HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS		QUICK SEARCH:   [advanced] Author: Keyword(s): Year:  Vol:  Page:

This Article Abstract Figures Only Full Text (PDF) Free Alert me when this article is cited Alert me if a correction is posted Services Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via HighWire Citing Articles via Google Scholar Google Scholar Articles by Youssef, M. A. Articles by Gilliam, J. W. Search for Related Content PubMed PubMed Citation Articles by Youssef, M. A. Articles by Gilliam, J. W. Agricola Articles by Youssef, M. A. Articles by Gilliam, J. W. Related Collections Pesticides Agricultural Pesticides

TECHNICAL REPORTS

Ground Water Quality

Field Evaluation of a Model for Predicting Nitrogen Losses from Drained Lands

^a Department of Biological and Agricultural Engineering., North Carolina State University, Raleigh, NC 27695
^b Department of Soil Science, North Carolina State University, Raleigh, NC 27695

^* Corresponding author (mohamed_youssef{at}ncsu.edu)

Received for publication June 22, 2005.

	ABSTRACT

TOP NOTES ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS AND DISCUSSION CONCLUSIONS REFERENCES

 
The N simulation model, DRAINMOD-N II, was field-tested using a 6-yr data set from an artificially drained agricultural site located in eastern North Carolina. The test site is on a nearly flat sandy loam soil which is very poorly drained under natural conditions. Four experimental plots, planted to a corn (Zea mays)–wheat (Triticum aestivum L.)–soybean (Glycine max.) rotation and managed using conventional and controlled drainage, were used in model testing. Water table depth, subsurface drainage, and N concentration in drain flow were measured and meteorological data were recorded continuously. DRAINMOD-N II was calibrated using the data from one plot; data sets from the other three plots were used for model validation. Simulation results showed an excellent agreement between observed and predicted nitrate-nitrogen (NO₃–N) losses in drainage water over the 6-yr period and a reasonable agreement on an annual basis. The agreement on a monthly basis was not as good. The Nash-Sutcliffe modeling efficiency (EF) for monthly predictions was 0.48 for the calibration plot and 0.19, 0.01, and –0.02 for the validation plots. The value of the EF for yearly predictions was 0.92 for the calibration plot and 0.73, 0.62, and –0.10 for the validation plots. Errors in predicting cumulative NO₃–N losses over the 6-yr period were remarkably small; –1.3% for the calibration plot, –8.1%, –2.8%, and 4.0% for the validation plots. Results of this study showed the potential of DRAINMOD-N II for predicting N losses from drained agricultural lands. Further research is needed to test the model for different management practices and soil and climatological conditions.

	INTRODUCTION

TOP NOTES ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS AND DISCUSSION CONCLUSIONS REFERENCES

 
DESPITE BEING THE FOCUS OF RESEARCH for many years, developing management practices that minimize the negative environmental impacts of crop production remains a challenge (Meisinger and Delgado, 2002; Dinnes et al., 2002; Gilliam et al., 1999). This complex task requires understanding N dynamics in the soil–water–plant system, which is regulated by a large number of interacting and sometimes highly dynamic physical, chemical, and biological processes. Relying only on experimentation to investigate the long term effects of different management scenarios on N fate and transport at the field scale for different soil types and climatic conditions is time-consuming and expensive. Computer models such as RZWQM (Ahuja et al., 2000), SOILN (Johnsson et al., 1987; Eckersten et al., 1998), ANIMO (Groenendijk and Kroes, 1997), DAISY (Hansen et al., 1993), CERES-N (Godwin and Jones, 1991), and LEACHM (Hutson, 2000) can be utilized to study N dynamics in agro-ecosystems when the experimental approach proves to be impractical. In fact, computer modeling and experimentation complement each other and, if used together, can lead to better understanding of N dynamics in agro-ecosystems, and thus improve the process of development and evaluation of best management practices for sustainable agriculture.

DRAINMOD-N II (Youssef, 2003; Youssef et al., 2005) is a field-scale, process-based model that was developed to simulate carbon and N dynamics in drained agricultural lands for a wide range of soils, climates, and management practices. Once validated, it can be used for the development and evaluation of methods that reduce N losses from drained croplands. DRAINMOD-N II simulations can be conducted to study the long term effects of drainage design and management, cropping systems, N fertilizer management, tillage and plant residue management, and their interactions on N losses from drained croplands for different soils and climates. This paper reports the first validation of DRAINMOD-N II using a 6-yr data set from an artificially drained agricultural research site located in eastern North Carolina.

	MATERIALS AND METHODS

TOP NOTES ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS AND DISCUSSION CONCLUSIONS REFERENCES

 
Predicting N fate, at the field scale, using DRAINMOD-N II is a two-step process. The water management model, DRAINMOD (Skaggs, 1978), first simulates water and heat flow and then DRAINMOD-N II simulates N transport and transformations. In the next two sections, both DRAINMOD and DRAINMOD-N II are briefly described. A detailed description of both models can be found elsewhere (Skaggs, 1999; Youssef et al., 2005).

The Water Management Model, DRAINMOD
DRAINMOD (Skaggs, 1978) was developed for design and evaluation of agricultural drainage and related water management systems. It uses functional algorithms to approximate the hydrologic components of shallow water table soils. It conducts a water balance on a day-by-day, hour-by-hour basis and calculates infiltration, evapotranspiration (ET), subsurface drainage, surface runoff, subirrigation, deep seepage, water table depth, and soil water status at each time step.

The model calculates infiltration using the Green–Ampt equation. Daily potential ET (PET) can be calculated by the Thornthwaite method or computed by the method of user's choice (e.g., Penman–Monteith) and read by the model as input data. Subsurface drainage is quantified using the Hooghoudt's equation for water table drawdown and Kirkham's equations for ponded surface conditions. Surface drainage is characterized by identifying a depressional storage that must be filled with water before runoff can begin (Skaggs, 1999).

The model numerically solves the heat flow equation and predicts soil temperature to simulate processes controlling field hydrology under cold conditions such as freezing, thawing, and snowmelt (Luo et al., 2000). It includes approximate methods to predict the effects of planting date delay, excessively wet or dry conditions, and soil salinity on relative crop yield (Evans et al., 1991). Over the past two decades, DRAINMOD has been extensively tested and proven to be a reliable model for simulating the hydrologic processes in artificially drained, high water table soils (Skaggs et al., 1981; Skaggs, 1982; Gayle et al., 1985; Rogers, 1985; Fouss et al., 1987; Cox et al., 1994).

The Nitrogen Model, DRAINMOD-N II
DRAINMOD-N II (Youssef, 2003; Youssef et al., 2005) is a field-scale, process-based model that simulates C and N dynamics in drained agricultural lands for a wide range of soil types, climatic conditions, and management practices. The model simulates a detailed N cycle (Fig. 1) that includes three soil N pools: nitrate-nitrogen (NO₃–N), ammoniacal-nitrogen (NH_x–N), and organic nitrogen (ON). The NH_x–N pool is set to be optional so it may be ignored if soil and environmental conditions do not favor its accumulation in the soil. Nitrogen processes and transformations considered in the model include atmospheric deposition, application of mineral N fertilizers and ON sources (plant residues/animal waste), plant uptake, N mineralization/immobilization, nitrification, denitrification, ammonia (NH₃) volatilization, and NO₃–N and NH_x–N losses via subsurface drainage and surface runoff (Youssef et al., 2005).

View larger version (20K): [in this window] [in a new window]   Fig. 1. The N cycle as modeled in DRAINMOD-N II.

View larger version (17K): [in this window] [in a new window]   Fig. 2. The carbon cycle as modeled in DRAINMOD-N II (MET = metabolic; STR = structural; LGN = lignin; CEL = cellulose; MCR = microbe; SURF = surface; ACT = active; SLO = slow; PAS = passive).

DRAINMOD-N II simulates the effects of tillage on C and N dynamics using a tillage intensity factor that reflects the level of soil disturbance caused by tillage operations. At the day of harvest, the model can return plant residues back to the system and update litter pools accordingly. The model can also simulate the application of animal waste to the soil–plant system. The mineral N portion of the animal waste is handled by the fertilizer component of the model. The organic portion of the animal waste is handled by the SOM component of the model.

The model simulates urea hydrolysis, nitrification, and denitrification using Michaelis-Menten kinetics. The effect of nitrification inhibitors on the nitrification process is simulated using a response function that modifies process rate according to inhibitor concentration which declines according to a first-order decay rate. The influence of OC decline with depth on the denitrification process is implicitly simulated using an exponential depth function that reduces potential process rate with soil depth (Youssef et al., 2005).

In DRAINMOD-N II the potential N uptake for each crop is computed (Youssef et al., 2005) and empirically distributed over the days of the growing season (Brevé et al., 1997a). Both NO₃ and NH₄ are assumed to be equally available to plants. If one form is used up before satisfying plant needs, the rest is taken from the other form. If total mineral N is insufficient and the crop is not a legume, actual N uptake is set to the available amount of mineral N. Legumes are assumed to fix atmospheric N only when a shortage in soil mineral N occurs (Youssef et al., 2005).

Model inputs include soil properties, crop and management parameters, and C and N processes and transformations parameters. Model output includes NO₃–N and NH_x–N concentrations in soil solution and drain flow, OC content of the top 20-cm soil layer, and rates of simulated N processes on daily, monthly, and annual bases.

The Field Study
The performance of DRAINMOD-N II was evaluated using a 6-yr (1992 to 1997) data set from an artificially drained agricultural research site located at the Tidewater Experiment Station (TES) near Plymouth, in the North Carolina lower coastal plain. The nearly flat, 13.8-ha site is divided into eight experimental field plots, four of which were used in this testing (Fig. 3).

View larger version (38K): [in this window] [in a new window]   Fig. 3. General layout of the research site at the Tidewater Experiment Stn., Plymouth, NC.

Two sets of parallel subsurface drains were installed on the site in 1985 and 1991. The drains in the first set were 0.8 to 1.0 m below ground surface and spaced 23 m apart. Drains in the second set were located midway between those in the first set, but at deeper depths of 1.1 to 1.3 m (Fig. 3). Each plot has six 10-cm diameter drains which outlet to an instrument house. Valves on drain outlets allowed each plot to be drained using the deep or the shallow drainage system. The center drain line of each drain set is instrumented to continuously measure drain flow rate and to provide flow-proportional water quality samples, collected biweekly or more frequently during high rainfall events. The outer drains on each plot serve as guard lines to hydraulically isolate the area drained by center drain line from the influence of adjacent experimental plots. Instrumentation in each house allows the individual field plots to be managed in conventional drainage, controlled drainage, or subirrigation modes. During 1992 to 1997, plots 2 through 5, which were used in this testing, were drained using the deep drainage system, managed in either conventional or controlled drainage (Table 1) (Munster et al., 1994; Chescheir et al., 1996).

Crop rotation on the study site consisted of three crops in 2 yr: corn, year 1, wheat, years 1 and 2, and soybean, year 2. This cropping sequence is common on many farmlands of the North Carolina lower coastal plains. Both conventional tillage and no-till practices were implemented on the site at different times during the observation period (Table 2). Conventional tillage operations included: one or two passes with a disk harrow, one or two passes with a chisel, and then bedding. Nitrogen fertilization followed common practices in the region with some exceptions for experimental reasons. Two N fertilization rates were used during some years as part of an experiment to study the effect of N fertilization on crop yield and N leaching losses (Table 2). Nitrogen fertilizer, mainly urea-ammonium nitrate (30% N) solution, was applied only to corn and wheat. Dolomitic lime was applied to the site three times from 1991 to 1997 (15 Nov. 1991, 8 Nov. 1993, 18 Nov. 1997) to control pH of the topsoil. Lime application rates followed soil test report recommendations (Chescheir et al., 1996).

Driving hydrologic parameters of DRAINMOD-N II are based on the output of the water management model, DRAINMOD. Therefore, DRAINMOD performance is expected to have a direct impact on the performance of DRAINMOD-N II. For this study, which was the first field evaluation of DRAINMOD-N II, it was decided to calibrate DRAINMOD for each of the four experimental plots during the 6-yr period. This decision was made to minimize the errors in the hydrologic "predictions" and their impact on DRAINMOD-N II performance. Excluding DRAINMOD from the field testing is further justified by the fact that DRAINMOD has been extensively tested over more than two decades for a wide range of soils, crops, and climatological conditions (Skaggs, 1999).

DRAINMOD was manually calibrated by visually and statistically comparing observed and simulated water table depths and drainage volumes. Statistical performance measures used for comparing observed and simulated water table depths are the mean error, ME, the mean absolute error, MAE, and the Nash-Sutcliffe modeling efficiency, EF (Table 4). Observed and simulated subsurface drainage volumes were statistically compared using the EF and the percent-normalized error, NE (Table 4).

DRAINMOD-N II was calibrated using the 6-yr data from plot 3; data sets from the other three plots were used for model validation. The model was manually calibrated by visual and statistical comparison of predicted and observed NO₃–N losses with drain flow. The EF and NE statistical performance measures were used in model calibration/validation (Table 4).

Hydrologic Parameters
DRAINMOD inputs include climatological data, soil properties, crop parameters, and drainage system parameters (Skaggs, 1980; Workman et al., 1994). An additional set of inputs is required for simulating heat flow and predicting soil temperature (Luo et al., 2000).

Climatological data collected at the site were used to generate DRAINMOD hourly rainfall and daily PET input files. Daily PET was estimated by the Penman–Monteith method (Jensen et al., 1990). Table 5 lists measured monthly rainfall and estimated monthly PET for the study period.

Soil temperature parameters are required only if the model is set to simulate heat flow, which is necessary to accurately simulate field hydrology under cold conditions (Luo et al., 2000) and N dynamics (Brevé, 1994; Youssef, 2003). Soil temperature input parameters used in this simulation were based on values reported by Luo et al. (2000).

Nitrogen Parameters
DRAINMOD-N II inputs include soil, crop, management, N transport and transformations, and organic matter parameters.

Soil inputs of DRAINMOD-N II for the TES site, summarized in Table 8, were based on previous measurements (Munster, 1992; Brevé, 1994). Soil properties of the TES site were reported in detail by Brevé et al. (1997b). The NH₄⁺ distribution coefficient was estimated as a function of clay content (Knisel et al., 1993).

View larger version (14K): [in this window] [in a new window]   Fig. 4. Nitrogen uptake functions for crops grown on the TES site from 1992 through 1997.

Plant HI is the ratio of grain yield to the total above-ground biomass (Hay, 1995). The RSR is the mass ratio between root dry matter and shoot dry matter (Hoad et al., 2001). DRAINMOD-N II uses HI and RSR to estimate non-grain above-ground dry matter and below-ground dry matter from DRAINMOD-predicted or field-measured grain yield (Youssef et al., 2005).

Both HI and RSR were included in the initial phase of the calibration process. Calibrated values of the HI and RSR for wheat, soybean, and corn (Table 9) were within the published ranges for the three crops (Youssef, 2003). With one exception, each crop in the rotation was assigned the same HI and RSR throughout the simulation period. A smaller HI of 0.48 and a larger RSR of 0.16 were obtained for corn grown in 1993, which reflects the effects of the drought stresses occurred in June through August of that year (Table 5) on both parameters. A summary of measured grain N for crops grown on the site during the test period is given in Table 3. Published values of shoot N and root N for wheat, soybean, and corn (Youssef, 2003) were used to parameterize DRAINMOD-N II (Table 9).

Plant residue management parameters include two flags to indicate whether or not to recycle plant shoots and roots back into the system at the day of harvest. The model requires the biochemical composition of each recycled plant residue including N, C, and lignin contents. Biochemical composition inputs used in this study (Table 9) were based on values reported in the literature (Youssef, 2003). Nitrogen and lignin contents of both plant shoots and roots were adjusted within the published ranges during model calibration.

Transport parameters used in simulating N dynamics for the TES site are listed in Table 10. The longitudinal dispersivity was the only transport parameter included in model calibration. Transformation parameters include parameters for processes of fertilizer dissolution, urea hydrolysis, nitrification, denitrification, pH control, and volatilization.

Inputs for urea hydrolysis, nitrification, and denitrification include Michaelis–Menten kinetics parameters (maximum reaction rate and half-saturation constant) for each process. They also include parameters that define, for each process, the soil temperature, soil water, and soil pH response functions. If nitrification inhibitors are used, additional inputs are required to parameterize the inhibitor response function. Input parameters for urea hydrolysis, nitrification, and denitrification used in simulating N dynamics for the site (Table 11) were based on values reported in the literature (Youssef, 2003). The effect of soil pH on denitrification was ignored (Youssef, 2003). Model calibration was primarily conducted by adjusting nitrification and denitrification parameters.

Organic matter parameters include the C/N ratio and potential rate of decomposition of each OM pool, the environmental response function parameters for OC decomposition, the initial OC content, and the initial percentage of soil OC assigned to each SOM pool. The C/N ratios and potential rates of decomposition used in this study (Table 12) were similar to those that are frequently used with the CENTURY model (Parton et al., 1993). The C/N ratios of the structural pools stay constant during the whole simulation period. The C/N ratios of the metabolic pools and the surface microbial pool are functions of the biochemical composition of the applied organic materials. The C/N ratios of the active, slow, and passive SOM pools vary with soil mineral N (NO₃–N + NH₄–N) content. Since OC decomposition is more sensitive to soil temperature and soil water than soil pH, the effect of pH on process rate was ignored (Youssef et al., 2005). The optimum temperature and the shape factor of the temperature response function were set to 36.9°C and 0.186, respectively (Kirschbaum, 1995). Organic C decomposition was assumed to proceed with a maximum at rate 0.5 to 0.6 water-filled pore space (WFPS) (Linn and Doran, 1984; Doran et al., 1988) and proceed with 15 and 50% of the maximum rate at wilting point and saturation, respectively (Stanford and Epstein, 1974; Linn and Doran, 1984). Organic C content of the top 20-cm soil layer, measured in 1992, was used to initialize the model. It ranged from 3.2 to 4.4% for the four plots considered in the study (Munster, 1992). The model was pre-run for 15 yr with arbitrarily chosen initial pool distributions allowing them to "equilibrate" before simulating the period of interest (1992 to 1997). The resulting initial percentages of total OC assigned to the active, slow, and passive SOM pools were 1, 37, and 62%, respectively.

	RESULTS AND DISCUSSION

TOP NOTES ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS AND DISCUSSION CONCLUSIONS REFERENCES

View larger version (28K): [in this window] [in a new window]   Fig. 5. Observed and simulated water table depths for plot 3 from 1992 through 1994.

View larger version (29K): [in this window] [in a new window]   Fig. 6. Observed and simulated water table depths for plot 4 from 1992 through 1994.

As indicated by the relatively small ME values (Table 13), model bias in predicting water table depth was generally small. Over the simulated period, the ME values were 7.2 cm for plot 2, 0.7 cm for plot 3, –1.0 for plot 4, and 3.7 for plot 5 (Table 13). The positive, relatively large ME values for plots 2 and 5 indicate that the simulated water table tended to be shallower than the observed (Table 13).

Subsurface Drainage
Simulated and observed subsurface drainage volumes were also in agreement (Fig. 7, 8 and Table 14). Over the simulated period, the value of the EF comparing simulated and observed monthly subsurface drainage was 0.67 for plot 2, 0.64 for plots 3 and 4, and 0.65 for plot 5. The absolute values of NE comparing simulated and observed annual subsurface drainage were less than 5% in 10 out of the 24 simulated plot-years and less than 15% in 19 out of the 24 simulated plot-years (Table 14). The absolute value of NE comparing simulated and observed cumulative drainage over the 6-yr period was 2.9% for plot 2, 0.7% for plot 3, 4.2% for plot 4, and 13.2% for plot 5 (Table 14). The model, however, tended to slightly underpredict drainage. Inspection of drainage scatter diagrams (not shown) indicated that DRAINMOD underpredictions of drainage volumes mostly occurred during high flow events where the water table reached the soil surface.

View larger version (26K): [in this window] [in a new window]   Fig. 7. Observed and simulated subsurface drainage for plot 3 from 1992 through 1994.

View larger version (25K): [in this window] [in a new window]   Fig. 8. Observed and simulated subsurface drainage for plot 4 from 1992 through 1994.

Nitrogen
Since both field measurements and model predictions showed very small drainage losses of NH₄–N (less than 0.5 kg N ha^–1 yr^–1 on average, data are not presented), this parameter was excluded from model testing and no comparison was made between observed and predicted NH₄–N leaching losses.

Monthly Nitrate-Nitrogen Drainage Losses
Overall, predicted monthly NO₃–N drainage losses were not in good agreement with those observed. The monthly EF values were considerably larger for the calibration plot (plot 3) than the three validation plots. Over the 6-yr period, the value of the EF was 0.48 for plot 3, 0.19 for plot 2, –0.02 for plot 4, and 0.01 for plot 5. On an annual basis, the monthly EF values ranged from –1.80 to 0.66 for plot 3, –1.57 to 0.61 for plot 2, –4.36 to 0.35 for plot 4, and –1.55 to 0.32 for plot 5 (Table 15). The small and negative EF values were caused by a few incidents (3 to 6 out of 72 per plot) with large discrepancies between observed and predicted monthly NO₃–N drainage losses (Fig. 9). The majority of these discrepancies occurred in October 1993, January and February 1994, June 1995, and June and July 1996. When these incidents (outliers) were excluded, the EF values were greatly increased (0.70 for plot 3, 0.63 for plot 2, 0.36 for plot 4, and 0.61 for plot 5). In many cases, monthly NO₃–N drainage losses were overpredicted during summer and early fall and underpredicted during winter and early spring.

View larger version (23K): [in this window] [in a new window]   Fig. 9. Scatter diagram comparing observed and predicted monthly NO3–N losses with drain flow for plots 2 through 5 from 1992 through 1997.

Annual Nitrate-Nitrogen Drainage Losses
Except for one experimental plot, the agreement between observed and predicted annual NO₃–N drainage losses was good. The value of the EF comparing observed and predicted annual NO₃–N loss with drain flow was 0.92 for plot 3, 0.62 for plot 2, –0.10 for plot 4, and 0.73 for plot 5. The absolute value of NE in predicting annual NO₃–N leaching loss was in the range of 2.0 to 35.9% for plot 3, 9.1 to 39.3% for plot 2, 6.6 to 63.8% for plot 4, and 14.5 to 30.7% for plot 5 (Table 15).

Errors in DRAINMOD-N II predictions of N losses were due to inaccuracies in characterizing both the hydrology and N dynamics. Inspection of results in Table 15 show that the absolute value of NE in predicting annual NO₃–N leaching losses was less than 15% in one-third of the simulated plot-years and less than 20% in half of the simulated plot-years. In three cases with high NE values, errors in predicting NO₃–N leaching losses appeared to be caused by errors in predicting drainage volumes. In 1994, for plot 3, drainage was underpredicted by 18.8% (Table 14) and NO₃–N drainage loss was underpredicted by 24.2% (Table 15). In 1997, drainage was underpredicted by 22.9% for plot 4 and 27.4% for plot 5 (Table 14), and NO₃–N drainage loss was underpredicted by 41.0% for plot 4 and 26.3% for plot 5 (Table 14). Two other cases can also be identified where errors in hydrologic predictions might have partly contributed to the high errors in predicting annual losses of NO₃–N with drain flow. DRAINMOD underpredicted drainage volumes for plot 2 by 10.6% in 1994 and by 10.1% in 1997. DRAINMOD-N II underpredicted NO₃–N leaching losses for the same plot by 33.1% in 1994 and 39.3% in 1997. While these instances indicate an influence of DRAINMOD hydrologic predictions on the performance of DRAINMOD-N II, simulation results for the 24 plot-yr does not indicate a statistically significant correlation between errors in hydrologic predictions and errors in predicting N leaching losses.

In late 1993, NO₃–N leaching losses were overpredicted and were underpredicted in early 1994. This caused relatively high errors on both monthly (low EF, Table 15) and annual bases (high NE, Table 15). For plots 3 through 5, and to a lesser extent plot 2, N leaching losses were considerably overpredicted in October through December 1993 and underpredicted in January through March 1994 (Fig. 10, 11). It was very dry in 1993; corn grew under water deficit conditions during the entire growing season (Table 5) and yield was low (Table 3). Thus, a large portion of fertilizer N applied in May and June (Table 2) was not recovered in the crop and remained in the soil profile. Part of that mineral N was immobilized during early stages of decomposition when the fresh corn residues were added to the system. The other part remained in the profile susceptible to leaching and denitrification losses. The rate and timing of those losses depended on the sequence of weather events.

View larger version (23K): [in this window] [in a new window]   Fig. 10. Observed and predicted NO3–N loss via subsurface drainage for plot 3 from 1992 through 1994.

View larger version (23K): [in this window] [in a new window]   Fig. 11. Observed and predicted NO3–N loss via subsurface drainage for plot 4 from 1992 through 1994.

The model might have predicted less N immobilization during the early stages of decomposition of corn residues than actually occurred, leaving more mineral N in the profile susceptible to leaching. The simulated duration of net immobilization during corn residue decomposition might also be shorter than the actual duration. In this case, the model may have predicted an early release of some of the mineral N that was previously immobilized during the first stages of decomposition. This would have increased the amount of mineral N in the soil profile available for leaching. Unfortunately, neither NO₃–N concentration in soil solution nor N mineralization or immobilization was measured, so it was not possible to directly test the accuracy of model predictions of these quantities.

Overall, DRAINMOD-N II overpredicted 1993 NO₃–N leaching losses by 64% for plot 4, 31% for plot 5, and 7% for plot 3 (Table 15). In reality, however, most of the mineral N accumulated in the soil profile remained until early 1994 when it was flushed out of the system by the rainfall of January and February. DRAINMOD-N II, though, underpredicted those NO₃–N losses because it had already predicted (erroneously) high losses in late 1993, which significantly lowered the NO₃–N concentration in the profile (Fig. 11). DRAINMOD-N II underpredicted the overall NO₃–N leaching losses during 1994 by 56% for plot 4, 31% for plot 5, and 24% for plot 3 (Table 15). Predicted and observed cumulative NO₃–N drainage losses from 1993 to 1994 were compared to quantify model performance during that period without the negative effects of the errors occurred in late 1993 and early 1994. Observed cumulative NO₃–N drainage loss during 1993 to 1994 was 56.6 kg N ha^–1 for plot 4 and 41.4 kg N ha^–1 for plot 5. Corresponding predicted values were 42.2 kg N ha^–1 for plot 4 and 36.5 kg N ha^–1 for plot 5. This means that DRAINMOD-N II underpredicted cumulative NO₃–N drainage losses from 1993 to 1994 by 25% for plot 4 and 12% for plot 5. This amounts to less than half the errors in yearly predictions for plot 4 and about one-third of the errors in yearly predictions for plot 5. This indicates that the discrepancies between model prediction and observation during late 1993 and early 1994 are the main sources of error in yearly model predictions for 1993 and 1994.

Cumulative Nitrate-Nitrogen Drainage Losses
DRAINMOD-N II did an excellent job in predicting cumulative NO₃–N leaching losses over the entire simulation period (Fig. 12 and Table 15). Cumulative NO₃–N leaching losses over the 6-yr simulated period were underpredicted by 1.3% for plot 3, 8.1% for plot 2, and 2.8% for plot 4 and overpredicted by 4.0% for plot 5 (Table 15). The strong dependence of DRAINMOD-N II performance in predicting NO₃–N leaching losses on the accuracy of subsurface drainage prediction was manifested in the close association between predicted cumulative drainage and NO₃–N leaching losses. This can be clearly observed for all experimental plots (Fig. 12).

View larger version (29K): [in this window] [in a new window]   Fig. 12. Observed and predicted cumulative rates of subsurface drainage and NO3–N leaching loss for experimental plots 2 through 5 from 1992 through 1997.

Simulation results (not presented) showed that predicted net mineralization was generally slow during winter due to low temperatures. In very dry summer seasons the process was very slow or ceased completely due to lack of available soil water in the biologically active topsoil layer. The effect of quality of added residues can be clearly observed by analyzing simulation results. Net mineralization was increased following the incorporation of the N-rich soybean residues. On the other hand, incorporating residues of corn and wheat, which are low in N, resulted in net immobilization.

Nitrogen plant uptake is a major process of the N cycle in agricultural ecosystems. Management practices such as the use of split applications, slow-release N fertilizers, and nitrification inhibitors have been developed to increase N fertilizer use efficiency and N plant uptake. The yearly average of simulated N uptake varied from a low of 159 kg N ha^–1 for plot 5 to a high of 172 kg N ha^–1 for plot 2 (Table 16). Simulated annual N uptake was a function of crop species, the amount of applied N fertilizer, and the soil water conditions. Simulated N plant uptake varied among the four plots for the same year (Table 16). This plot-to-plot variability in N uptake was primarily due to the differences in crop yield among the individual plots, which were caused by differences in soil and management factors.

The denitrification process is an important pathway for N loss from the soil–water–plant system. The performance of an N model is closely related to its success in making good predictions of denitrification rates in response to changes in NO₃–N availability as well as changes in soil aeration as reflected by the soil water status. On average, simulated annual denitrification varied over a small range of 44 to 47 kg N ha^–1 (Table 16).

The strong dependence of denitrification on mineral N availability and soil water conditions can be demonstrated by analyzing the simulation results. Both lowest and highest simulated annual denitrification rates corresponded with lowest and highest simulated amount of mineral N in the soil profile, respectively. In 1995, plots 3 and 5 received 60% more N fertilizer than plots 2 and 4 (Table 2). In the same year, plots 2 and 3 were in conventional-drainage mode and plots 4 and 5 were in controlled-drainage mode for the first 4 mo and then switched to conventional-drainage mode for the rest of the year (Table 1). The model predicted 20% more denitrification for plot 3 than for plot 2 and 33% more denitrification for plot 5 than for plot 4 (Table 16). These results underscore the close relationship between the availability of mineral N and the denitrification rate. It also indicates the influence of water table management treatment on the denitrification process, which can further be observed by inspecting the 1994 simulation results. In 1994, all plots received the same amount of N fertilizer (Table 2). Plots 2 and 3 were in conventional drainage, while plots 4 and 5 were in controlled drainage. DRAINMOD-N II predictions for 1994 showed 33% more denitrification, on average, for controlled drainage plots than for conventional drainage plots.

	CONCLUSIONS

TOP NOTES ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS AND DISCUSSION CONCLUSIONS REFERENCES

 
DRAINMOD-N II was field-tested using 6 yr of data from an artificially drained agricultural research site located in the North Carolina lower coastal plain. The test site is on a naturally poorly drained sandy loam soil with nearly flat topography. Four experimental plots, planted to a corn–wheat–soybean rotation and managed using both conventional and controlled drainage, were used; one plot for model calibration and the other three plots for model validation. Water table elevation, subsurface drainage, and N concentration in drainage water were measured and meteorological data were continuously recorded. Agreement between observed and predicted monthly NO₃–N drainage losses was relatively poor. The monthly EF was 0.48 for the calibration plot and 0.19, 0.01, and –0.02 for the validation plots. The small and negative EF values were caused by a few incidents with large discrepancies between observed and predicted NO₃–N drainage losses. Except for one experimental plot, agreement between observed and predicted annual NO₃–N drainage losses was good. The yearly EF was 0.92 for the calibration plot and 0.73, 0.62, and –0.10 for the validation plots. Large errors in predicting annual NO₃–N leaching losses in some years were in part due to errors in predicted drainage volumes. Errors in predicting cumulative NO₃–N leaching losses over the 6-yr period were remarkably small. Cumulative NO₃–N leaching losses over the 6-yr period were underpredicted by 1.3% for the calibration plot, 8.1 and 2.8% for two of the validation plots, and overpredicted by 4.0% for the third validation plot.

Results of this field-testing demonstrated the potential of DRAINMOD-N II as a model capable of simulating N dynamics in agroecosystems. However, this evaluation of DRAINMOD-N II using the TES data set should be regarded as an incomplete test since it relied primarily on literature rather than field and/or lab measurements to parameterize the model. A more rigorous field testing of the model should be conducted to validate its application for drained lands, a need common with nearly all simulation models.

	ACKNOWLEDGMENTS

 
The research reported herein is a product of the N.C. Agricultural Research Service, N.C. State University, Raleigh, NC 27695-7625. It was supported in part by funds provided by the USDA (NRI project 99-351028571) and by the USDA-NRCS.

	NOTES

TOP NOTES ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS AND DISCUSSION CONCLUSIONS REFERENCES

 
Presented in the 8th International Drainage Symp., ASAE.

	REFERENCES

TOP NOTES ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS AND DISCUSSION CONCLUSIONS REFERENCES

 

• Ahuja, L.R., K.W. Rojas, J.D. Hanson, M.J. Shaffer, and L. Ma (ed.) 2000. Root Zone Water Quality Model: Modeling management effects on water quality and crop production. Water Resources Publications, Highlands Ranch, CO.
• Brevé, M.A. 1994. Modeling the movement and fate of nitrogen in artificially drained soils. Ph.D. diss. North Carolina State Univ., Raleigh, NC.
• Brevé, M.A., R.W. Skaggs, J.E. Parsons, and J.W. Gilliam. 1997a. DRAINMOD-N, a nitrogen model for artificially drained soils. Trans. ASAE 40:1067–1075.
• Brevé, M.A., R.W. Skaggs, J.W. Gilliam, J.E. Parsons, A.T. Mohammad, G.M. Chescheir, and R.O. Evans. 1997b. Field testing of DRAINMOD-N. Trans. ASAE 40:1077–1085.
• Chescheir, G.M. R.W. Skaggs, J.W. Gilliam, M.A. Brevé, C. Munster, R.O. Evans, R.B. Leidy and J.E. Parsons. 1996. Effects of drainage and water table control on groundwater and surface water quality: Part II. Experimental results and simulation models. Tech. Rep. 301. Univ. of North Carolina Water Resour. Res. Inst., Raleigh, NC.
• Cox, J.W., D.J. McFarlane, and R.W. Skaggs. 1994. Field evaluation of DRAINMOD for predicting waterlogging intensity and drain performance in southwestern Australia. Aust. J. Soil Res. 32:653–671.[CrossRef]
• 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. 29:1568–1581.[ISI]
• Dinnes, D.L., D.L. Karlen, D.B. Jaynes, T.C. Kaspar, J.L. Hartfield, T.S. Colvin, and C.A. Cambardella. 2002. Nitrogen management strategies to reduce nitrate leaching in tile-drained Midwestern soils. Agron. J. 94:153–171.[Abstract/Free Full Text]
• Doran, J.W., L.N. Mielke, and S. Stamatiadis. 1988. Microbial activity and N cycling as regulated by soil water-filled pore space. p. 49–54. In Proc. 11th Int. Conf., Int. Soil Tillage Res. Organization, 11–15 July 1988, Edinburgh, Scotland.
• Eckersten, H., P.-E. Hansson, and H. Johnsson. 1998. SOILN model user's manual, version 9.2. Swedish Univ. of Agricultural Sciences, Uppsala, Sweden.
• Evans, R.O., R.W. Skaggs, and R.E. Sneed. 1991. Stress day index models to predict corn and soybean relative yield under high water table conditions. Trans. ASAE 34:1997–2005.
• Fouss, J.L., R.L. Bengston, and C.E. Carter. 1987. Simulating subsurface drainage in the lower Mississippi Valley with DRAINMOD. Trans. ASAE 30:1679–1688.
• Garrison, M.V., W.D. Batchelor, R.S. Kanwar, and J.T. Ritchie. 1999. Evaluation of the CERES-Maize water and nitrogen balances under tile-drained conditions. Agric. Sys. 62:189–200.[CrossRef]
• Gayle, G., R.W. Skaggs, and C.E. Carter. 1985. Evaluation of a management model for a Louisiana sugar cane field. J. Am. Soc. Sugar Cane Technol. 4:18–28.
• Gilliam, J.W., J.L. Baker, and K.R. Reddy. 1999. Water quality effects of drainage in humid regions. p. 801–830. In R.W. Skaggs and J. van Schilfgaarde (ed.) Agricultural Drainage. Agron. Monogr. 38. ASA, CSSA, and SSSA, Madison, WI.
• Godwin, D.C., and C.A. Jones. 1991. Nitrogen dynamics in soil-plant systems. p. 289–321. In J. Hanks and J.T. Ritchie (ed.) Modeling plant and soil systems. Agron. Monogr. 31, ASA, CSSA, and SSSA, Madison, WI.
• Groenendijk, P., and J.G. Kroes. 1997. Modelling the nitrogen and phosphorus leaching to groundwater and surface water: ANIMO 3.5. Rep. 144, DLO Winand Staring Centre, Wageningen, The Netherlands.
• Hansen, S., H.E. Jensen, N.E. Neilsen, and H. Svendsen. 1993. Description of the soil plant system model DAISY– Basic principles and modeling approach, The Royal Veterinary and Agricultural Univ., Copenhagen, Denmark.
• Hay, R.K. 1995. Harvest index: A review of its use in plant breeding and crop physiology. Ann. Appl. Biol. 126:197–216.[ISI]
• Hoad, S.P., G. Russell, M.E. Lucas, and I.J. Bingham. 2001. The management of wheat, barley, and oat root systems. Adv. Agron. 74:193–246.
• Hutson, J.L. 2000. LEACHM: Model description and user's guide. School of Chem., Physics, and Earth Sci., The Flinders Univ. of South Australia, Adelaide, Australia.
• Jabro, J.D., J.M. Jemison, Jr., L.L. Lengnick, R.H. Fox, and D.D. Fritton. 1993. Field validation and comparison of LEACHM and NCSWAP models for predicting nitrate leaching. Trans. ASAE 36:1651–1657.[ISI]
• Janssen, P.H.M., and P.S.C. Herberger. 1995. Calibration of process-oriented models. Ecol. Modell. 83:55–66.[CrossRef]
• Jensen, M.E., R.D. Burman, and R.G. Allen. 1990. Evapotranspiration and irrigation water requirements. ASCE Rep. 70. ASCE, New York.
• Johnsson, H., L. Bergstrom, and P.-E. Jan