| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Dep. of Natural Resources and Environmental Sci., Univ. of Illinois at Urbana-Champaign, W503 Turner Hall, 1102 South Goodwin Ave., Urbana, IL 61801 (Senior Author now at the School of Agronomy, Université de Lomé, B.P.1515 Lomé, Togo, West Africa)
* Corresponding author (gmcisaac{at}uiuc.edu)
Received for publication October 3, 2005.
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONSUMMARY AND CONCLUSIONSREFERENCES |
|---|
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONSUMMARY AND CONCLUSIONSREFERENCES |
|---|
Nitrate concentrations in Illinois surface waters have been a concern since the late 1960s (Kohl et al., 1971). Since 1980, the NO3-N concentrations in several major drinking water reservoirs in central Illinois (e.g., Danville, Decatur, Pontiac) have periodically exceeded the safe drinking water standard (10 mg N L–1) for NO3-N that was set to prevent incidence of methemoglobinemia (blue baby syndrome). In recent years, several additional concerns about the effects of NO3-N in surface waters have been identified: hypoxia in coastal marine ecosystems and non-Hodgkin's Lymphoma (Townsend et al., 2003). Therefore, N management practices policies are needed to reduce the ecological and health risks associated with NO3-N leached from tile-drained croplands. Identifying such management practices requires that the fate and transport of N in the soil-plant-atmosphere system be well understood.
Several simulation models have been developed to simulate the fate and transport of agricultural N. Examples of such models are GLEAMS (Groundwater Loading Effects of Agricultural Management Systems) (Leonard et al., 1987), LEACHM (Leaching Estimation And Chemistry Model) (Hutson and Wagenet, 1992), RZWQM (Root Zone Water Quality Model) (USDA-ARS, 1992), and DRAINMOD-N (Drainage Model for Nitrogen) (Brevé et al., 1997a; Brevé et al., 1997b). Most of these models are root zone models that do not address the processes below the root zone where fluctuating water table levels, subsurface drainage, and deep seepage might occur. In addition, these models were developed to simulate plot to field-scale processes, and, therefore, their use on a watershed scale may be limited by lack of equations that describe stream channel processes or lack of information about the location and characteristics of existing tile drainage systems. Furthermore, most of the currently available N models have received only limited independent evaluation, and more testing is needed to ensure model reliability.
Sogbedji and McIsaac (2002a, 2002b) demonstrated that the field-scale model ADAPT (Agricultural Drainage And Pesticide Transport) provided reasonably good simulations of monthly and weekly stream flow from watersheds ranging from 500 to 1500 km2 in central Illinois. Thus, field-scale models of tile drainage and chemical transport processes may be capable of representing the hydrology and water quality of large, relatively uniform areas of the Midwest that have extensive tile drainage.
The ADAPT model (Chung et al., 1992) is a daily time-step, field-scale, water table management model developed by extension and integration of GLEAMS, a root zone water quality simulation model, with DRAINMOD (Skaggs, 1978), a subsurface drainage model. In ADAPT, parts of GLEAMS and DRAINMOD were combined with algorithms the authors had selected to account for snowmelt, deep seepage, and preferential flow. The model has four components: hydrology, erosion, nutrient, and pesticide transport. Equations and descriptions of the processes in the nutrient component of the model (which is evaluated in this study) are presented in Leonard et al. (1987) and Desmond et al. (1995).
Gowda et al. (2002) applied ADAPT to a 365-ha watershed in Iowa and reported that the model accounted for 74% of the variation of monthly nitrate flux in the calibration period and 50% of variation during the validation period. Mulla et al. (2003) and Gowda et al. (2005) reported that the ADAPT model accounted for 70 and 67% of the variation of observed April to October nitrate flux in two different watersheds in southern Minnesota. Because of data limitations, this analysis did not include observations from November through March and involved calibration without validation on an independent data set.
Desmond et al. (1995) evaluated the nutrient component of the ADAPT model in a plot-scale study in Ohio, and found that a modification of the denitrification algorithms was needed to improve the model's prediction of NO3-N concentrations. Desmond and Ward (2000) incorporated a denitrification reduction factor in the model, which allows for adjusting the denitrication process (increasing or decreasing denitrification losses) to improve the model's predictions. Davis et al. (2000) reported on plot-scale simulations of continuous corn production in Minnesota.
The objective of this study was to evaluate the capability of ADAPT for predicting annual, monthly, weekly, and daily river NO3-N load from a 1500-km2 central Illinois watershed. In addition, we investigated the capability of the model for simulating crop yield (harvest N) and used the simulation results to establish partial field N budgets.
|
MATERIALS AND METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONSUMMARY AND CONCLUSIONSREFERENCES |
|---|
|
|
Nitrogen content of the harvested crops was estimated to be 1.3% for maize and 6% for soybean (Burkhart and James, 1999). Nitrogen in livestock manure in each county was estimated from the annual inventory of cattle and hogs reported by the Agricultural Statistics Service Bulletin published jointly by the Illinois and US Department of Agriculture (e.g., Illinois Department of Agriculture and US Department of Agriculture, 1995). Numbers of other livestock (e.g., poultry) in these counties were insignificant. Annual N excretion was estimated using daily N excretion values published in the National Resource Conversion Service Agricultural Waste Management Field Handbook (Barth et al., 1992). Waste production depends on animal type, size, and age. We assumed that average annual cattle weight was 360 kg and the average hog weight was 64 kg. This led to an estimate of 40 kg N yr–1 excreted for each cattle (beef or dairy) and 10 kg N yr–1 excreted for each pig or hog. The estimated manure N application for the watershed was averaged at 6 kg N ha–1 yr–1.
Annual maize and soybean harvest in each county was obtained from the National Agricultural Statistics Service (USDA National Agricultural Statistics Service, 2002). County-level crop harvest data and animal manure N estimates were divided by total county area to obtain estimates of N quantities per hectare in the county. Since the watershed covered portions of more than one county, weighted average values for N harvest and manure for the watershed were calculated based on the proportion of the watershed that was in the county.
Fertilizer N input to the watershed was estimated from county-level fertilizer sales data and from a variety of surveys. The National Agricultural Statistics Service (NASS) and the Illinois Department of Agriculture conduct annual surveys of fertilizer application rates used on maize, soybeans, and wheat fields in Illinois. In most years, data are collected from approximately 200 fields, and results are reported as state-level averages. In some years (1992 to 1994) approximately 500 fields were surveyed, which allowed reliable averages to be calculated for four different portions of Illinois. Surveys of N fertilizer application rates on maize were also conducted by the Soil and Water Conservation District (SWCD) and Natural Resource Conservation Service (NRCS) personnel in specific counties of east central Illinois in 1995, 1996, 1999, and 2000.
State-level fertilizer use surveys (USDA National Agricultural Statistics Service, 2002) indicated that in the 1990s approximately 90 kg N ha–1 yr–1 was applied to wheat, 2 kg N ha–1 yr–1 was applied to soybeans, and 175 kg N ha–1 yr–1 was applied to maize. The county-level surveys conducted by the SWCD and NRCS personnel in east central Illinois (Champaign County Soil and Water Conservation District, unpublished data, 1996) indicate that the average fertilizer application rates for maize were approximately 180 to 200 kg N ha–1 yr–1. Maize yields in the central and eastern portions of the state are greater than the state average, and thus fertilizer application rates are also likely to be higher than the state average. Because of uncertainties and inconsistencies in the county-level N fertilizer surveys, we also examined county-level fertilizer sales data.
Semi-annual N fertilizer sales at the county level were obtained from the Illinois Commercial Fertilizer Tonnage Report (e.g., Illinois Department of Agriculture, 1996). County-level fertilizer sales data includes fertilizer sold for all purposes and does not identify how much fertilizer is applied to different crops, or how much sold in one county is applied in another. We used fertilizer sales data from the seven counties that constitute the east-central crop-reporting district to estimate fertilizer application rates in the watershed. We assumed that the net fertilizer transport across the borders of this large area was a negligible portion of the total fertilizer use. To account for fertilizer applied to soybeans and wheat, we subtracted 2 kg N ha–1 for each soybean acre and 90 kg N ha–1 for each wheat acre planted in the region. This represented approximately 5% of the fertilizer sold in the district. We assumed that any other N fertilizer sold and used for other purposes, and other sources of N applied to maize such as sewage or industrial waste products, was negligible. The annual mass of fertilizer N sold in the district (minus the estimated application to soybeans and wheat) divided by the number of hectares of maize planted in the corresponding year ranged from 165 to 220 kg N ha–1 yr–1. The variability in annual estimates was high, partly because of inconsistencies in reporting frequency by some fertilizer dealers. To address the problem of late sales reports, we used a 3-yr moving average value for each year. This resulted in greater consistency of fertilizer sold per hectare of corn planted, which ranged from 185 to 205 kg N ha–1 yr–1 These estimated values were also similar to results of local surveys.
Model Calibration
We calibrated the ADAPT model to fit river NO3-N load determined from USGS streamflow and NIWC NO3-N concentration data collected from 1989 to 1993. This time period was selected for calibration because annual rainfall and stream flow were highly variable, which ensured the model was calibrated to a wide range of rainfall conditions.
We calibrated the model by adjusting the following input parameters in the following order: (1) denitrification reduction factor, (2) coefficient in the exponential function for crop N uptake, (3) dry matter ratio, (4) potentially mineralizable N, and (5) initial NO3-N concentration in soil. The last three parameters were calibrated within the range of values reported in relevant literature because they were not explicitly measured in this study. The denitrification reduction factor, which was incorporated into ADAPT by Desmond and Ward (2000), was adjusted to vary the amount of denitrification simulated in the soil and to improve the model's riverine NO3-N flux estimates. Leonard et al. (1987) reported that GLEAMS (which was combined with DRAINMOD to develop ADAPT) is very sensitive to the coefficient in the exponential function for crop N uptake when N fertilizer is applied to croplands.
The ADAPT model uses potential crop yield as an input parameter, and the simulated maize yield is based on estimation of environmental stress factors that can prevent the potential yield from being achieved. For soybeans, no environmental stress functions were programmed into the model and thus simulated soybean yield was identical to the potential yield. In this study, we used annual crop yield estimates from the county-level NASS annual reports as the potential crop yields. In most years, these yields reflect some environmental limitations, and therefore, this approach would tend to underestimate potential yield.
To simulate the watershed being approximately equally divided between soybean and maize production each year, two simulation scenarios in a soybean-maize rotation were used. One scenario began with soybean and the other began with maize. For a particular time period, the simulated nitrate N load (sum of NO3-N in surface runoff, tile drainage, and deep seepage) from the watershed was the average value from the two scenarios, and represented the ADAPT simulated average edge-of-field losses.
The ADAPT-simulated average edge-of-field losses were then adjusted to account for the in-stream losses of nitrate (i.e., denitrification and utilization by biota) using a first-order stream NO3-N decay approach (Whitehead and Toms, 1993) to give the simulated river NO3-N load for the watershed. The first order decay constant was determined by nonlinear fitting the following equation:
 | [1] |
The coefficients in the above model were estimated using the IEPA concentrations for 1978 to 1996 for the VP watershed. All coefficients were statistically significant at the 0.05 level of confidence, and accounted for 83% of the variation in nitrate concentration over the period. When the model was used to estimate daily concentrations, it accounted for 70% of the variation in NIWC nitrate concentrations. The exponential term in Eq. [1] provided an estimate of the fraction of edge-of-field nitrate losses that were lost due to in-stream processes for given conditions of flow and temperature. This in-stream loss term was calculated for each day and multiplied by the ADAPT estimate of edge-of-field nitrate losses, to simulate riverine nitrate fluxes after in-stream losses. The estimated fraction of nitrate transformed by in-stream processes was 0 to 5% during high flow events, but as high as 100% during summer low flows. On an annual basis, the average quantity of nitrate lost to in-stream processes estimated by this technique was approximately 15%, which is somewhat greater than in-stream denitrification values reported by Royer et al. (2004).
The ADAPT calibration consisted of performing multiple runs of the model for a 7-yr period (1987 to 1993) and varying the selected input parameters within a range of published values (Table 2) to achieve the closest agreement between predicted and measured values of cumulative, annual, monthly, weekly, and daily NO3-N load for the last 5-yr period (1989 to 1993). We did not consider the simulation output for the first 2 yr (1987 and 1988) to minimize the influence of the estimating initial soil conditions. To assess the accuracy of simulations, we used graphical and statistical methods (Loague and Green, 1991; Willmott, 1981; Addiscott and Whitmore, 1987). Simulated annual, monthly, weekly, and daily flow values were plotted against the corresponding measured values on a 1:1 scale. We assumed a linear relationship between measured and simulated data and a normal distribution of the data sets, and used PROC REG and t test procedures of the SAS software package (SAS Institute, 2002) to conduct least square regression analysis and to compare the slope and intercept to 1.0 and 0.0, respectively. The root mean square error (RMSE) was compared to the mean measured value to determine the prediction error. The statistical methods also included comparison of measured and simulated mean values and their standard deviation, and calculation of Willmott's index of agreement (d). The value of d reflects the degree to which the simulated variation accurately estimates the measured variation, and its value is 1.0 when there is a perfect agreement between simulated and measured values. We also performed a comparison between measured and simulated crop yield values.
We used the simulated N output data to develop a partial-field N budget for the 5-yr period in which the inputs were N applied in fertilizer, simulated fixed N, and N in rain, and outputs were simulated harvest N (N in grain yield), simulated NO3-N load (without adjustment to account for in-stream denitrification loss), and simulated in-field denitrified N. The balance of inputs and outputs may be attributed to mineralized or immobilized N.
Model Testing
The performance of the calibrated model was tested using river NO3-N load data determined from the USGS streamflow, and NIWC and IEPA NO3-N concentration data collected for a 4-yr period (1994 to 1997) for the VP watershed that were not involved in the calibration. However, we performed the simulations for the 1992 to 1997 period and, as with the calibration, did not include the first 2 yr (1992 and 1993) in the evaluation to minimize the effects of estimating initial conditions. The calibrated model was executed without any change to the calibrated input parameter values. For the 1994 to 1997 period, simulated cumulative, annual, monthly, weekly, and daily NO3-N load values were compared to measured values. The graphical and statistical methods described in the calibration section were used for evaluating the model simulation accuracy. Measured and simulated crop yield data were compared, and a partial field N budget was established as described in the calibration section.
|
RESULTS AND DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONSUMMARY AND CONCLUSIONSREFERENCES |
|---|
= 0.05.
|
|
= 0.05. The greatest prediction errors in monthly simulations occurred in January, March, and May (Fig. 2 ). The large discrepancy in January 1991 primarily resulted from heavy precipitation that occurred on 29 and 30 December 1990 (9.6 cm of rainfall in the 2 d) that flushed soil NO3-N (from residual NO3-N and fall N fertilizer application) out of the profile. The other discrepancies might result from the difficulties of simulating snow melt or hydrology during freeze-thaw cycles very well in some years (Sogbedji and McIsaac, 2002a). Additionally, in the simulations, N fertilizer was applied on two specific calendar days (one in the spring and one in the fall), and each crop was planted for the whole watershed in one day. In the actual watershed, these farming operations occur over several weeks with some variation from one year to the next, which the simulations did not represent.
|
Daily NO3-N load simulations were slightly less accurate than weekly simulations. Measured and simulated daily load values were reasonably well distributed around the 1:1 line (Fig. 1). Mean daily values, as well as associated standard deviations were identical (Table 3). RMSE value was identical to the measured mean value, indicating a 100% prediction error. However, the slope value (0.9) was not statistically different from 1.0 at = 0.05, the intercept value was 0.0 kg, the r2 value was 0.73 and the d value was 0.92, indicating that there was reasonably good agreement between measured and simulated daily load. The lower accuracy of daily load simulations compared with weekly and monthly simulations is partly a result of lower accuracy of daily water flow simulations (Sogbedji and McIsaac, 2002a). Surprisingly, however, daily nitrate load simulations were more accurate than daily water flow simulations.
Crop Nitrogen Harvest
Calibration resulted in satisfactory simulations of soybean N harvest. Simulated N harvest did not significantly deviate from measured values when plotted on a 1:1 scale (Fig. 3
). Measured and simulated annual mean values differed by 2.1% (mean difference = 3.2 kg N ha–1) and their standard deviations were similar (Table 3). Values for r2 and d were 0.98 and 0.99, respectively, and the RMSE value was 5.1 kg, leading to only a 3.2% prediction error for the measured mean value. The slope value was 1.0 and the intercept was –3.5 kg N ha–1, which was not statistically different from 0.0 kg at = 0.05.
|
= 0.05.
Partial Field Nitrogen Budget
A partial balance of N inputs and outputs indicated that a net of 53 kg N ha –1 yr–1 had accumulated in the soil during the 5-yr calibration period (Table 4). This simulated accumulation of N was largely due to a relatively large value of simulated biological fixation associated with soybeans: 248 to 280 kg N ha–1 yr–1 compared with rates of 80 to 100 kg N ha–1 yr–1 determined by Johnson et al. (1975), and more recent estimates of 100 kg N ha–1 yr–1 for Illinois conditions (Hoeft and Peck, 2002). In ADAPT, fixation of N by legumes is estimated by assuming a constant fixation rate after the soil inorganic N drops below a threshold value of 5 mg L–1 and fixation continues for the rest of the growing season, regardless of whether soil inorganic N increases. Improvement of the N fixation algorithms in the model to account for soil N fluctuations during the whole crop growth period is needed to allow for a better simulation of the N fixation process, and, consequently, a better estimate of the field N budget. In spite of the overestimation of biological N fixation, it was surprising that the model calibration suggested that there was relatively little denitrification occurring in the fields. In an empirical analysis of long-term N budgets for central Illinois, McIsaac and Hu (2004) found that riverine N flux in tile-drained watersheds was approximately equal to the sum of fertilizer, fixation, and manure N, minus the crop N harvest. If soil organic N was constant, as is often assumed after approximately 60 yr of continuous cultivation, then there would appear to be little N available for denitrification in these fields. Further research is needed to verify these aspects of the N cycle.
|
= 0.05.
|
|
= 0.05 (Table 5). Simulated and measured mean values were close to each other (mean difference value of 0.2 kg N ha–1 yr–1) and corresponding standard deviation was nearly identical. The prediction error was 36%, which was similar to the 40% value for the calibration data sets. The r2 (0.89) and d (0.97) values were greater than values obtained for the calibration period (r2 value of 0.83 and d value of 0.95). Simulations of weekly river NO3-N load were reasonably satisfactory (Fig. 4). Measured and simulated mean values, as well as the corresponding standard deviations, were similar (Table 5). The prediction error for the measured mean value was 80%, which was higher than the 62.5% value for the calibration data sets. The r2 value (0.83) and d value (0.95) were greater than values obtained in the calibration period (r2 value of 0.78 and d value of 0.94), and the slope and intercept were 1.0 and 0.0 kg N ha–1 yr–1, respectively.
Measured and simulated daily NO3-N load values agreed reasonably well (Fig. 4). The measured and simulated daily mean load values, as well as their standard deviations, were identical (Table 5). The RMSE values were identical to the measured mean value, leading to a 100% prediction error. The r2 and d values were 0.73 and 0.91, respectively, and were similar to values for the calibration period (r2 value of 0.73, and d value of 0.92). The slope was 1.0 and the intercept was 0.0 kg N ha–1 yr–1.
Crop Nitrogen Harvest
Simulations of crop N harvest during model testing were fairly satisfactory (Fig. 5
). For soybean, average measured and simulated crop N harvests were within 2% (Table 5). The RMSE values (5.5 kg N ha–1) led to a 3.3% prediction error, which was similar to the 3.2% value for the calibration period. The r2 value was 0.71 versus the 0.98 value for the calibration period, and the d value was 0.91 versus the 0.99 values for the calibration period. The slope value (1.1) and intercept value (–27.1 kg ha–1) were statistically different from 1.0 and 0.0 kg, respectively, at = 0.05.
|
= 0.05.
Partial Nitrogen Budget
During the 1994 to 1997 period, the model estimated an average of 72 kg N ha–1 yr–1 accumulated in the soil (Table 6), which was higher than the 53 kg N ha–1 yr–1 estimated during model calibration period. Simulated soybean N fixation was approximately 250 kg N yr–1 for each hectare of soybean, which is more than twice as large as that reported for the region (Hoeft and Peck, 2002). With a more realistic estimate of soybean N fixation (50 kg N ha–1 yr–1 for the watershed), the sum of N inputs would approximately equal the outputs, which would be consistent with soil organic N at steady state.
|
Related Studies
The correspondence between simulated and measured riverine NO3-N flux in our study appears to be superior to that reported by Gowda et al. (2002, 2005) for watersheds in Iowa and Minnesota. For April through October, monthly riverine NO3-N fluxes in two Minnesota watersheds, Gowda et al. (2005) reported indices of agreement of 0.88 and 0.90 and r2 values of 0.7 and 0.67. For monthly riverine NO3-N fluxes for an Iowa watershed (full year), Gowda et al. (2002) reported indices of agreement of 0.89 and 0.80 for their calibration period and validation period, respectively. Coefficients of determination were 0.71 and 0.50 for calibration and validation, respectively. The superior results in our study may result from larger absolute variations in stream flow and NO3-N flux, more uniform soils and land use, and less snow melt.
|
SUMMARY AND CONCLUSIONS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONSUMMARY AND CONCLUSIONSREFERENCES |
|---|
Annual river NO3-N load simulations were accurate during the calibration and testing periods. Monthly, weekly, and daily river NO3-N load simulations were satisfactory. Discrepancies between measured and simulated NO3-N-related data sets might have partly resulted from overestimation of soybean N fixation and from the fact that the model does not account for ammonia volatilization from N fertilizer in the soil. Discrepancies also presumably resulted from representing the spatial variability of the watershed as a single representative field in which fertilizer applications and planting operations occurred on a single day for the whole watershed, and areas in perennial vegetation in the watershed were not simulated.
Simulations of crop N harvest were fairly satisfactory, although observed yields were used as input for defining potential crop yield. For soybeans, the model does not simulate plant response to the environment and to soil conditions with sufficient detail for physiologically realistic simulation of fixation or crop harvest. This would seem to limit the use of the model to situations in which soybean yields are known. Results of the partial N field budgets showed a high level of N accumulation in the soil primarily due to the model's tendency to overestimate soybean N fixation.
The ADAPT model proved to be capable of simulating river NO3-N load. Incorporation of mechanisms that describe ammonia volatilization from N fertilizer in the soil and perennial vegetation N uptake may improve the model's capability of predicting N fate and transport at the watershed scale. Improvement of crop growth simulation, especially soybean N fixation, appears to be needed to allow for realistic crop response to changes in environment and soil conditions and to more realistically simulate N cycling.
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONSUMMARY AND CONCLUSIONSREFERENCES |
|---|
This article has been cited by other articles:
|
|
 |
|
  V. Nangia, P. H. Gowda, D. J. Mulla, and G. R. Sands Water Quality Modeling of Fertilizer Management Impacts on Nitrate Losses in Tile Drains at the Field Scale J. Environ. Qual., March 1, 2008; 37(2): 296 - 307. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| The SCI Journals | Agronomy Journal | Crop Science | |||
| Journal of Natural Resources and Life Sciences Education |
Vadose Zone Journal | ||||
| Soil Science Society of America Journal | Journal of Plant Registrations | The Plant Genome | |||
