JEQ Journal of Natural Resources and Life Sciences Education
HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
QUICK SEARCH:   [advanced]


     

Published online 6 July 2006
Published in J Environ Qual 35:1213-1226 (2006)
DOI: 10.2134/jeq2005.0377
© 2006 American Society of Agronomy, Crop Science Society of America, and Soil Science Society of America
677 S. Segoe Rd., Madison, WI 53711 USA
This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF) Free
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Services
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Citing Articles
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Tiktak, A.
Right arrow Articles by Vanclooster, M.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Tiktak, A.
Right arrow Articles by Vanclooster, M.
GeoRef
Right arrow GeoRef Citation
Agricola
Right arrow Articles by Tiktak, A.
Right arrow Articles by Vanclooster, M.
Related Collections
Right arrow Ground Water Quality
Right arrow Pesticides
Right arrow Vadose Zone Risk Assessment
Right arrow Vadose Zone Regulation
Right arrow Predictive soil mapping
Right arrow Agricultural Pesticides

TECHNICAL REPORTS

Ground Water Quality

Mapping Ground Water Vulnerability to Pesticide Leaching with a Process-Based Metamodel of EuroPEARL

A. Tiktaka,*, J. J. T. I. Boestenb, A. M. A. van der Lindenc and M. Vancloosterd

a Netherlands Environmental Assessment Agency (NEAA), P.O. Box 303, 3720 AH Bilthoven, the Netherlands
b Alterra, P.O. Box 47, 6700 AA Wageningen, the Netherlands
c RIVM, P.O. Box 1, 3720 BA Bilthoven, the Netherlands
d Université Catholique de Louvain, Croix du Sud 2, 1348 Louvain-la-Neuve, Belgium

* Corresponding author (aaldrik.tiktak{at}mnp.nl)

Received for publication September 30, 2005.

   ABSTRACT
TOP
 ABSTRACT
INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 CONCLUSIONS
 REFERENCES
 
To support EU policy, indicators of pesticide leaching at the European level are required. For this reason, a metamodel of the spatially distributed European pesticide leaching model EuroPEARL was developed. EuroPEARL considers transient flow and solute transport and assumes Freundlich adsorption, first-order degradation and passive plant uptake of pesticides. Physical parameters are depth dependent while (bio)-chemical parameters are depth, temperature, and moisture dependent. The metamodel is based on an analytical expression that describes the mass fraction of pesticide leached. The metamodel ignores vertical parameter variations and assumes steady flow. The calibration dataset was generated with EuroPEARL and consisted of approximately 60 000 simulations done for 56 pesticides with different half-lives and partitioning coefficients. The target variable was the 80th percentile of the annual average leaching concentration at 1-m depth from a time series of 20 yr. The metamodel explains over 90% of the variation of the original model with only four independent spatial attributes. These parameters are available in European soil and climate databases, so that the calibrated metamodel could be applied to generate maps of the predicted leaching concentration in the European Union. Maps generated with the metamodel showed a good similarity with the maps obtained with EuroPEARL, which was confirmed by means of quantitative performance indicators.

Abbreviations: EU, European Union • EU-15, European Union without the new Member States • PEARL, Pesticide Emission Assessment at Regional and Local Scales • SPADE, Soil Profile Analytical Database of Europe


   INTRODUCTION
TOP
 ABSTRACT
 INTRODUCTION
MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 CONCLUSIONS
 REFERENCES
 
THE SIXTH ENVIRONMENT ACTION PROGRAM, as adopted by the European Parliament and the European Council, provides for the development of a Thematic Strategy on the Sustainable Use of Pesticides. The principle objective of this program is the reduction of the impact of pesticides on human health and the environment and, hence, a sustainable use of pesticides. To report and monitor the progress made in achieving the objectives of this strategy, harmonized pesticide impact indicators are needed, which evaluate the effectiveness of the Action Program across EU Member States [COM (2002) 349 final, 2002]. Most indicators that are currently used deal only with changes in volumes and application frequencies of pesticides. However, because of the different physicochemical properties of pesticides, such parameters do not necessarily correlate with a decrease in risk. Therefore, other types of indicators are required; indicators that deal with effects on humans, non-target organisms, and environmental compartments.

Leaching to the ground water is one of the aspects considered in the Thematic Strategy, because ground water is a major source of drinking water in Europe and because pressures exerted by pesticides on European ground water bodies are high. The leaching potential of pesticides can be calculated with dynamic, multi-layer, mechanistic models at different spatial scales, including the field scale (Boesten and van der Linden, 1991; Vanclooster et al., 2000), the regional scale (Capri et al., 2000), and the national scale (Tiktak et al., 1996, 2002b). These models take into account transient flow, hydrodynamic dispersion, nonlinear adsorption, degradation, and uptake of pesticides by plant roots. The disadvantages of process-based pesticide-leaching models are (i) that they contain a large number of parameters, which may be difficult to identify directly or which may not be available at larger scales and (ii) that they are complex in nature, which make them hard to understand by non-expert stakeholders and also laborious to use.

To mitigate the above mentioned problems, simpler leaching models could be used. As an alternative to transient leaching models, analytical models may be proposed to assess leaching (Loague et al., 1989, 1996). These models try to describe the most important processes with minimal effort and data requirements (Rao et al., 1985; Jury et al., 1983, 1987; Van der Zee and Boesten, 1991). The above mentioned analytical models do not account for vertical heterogeneity and assume steady-state conditions, leading to an underestimation of the fraction leached (Van der Zee and Boesten, 1991).

An alternative to the direct use of simpler models and a way to maintain the dominant behavior of the more complex process-based model is to reduce the complex process-based leaching model into the mathematical form of the simple model in a modeling step referred to as metamodeling. In metamodeling, the model reduction is obtained by considering only those processes and parameters for which the considered simulation output is sensitive or for which input data are available. As such, a simpler model can be obtained which respects the behavior of the complex model, which is more compatible with available databases, and which can be more easily incorporated into policy evaluation and decision making tools. The simplification of the model structure in a metamodel improves also the transparency of the model and is, therefore, easier to use within the communication process with non-technical stakeholders, in particular if a process-based metamodel is proposed. Metamodeling theory and applications to emission modeling have recently been reviewed by Piñeros-Garcet et al. (2006).

Tiktak et al. (2004) implemented a spatially distributed version of the transient process-based pesticide leaching model PEARL (Tiktak et al., 2000) at the European scale, thereby using climate, soil, and land-use data, which were available at the European scale. They showed that the applicability of the spatially distributed leaching model was limited to 75% of the total agricultural area of the former EU-15 (i.e., the EU without the new Member States). The most important reason was that the first version of the Soil Profile Analytical Database of Europe (Madsen-Breuning and Jones, 1995) covered only part of the European Union.

Van der Zee and Boesten (1991) developed a simple process-based metamodel of a transient leaching model similar to PEARL for the pesticide fraction leached. The metamodel was based on the analytical solution for piston flow. They showed that the dependency of the leached fraction on pesticide properties as revealed with the metamodel was in good general agreement with the leaching patterns obtained with a dynamic numerical model based on the convection–dispersion equation.

In this paper, we present the metamodel of Van der Zee and Boesten (1991) in such a way that it describes concentrations instead of leached fractions and show how this metamodel can be used to assess the leaching risk at the European level. Risk indicators for pesticide leaching in the EU should preferably be based on the leaching concentration instead of the leached fraction, because the legal criterion for EU registration is a concentration (0.1 µg L–1). The process-based metamodel was fitted to leaching concentrations that were obtained with a spatially distributed, dynamic, multi-layer, mechanistic model of pesticide leaching, referred to as EuroPEARL (Tiktak et al., 2004). We intend to show that, if parameterized in this way, the metamodel gives comparable results as the original EuroPEARL model. The calibrated metamodel was applied to assess the vulnerability of European ground water bodies at the entire area of the European Union. Based on this assessment, we will show how pesticide degradation and sorption parameters affect the spatial pattern of pesticide leaching at the European scale.


   MATERIALS AND METHODS
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
RESULTS
 DISCUSSION
 CONCLUSIONS
 REFERENCES
 
The EuroPEARL Model
The EuroPEARL model (Tiktak et al., 2004) was used to generate the leaching data, which are required to calibrate the metamodel. EuroPEARL consists of a link between the one-dimensional, multi-layer, mechanistic pesticide leaching model PEARL (Tiktak et al., 2000) and a Geographical Information System (GIS). Calculations are performed for 1062 unique combinations of soil and climate, which together represent 75% of the total agricultural area of the former EU-15. Basic soil data, including horizon designations, were taken from the Soil Profile Analytical Database of Europe (Madsen-Breuning and Jones, 1995). Time series of temperature and precipitation were obtained from the European Climate database (Vossen and Meyer-Roux, 1995). Results are presented with a resolution of 10 x 10 km2, which is the maximum justifiable resolution that can be derived from the 1:1000000 soil map of Europe. Figure 1 shows maps of the most important soil and climatic parameters.


Figure 1
View larger version (66K):
[in this window]
[in a new window]
  		Fig. 1. Basic maps for EuroPEARL. Areas without agricultural land use and areas where insufficient soil information was available are not shown. (a) Organic matter content of the upper meter as derived from the SPADE database, (b) soil texture from the 1:1 000 000 Soil Map of Europe, (c) mean annual rainfall, and (d) mean annual temperature. Temperature and rainfall were taken from the Pan-European climate database.

 
Water flow is described with a submodel (Van Dam, 2000), which uses the Richard's equation to describe soil-water flow. The upper boundary of the model is situated at the top of the crop canopy. Daily rainfall fluxes are input to the model; the reference evapotranspiration is calculated with the FAO modified Penman–Monteith approach (Allen et al., 1998). The actual soil evaporation is calculated according to Black et al. (1969). Potential transpiration is distributed over the root zone using the root density distribution as a weighing factor. Water uptake is reduced for layers with low pressure heads or anaerobic conditions. In EuroPEARL, the ground water level is fixed at 2-m depth. This simplification is considered acceptable, because sensitivity analyses showed that the ground water depth had little influence on the leaching concentration at a given depth as long as the ground water level was below 1-m depth (Diels et al., 1996; Tiktak et al., 2004). EuroPEARL ignores losses by field drains, because information on drainage is not available at the European scale. This may result in an overestimation of the fraction leached, particularly if a shallow drainage system is present. However, the influence of drainage on the leaching concentration is generally less significant than the influence on the fraction leached (Tiktak et al., 2002b).

Pesticide transport is described with the convection–dispersion equation. Sorption onto the soil solid phase is described with a Freundlich isotherm. The Freundlich coefficient is assumed to be directly proportional to organic matter. The degradation of pesticides is described with a first-order rate equation and a number of reduction factors, which account for the influence of temperature, soil moisture, and depth in soil (Boesten and van der Linden, 1991). The uptake of pesticides by plant roots is taken proportional to the water uptake by plant roots and an empirical transpiration stream concentration factor. In this study, an initially pesticide-free soil was assumed. The pesticides were annually applied to the soil surface, 1 d after crop emergence. The annual dose was 1 kg ha–1.

We selected the leaching concentration at 1-m depth as the target variable. The target depth of 1 m is in line with the new EU Groundwater Directive, which considers this depth as the depth where ground water contamination can be detected in an early stage (the so-called "early-warning level"). The selection of the leaching concentration results from European pesticide registration procedures (FOCUS, 2000), which also require that the 80th percentile of the leaching concentration due to weather conditions is used. This percentile is calculated from a time series of 20 yr using a two-step approach. First, for each year, the mass flux of pesticide leached was divided by the annual precipitation surplus. Second, from the so-obtained 20 leaching concentrations, the 80th percentile was chosen.

The Metamodel
The aim of the metamodel is to predict the 80th percentile of the leaching concentration at 1-m depth. Let us first consider the analytical solution of the mass fraction of a pesticide dose that leaches below a certain depth in a homogeneous system, based on (i) the convection–dispersion equation (ignoring diffusion), (ii) steady-state water flow, (iii) a linear adsorption isotherm, and (iv) first-order degradation kinetics. Jury and Gruber (1989) derived this solution and their equation can be rewritten as:

Formula 1[1]
in which F (unitless) is the mass fraction leached, L (m) is the depth considered, Ldis (m) is the dispersion length, µ (d–1) is the first-order degradation rate coefficient, {theta} (m3 m–3) is the volume fraction of water, {rho} (kg dm–3) is the dry bulk density of the soil, fom (kg kg–1) is the organic matter content, Kom (dm3 kg–1) is the coefficient for distribution over organic matter and water, and q (m d–1) is the volume flux of water.

According to its definition, the leaching concentration is a flux concentration. Jury and Roth (1990) describe the solution for the flux concentration, on which Eq. [1] is based. Their Eq. [2.17], [2.51], and [4.68] indicate that this solution is given by:

Formula 2[2]
in which CL (kg m–3) is the flux concentration at the lower boundary, M (kg m–2) is the pesticide dose, t (d) is the time, and R (unitless) is the retardation factor, which is defined by:

Formula 3[3]

For pesticide leaching, one may expect that the flux concentration evaluated at a certain depth in soil is more or less proportional to the fraction of the dose that leaches beyond that depth: low fractions leached can only be achieved by low leaching concentrations and similarly high fractions leached can only be achieved by high leaching concentrations. We made Monte Carlo simulations with Eq. [1] and assumed uniform distributions for {theta}, R, q, and the half-life of the pesticide. The range of {theta} was 0.1 to 0.3 m3 m–3, the range of R was 1 to 100, the range of q was 0.25 to 2.5 mm d–1, and the range of the half-life was 20 to 100 d. The pesticide dose was 1 kg ha–1, the depth in soil was 1 m, and the dispersion length was 0.05 m. For each combination of the stochastic variables the maximum in time of the concentration was calculated and it was compared with the fraction leached from Eq. [1]. Figure 2 shows that the maximum concentration is indeed more or less directly proportional to the fraction leached. This suggests that the metamodel for the 80th percentile concentration can be based on the simpler equation for the fraction leached.


Figure 2
View larger version (21K):
[in this window]
[in a new window]
  		Fig. 2. The maximum in time of the flux concentration, CL, at 1-m depth (µg L–1) as calculated with Eq. [2] as a function of the leached fraction F (dimensionless) as calculated with Eq. [1]. The points are Monte Carlo calculations based on random values for the volume fraction of water, retardation factor, water flux, and half-life of the pesticide. The lines are at arbitrary concentration levels but have a slope of 1 implying that the concentration is proportional to the leached fraction.

 
Van der Zee and Boesten (1991) adapted Eq. [1] slightly to include also pesticide uptake by plant roots:

Formula 4[4]
where g (unitless) is the transpiration stream concentration factor and S (d–1) is the water uptake by plant roots. Van der Zee and Boesten (1991) made calculations with a model similar to EuroPEARL for a single Dutch soil (Ldis = 0.05 m), but for a range of degradation half-lives and Kom values. They fitted the fraction leached to Eq. [4] with {theta}, S, and q as regression parameters. They found that Eq. [4] was a suitable metamodel to describe the output of the simulation model. Moreover, they found that the fitted values of {theta}, S, and q were physically realistic.

Van der Zee and Boesten (1991) used also the fraction leached for the same system, but now assuming piston flow instead of the convection–dispersion equation:

Formula 5[5]

They fitted the calculated fractions leached also to Eq. [5] using again {theta}, S, and q as adjustable parameters. Equation [5] appeared to describe the calculated fractions equally well as Eq. [4], but the fitted value of q was less realistic. Van der Zee and Boesten (1991) attributed this to the fact that Eq. [5] ignores dispersion and thus the fitted values of q include the effect of the dispersion process.

For our metamodel we prefer an equation of the type of Eq. [5] over an equation of the type of Eq. [4], because Eq. [4] describes the calculated fractions equally well with fewer parameters (a metamodel should be as simple as possible by its nature). So we combine Eq. [5] with the phenomenon that the flux concentration is approximately directly proportional to the fraction leached (Fig. 2). This gives the following metamodel for the 80th percentile leaching concentration:

Formula 6[6]
where C0 (kg m–3) is the concentration at the upper boundary of the column.

One cannot expect that Eq. [6] gives as accurate predictions of leaching concentrations as a model such as EuroPEARL, which accounts for vertical heterogeneity of soil physical and chemical properties, nonlinearity in sorption, daily variations of water fluxes, etc. Therefore, we rewrote Eq. [6] as a multiple linear regression model and fitted the leaching concentration to the leaching concentration obtained by EuroPEARL:

Formula 7[7]
in which {alpha}0, {alpha}1, {alpha}2, and {alpha}3 are the regression coefficients and where X1 (unitless), X2 (unitless), and X3 (unitless) are independent regression variables, which are defined as follows:

Formula 8[8]

Formula 9[9]

Formula 10[10]

Rewritten in this way, a process-based metamodel of EuroPEARL results. The proposed methodology for model reduction (combination of process knowledge and regression) has much in common with the data-based mechanistic models of Young (1998) and the transfer function models of Stewart and Loague (2003, 2004).

Parameterization of the Metamodel
EuroPEARL was used to generate the leaching data required to calibrate the metamodel. The time of application has an important effect on pesticide leaching. To investigate the effect of the time of application on the metamodel parameterization, we generated two leaching datasets, namely one resulting from an annual surface application in winter wheat and one resulting from an annual surface application in maize. Because the pesticides are applied after crop emergence, we have one leaching dataset with autumn applications and one leaching set with spring applications. For each of the two leaching datasets, EuroPEARL runs were done for 56 different pesticides. The degradation half-life (DT50) ranged from 10 to 200 d and the organic matter–water partition coefficient (Kom) was between 0 and 200 dm3 kg–1. The 56 combinations of DT50 and Kom cover the full range of relevant pesticides as described by Boesten and van der Linden (1991). For each leaching dataset, the total number of points for calibration amounted to 1062 x 56 = 59 808. Because EuroPEARL is rather demanding with respect to computer resources, we used a computer cluster consisting of 256 loosely coupled CPUs to perform the simulations.

For each of the 59 808 points, the independent regression variables X1, X2, and X3 in Eq. [8], [9], and [10] were calculated as well. These equations contain two substance parameters (µ and Kom), two soil parameters ({rho} and fom), three dynamic parameters ({theta}, S, and q), and two constants (L and g). All parameters were inferred from the EuroPEARL database. In the case of dynamic properties, 20-yr averages were taken. In the case of depth-dependent soil properties, we took the average over the top 1 m, using the horizon thickness as a weighing factor. The degradation rate coefficient, µ, is not directly available in the EuroPEARL database, because it is temperature dependent. The Arrhenius equation was applied to account for this effect:

Formula 11[11]
where DT50 (d) is the degradation half-life at reference temperature, Ea (J mol–1) is the molar activation energy, R (J mol–1 K–1) is the molar gas constant, T (K) is the 20-yr average air temperature, and Tr (K) is the temperature at reference conditions, which was set to 20°C. The molar activation energy was fixed to 54 kJ mol–1 (FOCUS, 2000), which is the same value as used in the EuroPEARL model.

The organic matter content, fom, was averaged over the top 1 m, using the horizon thickness as a weighing factor. Both parameters were taken from the EuroPEARL input files. A continuous pedotransfer approach was used to relate the bulk density, {rho} (kg dm–3), to the organic matter content (Tiktak et al., 1996):

Formula 12[12]

Dynamic properties ({theta} and S) were taken from the EuroPEARL output files. These parameters were first averaged over the top 1 m of the soil and then averaged over the 20-yr simulation period. The water flux, q, was calculated as the represented by the excess rainfall over evapotranspiration and run-off.

The actual fitting was done in S-Plus (Venables and Ripley, 1994), using the robust MM-regression algorithm (Yohai and Zamar, 1997). Robust regression techniques generate answers similar to the classical least-squares regression when the data are linear with normally distributed errors, but differ significantly from the least-squares fit when the data contain significant outliers.

For each of the two leaching sets (autumn and spring applications), the metamodel was calibrated. The calibrated metamodel was used to generate maps of the predicted leaching concentration at the agricultural area of the EU-15. These maps were compared with the leaching concentration obtained with the reference model (EuroPEARL).


   RESULTS
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
DISCUSSION
 CONCLUSIONS
 REFERENCES
 
Calibration of the Metamodel
Fitting of Eq. [7] to the two leaching sets yielded positive values for {alpha}1 and {alpha}2 and negative values for {alpha}3 (Table 1; Model I). A negative value for {alpha}3 would imply that the leaching concentration increases with increasing uptake by plant roots (Eq. [7] and [10]), which is physically incorrect. Further analyses also showed that {alpha}3 was strongly correlated with {alpha}1 (r = 0.84). Apparently, a leaching set based on simulations for a large range of pesticides cannot be used to distinguish between the effect of plant uptake and the effect of other processes. The explanation might be that the uptake of pesticides by plant roots is an important process only when Kom fom approaches zero (Boesten, 1991). In all other cases, the pesticide will become retarded and therefore reside in the soil column for a longer time. Since a longer residence time favors losses by degradation, the actual concentrations in the soil solution will be smaller than for unretarded cases. This affects mainly the uptake, which is reduced considerably.


View this table:
[in this window]
[in a new window]
  		Table 1. Coefficients resulting from calibration of the metamodel to the two leaching sets.

 

View this table:
[in this window]
[in a new window]
  		Table 2. Major climate zones of the European Union, based on mean annual rainfall and mean annual temperature. Zones are a reclassification of the zones described in FOCUS (2000).

 
Because we do not want a process-based metamodel with physically unrealistic regression coefficients, the regressions were repeated for a model with two regression variables, namely X1 and X2. Results are also shown in Table 1 (Model II). All coefficients of Model II are physically realistic (i.e., positive), while the proportion of variation explained by the metamodel is still high. The most important difference between the two leaching sets is in {alpha}1, which is lower in the case of autumn applied pesticides. The X1 term of Eq. [7] reflects the retardation of solute resulting from the volume fraction of water in soil (it is the {theta} term of the retardation factor), while the X2 term reflects the retardation resulting from sorption. Apparently, sorption is the key factor for the leaching concentration in the case of autumn applied pesticides. A possible explanation is that autumn applied pesticides become only subject to degradation if the residence time in the topsoil is long enough: directly after application, the temperature is low and degradation rates are small.

Model II was used to construct Fig. 3a and 3b. These figures show the leaching concentration predicted by EuroPEARL as a function of the leaching concentration predicted by the metamodel. The number of leaching points in each figure equals 59 808. The concentrations were plotted on a log10 scale and the lines represent the fit. The figure shows that, despite the high proportion of variation explained by the metamodel (Table 1), there is large scatter around the 1:1 line. Further inspection of the leaching sets suggests that the deviation from the 1:1 line is related to the annual precipitation: under dry conditions, the metamodel tends to underestimate the leaching concentration, whereas the leaching concentration is overestimated in those cases where the mean annual precipitation is high.


Figure 3
View larger version (53K):
[in this window]
[in a new window]
  		Fig. 3. Leaching concentration CL (µg L–1) at 1-m depth as calculated with the EuroPEARL model plotted against leaching concentrations predicted with the metamodel (Eq. [5]). The points are leaching concentrations and the line represents a 1:1 correspondence. Model II: one regression for the EU-15 as a whole. Model III: regressions for individual climate zones as described in Table 2.

 
To reduce the systematic differences due to climate, the leaching sets were split in four subsets, namely one for each climate zone in Table 2. The underlying assumption is that the climate zones are more homogeneous with respect to seasonal dynamics of weather than Europe as a whole. Figures 3c and 3d show that the systematic errors are indeed reduced. The regression coefficients as shown in Table 1 (Model III) are generally low in dry climate zones and high in wet climate zones: {alpha}1 increases in the order WD < TD {approx} TW < WW, while coefficient {alpha}2 increases in the order WD < TD < TW {approx} WW. This suggests that in the case of dry climates, the effective model parameters deviate more from realistic values than in the case of wet climates. Averaging causes bias in the results of the analytical model, which is reflected in the coefficients of the metamodel. It may be expected that the effect of averaging is more pronounced in the case of dry climates, because the seasonal variability of the water flow pattern is generally higher in those climates. Van der Zee and Boesten (1991) found that the bias between realistic and effective model parameters appears most in the water flow velocity. They attributed this to the fact that Eq. [5] ignores dispersion and thus the fitted values of q include the effect of the dispersion process. Using their findings, one can make an estimation of the ratio between the true water flow velocity and the apparent water flow velocity for the four climate zones by substituting {alpha}1 and {alpha}2 into Eq. [7], [8], and [9]. The calculated ratios (approximately two for wet climates and four for dry climates) confirm that the effect of averaging is most pronounced in the dry climate zones. Notice that the above exercise yields only a crude estimate of the apparent flow velocity, because the averaging of the degradation rate (µ) and the nonlinearity of the sorption process affects the leaching as well.

Summarizing, one can state that the time of application mainly affects the ratio between {alpha}1 and {alpha}2, while the absolute values are affected mainly by the seasonal dynamics of water flow (represented here by the climate zone). We can conclude that Model III explains a high proportion of variation of the original model, while also adequately describing the dependency of the leaching concentration on the main processes (i.e., retardation, degradation, and hydrology). Hence, we can use Model III to map the leaching concentration at the European level.

Validation of the Metamodel
EuroPEARL and Metamodel III were both used to generate maps of the leaching concentration at 1-m depth, which is the compliance depth for the first tier of the European pesticide registration procedure (FOCUS, 2000). As described before, we mapped the 80th percentile of the leaching concentration due to weather conditions. To avoid possible bias that might result from using different datasets, we applied the metamodel to the same dataset as used for the parameterization of EuroPEARL; the most important soil and climate properties are shown in Fig. 1. The comparison was done for three example substances as described in FOCUS (2000). A summary of the most important pesticide properties is given in Table 3.


View this table:
[in this window]
[in a new window]
  		Table 3. Overview of the most important properties of the pesticides considered in this study.

 
The comparison was done with a combination of qualitative (visual) methods and quantitative methods. Quantitative methods try to express the agreement in performance criteria, while qualitative methods are based on subjective visual methods. The performance criteria were selected to reflect the objectives of the metamodel, namely the ability to predict the leaching concentration at multiple sites and the ability of the metamodel to predict the target variable for European registration procedures. The target variable is defined as the leaching concentration at an 80th percentile vulnerable grid cell (i.e., 80% of the area of the European Union has a lower leaching concentration than the grid cell). The three selected indicators are the normalized average error (NAE), the normalized root mean square error (NRMSE), and the model efficiency (ME). The NAE measures the bias in the target variable, which is the difference between the metamodel predictions and the EuroPEARL "observations". The NRMSE measures the deviation between the predicted and "observed" leaching concentrations. The modeling efficiency quantifies the improvement of the metamodel over the mean of the EuroPEARL "observations". Any positive value of ME can be interpreted as an improvement. A value of 1 is best. The indicators are defined as:

Formula 13[13]

Formula 14[14]

Formula 15[15]
where Pi and Oi denote the predicted and observed value in grid cell i, respectively, Formula 15 and Formula 15 are the mean values, P80 and O80 are the 80th percentiles of the leaching concentrations in the maps, and N is the number of grid cells.

Maps of the predicted leaching concentration are shown in Fig. 4 (autumn applications) and Fig. 5 (spring applications). The maps generated by EuroPEARL and the maps generated by Metamodel III show a striking similarity. Both models simulate in a consistent way higher leaching concentrations in response to autumn applications, which was expected. Differences between autumn applications and spring applications are also generally higher in southern Europe, where there is a distinct dry and hot season (see explanation in the previous section). The two models also consistently predict that the leaching concentration increases in the order Substance D < Substance A < Substance B. Despite the similarity between the maps, there are also regions where there are significant differences. In Denmark and northeastern Germany, for example, the metamodel predicts lower leaching concentrations than EuroPEARL, while the opposite is true for the Netherlands. Analysis of the SPADE database showed that the soil profiles in Denmark and Germany are more heterogeneous with respect to the vertical distribution of organic matter than the average profile, while the opposite is true for the Netherlands. The phenomenon that the leaching concentration is underestimated when vertical heterogeneity is underestimated is in line with earlier findings reported by Tiktak et al. (2002a).


Figure 4
View larger version (55K):
[in this window]
[in a new window]
  		Fig. 4. Predicted leaching concentration in response to annual applications in autumn, as calculated with EuroPEARL (left) and the metamodel (right). Areas without agricultural land use and areas where EuroPEARL could not be parameterized are not shown.

 

Figure 5
View larger version (55K):
[in this window]
[in a new window]
  		Fig. 5. Predicted leaching concentration in response to annual applications in spring as calculated with EuroPEARL (left) and the metamodel (right). Areas without agricultural land use and areas where EuroPEARL could not be parameterized are not shown.

 
The performance criteria are listed in Table 4. Using the classification proposed by Henriksen et al. (2003), the ME scores "excellent" for Substances A and D, and "good" for Substance B. Table 3 shows that Substance B has a lower sorption coefficient than Substances A and D. Apparently, the metamodel performs better for non-mobile substances. This was expected because short-term variations due to weather conditions are attenuated in the case of substances with a high Kom. The NRMSE is generally lower than 10%. Highest values are found for Substance D. This was also expected: Substance D has the lowest leaching potential, and the scatter around the 1:1 line increases at low concentration ranges (Fig. 3). The 80th percentile of the leaching concentration is best predicted for Substance A. The largest error is found for Substance D applied in spring, which confirms results shown in the maps (Fig. 5).


View this table:
[in this window]
[in a new window]
  		Table 4. Summary of Metamodel III performance indicators{dagger} (unitless).

 
Both the visual inspection of the leaching maps and the quantitative indicators reveal that the performance of the metamodel is generally good. The performance indicators also show, however, that the application of the metamodel should be done with care.

Leaching Assessments for the Entire EU-25
So far, we applied the metamodel to the same dataset as EuroPEARL, which is limited to 75% of the former EU-15. In this section, we will apply the metamodel to generate leaching maps (resolution 10 x 10 km2) for the entire area of the EU. To be able to generate leaching maps for the entire EU-25, a slightly different parameterization scheme has to be used, which avoids the use of the SPADE database:


Figure 6
View larger version (13K):
[in this window]
[in a new window]
  		Fig. 6. Relation between (a) mean annual precipitation (P) and the mean annual flux q100 at 100-cm depth and (b) relationship between water content at field capacity ({theta}fc) and long-term average water content ({theta}). Both relationships were obtained from EuroPEARL simulations.
 
Parameterized in this way, the metamodel uses only four independent spatially distributed model inputs (organic matter, texture, annual precipitation, and mean annual temperature). These four parameters are available in georeferenced databases that cover the entire area of the EU-25 (Jamagne et al., 1995; Vossen and Meyer-Roux, 1995; Jones et al., 2003).

With the above described dataset, leaching concentrations were calculated for the entire area of the EU-25 for Substances A and B. Maps of the leaching assessment are shown in Fig. 7. To facilitate the interpretation of the predicted spatial patterns, maps of organic matter and precipitation surplus are presented as well. The predicted leaching concentrations generally increase with precipitation and decreases with increasing organic matter content (Fig. 7c and 7d), which was expected. The leaching maps also show that the variability of the leaching concentration at short distances is considerable. This is caused by the strong sensitivity of pesticide leaching to the organic matter content, which shows a strong variability at short distances. The leaching maps also show that the predicted leaching concentrations in certain areas of southern Europe are relatively high. This was not expected, because in these countries degradation rates are generally high and precipitation surplus low. The explanation is found in the extremely low organic matter contents in Mediterranean countries, which can be lower than 1% (Fig. 7a). The EU considers decline of organic matter as one of the most important threats to soil quality, particularly in southern Europe [COM (2002) 179 final, 2002].


Figure 7
View larger version (59K):
[in this window]
[in a new window]
  		Fig. 7. Results of the metamodel application at the entire EU-25. Leaching set autumn application was used. (a) Organic matter content of the upper meter of the soil profile, (b) annual mean precipitation surplus, (c) predicted leaching concentration for Substance A, (d) predicted leaching concentration for Substance B, (e) normalized vulnerability score for Substance A, and (f) normalized vulnerability score for Substance B.

 
In Fig. 7e and 7f, the leaching concentration is normalized: the grid cell with the highest leaching scores 100%, while the grid cell with the lowest leaching scores 0% (the normalized vulnerability score). As expected, the vulnerability score is generally high where precipitation surplus is high and organic matter is low. There are, however, important differences between the two substances. The vulnerability score of the very mobile Substance B shows much more resemblance with the precipitation surplus map than the vulnerability score of Substance A. The vulnerability score of Substance A is strongly correlated with the organic matter content map. These differences are in line with results obtained with EuroPEARL (Tiktak et al., 2004), which suggests that the metamodel captures the main features with respect to the dependency of the leaching concentration on the various processes. This analysis further shows that mapping the vulnerability to pesticide leaching without considering pesticide properties, as is done in procedures where the vulnerability is considered to be only a function of weather and climate properties (for example FOCUS, 2000), may be subjected to bias.


   DISCUSSION
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
CONCLUSIONS
 REFERENCES
 
A process-based metamodel of the leaching model EuroPEARL has been developed, which was successfully used to obtain quantitative leaching assessments for the entire area of the EU-25. Based on common knowledge of the leaching process, the behavior of the model can be judged "plausible". Nevertheless, the model predictions are subject to a high degree of uncertainty. Errors result from the way the system is conceived in the selected model and from the way the model inputs and parameters have been generated (Loague and Corwin, 1996).

Model errors at the conceptual level arise when processes are inappropriately described by the model or when process descriptions are forced to be used in an application for which they were not initially intended. The metamodel inherits all the uncertainties associated with the original EuroPEARL model (Tiktak et al., 2004). A conceptual limitation of this model is for example related to the spatial schematization of the system. The properties of the environmental system vary extremely in space and time and this variability is now encoded by spatially distributing the environmental properties in a discrete way. Thereby, it is considered that the transport of pesticides from the land surface to the compliance depth passes through a set of 10- x 10-km2 parallel soil columns. Variability of fate and transport processes at the surface and within these columns is completely ignored. Techniques for assessing the small-scale variability are still poorly developed and cannot be implemented at the European scale. An extreme example of this small scale variability is the ignorance of preferential flow, a process for which consensus exists that is extremely important for correctly describing pesticide transport in soils (Flühler et al., 2001). Basic soil information for preferential flow models such as quantitative soil structure information (Rawls et al., 1996) is not yet available at the European scale, so it remains questionable whether a regional-scale version of preferential flow models will become available shortly.

Input and parameter generation errors depend on the quality of the underlying databases and the quality of the parameter generation techniques, such as the quality of the applied pedotransfer functions (Tiktak et al., 1999). For characterizing the spatial patterns of soil properties throughout Europe, the European Soil Map at the scale 1:1 000 000 was used in combination with the Soil Profile Analytical Database of Europe (Jamagne et al., 1995). The Soil Profile Database has serious limitations. The most serious limitations are that soil profile data is available for only 75% of the agricultural area of the EU-15, and that the soil profiles are not uniformly distributed across the continent. Jamagne et al. (1995) showed, however, that all major soil types are included. The metamodel was used to extend the simulations toward the entire EU-25. This can be done, as long as the metamodel is not applied beyond the range of values in the original database. Analysis of the EuroPEARL database revealed that only 6% of the total agricultural area of the EU-25 was outside the range of model inputs of the original model. The missing area is mainly in cold climates, where the mean annual temperature is below 5°C. The effect of important processes for these regions, like snow accumulation and the effect of frost on waterflow, may therefore be underestimated. Predictions for the Nordic and Baltic countries should, therefore, be treated with extra care.

The metamodel validation in this study pertains only to the comparison of the metamodel with the original model; no comparison with field observations was made in this study. So far, leaching models have primarily been validated at the field scale (e.g., Vanclooster et al., 2000; Trevisan et al., 2003) and very few studies, if any, have looked at the validity of the spatial leaching patterns simulated by spatially distributed leaching models. Analyzing the validity of the predicted spatial patterns needs detailed information on the occurrence of pesticides within ground water bodies. Unfortunately, high quality regional data sets that allow to make such an assessment are only available for some limited cases (e.g., Leterme et al., 2004; Tiktak et al., 2005). It is expected that the EU Groundwater Directive will call for monitoring data on pesticide concentrations in the ground water, yet it should still be analyzed how data of such monitoring programs could be used to assess the validity of spatial predictions of pesticide leaching.

When applying the metamodel, an additional error is added on top of the model error of the original model. Although the performance of the metamodel is generally good, the predicted concentration can deviate significantly in specific cases, particularly in the low concentration range. Recently, a study was started to quantify the error propagation in the chain EuroPEARL metamodel.


   CONCLUSIONS
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 CONCLUSIONS
REFERENCES
 
A process-based metamodel of the recently developed European leaching model EuroPEARL has been developed, which explains more than 90% of the variation of the original model using only four independent spatially distributed parameters that are available from general soil and climate databases. The calibrated metamodel was applied to generate maps of the leaching concentration in the European Union. Maps generated with the metamodel showed a striking similarity to maps obtained with EuroPEARL. The predicted leaching concentration generally increases with precipitation and decreases with increasing organic matter content. The short-distance variability of the leaching concentration due to organic matter overruled the north–south gradient caused by climatic differences, a leaching pattern that is also simulated by the original model. Quantitative performance indicators confirmed that the metamodel gives comparable results as the original model.

The work presented in this paper can be seen as the first quantitative and process-based leaching assessment at the European scale. Based on common knowledge of the pesticide leaching process, the behavior of the model can be judged plausible. Nevertheless, the model predictions are subject to a high degree of uncertainty. An important reason is that the metamodel inherits all the uncertainties associated with the original model. Given these uncertainties, we stress that the predicted concentrations should be considered as proxy variables of the actual concentrations found in ground water and should be used in conjunction with results from monitoring of the ground water system. Notwithstanding the intrinsic high uncertainty with the predicted concentrations, we believe that the presented methodology makes a positive contribution to the modeling of ground water contamination by the use of pesticides, in particular in view of European harmonized risk-assessment procedures.

The metamodel was primarily developed to support the Pesticides Strategy. However, the maps generated by the metamodel can serve many other practical purposes. In many European pesticide registration procedures, for example, field or lysimeter studies are required if a pesticide fails to pass the first tier of the registration. The registration procedures require that additional studies are carried out under high-risk conditions. The metamodel could be used to identify high-risk areas, where these studies should preferably be performed.


   ACKNOWLEDGMENTS
 
The reported work was carried out within the framework of the project "HArmonized environmental Indicators for pesticide Risk" (HAIR), which was supported by the European Commission under the sixth framework program (Project no. SSPE-CT-2003-501997).


   REFERENCES
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 CONCLUSIONS
 REFERENCES
 





This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF) Free
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Services
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Citing Ar