpredicted subsurface drainage for wet months. This is primarily due to the difficulty in predicting the timing and magnitude of subsurface drainage during spring snowmelt runoff. Consequently, the standard deviation of measured data (6.1 cm) was 40.8% higher than that of predicted drainage, although the predicted mean monthly and total drainage flows were in very close agreement with the measured data. A statistical evaluation of the measured and observed data gave an r 2 of 0.90, with a slope and intercept of 0.67 and 1.50 cm, respectively. The d value was about 0.89, and the RMSE was only 1.5% of the measured mean monthly subsurface drainage (table 4). From these statistical results, we can conclude that the ADAPT model performed reasonably well in predicting subsurface drainage during the non-snowmelt period.

Table 4. Model performance statistics for predicted versus observed nitrate losses in subsurface drainage during April-August in calibration and validation years.

Statistic

Calibration Period

Validation Period

Subsurface

Drainage

(cm)

Nitrate-N in

Subsurface

Drainage

(kg h -1 )

Subsurface

Drainage

(cm)

Nitrate-N in

Subsurface

Drainage

(kg h -1 )

RMSE

1.50

2.73

1.79

3.02

R 2

0.90

0.71

0.70

0.47

Slope

0.67

0.49

0.74

0.41

Intercept

1.50

3.31

1.30

3.45

d

0.89

0.74

0.83

0.60

SW9426_files/image1.gif

Figure 1. Comparison of predicted versus observed subsurface tile drainage for the calibration years.

SW9426_files/image2.gif

Figure 2. Comparison of predicted versus observed nitrate-N losses through subsurface tile drainage for the calibration years.

Figure 2 shows the 1:1 relationship between predicted and measured monthly nitrate-N losses for calibration years. As in the case of drainage, predicted mean monthly nitrate-N losses were in close agreement with the observed data. However, the standard deviation of the measured data (11.2 kg ha -1 ) was 71.9% higher than that of the predicted nitrate-N losses. Errors in the prediction of nitrate-N losses were due to a gross prediction error in just one month (May 1991), when a heavy rainstorm flushed residual soil nitrate-N, which had accumulated in several previous years of drought, out of the profile. Statistical comparison of predicted and observed monthly nitrate-N losses gave an r 2 of 0.71, with a slope and intercept of 0.49 and 3.3 kg ha -1 . The d value was about 0.74, and the RMSE was only 2.7% of the measured mean monthly nitrate-N loss (table 4). Thus, except during heavy spring storms in a year following drought, the predictions of nitrate-N losses in the drainage were satisfactory.

SW9426_files/image3.gif

Figure 3. Comparison of predicted versus observed subsurface tile drainage for the validation years.

SW9426_files/image4.gif

Figure 4. Comparison of predicted versus observed nitrate-N losses through subsurface tile drainage for the validation years.

Figure 3 shows a 1:1 relationship between predicted and measured monthly subsurface drainage for the validation years. These results show a trend similar to that observed for the calibration years. However, the model overpredicted the total subsurface drainage by 8.8%. The standard deviation of the measured data (4.3 cm) was only 13.6% higher than that of the predicted subsurface drainage, compared with 40.8% for the calibration period. The RMSE was only 1.8% of the measured mean subsurface drainage for the validation period (table 4), which is slightly higher than that of the calibration period. Comparison of predicted and measured monthly subsurface drainage gave an r 2 of 0.7, with a slope and intercept of 0.74 and 1.3 cm, respectively. The d value was about 0.74. Differences in the statistical results between the calibration and validation years are partly due to very large rainfall events that occurred in the wettest years of 1991 and 1993.

Figure 4 shows a 1:1 relationship between predicted and measured monthly nitrate-N losses for the validation period. ADAPT underpredicted nitrate-N losses by 7.0%, compared with 2.7% for calibration years. This underprediction was primarily due to flushing of large quantities of residual soil nitrate-N out of the profile during three storms in 1990 that followed two years of severe drought. The model could not account for the enhanced mineralization of organic matter in wet years following severe drought. The standard deviation in measured monthly nitrate-N losses (9.5 kg ha -1 ) was 67.8% higher than that for the predicted monthly nitrate-N losses (5.7 kg ha -1 ). Consequently, the d value was about 0.6, with an RMSE of 3.0% of the measured mean monthly nitrate-N loss (table 4). A comparison of predicted and measured monthly nitrate-N losses gave an r 2 of 0.47, with a slope and intercept of 0.41 and 3.5 kg ha -1 , respectively.

Table 5. Predicted and observed components of water balance: Annual averages for the period 1983-1996.

Model Predictions

Field Observations

Depth

(cm)

% of

Precipitation

Depth

(cm)

% of

Precipitation

Precipitation

90.7

100.0

90.7

Evapotranspiration

53.0

58.4

-

Tile drainage

29.8

32.9

29.7

32.8

Runoff

2.9

3.2

-

Deep seepage

4.5

5.0

-

Table 6. Predicted and observed components of the nitrogen budget: Annual averages for the period 1983-1996.

Model Predictions

Field Observations

Nitrogen

(kg ha -1 )

% of N

Applied

Nitrogen

(kg ha -1 )

% of N

Applied

N inputs

Fertilization

194.9

194.4

Mineralization

69.8

-

Rainfall

10.9

-

Nitrogen fixed

0.00

-

Total N input

275.6

N losses

Total plant uptake

174.4

89.48

143.0

73.36

Subsurface drainage

44.6

22.87

41.0

21.03

Denitrification

13.3

6.82

-

Sediment

1.2

0.64

-

Runoff

0.8

0.43

-

Deep seepage

6.1

3.10

-

Ammonia volatilized

0.0

0.00

-

Total N losses

240.4

Tables 5 and 6 present comparisons of the measured and predicted average annual water and nitrogen budgets, respectively, for the study period (1983-1996). The predicted annual subsurface drainage was about 32.9% of the total precipitation that occurred during that period, which closely matched the measured value of 32.8%. The predicted ET was 58.4% of the total precipitation and is comparable with measured values (64.1% in 1992 and 72.0% in 1992) on a fine-textured subsurface-drained soil cropped with corn in central Iowa (Moorman et al., 1999). Measured ET values were not available for our study site. This comparison indicates that the model partitioned water reasonably well. The predicted annual average nitrate-N loss through subsurface drains (44.6 kg ha -1 ) was about 22.9% of the applied N and about 8.7% higher than the measured nitrate-N losses (41.0 kg ha -1 ). This error might be due to underprediction of nitrate-N loss by denitrification. The predicted nitrate-N loss by denitrification was about 6.8% of the total N applied, which is less than the estimated values (10% to 25%) reported by Meisinger and Randall (1991). Plant N uptake predicted by the model was about 15% greater than measured plant N uptake. If predictions of plant N uptake were improved, this would most likely result in more soil N and increases in denitrification and nitrate-N losses. More information on this case study is presented by Davis et al. (2000).

Discussion

The strengths of the ADAPT model include: (1) its ability to predict both hydrology and water quality components; (2) as a field-scale model, there are measured data for calibrating and validating most of the simulated components; and (3) ADAPT has been calibrated and successfully applied in numerous studies throughout the Midwest U.S., where subsurface drainage systems are common. These studies have indicated that the performance of ADAPT is comparable to that of other widely used models, such as SWAT and DRAINMOD. At present, ADAPT is the only hydrologic model that is well calibrated for estimating water quantity and quality for subsurface drainage systems. Although the model is designed for plot or field scale applications, it has been used at the watershed level using the concept of hydrologic response units (Gowda et al., 2007). The source code for this model is available to modelers interested in enhancing the capability of the model. However, it is a DOS-based model with a text editor for creating input files. It is somewhat difficult to identify errors in the format of the input, and user support for the model is limited.

Future Developments

There are no new developments planned in the near future. However, a user-friendly interface for creating model inputs from GIS datasets would reduce formatting errors in the input files.

REFERENCES

Anand, S., K. R. Mankin, K. A. McVay, K. A. Janssen, P. L. Barnes, and G. M. Pierzynski. 2007. Calibration and validation of ADAPT and SWAT for field-scale runoff prediction. J. American Water Resources Assoc. 43(4): 899-910.

Anderson, E. A., and N. H. Crawford. 1964. The synthesis of continuous snowmelt runoff hydrographs on a digital computer. Tech. Report No. 36. Stanford, Cal.: Stanford University, Department of Civil Engineering.

Benoit, G. R., and S. Mostaghimi. 1985. Modeling frost depth under three tillage systems. Trans. ASAE 28(5): 1499-1505.

Buhler, D. D., G. W. Randall, W. C. Koskinen, and D. L Wyse. 1993. Atrazine and alachlor losses from subsurface drainage of a clay loam soil. J. Environ. Qual. 22(3): 583-588.

Chung, S. O., A. D. Ward, and C. W. Schalk. 1992. Evaluation of the hydrologic component of the ADAPT water table management model. Trans. ASAE 35(2): 571-579.

Cully, J. L. B. 1986. Water regimes and shear strengths of typic haplaquoll under conventional and no tillage. PhD diss. St. Paul, Minn.: University of Minnesota, Department of Soil Science.

Dalzell, B. J., P. H. Gowda, and D. J. Mulla. 2004. Evaluating feasibility of TMDLs with alternative management practices on an agricultural watershed. J. American Water Resources Assoc. 40(2): 533-543.

Davis, D. M., P. H. Gowda, D. J. Mulla, and G. W. Randall. 2000. Modeling nitrate nitrogen leaching in response to nitrogen fertilizer rate and tile drain depth or spacing for southern Minnesota, USA. J. Environ. Qual. 29(5): 1568-1581.

Desmond, E. D., A. D. Ward, N. R. Fausey, and S. R. Workman. 1996. Comparison of daily water table depth prediction by four simulation models. Trans. ASAE 39(1): 111-118.

Foster, G. R., L. J. Lane, J. D. Nowlin, J. M. Laflen, and R. A. Young. 1980. A model to estimate sediment yield from field-size areas: Development of model. In CREAMS: A Field-Scale Model for Chemicals, Runoff, and Erosion from Agricultural Management Systems , 36-64. W. G. Knisel, ed. Conservation Research Report No. 26. Washington, D.C.: USDA Science and Education Administration.

Gowda, P. H., A. D. Ward, D. A. White, J. G. Lyon, and E. Desmond. 1999a. The sensitivity of stream flows to model input parameters used to define hydrologic response units. Trans. ASAE 42(2): 381-389.

Gowda, P. H., A. D. Ward, D. A. White, D. B. Baker, and J. G. Lyon. 1999b. An approach for using field-scale models to predict daily peak flows on agricultural watersheds. J. American Water Resources Assoc. 35(5): 1223-1232.

Gowda, P. H., B. J. Dalzell, and D. J. Mulla. 2007. Model-based nitrate TMDLs for two agricultural watersheds of southeastern Minnesota. J. American Water Resources Assoc. 43(1): 1-10.

Gowda, P. H., D. J. Mulla, and D. B. Jaynes. 2008. Simulated long-term nitrogen losses for a Midwestern agricultural watershed in the United States. Agric. Water Mgmt. 95(5): 616-624.

James, L. G. 1988. Principles of Farm Irrigation System Design. New York, N.Y.: John Wiley and Sons.

Johansson, R. C., P. H. Gowda, D. J. Mulla, and B. J. Dalzell. 2004. Metamodeling phosphorus best management practices for policy use: A frontier approach. Agric. Economics 30(1): 63-74.

Knisel, W. G. 1980. CREAMS: A field-scale model for chemicals, runoff, and erosion from agricultural management systems. USDA Conservation Research Report 26. Washington, D.C.: USDA.

Knisel, W. G., R. A. Leonard, and F. M. Davis. 1993. The GLEAMS model plant nutrient component: Part I: Model documentation. Tifton, Ga.: University of Georgia, Department of Agricultural Engineering.

Leonard, R. A., W. G. Knisel, and D. A. Still. 1987. GLEAMS: Groundwater loading effects of agricultural management systems. Trans. ASAE 30(5): 1403-1418.

Mein, R. G., and C. L. Larson. 1971. Modeling infiltration component of the rainfall-runoff process. Bulletin No. 43. Minneapolis, Minn.: University of Minnesota, Water Resources Research Center.

Meisinger, J. J., and G. W. Randall. 1991. Estimating nitrogen budgets for soil-crop systems. In Managing Nitrogen for Groundwater Quality and Farm Profitability , 85-124. R. F. Follet et al., eds. Madison, Wisc.: SSSA.

Moorman, T. B., D. B. Jaynes, C. A. Cambardella, J. L. Hatfield, R. L. Pfeiffer, and A. J. Morrow. 1999. Water quality in Walnut Creek watershed: Herbicides in soils, subsurface drainage, and groundwater . J. Environ. Qual. 28(1): 35-45.

Moriasi, D. N., J. G. Arnold, M. W. Van Liew, R. L. Bingner, R. D. Harmel, R. D., and T. L. Veith. 2007. Model evaluation guidelines for systematic quantification of accuracy in watershed simulations. Trans. ASABE 50(3): 885-900.

Mulla, D. J., and T. M. Addiscott. 1999. Validation approaches for field-, basin-, and regional-scale water quality models. In Assessment of Non-Point Source Pollution in the Vadose Zone, 63-78. Geophysical Monograph Series, Vol. 108. Washington, D.C.: American Geophysical Union.

Nangia, V., P. H. Gowda, D. J. Mulla, and G. R. Sands. 2008. Water quality modeling of fertilizer management impacts on nitrate losses in tile drains at the field scale. J. Environ. Qual. 37(2): 296-307.

Nangia, V., P. H. Gowda, D. J. Mulla, and G. R. Sands. 2010a. Modeling impacts of tile drain spacing and depth on nitrate-nitrogen losses. Vadose Zone J. 9(1): 61-72.

Nangia, V., P. H. Gowda, and D. J. Mulla. 2010b. Evaluation of predicted long-term water quality trends to changes in N fertilizer management practices for a cold climate. Agric. Water Mgmt. 97(11): 1855-1860.

Nangia, V., D. J. Mulla, and P. H. Gowda. 2010c. Precipitation changes impact stream discharge, nitrate-nitrogen load more than agricultural management changes. J. Environ. Qual. 39(6): 2063-2071.

NRCS. 1994. MUUF v. 2.13 User s Manual. Ames, Iowa: USDA Natural Resources Conservation Service.

Petrolia, D. R., and P. H. Gowda. 2006a. Missing the boat: Midwest farm drainage and Gulf of Mexico hypoxia. Rev. Agric. Economics 28(2): 240-253.

Petrolia, D. R., and P. H. Gowda. 2006b. An analysis of the role of tile-drained farmland under alternative nitrogen abatement policies. J. Agric. and Resource Economics 31(3): 580-594.

Randall, G. W. 1983. Nitrogen loss to tile drains as affected by tillage. In A Report on Field Research in Soils , 112-113. Miscellaneous Publication 113. St. Paul, Minn.: University of Minnesota, Minnesota Agricultural Experiment Station.

Randall, G. W., and T. K. Iragavarapu. 1995. Impact of long-term tillage systems for continuous corn on nitrate leaching to tile drainage. J. Environ. Qual. 24(2): 360-366.

Randall, G. W., D. R. Huggins, M. P. Russelle, D. J. Fuchs, W. W. Nelson, and J. L. Anderson. 1997. Nitrate losses through subsurface tile drainage in conservation reserve program, alfalfa, and row crop systems. J. Environ. Qual. 26(5): 1240-1247.

Sands, G. R., C. X. Jin, A. Mendez, B. Basin, P. Wotzka, and P. H. Gowda. 2003. Comparing subsurface drainage flow prediction of the DRAINMOD and ADAPT models for a cold climate. Trans. ASAE 46(3): 645-656.

Skaggs, R. W. 1980. A water management model for artificially drained soils. Tech. Bulletin No. 267. Raleigh, N.C.: North Carolina State University, North Carolina Agricultural Research Service.

Skaggs, R. W. 1982. Field evaluation of a water management simulation model. Trans. ASAE 25(3): 666-674.

Sogbedji, J. M., and G. F. McIsaac. 2006. Evaluation of the ADAPT model for simulating nitrogen dynamics in a tile-drained agricultural watershed in central Illinois. J. Environ. Qual. 35(5): 1914-1923.

Updegraff, K., P. H. Gowda, and D. J. Mulla. 2004. Watershed-scale modeling of the water quality effects of cropland conversion to short-rotation woody crops. Renewable Agric. and Food Systems 19(2): 1-11.

Viessman, W., Jr., G. L. Lewis, and J. W. Knapp. 1989. Introduction to Hydrology. 3rd ed. New York, N.Y.: Harper and Row.

Ward, A. D., E. Desmond, N. R. Fausey, T. J. Logan, and W. G. Logan. 1993. Development studies with the ADAPT water table management model. In Proc. 15th Intl. Congress on Irrigation and Drainage , 235-245. New Delhi, India: International Commission on Irrigation and Drainage (ICID).

Willmott, C. J. 1981. On the validation of models. Physical Geography 2(2): 184-194.