| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
a Istituto Sperimentale Agronomico (ISA), Sezione di Modena, Ricerche agronomiche applicate all'ambiente settentrionale, Viale Caduti in Guerra 134, I-41100 Modena, Italy
b Regione Emilia Romagna, Soil Bureau, Viale Silvani 4/3, 40122, Bologna, Italy
* Corresponding author (rosamar{at}pianeta.it).
Received for publication March 18, 2003.
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
Abbreviations: F(LOFIT), F test for the lack of fit • MD, mean difference • RMSE, root mean square error
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
A simulation model was chosen, within the framework of an interregional project, SINA (National Environmental Information System, soil mapping in areas at high environmental risk), to estimate soil protective capacity against pollutants on a regional scale. Local Soil Survey Services are going to use it as a tool to facilitate the drawing of regional vulnerability maps, according to the 91/676 European Union directive of 1991. This model is a coupling (Larsson and Jarvis, 1999) of the MACRO model (Jarvis, 1994), which simulates water flow and solute transport in soils with macropores, with SOILN (Johnsson et al., 1987), which simulates N dynamics in agricultural soils. It was chosen because the two model components, MACRO and SOILN, were specifically developed to simulate soil water and nitrogen dynamics, respectively. Even though several crop and cropping system models are able to simulate soil water and N dynamics, the soil compartment rarely plays a determinant role, since the prime goal of these models is the simulation of crop growth. In contrast, MACRO's hydrological component is able to simulate water flow in soils with macropores, which are common in the Po Valley.
The model's components, MACRO and SOILN, have been individually and widely tested. MACRO was tested for simulation of water movement and salt leaching (Andreu et al., 1994; Jabro et al., 1994) and was used for assessing pesticide fate and mobility in soils (Jarvis et al., 1994; Bergström, 1996). Whereas substantial agreement exists in the literature on MACRO's ability to simulate water flow, provided that the two-domain flow option is used, the judgment of the authors who evaluated the SOILN model is less unanimous. Bergström and Jarvis (1991) evaluated the SOILN model for prediction of nitrate leaching losses from arable land under different mineral N application rates. They found large discrepancies (up to 100%) between measurements and model predictions in some years and attributed them to important soil processes that were either not included in the model (such as macropore flow) or were difficult to model satisfactorily, such as crop N uptake. Katterer and Andrén (1996) compared observed and predicted values of soil mineral N under wheat (Triticum aestivum L.) with different soil moisture and mineral fertilizer N regimes. Simulated soil mineral N levels agreed with measurements on a yearly time scale, whereas their short-term dynamics were less well described by the simulations. Borg et al. (1990) compared simulated and measured N levels in plots with grain crops that had been fertilized with liquid manure and commercial fertilizers. The simulated temporal variations of the nitrate and ammonium storage were not always in agreement with measurements. In particular, simulated nitrate storage showed a tendency to be higher than measured storage after the occasions when liquid manure was applied, while the simulated ammonium content tended to be sometimes much lower than the measured content. According to Jabro et al. (2001), who evaluated the model in an experiment to study the effect of dairy urine and feces distribution on nitrate leaching, the SOILN model adequately predicted nitrate losses for three urine treatments but failed to produce accurate simulations for two control treatments and for the feces treatments. They related model inaccuracy in the simulation results to an inadequate modeling of the N transformation processes. In all these applications, SOILN was coupled with another hydrological model, SOIL, which does not simulate macropore flow. In our work the coupling of MACRO and SOILN was, therefore, preferred over the coupling of SOIL and SOILN.
The aim of our study was to evaluate the model to better identify the key parameters and processes that influence N losses from agricultural soils. The soil water and nitrate content measured under corn fertilized with urea and pig slurry in two field experiments were compared with those predicted by the model.
|
MATERIALS AND METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
The MACRO model simulates vertical water movement in two flow domains, macro- and micropores. The division between the two flow domains is defined by a boundary water potential (
b) and a corresponding water content (
b), and hydraulic conductivity (Kb). Lateral water exchange between macro- and micropores is regulated by an "effective diffusion pathlength," d. The value of d depends on soil structure, that is, shape and size of the aggregates and degree of structural development (Jarvis et al., 1997). Water flow in the micropores is described by the Richards' equation. Soil water pressure head is estimated by the Brooks and Corey (1964) equation, while hydraulic conductivity is given by Mualem's (1976) model.
Plant N uptake is calculated in SOILN with a logistic function and depends on the availability of inorganic N in the layers where roots have been simulated. Roots after harvesting, unless alive, become soil litter. Plant residues, which are left on the field, enter the litter pool after soil cultivation. Organic N in soil is divided into three pools: (i) litter (undecomposed soil residues, dead roots, microbial biomass), (ii) feces (originating from manure), and (iii) humus (resistant decomposition products characterized by a stable C to N ratio). The litter and feces N pools are coupled to pools of C for control of mineralization and immobilization rates. A more detailed description of model functioning may be found in Johnsson et al. (1987). The model does not simulate ammonia emissions to the atmosphere that are produced by land-spreading of animal wastes.
Dataset
Two experiments were considered in this model evaluation. One of them will be called the "Piacenza" experiment; the other the "San Prospero" experiment. Both of them were performed to compare the effect of the fertilization with urea and pig slurry on both crop productivity and inorganic N levels in soil.
The study area is located in the Emilia Romagna region (northern Italy; Fig. 1) and morphologically prominent within an alluvial plain with an elevation of 2 to 70 m. The thermal regime is temperate subcontinental. Mean annual temperatures vary between 12 and 14°C. Mean annual precipitation range is 650 to 800 mm. Rainfall is concentrated mainly in the fall–spring period. Soils of this area are flat (with 0.1 to 0.5% slopes), deep, calcareous, and moderately alkaline, and have good oxygen availability (Filippi and Sbarbati, 1994). For both experiments, measured values of selected soil properties are reported in Table 1. The soil series are described in the Emilia Romagna regional soil catalog (Regione Emilia Romagna, 2000).
|
|
At each site, about the same amount of pig slurry N, that is, 60 g N m–2 yr–1, was spread in five different distribution periods: (i) the whole amount (60 g N m–2) was supplied in February, (ii) the whole amount (60 g N m–2) was supplied in June, (iii) 30 g N m–2 was supplied in February and 30 g N m–2 in June, (iv) 30 g N m–2 in February and 30 g N m–2 in October, and (v) 20 g N m–2 was supplied in February, 20 g N m–2 in June, and 20 g N m–2 in October. The effect of the five pig slurry spreading calendars on the seasonal soil nitrate content was assessed together with that of three urea application rates for a total of eight treatments in a completely randomized experimental design.
Climatic data for the Fienili site were collected at the meteorological station of Monticelli (45°05' N, 9°56' E), 4 km from the experimental field; those for Del Fiducia and Sant'Omobono, at Bacedasco (44°51' N, 9°54' E), 8 and 16 km from the experimental field, respectively. Soil samples for the measurement of soil water and NO3–N content were taken on eight occasions, at the beginning and end of the winter period of each year, from 7 Dec. 1988 to 24 May 1991, in 0.2-m increments to a depth of 1.2 m (1152 samples in all). Soil water content was determined gravimetrically. Nitrate N content was determined by steam distillation with MgO and Devarda alloy (Keeney and Nelson, 1982). The water table level was monitored every month in all eight plots of the three sites from December 1987 until May 1989 by means of observation wells with perforated casing. It was detected in the soil profile only at the Fienili site (eight positions, one for each plot).
The San Prospero experiment was performed in the ISA experimental farm of San Prospero (44°47' N, 11°01' E), in the Modena province, from 1993 to 1995 in 224-m2 plots on a La Boaria silty clay soil (fine, mixed, mesic Vertic Ustochrepts). Within the framework of a strip plot design comparing four urea N and four pig slurry N levels in factorial combination with two replicates, four treatments were selected: (i) the highest pig slurry N rate, (ii) the highest urea N rate, (iii) the highest N rate from combined pig slurry and urea, and (iv) the nonfertilized control. Climatic data were collected at the meteorological station near the experimental field. The water table level was measured weekly starting from August 1993 until the end of 1995 in five positions of the experimental field at three depths (0.5, 1.1, and 1.8 m) by means of piezometers. Soil sampling to a 1.8-m depth in 0.2-m increments for the measurement of soil water and NO3–N covered the whole experimental period and was more frequent during the crop growth season and in the 0- to 0.6-m soil layer. It was usually performed in only one of the two available plot replicates. Soil water content was determined gravimetrically. Nitrate N content was determined colorimetrically with an automatic analyzer (AutoAnalyzer 3; Bran + Luebbe GmbH, Norderstedt, Germany) after Cd reduction according to Keeney and Nelson (1982). The other soil chemical analyses were performed according to the ASA and SSSA methods of soil analysis (Page et al., 1982). Plant biomass and pig slurry N content were determined by the Kjeldahl method.
Corn management details are reported in Tables 2 and 3. Winter wheat, in the last year of the Piacenza experiment, was seeded in October 1990 and harvested in June 1991. Crop residues were removed after harvesting (except in 1987 in the Piacenza sites), and soil was plowed every following fall. Aboveground biomass production and N removal were measured at each harvesting.
|
|
Driving Variables
Daily precipitation, maximum and minimum air temperature, and potential evapotranspiration were used as climatic driving variables. Potential evapotranspiration was estimated according to Hargreaves' method (Hargreaves and Samani, 1982).
Soil Properties
The thickness of the simulated soil layers varied from 0.03 m in the top layer to 0.5 m in the deeper layers. The soil hydrological parameter values are reported in Table 4. Water retention parameters, for which measurements were lacking, and the hydraulic conductivity value at the boundary between the two flow domains were estimated for all soils from measured texture, bulk density, and organic C content by means of pedotransfer functions (Jarvis et al., 1997). These were validated for benchmark soils in the Po Valley by Ungaro and Calzolari (2001). The morphological description of the soil structure was used for defining the d parameter value.
|
As far as the N simulation is concerned, the initial organic N content in the soil profile was measured for the sites at the beginning of the experiment. We also assumed an amount of C and N to be present from residues of the previous crop (corn).
Nitrogen Inputs
Nitrogen from pig slurry was divided between feces ammonium and feces organic N according to measurements; the feces C to N ratio, varying between 2 and 6, was also measured. These values, while being lower than the minimum suggested by the model (i.e., 6), did not cause any simulation error. Given that local data were not available, N dry deposition from the atmosphere was set at 0.001 g m–2 d–1 according to the model user's manual (Jansson et al., 1991). Nitrogen wet deposition was equal to 0.6 mg N L–1. These values are presumably different from those occurring in Italy. Nevertheless, we considered this approximation acceptable because N deposition, as high as it may be, has very little influence on the whole N balance compared with other sources, such as fertilizer N or N deriving from the organic matter mineralization. Since the model does not simulate ammonia emissions, we estimated their amount according to Meisinger and Randall (1991). They were set equal to 44% of the pig slurry ammonium N based on the fact that the pig slurry ammonium N fraction was 70% of the total N, and the manure application method was broadcast with no incorporation. In case of rainfall within 24 h following pig slurry spreading, the emission amount was reduced to 32% of the pig slurry ammonium N. These percentages agreed well with those measured in previous field-based experiments (unpublished data, 1995). Estimated ammonia losses were subtracted from the input feces ammonium N (Table 3). No emission was considered to have occurred from urea, since urea was incorporated into the soil at the time of distribution. The model assumes that the mineral fertilizer N (including urea) enters the system as nitrate N.
Soil and Plant Management
Seventy-five percent of the total plant N was removed at each harvest, whereas 5% was left in the standing stubbles. The root N was assumed to be 20% of the whole plant N. The C to N ratio of corn residues was estimated from measurements and set equal to 42 and 100 for the fertilized and unfertilized plots, respectively. It was equal to 36 for the wheat residues. The C to N ratio of the roots was set equal to that of the aboveground residues.
Plant Water and Nitrogen Uptake
The annual maximum leaf area index (LAImax) was estimated for each site from aboveground biomass (AGB) measurements assuming a roughly proportional relationship between plant LAI and plant AGB. For each year and site the following relationship was applied:
 |
Root depth was assumed to linearly increase from zero at crop emergence to a maximum value (Rootmax) when the crop reaches the LAImax value. Rootmax was estimated by the same procedure used for LAImax taking the 1.6-m value as the corn potential maximum rooting depth. An exponential decrease of the root density from the soil surface to the maximum root depth was assumed.
For each year and site, the plant potential N uptake value was selected as the maximum measured amount of AGB N removed at harvest among the various treatments, in nonlimiting N conditions. This amount was increased by 25% to include N in root and stubble residues. The plant potential N uptake values varied between 12 and 48 g N m–2. The values of the other parameters regulating the shape of the LAI and of the N uptake curves were those previously obtained for the same crop (Marchetti et al., 2001).
Nitrogen Transformations
The humus mineralization rate (kh) was estimated on the basis of measurements of crop N removal in the less-fertilized plots, assuming that the N taken up by the crop, in the absence of fertilization, was derived from humus mineralization. The rate value was adjusted until the estimated N removal reached the highest measured value. It was equal to 2.78 x 10–5 d–1 for Fienili, 1.95 x 10–4 d–1 for Del Fiducia, 3.41 x 10–4 d–1 for Sant'Omobono, and 6 x 10–6 d–1 for San Prospero. The other parameters for simulating N transformations were selected within the range reported as reasonable by the developers of the model. In particular, litter and feces specific decomposition rate was set at 0.035 d–1, specific nitrification rate was equal to 0.1 d–1, and potential denitrification rate was equal to 0.2 g N m–2 d–1.
The feces turnover efficiency (fe) and feces carbon humification constant (fh) parameters also affect the feces decomposition process. According to the model user's manual the fe parameter specifies the fraction of organic C that, after respiration, remains as organic C. It may normally range between 0.2 and 0.7, but the model accepts higher values up to 0.9. The fh parameter determines the fraction of feces N that enters the humus pool. Low fh values (0.1–0.3, "fast" feces turnover) result in a major part of the feces organic N being remineralized, while a minor part is humified. High values (0.6–0.9, "slow" feces turnover) result in the reverse. These parameters were both set equal to 0.9 to favor N feces immobilization and humification. In fact, model performances in simulating nitrate leaching in previous lysimeter experiments had been remarkably improved following the adjustment of these two parameters (Marchetti et al., 2001).
Model Evaluation
Model efficiency indexes were used to estimate either the quality of the calibration results or the model predictive capability. The observed vs. predicted mean values and the coefficients of linear regression were used for the evaluation of calibration results concerning the plant N uptake. The observed vs. predicted mean values, the root mean square error (RMSE), and the mean difference (MD) were used for evaluating the simulation of soil water and NO3–N content. The RMSE may be considered as an index of the total error, while MD is an index of the model bias.
Because the experimental design was complex (three farm fields x eight treatments, with several annual measurements on crop traits and hydrological parameters), no treatment replicates were conducted. Since the water amount supplied at pig slurry spreading was on average equal to 1.7% of the total water inflow in the Piacenza experiment and 1.4% at San Prospero, we assumed that the inflow increase due to pig slurry spreading was unimportant. Therefore, to evaluate the hydrological model at each site, the experimental plots were considered as replicates from a hydrological point of view, and the lack-of-fit method could be applied. For each soil, in the Piacenza experiment, the F test for the lack of fit, F(LOFIT), was used in the model evaluation of soil water content and was based on 48 observations (eight dates x six soil layers), each of them with eight replicates. At San Prospero, the available measurements allowed the F(LOFIT) to be calculated only for the top-0.6-m soil depth (61 dates for the 0- to 0.2- and 0.2- to 0.4-m soil layers, and 59 dates for the 0.4- to 0.6-m layer, for a total of 181 data pairs, with three or four replicates for each observation). For the F(LOFIT) in the water table simulation, eight and five replicates were available at the Fienili and San Prospero sites, respectively.
For the evaluation of the SOILN model at the Piacenza sites, since the total amount of N supply was the same for each treatment, data relevant to the five pig slurry treatments were analyzed altogether, and the standard deviation included the variability due to the land-spreading time differences. Therefore, in the graphs showing the simulation results, the measured nitrate levels in soils fertilized with pig slurry are the mean of those measured in the five treatments, and error bars include the variability due to the differences in pig slurry spreading time. Also, the simulated nitrate levels are the mean of those simulated for each treatment, and the continuous thin lines are the standard deviation of the estimate. The meanings and the methods for the calculation of the RMSE, MD, and F(LOFIT) tests are reviewed by Smith et al. (1996)(1997).
Sensitivity Analyses
Ammonia Emissions
To quantify the effect of an error in the ammonia emission estimate on the model response, we compared measured and predicted soil nitrate levels in simulations in which the ammonia losses were set equal to 20 or 60% of the pig slurry ammonium N content. The evaluation was performed for the pig slurry and pig slurry + urea treatments of the San Prospero experiment.
Feces Decomposition
A sensitivity analysis was performed to evaluate the effect on the output (model response) of changing the level of the parameters controlling the feces decomposition process, that is, feces decomposition rate (kf), feces turnover efficiency (fe), and feces carbon humification constant (fh). The sensitivity analysis was applied to the simulation of the pig slurry treatment of the San Prospero site. As no information was available on possible interactive effects between the above-mentioned parameters, a factorial design (Box et al., 1978) was chosen to study the effects of the three parameters, kf, fe, and fh, hereafter called "factors," on selected model responses. The experimental design was a composite one, including a two-level factorial design, which allows the evaluation of first-order effects (main factor effects) and interaction effects between factors, and a "star" design for the evaluation of additional second-order curvature effects, such as those represented by the quadratic terms in a polynomial model. Since the center of these two experimental designs was allowed to coincide, this kind of design combination is called central composite design (Deming and Morgan, 1987). The factorial design required eight simulations ("runs"), as three factors at two levels produce 23 combinations, whereas the star design required seven runs because it generates 2k + 1 combinations, with k being the number of factors included in the model, for a total of 15 runs.
The factor-level combinations used in the simulation runs, according to the described central composite design, are reported in Table 5. Factor levels were coded to –2, –1, 0, +1, and +2, respectively, and used for the calculation of the model responses. The –1 and +1 levels are for the factor design, the –2 and +2 levels are for the star design, and the 0 level is needed for the evaluation of the central point in the central composite design. Model responses were (i) the RMSE obtained from the comparison of measured and simulated crop N uptake, (ii) the RMSE obtained from the comparison of measured and simulated soil nitrate content, (iii) the simulated N leaching values, and (iv) the simulated changes in mineral N and organic N storage. For each run, the RMSE for N uptake was calculated for three data pairs, corresponding to 3-yr measurements of crop N removals at the San Prospero site. The RMSE for soil nitrate N content was calculated for 46 data pairs, corresponding to 3-yr measurements (46 dates) of the accumulated nitrate level in the top-0.6-m soil profile. The model response values were fitted to a full second-order polynomial model of the form:
 |
|
|
|
RESULTS AND DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
|
|
|
|
The model simulated also an inorganic N buildup (mainly as nitrate N) that was higher at the higher N rates in all soils. It was more evident for Sant'Omobono and La Boaria soils. This buildup was due to the progressive model error in simulating soil nitrate content, as shown in Fig. 2 and 3. In the hypothesis that the inorganic N accumulation could derive from the adoption of an excessively high humus mineralization rate, we made an attempt to reduce the humus depletion by reducing the kh value. This attempt, while allowing an improvement of the soil nitrate content at Sant'Omobono, brought about a remarkable reduction of the N taken up by the crop at the Fienili and Del Fiducia sites and was, therefore, abandoned.
To identify possible sources of model error the following points were examined: (i) simulation of water flow, (ii) simulation of crop N uptake, (iii) estimate of the ammonia emissions, and (iv) setting of the N transformation parameters.
Soil Water Flow
The correct reproduction of water flow in models simulating soil N dynamics is essential to appropriately describe nitrate leaching, as the solute fate is tied to water movement in the soil profile. We evaluated the model performances in simulating water flow by comparing measured and predicted values of soil water content and water table level (Table 9).
|
The simulated mean annual actual evapotranspiration at the Piacenza sites was 677 mm at Del Fiducia, 605 mm at Fienili, and 625 mm at Sant'Omobono, and 737 mm at the San Prospero site. It constituted between 79 and 86% of the mean annual total inflow (precipitation + irrigation). At the Fienili and San Prospero sites 59 and 118 mm yr–1, respectively, ran off. No deep percolation was simulated in these sites, due to the presence of a water table. At the Del Fiducia and Sant'Omobono sites, 36 and 37 mm yr–1 of water, respectively, ran off, whereas 67 and 105 mm yr–1, respectively, drained on average from the deep soil profile. These low percolation amounts (or the lack of percolation) were responsible for the simulated low nitrate leaching amounts (Table 8).
Crop Nitrogen Uptake
The model parameters regulating crop N uptake were calibrated on the basis of crop N removal measurements. This calibration resulted in good agreement between measured and simulated crop N uptake values (Table 10). In fact the R2 value was significant in all cases, except for the pig slurry treatment at Fienili. For this site, the R2 lack of significance for the pig slurry treatment may be explained by the fact that measured values varied in a narrower range than in the other sites and treatments. Therefore, the model was probably unable to reproduce the minor differences existing among the observed values. Despite this lack of significance, however, the prediction of soil nitrate content in the pig-slurry-treated plot at the Fienili site was still reasonable. On the other hand, at this same site the underestimation of the soil nitrate content under corn fertilized with urea could not be attributed to an overestimation of crop N uptake, because in the urea-treated plots the crop N uptake was underestimated.
|
In our sensitivity analysis, a 20% ammonia emission raised the RMSE by 15% and the MD by 45% in comparison with the value that had been obtained by setting the ammonia emission to 44% (default); a 60% ammonia emission reduced the RMSE by 10% and the MD by 21% in comparison with the default. The R2 value for the crop N removal was reduced by 12.1% for an emission equal to 20% of the pig slurry ammonium N content, because the simulated crop took up more N (which was more abundant in the soil) than did the real crop; it increased by 8.3% for a 60% emission. We conclude that the initially chosen ammonia emission percentages could be underestimated.
Nitrogen Transformation Parameters
Laroque and Banton (1994) evaluated the influence of the SOILN parameter values on nitrate leaching in systems with inorganic fertilization in nordic climates and they found that parameters favoring the N uptake and the mineralization processes have the highest impact on nitrate leaching. In our work the error that can derive from the setting of inadequate crop parameter values was controlled by calibrating them on the basis of measurements. The humus N mineralization rate was also set on the basis of (even though indirect) measurements. In fact we estimated it on the basis of the measured N removal by the unfertilized crop. Although arguable, as it does not consider the possibility of a fertilizer priming effect (Jansson and Persson, 1982), this procedure does allow a better evaluation of this parameter value (Jarvis et al., 1996) than a completely empirical one. In this way the humus N mineralization turned out to be much slower in the more clayey soils of the Fienili and San Prospero sites than in the other two soils. This is in agreement with the experimental evidence that fine-textured soils are endowed with protective capacity toward organic matter (Jenkinson, 1988) and that mineralization rate decreases for an increasing clay content (Bosatta and Ågren, 1997).
To improve the understanding of the influence of N transformation parameter values on the simulation of the soil N dynamics we examined the effect of the feces decomposition parameter group on selected model outputs (Table 6). The intercept value for the RMSE response was quite high, especially in the case of the simulation of soil nitrate content, and changes in the selected parameter levels could only partially modify the overall model error. As far as the main factor effects are concerned, the kf factor showed the highest effect (ßkf = 3.80) on the RMSE for crop N uptake, whereas the fe factor had the highest effect on the other model responses. According to the negative sign of their coefficients, the increase from the –1 to the +1 level of the fe and fh factor values allows a reduction of the RMSE, thus improving the model efficiency. An RMSE reduction can be obtained also by reducing the kf value (positive sign of the coefficient). Changes in the fh value only slightly affected the RMSE for the crop N uptake. The 15 runs simulated on average a limited loss of N to streams (model's intercept equal to +1.44 g N m–2), a reduction of the mineral N reserve (–4.37 g N m–2), and a remarkable increase of the organic N storage (+15.36 g N m–2). Changing the kf value from the –1 to the +1 level produces a change in the same direction of the N leaching to streams and also of the mineral N reserve in soil, as shown by the positive sign of the relevant main factor coefficients. On the contrary, the organic N reserve decreases (negative sign of the coefficient). The opposite is true for the fe and fh factor coefficients. The most significant model component was the linear component, whereas the quadratic component was significant only for the RMSE for N uptake and soil nitrate content simulation. The interaction model component was significant only for the RMSE in the soil nitrate content simulation.
The sensitivity analysis results showed that better model performances could be obtained not only with high values of the fe and fh parameters, but also by a reduction of the feces decomposition rate (kf). Although we selected really high values of the fe and fh parameters for our simulations to reduce the RMSE, we should have chosen a lower kf value (e.g., 0.001 d–1, corresponding to the –2 level in the central composite design; Table 5) than the one we set initially for our simulations (0.035 d–1). These results suggest the possibility of a reduced turnover or else an immobilization of the soil organic matter from pig slurry spreading. Feces N immobilization is recognized as occurring on the short run in soil after pig slurry application (Flowers and Arnold, 1983; Kirchmann and Lundvall, 1993; Morvan et al., 1996).
|
CONCLUSIONS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
The sensitivity analyses, which were performed on some model parameters, showed that the choice of their values, more than any intrinsic model limitation, may also be partly invoked to explain the model's poor performances in our simulations. On the other hand, the implementation of a submodel for ammonia emissions should allow a better representation of the real system to be simulated.
Since the measured soil nitrate content varied not only in relation to the fertilizer N source and rate, but also according to the soil, this demonstrates that the soil type may have a specific influence not only on the water flow pattern, but also on the N transformation processes. As no rule exists indicating which values the model user should adopt for the N transformation parameters, according to soil type, a better quantification of soil type–N transformation relationships is needed.
Even though the model error could be partly reduced by a more accurate setting of the ammonia emissions and of the parameters regulating the feces N transformations, the model efficiency index values remained unsatisfactory. It is possible for the model not to include the simulation of processes that could have relevant effects on the soil N dynamics.
|
ACKNOWLEDGMENTS |
|---|
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
Related articles in JEQ:
This article has been cited by other articles:
|
|
 |
|
  J. D. Jabro, A. D. Jabro, and R. H. Fox Accuracy and Performance of Three Water Quality Models for Simulating Nitrate Nitrogen Losses under Corn. J. Environ. Qual., July 1, 2006; 35(4): 1227 - 1236. [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 | |||
