Sensitivity analysis and field testing of the RISK-N model in the Central Valley of Chile

agricultural water management 87 (2007) 251–260

available at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/agwat

Sensitivity analysis and field testing of the RISK-N model in the Central Valley of Chile Ricardo Oyarzun a,*, Jose Arumı´ b,1, Luis Salgado b,2, Miguel Marin˜o c,3 a

Centro de Estudios Avanzados en Zonas Aridas (CEAZA), Benavente 980, La Serena, Chile Departamento de Recursos Hı´dricos, Universidad de Concepcio´n, Av. Vicente Me´ndez 595, Chilla´n, Chile c Hydrology Program and Department of Civil and Environmental Engineering, University of California, Davis, CA 95616, USA b

article info

abstract

Article history:

We present the results from a sensitivity analysis and a preliminary short-term, site-scale

Accepted 30 July 2006

performance assessment of the analytical soil and groundwater nitrate transport RISK-N.

Published on line 18 September 2006

The study was carried out in the Central Valley of Chile, on a 2.6 ha corn (Zea mays L.) field underlain by a shallow unconfined aquifer during the cropping season 2000–2001. Nitrogen

Keywords:

levels in soils as well as NO3–N irrigation water and groundwater concentrations were

Nitrogen transport

monitored through the crop-growing period, the latter by a network of 16 monitoring wells.

Groundwater quality

A sensitivity analysis shows that both the nitrate flux from the vadose zone and NO3–N

Vadose zone

groundwater concentration are mainly influenced by the initial soil nitrogen levels, water

Agricultural management

input, and soil porosity. Also, simulated groundwater NO3–N levels are sensitive to changes

Analytical model

on the saturated zone denitrification constant. An additional analysis further reveals the significance of the latter parameter, in conjunction with the amount of applied nitrogen fertilizer. We obtained a good agreement between observed average and simulated values. While the model performs well when spatially averaged values are used (root mean square error, RMSE = 1.4 mg l1 of NO3–N), the prediction error increases (RMSE = 1.9 mg l1 of NO3–N) when the concentration in each well is considered. This fact could be explained by the time and space scale of the experiment and the characteristics of the RISK-N model. The model is easy to use and seems appropriate for mid- and long-term studies of nitrogen contamination in groundwater for agricultural conditions in the Central Valley of Chile and under limited field data availability conditions. However, it needs to be tested for longer periods and under different climatic conditions, soil types, and aquifer characteristics, before its range of applicability can be fully established and recognized. # 2006 Elsevier B.V. All rights reserved.

1.

Introduction

Modern agriculture is characterized by the intensive use of irrigation water and agrochemicals, such as fertilizers and

pesticides, that can become important environmental pollutants when applied inadequately. Furthermore, the use of lowefficiency irrigation methods, i.e., an excess of applied water, increases the leaching and transport of these substances to

* Corresponding author. Tel.: +56 51 204378; fax: +56 51 334741. E-mail addresses: [email protected] (R. Oyarzun), [email protected] (J. Arumı´), [email protected] (L. Salgado), [email protected] (M. Marin˜o). 1

Tel.: +56 42 208804; fax: +56 42 275303. Tel.: +56 42 208810; fax: +56 42 275303. 3 Tel.: +1 530 752 0684; fax: +1 530 752 5262. 0378-3774/$ – see front matter # 2006 Elsevier B.V. All rights reserved. doi:10.1016/j.agwat.2006.07.008 2

252

agricultural water management 87 (2007) 251–260

both surface waters and aquifers (Ng et al., 2000). In fact, the agricultural activity is recognized as a major nonpoint contamination source of both nutrients and pesticides to groundwater (Goderya et al., 1996; Breve´ et al., 1997; Antonopoulos, 2001). In Chile, the only country producing natural nitrate fertilizer, and an important consumer as well, little research has been carried out on the topic of nitrate contamination. Besides, the amount of data published on surface and groundwater nitrate concentrations is scarce (Arumı´ et al., 2005). Nevertheless, according to these data, industrial spills instead of agricultural activities seem to be the main source of NO3–N groundwater contamination problems identified in some areas of the country (Schalscha et al., 1979; Falco´n and Matutano, 2000). Moreover, there seems to exist special local characteristics in soils and aquifers that have prevented an extensive occurrence of high groundwater NO3–N levels, despite the intensive use of fertilizers in Chilean agriculture (Arumı´ et al., 2005). In contrast, in the USA and Europe, the problem of groundwater contamination has been studied for more than 25 years, identifying the nitrate ion as the most ubiquitous chemical contaminant in the aquifers (Spalding and Exner, 1993; Canter, 1997; Gusman and Marin˜o, 1999). Thus, it as been reported the occurrence of NO3–N groundwater contamination problems in agricultural areas, such as the so-called ‘‘corn belt’’ of Central USA, developed in relatively brief periods of time. The main factors producing these problems correspond to inadequate and excessive water and fertilization strategies (Watts et al., 2001). Thus, the evaluation of agricultural management practices and their consequences, with the use of tools, such as physically based models, can be of great importance in the prevention of nitrate contamination problems and in the assessment of soils and groundwater pollution risks (Antonopoulos, 2001; Mahmood et al., 2002). Nitrogen transport and fate models are numerous and diverse, ranging from simple leaching equations to complex mechanistic models (Canter, 1997). These models usually describe nitrate transport either in the vadose zone or in the saturated zone, but there have been few attempts to link both zones in a single model. The models considering the whole system are usually of the numerical type, and therefore, tend to require large amount of input data, which restricts their utilization (Canter, 1997). However, Mehran et al. (1984), in Canter (1997), pointed out that ‘‘for field-scale problems, onedimensional transport in the vadose zone with two-dimensional flow in the saturated zone allows a realistic description of transport processes in the overall flow region’’. They remarked that this type of modeling approach provides an efficient and practical tool for long-term predictions of the impact of agricultural activities on aquifer systems and evaluation of potential nitrate management alternatives. Similarly, Hutson and Wagenet (1991) indicated that given the inherent variability of many field soils, a simple model including all the relevant processes at an appropriate scale, is easier to use, requires few and simple data, and can be as accurate as a more complex model. Taken all this into account, the analytical RISK-N model of Gusman and Marin˜o (1999) was developed to simulate nitrogen cycling in soils and nitrate transport and fate in

soils and groundwater. This model was used by these researchers to simulate N-fate in a hypothetical corn plot, using meteorological, soil, hydrologic, and hydrogeological data from the South Platte River region of northeastern Colorado, and a conventional fertilizer management scheme. They concluded that the RISK-N model was capable of simulating nitrate leaching rates and groundwater concentrations that are consistent with those obtained by numerical models, while requiring fewer input variables. Also, Tabachow et al. (2001) presented a comparison between four biogeochemical models, APS, DAISY, NLEAP, and RISK-N, which simulated N cycling in the plant–soil–water–atmosphere environment. These authors concluded that RISK-N ‘‘seems best suited for modeling biogeochemical N cycles that are associated with N losses’’. However, at the same time they suggested that ‘‘future work must include comparison of RISKN simulations to laboratory and field studies at agricultural sites’’. Indeed, to the best of our knowledge, RISK-N has not been tested yet in its performance and sensitivity behaviour considering actual field cases. Therefore, the purpose of this paper is to test the RISK-N model usefulness under representative conditions of the Central Valley of Chile (33–408S), which concentrates most of the Chilean agricultural activity (ca. 90%), using easy to obtain field-data and literature-derived information. Also, the effect of uncertainty in model input was studied, to assess the suitability of using RISK-N as a practical tool at the field-scale when limited field information is available, that is, a very frequent case in most agricultural sites. Thus, this paper presents a sensitivity analysis of the model, and compares observed and predicted groundwater nitrate levels for the agro-environmental conditions of Central Chile.

2.

Materials and methods

2.1.

Model description

RISK-N is a physically based, analytical nitrate transport model that includes both the unsaturated and saturated zones. In the following paragraphs, the model is briefly described. For a more detailed explanation on water transport and nitrogen processes covered, see Gusman and Marin˜o (1999). Instead of using the Richards equation commonly found in numerical models, a simplified water-balance approach for soil water transport, i.e., infiltration and percolation, is used for the unsaturated soil zone. All water fluxes are assumed as one-dimensional, in a steady-state condition, and calculated as seasonal averages. Soil properties are assumed to be uniform in each zone. Nitrogen transport in each unsaturated soil zone is simulated on the premise of complete mixing, i.e., spatial average, of nitrogen concentrations. In the saturated zone, complete mixing is not assumed. Instead, a two-dimensional advective–dispersive equation is solved analytically. In simulating nitrogen-related processes, RISK-N separates the unsaturated soil into (1) upper root, (2) lower root, and (3) intermediate-vadose zones, considering that nitrogen transformation processes predominantly occur in the top 0.3 m of the soil, while roots often extend to a deeper level (Shaffer et al., 1991; Gusman and Marin˜o, 1999). Nitrogen

253

agricultural water management 87 (2007) 251–260

Table 1 – Time period, length, and crop development for each time period Season 1 2 3 4

Time period

Length (years)

Crop development stage

November 22–January 6 January 7–February 26 February 27–May 4 May 5–November 21

0.126 0.140 0.184 0.550

Corn planting and growth Vegetative growth and flowering Grain development and maturity No crop

fate processes represented include mineralization, ammonium immobilization, ammonium adsorption, ammonium volatilization, nitrification, denitrification, and nitrate leaching. For the saturated zone, and in addition to advection, nitrate is considered to be influenced by diffusion, dispersion, and denitrification.

2.2.

Experimental site

The RISK-N model was used to simulate the nitrate leaching as a consequence of agricultural activities in a 2.6 ha crop field located in the Central Valley of Chile, at 368230 S, in the Experimental Irrigation Station belonging to the Agricultural Engineering College (Facultad de Ingenierı´a Agrı´cola), University of Concepcio´n at Chilla´n, Chile. The soil belongs to the Tiuquilemu series, and is classified as loam, fine, thermic dystric fluventic xerochrepts, presenting clay to silty clay textures and low permeability (CNR, 1987). It includes a very low permeability layer at 1.4–1.8 m beneath the surface, which favors the existence of a shallow water table ca. 1.0 m depth in average. The climate of this region of Chile is classified as temperate Mediterranean. The thermal regime is characterized by an annual average temperature of 14 8C, with a maximum monthly average in January, the warmest month, of 28.8 8C. The annual hydrologic regime consists of a precipitation of 1025 mm and a pan evaporation of 1331 mm, which has a monthly peak of 245 mm in January (Novoa et al., 1989). The model was tested considering actual agricultural management: after plowing, a commercial hybrid Mycogen 2722 grain corn crop was sowed (65,000 plants ha1). Typical agricultural practices were followed, including furrow irrigation, fertilization levels, and pest management. A record of these activities was kept during the growing period of the crop, ca. 6 months. The first fertilization was accomplished during sowing (November 22, 2000), using two commercially available fertilizers: BZ Cargill special grain (9% N) and Hydro Premium (8% N). Nitrogen in both fertilizers is available in a fast release form, as NO3. They were mixed in a 2:1 ratio and were applied at a rate of 500 kg ha1. On January 7, 2001, the crop was fertilized again using granulated urea (45% N), a slow N release fertilizer, at a rate of 250 kg ha1. The harvest was initially

scheduled for March 2001, but it had to be delayed due to unfavorable climate conditions and agronomical reasons, and it was only accomplished by early November 2001. Given the fact that field measurements were scheduled and carried out until May 2001, a noncrop condition (bare soil) was assumed for the period May 5–November 21 (Table 1), in order to cover the minimum 1-year period required by RISK-N to perform the simulations (Gusman and Marin˜o, 1999).

2.3.

Field measurements

Before sowing (early November 2000), three replicated soil samples were taken in the unsaturated zone (0–90 cm) at 30 cm incremental depth, using an Eijkelkamp blade hand-drill. Also, both surface and groundwater samples were obtained. All these samples were analyzed to determine the initial nitrogen levels (nitrate and ammonium) in both soil and water. The soil samples were further analyzed for the determination of texture, organic matter content, and pH. Also, soil cores were obtained for the same depths (0–30, 30–60, and 60–90 cm) using 5 cm  5 cm Eljelkamp sampling steel cylinder, for the determination of field capacity and wilting point (Klute, 1986), bulk density, and, therefore, soil porosity (Table 2). In addition, the existing stubble of a previous crop was sampled, and its C/N ratio determined, in order to evaluate it as a possible additional nitrogen source or sink. Finally, a topographic survey of the experimental site was performed. After sowing, 16 observation wells were installed on January 19, 2001, using 5.9 cm inside diameter polyvinyl chloride (PVC) pipes, regularly bored and covered with geotextil material. They were positioned each 12 m apart on a regular square grid at the central part of the crop field, covering a sub-area of 1300 m2. Water table depths on each observation well were measured on a weekly basis using an electrical probe, allowing the determination of a fairly constant hydraulic gradient of 0.6%, on a westward direction (Fig. 1). The subsequent soil and water sampling was limited to this subarea, as well as the model domain defined for the simulations. Thus, soil-saturated hydraulic conductivity was field measured within this sub-area at three locations using the auger-hole method and the Hooghoudt’s solution (Porta et al., 1999), and an average value of 4.24 m day1 was determined, which was used as input for the model.

Table 2 – Soil texture, physical, and chemical properties Depth (cm)

Sand (%)

Silt (%)

Clay (%)

ufc (cm3 cm3)

uwp (cm3 cm3)

r (g cm3)

Om (%)

pH (water)

0–30 30–60 60–90

21.6 29.6 45.9

35.2 32.2 28.6

43.2 38.2 25.5

0.45 0.46 0.37

0.29 0.33 0.27

1.21 1.07 1.08

3.68 1.35 0.74

5.8 6.2 6.5

Note: ufc, soil moisture content at field capacity; uwp, soil moisture content at permanent wilting point; r, bulk density; Om, organic matter.

254

agricultural water management 87 (2007) 251–260

Fig. 1 – Layout map of the experimental site showing the network of observation wells (1–16) and (panel A) topography (contours every 25 cm) and (panel B) groundwater levels (contours every 2.5 cm) and flow direction (segmented arrow) for a representative day (January 19, 2001).

Approximately every 3 weeks, groundwater samples were obtained from each well for NO3–N, NH4–N, and complementary (pH, Eh) analyses. At the same time, three replicated soil samples at three depths (0–30, 30–60, and 60–90 cm) were taken in order to monitor soil nitrogen (NO3, NH4+) levels as well as water content, the latter determined by the gravimetric method, oven-drying the samples at 105 8C for 24–40 h (Jury et al., 1991). All the analyses were performed at the Soils Laboratory and Quality Control Laboratory of the Agronomy and Agricultural Engineering Colleges, respectively, of the University of Concepcio´n at Chilla´n. For each soil sample, NO3–N and NH4+–N were determined by the colorimetric and the Nessler method (Longeri et al., 1979; Robarge et al., 1983), and organic matter by moist combustion. The N levels of crop stubble were determined by the Kjeldhal method. Both, NO3–N and NH4+– N in surface and groundwater, were determined colorimetrically, and Eh and pH were determined using the potentiometric method.

Meteorological data during the experiment were obtained from a semiautomatic station at the research site, about 200 m from the experimental field.

2.4.

Input data estimation

2.4.1.

Nitrogen dynamics parameters

Mineralization, nitrification, and denitrification rates coefficients for the unsaturated soil zones were calculated using the internal algorithms available on RISK-N. A detailed description of both equations and the assumptions involved are presented in detail in Gusman and Marin˜o (1999). Although the ammonium sorption coefficient is included in several models, there is not detailed information on this parameter in the literature. Published values range between 0.5 and 3.0 cm3 g1 (Lafolie, 1991; Vereecken et al., 1991; Jemison et al., 1994; Antonopoulos, 2001). For this study, an average value of 1.5 cm3 g1 was considered.

agricultural water management 87 (2007) 251–260

Regarding the first-order denitrification rate for the saturated zone, published values range between 0.001 and 0.1 day1 (Jemison et al., 1994; Lafolie et al., 1997; Ng et al., 2000; Antonopoulos, 2001). Herein, an average value of 0.01 day1 was considered initially, although as explained below this figure was later reduced to 0.001 day1 during the trend calibration process. Finally, as the model does not have the capability to predict plant nitrogen uptake rates, the user must define them from own data or from the literature (Gusman and Marin˜o, 1999). Thus, we used data presented by Longeri et al. (1987), for a corn crop grown on similar soil, climatic, and time of the year conditions as our research. This information was adjusted to an S-shaped uptake rate as described by Watts and Hanks (1978) during the calibration process.

2.4.2.

Water flow and meteorological parameters

The water applied by irrigation was estimated using the option available in RISK-N (Gusman and Marin˜o, 1999). The leaching fraction was assumed at 30%, considering the average efficiency assigned to the furrow irrigation practice given by studies carried out by researchers of the Irrigation and Drainage Department of the University of Concepcion in crop fields of the Central Valley of Chile between 35 and 368S (Holzapfel, 2001). Evapotranspiration rates were obtained from evaporation measurements from a class A pan, existing at the experimental site using a pan coefficient of 0.8, considering the meteorological and environmental conditions around the evaporation pan on the field (Jensen et al., 1990). Crop coefficients for grain corn were obtained from Doremboos and Pruitt (1977). The upper root evapotranspiration fraction was assumed equal to the root fraction of this soil zone. Surface runoff was determined using the curve-number methodology (SCS 1972, in Rawls et al., 1995). Finally, the unsaturated zone average soil temperature for each season was determined from measured soil temperatures at 5 cm depth, adjusted to different depths by the use of the sinusoidal wave approach (Rijtema and Kroes, 1991; Porta et al., 1999).

2.4.3.

Saturated zone and nitrate transport parameters

From measurements of the water table depths and well installation logs, the mean thickness of the soil-saturated zone over the low-permeability layer was estimated to be of 70 cm with a porosity of 0.56. A value of 0.034 cm2 h1 was used for the nitrate molecular diffusion coefficient, which corresponds to an average value of the data presented by Watts (1975). Longitudinal dispersivity of the saturated medium was estimated considering the expressions proposed by Xu and Eckstein (1995). Transverse dispersivity was assumed as 1/10 of the longitudinal dispersivity. The values were 4.0 and 0.4 m, respectively.

2.5.

Sensitivity analysis

To examine the model response to changes of specific input data, i.e., to have an indication of the required accuracy at which each parameter should be available, a sensitivity analysis of the model was performed. From the original

255

dataset, obtained from field measurements and literature as explained above, the base simulation was established. From this, the sensitivity analysis was performed, assessing the effect produced by a given variation of the input data on the model output. Specifically, the change of both the cumulative NO3–N flux from the vadose zone (g m2) as well as the maximum NO3–N groundwater concentration (mg l1), for 1 and 20 years, were the model outputs of interest. The reason of choosing both variables lies in the fact that the RISK-N model may be used either to estimate the nitrate leaching from the unsaturated soil and then export the results into a more complex groundwater transport model like MODFLOW–MT3D, or be used independently to simulate nitrate transport in the saturated zone (Gusman and Marin˜o, 1999). Spider plots were used for the representation of the results (Eschenbach, 1992). Furthermore, in order to scope for interaction between parameters within the model, a combinatory analysis was carried out, considering the simultaneous change of two selected input variables with a special interest for the problem under study: the nitrogen fertilization rate and the denitrification rate.

2.6.

Statistical analysis

The model performance analysis included the determination, after the preliminary model calibration process, of the root mean square error (RMSE), the relative RMSE (RRMSE) and the coefficient of residual mass (CRM), for selected dates within the experimental period considered, i.e., end of seasons 2 and 3. These relationships have the following expressions (Loague and Green, 1991; Antonopoulos, 2001): sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Pn 2 i¼1 ðPi  Oi Þ RMSE ¼ n RRMSE ¼

RMSE  100 Oavg

Pn CRM ¼

i¼1

O  Pin

i¼1

Pn

i¼1

(1)

(2)

Pi

Oi

(3)

where Pi and Oi are the simulated and measured values of groundwater Nitrate–N concentration, respectively, n the number of pairs of data (observed and measured values), and Oavg is the mean of measured values. The optimal values of RMSE, RRMSE, and CRM criteria are zero. RMSE is expressed in mg l1 and RRMSE in percentage. Positive values of CRM indicate that the model underestimates the measurements, and negative values indicate overestimates.

3.

Results and discussion

3.1.

Model sensitivity to changes in input data

The 1-year cumulative effect on the total NO3–N flux from the vadose zone due to the change in values of different input data is shown in Fig. 2 (panels A–C). Among the N dynamics related parameters, the starting nitrogen levels in the soil (NO3–N, NH4+–N, and organic N) have major influence on the nitrate flux from the vadose zone (panel A). Also, the model shows

256

agricultural water management 87 (2007) 251–260

Fig. 2 – Model sensitivity for vadose zone N–NO3S flux to changes in (panel A) nitrogen-related parameters (Ic, soil initial nitrogen; Wl, nitrate concentration in irrigation water and leaching fraction; Ni, nitrogen input; Np, nitrogen plant uptake; Kd, ammonium distribution coefficient), (panel B) changes in water fluxes (Rf, rainfall; Et, crop evapotranspiration; Ur, upper root evapotranspiration fraction; Ro, surface runoff), and (panel C) soil properties (St, average temperature; Fu, field capacity upper root zone; Un, unsaturated soil porosity; Ml, moisture level).

higher sensitivity to the amount of plant nitrogen uptake instead of the amount of nitrogen applied (fertilization and residual N). Thus, a proper assessment of nitrogen plant uptake makes the use of the RISK-N model more difficult on cropping systems where this variable needs to be estimated and rarely can be directly measured. A clear example of this case corresponds to fruit tree orchards. Finally, special care must be taken into account regarding model response to nitrogen applications, as is presented later. The simulated NO3–N flux shows low sensitivity to the ammonium adsorption coefficient. This fact is in agreement with Lafolie (1991) in that an arbitrary but plausible value could be assumed for this parameter and that it has a minor effect on the nitrate balance results. Regarding water-fluxes parameters (Fig. 2(panel B)), vadose zone nitrate leaching is more sensitive to water inflow to the system, through rainfall, than to surface water outflow from the system through evapotranspiration and runoff. However, it must be recalled that the model was used (and, therefore, this analysis performed) considering the available option for automatic irrigation input determination. Thus, the eventual effect of the

irrigation factor could not be analyzed individually. Therefore, it is likely as well that an excessive or inefficient irrigation rate, that favors percolation, may have also an important effect on the leaching processes in the vadose zone, however, this hypothesis could not be evaluated in the present study because of the above-mentioned reasons. Considering the model sensitivity to changes in soil parameters and conditions (panel C), the important effects of both soil moisture level and soil porosity are evident. Even more, RISK-N shows the greatest sensitivity to those parameters in the overall analysis (i.e., panels A to C). This can be explained in terms of the relationships between these variables and the determination method for the unsaturated soil zone denitrification rate used in the RISK-N model. This case confirms the statement of Marchetti et al. (1997), in that both a decrease of the soil moisture level as well as an increase in soil porosity cause a decrease on the denitrification rate, a rise in nitrate availability in the unsaturated soil zone and thus an increase of the NO3–N flux from the vadose zone. The effect of input data changes over the spatially averaged NO3–N groundwater concentrations, for 1-year simulations, is presented in Fig. 3 (panels A–C). In addition to the parameters discussed earlier (presented in Fig. 2), the firstorder denitrification rate for the saturated zone and the soil porosity of the saturated zone were included and considered. Since the model did not show sensitivity to the soil ammonium adsorption coefficient, this parameter was excluded from Fig. 3 for the sake of clarity. The effect of the variation of the different parameters on the groundwater nitrate concentration (Fig. 3) is rather similar to the model response in terms of nitrate flux from the vadose zone (Fig. 2). However, the sensitivity exhibited by the RISK-N model to the saturated zone denitrification rate is also important (panel A), thus an incorrect value may lead to erroneous conclusions, confirming what was previously suggested by Gusman and Marin˜o (1999). This case must be taken into account in future uses of the model. When the parameters related to water fluxes are considered (panel B), Nitrate–N groundwater concentrations appear to be a direct consequence of the N flux from the vadose zone, a fact that explains the shape similarities between panels B in Figs. 2 and 3. The model sensitivity to soil porosity in the saturated zone (panel C) shows a different result compared to what was obtained for the vadose zone. The reason for this difference lies in the fact that the saturated zone denitrification rate is a constant value, fixed by the model’s user, so it does not depend on porosity. In exchange, according to the Eq. (33) of Gusman and Marin˜o (1999), the nitrate concentration in groundwater calculated by the RISK-N model is inversely proportional to the aquifer porosity, which explains the results obtained. The sensitivity analyses on a long-term basis (20-year simulations) show a similar curve shape for both the nitrate flux from the vadose zone and the nitrate groundwater concentrations. Thus, to conserve space, these plots were not included in this paper. However, changes in input parameter values resulted in a greater change, in percentage, both in the nitrate flux and in the nitrate groundwater levels. Regarding model sensitivity, special attention should be paid to the simultaneous change in two model inputs mentioned earlier: the nitrogen fertilization and the saturated

agricultural water management 87 (2007) 251–260

257

Fig. 4 – Model sensitivity for maximum Nitrate–N concentration in groundwater in 1-year simulation as resulted from simultaneous changes in applied nitrogen fertilizer and first-order denitrification constant for the saturated zone.

Fig. 3 – Model sensitivity for maximum N–NO3S groundwater concentrations to changes in (panel A) nitrogen-related parameters (Ic, soil initial nitrogen; Wl, nitrate concentration in irrigation water and leaching fraction; Dr, first-order denitrification constant for the saturated zone; Ni, nitrogen input; Np, nitrogen plant uptake), (panel B) changes in water fluxes (Rf, rainfall; Et, crop evapotranspiration; Ur, upper root evapotranspiration fraction; Ro, surface runoff), and (panel C) in soil properties (St, average temperature; Fu, field capacity upper root zone; Un, unsaturated soil porosity; Ml, moisture level; Sn, saturated zone soil porosity).

fertilizer applications. Moreover, NH4+–N levels were lower, generally about one-half of the NO3–N contents (data not shown). This phenomenon of ‘‘temporal disappearance’’ of nitrogen fertilizer and its subsequent reappearance some months later has been described by other investigators in similar kind of studies (Kersebaum and Richter, 1991; Lafolie, 1991). These authors have attributed this situation to the occurrence of a reversible microbial immobilization process. In our case, considering that the previous crop stubble had a C/ N ratio over 100, it is likely that microbial activity played an important role in short-term soil nitrogen dynamics (e.g., immobilization), although we acknowledge the lack of additional information to support this assumption.

zone denitrification rate. In fact, the effect of applied N fertilizers on the nitrate leaching and groundwater levels cannot be fully appreciated if the model’s user does not have a reliable estimation or confident measurement of the aquifer denitrification potential. Indeed, as shown in Fig. 4, the effect of the simultaneous change of these variables is nonlinear. An increase in the nitrogen application rate by fertilization together with a lower denitrification rate promotes an important increase on the aquifer NO3–N levels. Thus, the importance of having confident data or making a good estimation of these parameters is clear, especially for longterm simulations.

3.2.

Nitrate levels in soils and groundwater

The time variation of both soil and water field measured NO3– N concentrations is shown in Fig. 5. Regarding soil nitrate, except for the last sampling date (May 4, 2001), the levels are generally low, and do not clearly reflect the effect of the

Fig. 5 – Soil nitrate concentrations at different depths (panel A, in ppm) and water (Iw, irrigation; Gw, groundwater) nitrate concentrations (panel B, in mg lS1) during the course of the experiment. Note: The first measurement date is actually 8 days before sowing.

258

agricultural water management 87 (2007) 251–260

The rise in the soil nitrate content at the end of the experiment coincides with a minor increase in the groundwater NO3–N concentrations (panel B). Nevertheless, an overall decreasing trend of groundwater nitrate levels took place during the course of the experiment. All these values are below 10 mg l1, the maximum contaminant level for drinking water according Chilean health standards (INN, 1984). This case can be explained as follows. First, the time period when the measurements were made was coincident with the season of lower rainfall and major crop soil cover, so there was only a minor nitrate leaching potential from the soil to the aquifer. Certainly, in winter, with an increase in precipitation and no crop development and activity, the nitrate flux from the vadose zone might be larger, as pointed out elsewhere (Longeri et al., 1987; Moreno et al., 1996). Secondly, and besides the effect of plant nitrate uptake, the possibility of the occurrence of denitrification processes should be considered. Indeed, as pointed out by Lafolie et al. (1997), under crop conditions that combine intensive irrigation with high soil temperatures, short periods with a high water saturation ratio may induce brief but intensive microbiological activity leading to high denitrification rates. Likely, those conditions occurred frequently during this experiment, carried out in the spring–summer crop season and furrow-irrigated. In addition, groundwater pH and Eh values were between 5.97 and 7.25 and between 39.2 and 42 mV, respectively, which represent qualitatively unfavorable geochemical conditions for the existence of NO3 (Lowrance and Pionke, 1989). However, it must be acknowledged that these figures, specially regarding Eh, are illustrative rather than conclusive, given some delays occurred from sample taking to measurement in laboratory, and therefore they must be only considered as indicative. With respect to NH4+–N groundwater concentrations, they were negligible during the experiment, always lower than 0.5 mg l1.

3.3. Trend calibration process and model performance assessment Given the aims of this research, the RISK-N model characteristics, and the time duration of the experiment, the purpose of

Table 3 – Average values and standard deviation (in parenthesis) of groundwater Nitrate–N concentrations (mg lS1) Date February 26, 2001 May 4, 2001 Overall average

Field measured

Model predicted

2.57 (1.94) 1.84 (0.91) 2.14 (1.42)

0.81 (0.71) 1.10 (0.85) 0.98 (0.78)

the study was the assessment of the model’s capability to reproduce seasonal trends in NO3–N concentrations in groundwater. Nevertheless, considering the well-established unsuitability of using short-time data for the calibration and validation of long-term simulation models (Grant, 1991; Jemison et al., 1994) a rigorous RISK-N model calibration was not done. Thus, a limited change is made in the range of values of both the nitrogen plant seasonal uptake and the firstorder denitrification rate for the saturated zone, as selected from the sensitivity analysis. This operation favors a good agreement between simulated and observed groundwater NO3–N levels on the two dates selected for this purpose (coincidental with field measurements): February 26, 2001 (end of season 2) and May 4, 2001 (end of season 3). Finally, the results presented on Tables 3 and 4 were obtained selecting a 0.001 day1 value for the denitrification rate (a variable playing a major role in the adjustment process) and a N plant uptake ratio of 3:9:4 (kg ha1) for cropping seasons 1, 2, and 3, respectively (Table 1). It should be stated that the model tends to simulate lower values of NO3–N concentrations in groundwater than field measurements (CRM > 0). Regarding model performance, the results found in this study are comparable to those reported by other studies using more complex numerical models on similar crop systems and time duration of experiments and simulations (e.g., Jemison et al., 1994; Antonopoulos, 2001; Mahmood et al., 2002). A spatial analysis of the differences between simulated and observed values for May 4 shows that the major disagreement was observed in the test wells located in the westernmost column of the grid (Fig. 6(A)). This case can be explained in terms of the time span of the experiment, the nitrate transport aquifer parameters

Fig. 6 – Spatial distribution of groundwater N–NO3S concentrations for May 4, 2001: (A) differences (absolute value) between measured and simulated values, (B) measured values, and (C) simulated values. Circles indicate position of observation wells. Figure is oriented so groundwater flow is from left to right.

agricultural water management 87 (2007) 251–260

Table 4 – Overall (February 26 and May 4) model performance considering each of the 16 observation wells individually Indicator

Magnitude of error

RMSE (mg l1) RRMSE (%) CRM

1.94 90.80 0.54

Table 5 – Average Nitrate–N concentrations in groundwater (mg lS1) when westernmost column of observation wells is not considered Date February 26, 2001 May 4, 2001 Overall average

Average observed values

Average predicted values

2.2 1.9 2.0

1.1 1.5 1.3

(e.g., saturated zone Darcy flux velocity), and the longitudinal and transverse dispersivity parameters. Indeed, while groundwater nitrate concentrations are rather heterogeneously distributed throughout the experimental plot (Fig. 6(B)), model simulated values exhibit a marked decreasing trend in the direction of the groundwater flow (Fig. 6(C)). Thus, the combined effect of the mentioned factors would result in the model’s difficulty to simulate NO3–N concentrations in the westernmost observation wells, for a time span shorter than 1 year. When data from the westernmost observation wells are not considered, a better agreement between simulated and observed values is obtained, decreasing the error magnitude, as shown in Tables 5 and 6. If only the average from the 16 wells (both observed and simulated values) is considered, the magnitude of the error diminishes (RMSE = 1.40; RRMSE = 61.02%). This is consistent with the fact that the RISK-N model considers as input data the spatial averages of nitrogen concentrations, and therefore it fails to perform an accurate simulation of groundwater NO3–N concentrations for each of the 16 observation wells individually. Also, the inclusion in the RISK-N model of several subroutines adds an additional inaccuracy source on the simulation performed, so better results are obtained when average values are considered (Marchetti et al., 1997). A good example of this is the method for determining the seasonal denitrification rate in the unsaturated zone considered by the RISK-N model and its importance on the simulated NO3–N groundwater concentration (Fig. 4). Thus, the model is less accurate at a more detailed spatial and temporal scale, a trait that is characteristic of analytical-type models.

Table 6 – Model performance when westernmost column of observation wells is not considered Indicator

Magnitude of error 1

RMSE (mg l ) RRMSE (%) CRM

1.3 66.2 0.4

4.

259

Conclusions

The analytical RISK-N model, a quasi two-dimensional model, was used to simulate nitrate groundwater concentrations as a consequence of agricultural practices in the Central Valley of Chile. In the model, all water fluxes through the unsaturated soil were taken to be one-dimensional, steady-state, and calculated as seasonal averages, while the groundwater system was represented by a one-dimensional flow but twodimensional dispersion model. A sensitivity analysis highlighted the importance of the initial soil N levels, soil porosity, and water input amounts on both nitrate leaching from the vadose zone and the groundwater NO3–N concentrations simulated by the model. A combinatory analysis showed interaction between the selected model parameters, denitrification rate and fertilizer application, which suggests that further attention should be paid to these variables on future uses of this model. The model performance assessment showed best results when average observed and simulated values were considered. Less robust results were found when individual values were used. Results from this 6-month exploratory field study confirmed that the RISK-N model is relatively easy to use, does not need an extensive input data set, and seems adequate for midand long-term simulations of nitrogen concentrations in soils and groundwater as consequence of agricultural management practices and where field data is limited. However, it needs to be tested for longer time periods and different climatic conditions, soil types, and aquifer characteristics, before its range of applicability can be fully established and recognized.

Acknowledgments We thank the Research Office of the University of Concepcion for funding this project (DIUC 200.133.005-1.0). Also, field and technical assistance provided by Carlos Cea and Juan Navarro is greatly appreciated. The paper benefited from the comments of two anonymous reviewers.

references

Antonopoulos, V.Z., 2001. Simulation of water and nitrogen balances of irrigated and fertilized corn-crop soil. J. Irrig. Drain. Eng. 127 (2), 77–83. Arumı´, J.L., Oyarzun, R., Sandoval, M., 2005. Natural protection against groundwater pollution by nitrates in the Central Valley of Chile. Hydrol. Sci. J. 50 (2), 331–340. Breve´, M.A., Skaggs, R.W., Gilliam, J.W., 1997. DRAINMOD-N, a nitrogen model for artificially drained soils. Trans. ASAE 40 (4), 1067–1075. Canter, L.W., 1997. Nitrates in Groundwater. CRC, Boca Raton, Florida. CNR, 1987. Estudio de suelos, Proyecto Itata, etapa 1. Comisio´n Nacional de Riego, Agrolog and R & Q Ingenierı´a. Tomo II, 541 pp. (unpublished). Doremboos, J., Pruitt, W.O., 1977. Guidelines for Predicting Crop Water Requirements, Irrig. Drain. Paper 24, 2nd ed., FAO, Rome, 114 pp.

260

agricultural water management 87 (2007) 251–260

Eschenbach, T.G., 1992. Spiderplots versus tornado diagrams for sensitivity analysis. Interfaces 22 (6), 44–46. Falco´n, M.F., Matutano, G., 2000. Contaminacio´n de nitratos en las aguas subterra´neas del sector poniente de la cuenca de Santiago, Regio´n Metropolitana Chile. In: Fajardo, A. (Ed.), Proceedings of the IX Congreso Geolo´gico Chileno. pp. 37–41. Goderya, F.S., Woldt, W.E., Dahab, M.F., Bogardi, I., 1996. Comparison of two transport models for predicting nitrates in percolating waters. Trans. ASAE 39 (6), 2131–2137. Grant, R.F., 1991. The distribution of water and nitrogen in the soil–crop system: a simulation study with validation from a winter wheat field trial. Fert. Res. 27, 199–213. Gusman, A.J., Marin˜o, M.A., 1999. Analytical modeling of nitrogen dynamics in soils and groundwater. J. Irrig. Drain. Eng. 125 (6), 330–337. Holzapfel, E.A., 2001. Optimizacio´n de los recursos hı´dricos en la agricultura. In: Proceedings of the III Encuentro de Aguas. Santiago, Chile (Cd Rom). Hutson, J.L., Wagenet, R.J., 1991. Simulating nitrogen dynamics in soils using a deterministic model. Soil Use Manage. 7 (2), 74–78. INN, 1984. Norma Chilena Oficial para Agua Potable, Nch 409. Instituto Nacional de Normalizacio´n, Santiago, Chile, 9 pp. (unpublished). Jemison, J.M., Jabro, J.D., Fox, R.H., 1994. Evaluation of LEACHM: II simulation of nitrate leaching from nitrogen-fertilized and manure corn. Agron. J. 86, 852–859. Jensen, M.E., Burman, R.D., Allen, R.G. (Eds.), 1990. Evapotranspiration and Irrigation Water Requirements. ASCE Manuals and Reports on Engineering Practice No. 70, 249 pp. Jury, W.A., Gardner, W.R., Gardner, W.H., 1991. Soil Physics, 5th ed. John Wiley and Sons, Inc., New York. Kersebaum, K.C., Richter, J., 1991. Modelling nitrogen dynamics in a plant–soil system with a simple model for advisory purposes. Fert. Res. 27, 273–281. Klute, A., 1986. Water retention: laboratory methods. In: Klute, A. (Ed.), Methods of Soil Analysis, Part 1, 2nd ed., vol. 9. Agronmy, pp. 635–662. Lafolie, F., 1991. Modelling water flow, nitrogen transport and root uptake including physical non-equilibrium and optimization of the root water potential. Fert. Res. 27, 215–231. Lafolie, F., Bruckler, L., de Cockborne, A.M., Laboucarie´, C., 1997. Modeling the water transport and nitrogen dynamics in irrigated salad crops. Irrig. Sci. 17, 95–104. Loague, K., Green, R.E., 1991. Statistical and graphical methods for evaluating solute transport models: overview and application. J. Contam. Hydrol. 7, 51–73. Longeri, L., Etchevers, J., Venegas, J., 1979. Metodologı´a de perfusio´n para estudios de nitrificacio´n en suelos. Cienc. Inv. Agraria (Chile) 6, 295–299. Longeri, L., Vidal, I., Lo´pez, I., 1987. Influencia de la nitrapirina sobre la transformacio´n del nitro´geno de la urea y rendimiento del maı´z. Cienc. Inv. Agararia (Chile) 14, 15–22. Lowrance, R.R., Pionke, H.B., 1989. Transformations and movement of nitrate in aquifer systems. In: Follet, R.F. (Ed.), Nitrogen Management and Groundwater Protection. Elsevier, Amsterdam, pp. 373–392.

Mahmood, B., Russel, J.M., Wall, G.B., 2002. Field-scale nitrate simulation. Trans. ASAE 45 (6), 1835–1842. Marchetti, R., Donatelli, M., Spallaci, P., 1997. Testing denitrification functions of dynamics crop models. J. Environ. Qual. 26, 394–401. Moreno, F., Cayuela, J.A., Ferna´ndez, J.E., Ferna´ndez-Boy, E., Murillo, J.M., Cabrera, F., 1996. Water balance and nitrate leaching in an irrigated maize crop in SW Spain. Agric. Water Manage. 32, 71–83. Ng, H.Y.F., Drury, C.F., Serem, V.K., Tan, C.S., Gaynor, J.D., 2000. Modeling and testing of the effect of tillage, cropping and water management practices on nitrate leaching in clay loam soil. Agric. Water Manage. 43, 111–131. Novoa, R., Villaseca, S., del Canto, P., Rouanet, J., Sierra, C., del Pozo, A., 1989. Mapa Agroclima´tico de Chile. INIA, Ministerio de Agricultura. Porta, J., Lopez-Acevedo, M., Roquero de Laburu, C., 1999. Edafologia para la Agricultura y el Medio Ambiente, 2nd ed. Mundi Prensa, Bilbao. Rawls, W.J., Ahuja, L.R., Brakensiek, D.L., Shirmohammadi, A., 1995. Infiltration and soil water movement. In: Maidment, D.R. (Ed.), Handbook of Hydrology. McGraw-Hill, New York, pp. 5.1–5.51. Rijtema, P.E., Kroes, J.G., 1991. Some results of nitrogen simulations with the model ANIMO. Fert. Res. 27, 189–198. Robarge, W.P., Edwads, A., Johson, B., 1983. Water and waste water analysis for nitrate via nitratation of salicylic acid. Comm. Soil Sci. Plant Anal. 14, 1207–1215. Schalscha, E.B., Vergara, I., Schirado, T., Morales, M., 1979. Nitrate movement in a Chilean agricultural area irrigated with untreated sewage water. J. Environ. Qual. 8, 27–30. Shaffer, M.J., Halvorson, A.D., Pierce, F.J., 1991. Nitrate leaching and economic analysis package (NLEAP): model description and application. In: Follet, R.F., Keeney, D.R., Cruse, R.M. (Eds.), Managing Nitrogen for Groundwater Quality and Farm Profitability. SSSA, Madison, pp. 285–322. Spalding, R.F., Exner, M.E., 1993. Occurrence of nitrate in groundwater—a review. J. Environ. Qual. 22, 392–402. Tabachow, R.M., Peirce, J.J., Richter, D.D., 2001. Biogeochemical models relating soil nitrogen losses to plant available N. Environ. Eng. Sci. 18 (2), 81–89. Vereecken, H., Vanclooster, M., Swerts, M., Diels, J., 1991. Simulating water and nitrogen behaviour in soils cropped with winter wheat. Fert. Res. 27 (2–3), 233–243. Watts, D.G., 1975. Soil–water–nitrogen–plant model for irrigated corn on coarse textured soils. PhD dissertation, Department of Agricultural and Irrigation Engineering, Utah State University, Logan, Utah, USA. Watts, D.G., Hanks, R.J., 1978. A soil–water–nitrogen model for irrigated corn on sandy soils. Soil Sci. Soc. Am. J. 42 (3), 492–499. Watts, D.G., Schepers, J.S., Spalding, R.F., 2001. Evolucio´n y control de la contaminacio´n por nitrato en una zona bajo riego. In: Proceedings of the III Encuentro de Aguas. Santiago, Chile (Cd Rom). Xu, M., Eckstein, Y., 1995. Use of weighted least-squares method in evaluation of the relationship between dispersivity and field scale. Ground Water 33 (6), 905–908.