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


     

Published online 1 March 2006
Published in J Environ Qual 35:628-640 (2006)
DOI: 10.2134/jeq2005.0257
© 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 ISI Web of Science
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 ISI Web of Science (4)
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Dann, R. L.
Right arrow Articles by Pang, L.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Dann, R. L.
Right arrow Articles by Pang, L.
GeoRef
Right arrow GeoRef Citation
Agricola
Right arrow Articles by Dann, R. L.
Right arrow Articles by Pang, L.
Related Collections
Right arrow Pesticides
Right arrow Organic Compounds
Right arrow Agricultural Pesticides

TECHNICAL REPORTS

Vadose Zone Processes and Chemical Transport

Impact of Data Quality and Model Complexity on Prediction of Pesticide Leaching

R. L. Danna, M. E. Closea,*, R. Leeb and L. Panga

a Institute of Environmental Science and Research, PO Box 29-181, Christchurch, New Zealand
b Landcare Research NZ Ltd, Private Bag 3127, Hamilton, New Zealand

* Corresponding author (murray.close{at}esr.cri.nz)



   ABSTRACT
TOP
 ABSTRACT
INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 CONCLUSIONS
 REFERENCES
 
Accurate input data for leaching models are expensive and difficult to obtain which may lead to the use of "general" non-site-specific input data. This study investigated the effect of using different quality data on model outputs. Three models of varying complexity, GLEAMS, LEACHM, and HYDRUS-2D, were used to simulate pesticide leaching at a field trial near Hamilton, New Zealand, on an allophanic silt loam using input data of varying quality. Each model was run for four different pesticides (hexazinone, procymidone, picloram and triclopyr); three different sets of pesticide sorption and degradation parameters (i.e., site optimized, laboratory derived, and sourced from the USDA Pesticide Properties Database); and three different sets of soil physical data of varying quality (i.e., site specific, regional database, and particle size distribution data). We found that the selection of site-optimized pesticide sorption (Koc) and degradation parameters (half-life), compared to the use of more general database derived values, had significantly more impact than the quality of the soil input data used, but interestingly also more impact than the choice of the models. Models run with pesticide sorption and degradation parameters derived from observed solute concentrations data provided simulation outputs with goodness-of-fit values closest to optimum, followed by laboratory-derived parameters, with the USDA parameters providing the least accurate simulations. In general, when using pesticide sorption and degradation parameters optimized from site solute concentrations, the more complex models (LEACHM and HYDRUS-2D) were more accurate. However, when using USDA database derived parameters, all models performed about equally.

Abbreviations: CRM, coefficient of residual mass • GOF, goodness-of-fit • Koc, organic carbon distribution coefficient • Ksat, saturated hydraulic conductivity • SSres, sum of squares residual • T1/2, half-life


   INTRODUCTION
TOP
 ABSTRACT
 INTRODUCTION
MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 CONCLUSIONS
 REFERENCES
 
THE LEACHING of pesticides through soil and into ground water has been observed in New Zealand ground waters (Close, 1996; Close and Rosen, 2001; Close and Flintoft, 2004) and internationally (Barbash et al., 2001; Kolpin et al., 2000; Sankararamakrishnan et al., 2005; Worrall and Besien, 2005). Ground water is a very important drinking water source in many countries with about 50% of New Zealand's community water supplies being sourced solely or partially from ground water (Davies, 2001). The leaching of potentially harmful pesticides into ground water is, therefore, of great concern. Understanding the movement and persistence of these chemicals is fundamental to their effective environmental management. Computer simulation models now play a major role within research and resource management in predicting the behavior of various pesticides within the soil, vadose zone, and ground water systems. Numerous models of differing complexity have been developed over the past few decades to address a range of issues within the soil–water–chemical system including predicting fluid flow, contaminant adsorption, and dispersion processes, and determination of the persistence of chemicals which may impact on ground water and surface water quality.

Computer simulation models range in complexity from simple water-capacity-type models, such as GLEAMS (Davis et al., 1990) and CALF (Nicholls et al., 1982), to more sophisticated models (which have a greater demand in data inputs and model construction) that use numerical approximations to solve equations for water movement and contaminant transport, for example, HYDRUS-2D (Simunek et al., 1996).

Models can generally be grouped together according to their functional attributes and applications. Addiscott and Wagenet (1985) developed a hierarchical scheme for classification of leaching models. This included a primary distinction between deterministic (single set of variables) and stochastic models (statistical distributions of variables), underlain by a second distinction of mechanistic (complex description of fundamental processes) and functional (simplified often empirical) classes of models. Models are generally used for two purposes, research or resource management.

An ongoing question for model users is, What complexity and type of model is required to best achieve the study objectives? In making this decision the modeler must also assess which model(s) will function most effectively with the available amount and quality of input parameter data. Generally, the more simplified the model representation of the leaching process is, the less the demand for input data characterizing field soil-water properties (Hutson and Wagenet, 1993). More complex models require more complex, and usually more difficult and expensive to obtain, input parameters than simpler models. Parameters required by computer simulation models may include, but are not limited to, porosity, field capacity, wilting point, bulk density, organic carbon content, clay and silt content, saturated hydraulic conductivity (Ksat), climatic data including rainfall and temperatures, water retention curve characteristics, and contaminant specific data such as distribution coefficients (Kd and/or Koc) and half-lives (T1/2) or degradation rates ({lambda}). Many studies have been performed to determine transport-related parameters of pesticides, particularly the organic carbon distribution coefficient (Koc), T1/2, and {lambda} (Close et al., 2003a; Hornsby et al., 1996; Patterson et al., 2000; Weber, 1994) due to their importance in understanding and estimating the mobility and persistence of pesticides in the subsurface environment.

Obtaining input data can be a time consuming and costly process and site-specific parameters describing hydraulic and pesticide properties are often unavailable, which may affect the use of the more "data hungry" models such as LEACHM (Dust et al., 2000). Loague and Green (1991) highlighted the importance of recognizing the appropriateness of a model for each application to avoid oversimulation when simpler, less data-intensive models may be equally well suited. Brown et al. (1995) noted that given the range of models for pesticide fate in soil that are available, there is a clear need for evaluation of the capabilities of models in a range of scenarios.

If simulation models are to be used effectively in the management of natural resources on a large scale, readily accessible and affordable data is required as input parameters. There are a number of sources of soil, climate, and pesticide characteristic data. A major source of pesticide characteristic data is the USDA-ARS Pesticide Properties Database (USDA-ARS, 2005). The Pesticide Properties Database provides a list of the pesticide properties that are most important for predicting the potential of pesticides to move into ground water and surface waters under a range of weather and soil conditions. In New Zealand, computer simulation models and digitized soil maps are increasingly being used by research institutes and territorial authorities to assess environmental and land-use issues. However, the lack of accurate data to estimate soil physical properties for soil types is limiting the wide application of simulation models to address current environmental and land-use issues (Webb et al., 2000). The suitability of parameter data sourced from large general databases and transferred to localized models as input data is a significant problem because experimental conditions used to derive values in the database are generally very different from the site or area of interest.

Most pesticide leaching studies have involved the use of one, or a number of models to assess contaminant transport characteristics using one set of specific field data (Dust et al., 2000; Garratt et al., 2002; Gottesburen et al., 2000). However, little research to date has been done on comparing model-simulated results against observed data, using a variety of parameter data ranging from high quality site-specific data to less site-specific data sourced from an external database such as the Pesticide Properties Database or manual-derived information such as that available in GLEAMS (Knisel et al., 1994).

This research was part of a series of studies into the leaching behavior of pesticides in New Zealand (Close et al., 1999, 2003a, 2003b). This paper aims to investigate the effects of data sparseness (reducing the amount of site specific data) on how well models of differing complexity perform. The objectives of this paper are to assess: (i) the effect of using three different sets of Koc and T1/2 parameters (model optimized, laboratory determined, and sourced from the USDA database) on model accuracy; (ii) the effects of using variable soil characteristics data on model outcomes; and (iii) the ability of three leaching models of differing complexity to accurately determine leaching parameters with different levels of input. The three soil datasets used vary in their amounts of relevant site information, from data collected at the actual field site, to data collected from the same soil type, to data derived from the same soil textural class. Likewise the pesticide properties datasets vary from data derived from the field observations, to laboratory data derived from similar soils, to data taken from a wide range of soil and experimental conditions.


   MATERIALS AND METHODS
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
RESULTS
 DISCUSSION
 CONCLUSIONS
 REFERENCES
 
Field Site and Trials
A detailed description of the study site, trial design, sampling program, analysis methods, and inverse modeling results is presented in Close et al. (2003a, 2003b). The experimental site was located on a large alluvial fan about 10 km south of Hamilton (37.8° S, 175.3° E), in the North Island of New Zealand, on a Horotiu silt loam, classified as a Typic Orthic Allophanic Soil [New Zealand Soil Classification; (Hewitt, 1992)]. The topsoil contained up to 12% allophane.

Two trials were conducted at the site involving the application of four pesticides: hexazinone [3-cyclohexyl-6-(dimethylamino)-1-methyl-1,3,5-triazine-2,4(1H,3,H)-dione], procymidone [N-(3,5-dichlorophenyl)-1,2-dimethylcyclopropane-1,2-dicarboximide], picloram (4-amino-3,5,6-trichloropyridine-2-carboxylic acid), and triclopyr (3,5,6-trichloro-2-pyridyloxyacetic acid). The pesticides in the field trial were selected, from those that had been detected in ground water systems in New Zealand, to give a range of mobility and degradation characteristics. Seven pesticides were applied in the field trial but only the four mentioned had a sufficient number of either soil or suction cup observations to be suitable for the analysis reported in this paper. The pesticides and a bromide tracer were applied in November 1997 and monitored for 2 yr. Soil samples and water samples (using nine suction cups located from 0.2 to 2.5 m depth) were collected for analysis on a regular basis. The soil water samples were collected from each suction cup every 1 to 4 wk, with the total number of samples being 106, while the soil samples were taken every 3 to 4 mo to a maximum depth of 1 m. Five cores were taken on each sampling occasion, separated into 10-cm depth increments, and composited for analysis, giving a total number of 43 soil samples. The bromide tracer, together with soil moisture measurements, were used to determine parameters, such as dispersivity, for the simulation models and to check that the models were performing as expected, before the inverse modeling (Close et al., 2003b).

Simulation Models Characteristics and Input Requirements
Modeling of pesticide movement was performed using three commonly used computer simulation models: GLEAMS (Groundwater Loading Effects of Agricultural Management Systems; Leonard et al., 1987), LEACHM (Hutson and Wagenet, 1995), and HYDRUS-2D (Simunek et al., 1996). These models were selected to give a range of simulated water and soil processes as well as some differences in the representation of the soil layers. A brief comparison of models and input parameters used for each model is given in Table 1.


View this table:
[in this window]
[in a new window]
  		Table 1. Comparison of model requirements for GLEAMS, LEACHM, and HYDRUS-2D.

 
The GLEAMS model is a water-capacity field-scale model, which has been developed to evaluate the effects of land management practices on the movement of chemicals associated with runoff and leaching through the soil and unsaturated zone. The water-capacity method involves the transfer of water from one soil layer to the adjacent downward layer in the soil profile once the layer above exceeds its water-holding capacity (i.e., field capacity). The GLEAMS model allows soil and climatic characteristics and land management practices to be varied over time. If not available from site investigations soil moisture data may be estimated using tables in the GLEAMS manual from soil texture information allowing for minimal site-specific data (Knisel et al., 1994). The GLEAMS model uses a single value of Koc for the whole profile with varying organic matter content for each soil layer. Organic carbon values are converted to organic matter content for use in GLEAMS. More detailed descriptions of GLEAMS can be found in Leonard et al. (1987) and Rekolainen et al. (2000).

The LEACHM model simulates general vertical water and solute movement, and LEACHP is the module of the LEACHM model used to simulate pesticide degradation and transport. It is a more complex model than GLEAMS in that it requires more detailed input variables to model water movement using the Richards equation and the convection–dispersion equation (CDE) for solute (pesticide) transport. The model must be either supplied with a function to describe water retention, matric potential, and hydraulic conductivity or LEACHM may derive this information from supplied silt and clay percentages using in-built subroutines. It uses a modification of Campbell's function (Hutson and Cass, 1987) where the exponential function is replaced by a parabolic function at high potentials to give a better representation of water retention in real soils. Pesticide degradation rates are modeled using first-order kinetics and sorption isotherms are assumed to be in equilibrium and linear. More detailed descriptions of the LEACHM model can be found in Hutson and Wagenet (1995) and Dust et al. (2000). A description and explanation of the model parameters used in this study can be found in Close et al. (2003b).

The HYDRUS-2D model solves the Richards equation for saturated–unsaturated water flow, uses a convection–dispersion equation for solute transport, and includes provisions for linear and nonlinear equilibrium or non-equilibrium adsorption, and first-order degradation (Simunek et al., 1996). The HYDRUS-2D model was used for simulating pesticide transport through both the unsaturated zone and saturated zone (incorporating observations from monitoring wells down-gradient of the plot) at the same experimental site (Close et al., 2003b). Therefore, we have used the same model setup in this study for soil profile simulations, with linear equilibrium adsorption and first-order degradation. There is a significant difference between LEACHM and HYDRUS-2D in the definition of soil layers (fixed thickness in LEACHM compared to variable thickness in HYDRUS-2D), which results in differing complexity in the model setup. The variable thickness of the soil layers in HYDRUS-2D should result in a better match of the model to the simulated soil. Only the liquid phase resident concentrations are output in HYDRUS-2D, and thus direct calibrations can be made only against soil water data. The calibration of soil concentrations was indirectly performed through calibration of soil mass in each subregion of the model domain.

Simulation Comparisons
A number of comparisons were made in this study to assess the effects of increasing data quality on the accuracy of three leaching models. These include a comparison of simulations run with three different sources of pesticide parameters and three sets of soil profile data. Both soil and water (suction cup) solute concentration data were compared where suitable data were available. Unfortunately, most suction cup data obtained for procymidone were at concentrations below detection limits. This limited model comparisons using suction cup data to hexazinone only, as no suction cup samples were collected for picloram or triclopyr.

Pesticide Parameters
Pesticide sorption and degradation data (i.e., Koc, T1/2, and {lambda}) for the pesticides used in this study (hexazinone, procymidone, picloram, triclopyr) were obtained from three separate sources. The first set of pesticide sorption and degradation data was obtained from previous inverse modeling of the field data (Close et al., 2003b). In their study the PEST optimization package (Doherty, 1994) was used to optimize the pesticide sorption and degradation parameters in each of the models using field observed soil and water solute concentration data. Median values from the three models and three combinations of observed data (soil, water, and soil and water combined) were used as the optimized Koc and degradation rate parameters in this study. The second source of pesticide leaching values was from batch test laboratory studies for procymidone performed using a range of eight silt loam soils from New Zealand (McNaughton et al., 1999). The authors measured sorption on fresh soil samples at typical field application rates and measured degradation at 20°C over a period of 56 d. The third source of sorption and degradation parameters were the "best available literature values" (BALVs) published by the USDA-ARS website http://www.arsusda.gov/acsl/services/ppdb/ppdb3.html (verified 7 Dec. 2005). These parameters are derived from a combination of laboratory batch studies, both from manufacturer and other sources, field experiments,and calculated values where more reliable values are not available. A comparison of the pesticide sorption and degradation parameters is presented in Table 2.


View this table:
[in this window]
[in a new window]
  		Table 2. Sources of pesticide sorption and degradation parameter values, organic carbon distribution coefficient (Koc) and half-life (T1/2), for the selected pesticides.

 
Soil Data
Soil chemical and physical characteristic data for this study was obtained from three sources. These were (i) site investigations, named "best" data; (ii) the New Zealand National Soils Database (NSD), named "good" data; and (iii) the textural analysis table, named "limited" data. These three levels of data span the range of what would typically be available in the New Zealand setting. The "best" soil data characteristics include soil pH, clay and silt %, organic carbon %, allophane %, bulk density, total porosity, saturated conductivity (Ksat), and water retention data. The "good" soil data consists of NSD "Horotiu soil" particle size and moisture release data but does not include Ksat. The "limited" soil data has particle size estimates from the textural soil name of each horizon using the New Zealand textural class estimates as outlined in the Manual for National Soils Database (Wilde, 2003). The data sourced from here is very similar to that found in the GLEAMS manual (Knisel et al., 1994).

The limited soil dataset has no site-specific Ksat or moisture release data. For GLEAMS the manual tables were used to obtain bulk density, wilting point, and field capacity data. The LEACHM model uses the utility RETFIT to estimate Campbell's a and b parameters from particle size data. The water release characteristics of the soil classes, required by HYDRUS-2D, were obtained using the RETC program (deriving hydraulic parameters from measured water retention curve data) for the best and good soil and the ROSETTA program (deriving hydraulic parameters from % silt, sand, and clay) for the limited soil group. The soil input parameters are shown for the topsoil layer in Table 3.


View this table:
[in this window]
[in a new window]
  		Table 3. Soil parameters used for the three models (only topsoil horizon shown).

 
Goodness-of-Fit Statistics and Analysis of Variance
The performance of the models was quantitatively evaluated by comparing the simulated pesticide leaching results from the three models to the observed data using three goodness-of-fit (GOF) parameters, namely sum of squares residuals (SSres), coefficient of residual mass (CRM), and coefficient of determination (r2). Goodness-of-fit parameters provide a suitable mechanism to quantitatively compare the results of simulated models to actual values and to characterize systematic under- or overprediction (Loague and Green, 1991). The following GOF statistics were selected for evaluation of the parameter estimates and the leaching models:

Formula 1[1]

The residual sum of squares is the sum of the squared deviations of the observed values for sample i (Oi) from the simulated values of sample i (Si). The SSres value is a commonly used parameter that is unit dependent with a lower limit, which is an ideal value of zero, and no upper limit.

Formula 2[2]

The coefficient of determination used here is the square of the correlation coefficient. The variables Om and Sm represent the means of the observed and simulated data, respectively. The upper limit and ideal value for r2 is 1.

Formula 3[3]

The coefficient of residual mass (CRM) value provides a comparison of the mass of the pesticide observed and simulated within the profile irrespective of its distribution. The CRM value can be positive or negative and gives the ideal value of zero when the observed and the predicted concentrations of the contaminant throughout the profile are equal.

A single analysis of variance (ANOVA) was used to compare GOF values for models run with (i) the three different models; (ii) the differing sorption and degradation parameters, and (iii) the different soil quality data, for each pesticide and sample medium (e.g., hexazinone–water solute concentration data).


   RESULTS
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
DISCUSSION
 CONCLUSIONS
 REFERENCES
 
An example comparison of the observed versus simulation data, using LEACHM with the different inputs of pesticide mobility and degradation parameters, is given for the procymidone soil concentration data (Fig. 1). The better fit of the simulations using the optimized pesticide parameters is clearly shown, particularly for the later sampling dates.


Figure 1
View larger version (24K):
[in this window]
[in a new window]
  		Fig. 1. Comparison of observed procymidone soil concentration data for LEACHM simulations using optimized, laboratory, and USDA derived sorption and degradation parameters with site derived soil data. Sampling depths were 15 cm on Day 43, 50 cm on Day 149, 60 cm on Day 267, and 1 m on the remaining dates (detection limit = 10–2 mg kg–1). Simulated values below 10–4 are plotted at 10–4.

 
Goodness-of-fit values for all comparisons are presented in Tables 4 and 5. Single analysis of variance of GOF values was also performed comparing the three models, the four soil groups, and the three sources of pesticide parameters (Table 5). As SSres provides a direct indication of the "match" between observed and simulated data, this GOF parameter will be mainly used in the discussion of the results. The results for the other GOF parameters, CRM and r2, follow similar trends to SSres (Tables 4 and 5). Comparisons of GOF parameters between different models and soil quality data generally showed no significant differences (Table 6), but comparison of models run with different pesticide half-lives and Koc showed significant differences in each case.


View this table:
[in this window]
[in a new window]
  		Table 4. Goodness-of-fit values for hexazinone, picloram, and triclopyr simulations.

 

View this table:
[in this window]
[in a new window]
  		Table 5. Goodness of fit values for procymidone–soil simulations.

 

View this table:
[in this window]
[in a new window]
  		Table 6. Analysis of variance p values for comparison of goodness-of-fit (GOF) values from simulations run with the three different models (GLEAMS, LEACHM, and HYDRUS-2D), the different sorption and degradation parameters, and the different soil quality data.

 
Comparison between Pesticide Data Sources—Organic Carbon Distribution Coefficients and Half-Lives
Comparisons of SSres values for each of the simulations with the different pesticide sorption and degradation are shown in Fig. 2. Models run with the optimized pesticide sorption and degradation parameters (Koc and degradation rates) provided better GOF results than models run with the USDA sorption and degradation parameters (Fig. 1 and 2). There was a consistent pattern of SSres values ranging from low (for optimized input data) to medium (for procymidone laboratory data) to highest (for USDA pesticide data). In addition, it appears that there were fewer differences between models run using optimized and USDA data for the more mobile hexazinone as opposed to the other less mobile pesticides, reflecting the relative similarity of sorption and degradation parameters in comparison to the other pesticides (Table 2). Analysis of variance (p values) of GOF values for the different pesticide–sample media combinations indicated that there was a significant difference between the modeled results using different leaching parameters for all pesticides (Table 6), with optimized Koc and half-lives providing a significantly better "match" than either laboratory or USDA data. For procymidone soil data results (SSres) there was an expected gradient of results with models using optimized data providing better results than those using laboratory data, which in turn provided better results than those using USDA data. A comparison of correlations between observed and modeled data (r2) and CRM values for procymidone soil reflect this trend (Fig. 3).


Figure 2
View larger version (34K):
[in this window]
[in a new window]
  		Fig. 2. Comparisons of the range of SSres values (from "best" soil value to "limited" soil quality value) for each of the models run with optimized, USDA, and laboratory (for procymidone) pesticide parameters. Boxes show maximum and minimum range with line through box indicating median value. The term SSres is the sum of squares residual, and Hex, Pro, Pic, and Tri are hexazinone, procymidone, picloram, and triclopyr, respectively.

 

Figure 3
View larger version (26K):
[in this window]
[in a new window]
  		Fig. 3. (a) Comparison of ranges ("best" to "limited") of r2 values for procymidone soil results with each model and each pesticide leaching parameter source, showing better correlations for optimized than laboratory or USDA sourced model results; and (b) a gradient of coefficient of residual mass (CRM) values from low (optimum value of 0) using optimized data to high (near 1) using USDA data.

 
Comparison Between Soil Quality Data
Analysis of variance of GOF values indicates that there was generally no significant difference between the three soil parameter groups when combining all GOF data (Table 6) or when comparing soil data groups with optimized and USDA parameter data individually (p = 0.27 and 0.98, respectively, for SSres). The one exception was the r2 value for hexazinone–water comparison for the soil groups. The p value of 0.00 for the r2 indicates a significant difference between models compared. However, the highest r2 value was 0.28 (see Table 4) indicating little correlation between any of the modeled and observed datasets. We observed that, when using USDA leaching data, models run with the different soil data groups provided very similar results, with more variation in the optimized results. This is reflected qualitatively in Fig. 4 where there is a trend for the two better soil data groups ("best" and "good") to provide similar and better results than models run with the limited dataset.


Figure 4
View larger version (37K):
[in this window]
[in a new window]
  		Fig. 4. Comparison of models run with different soil quality inputs using optimized pesticide sorption and degradation data (y axis values are normalized SSres values). The term SSres is the sum of squares residual.

 
The results of GLEAMS model simulations run with optimized Koc and degradation rate values indicated that the "best" and "good" soil data generally (with the exception of hexazinone) provided better results than those using the "limited" soil dataset. The GLEAMS results for procymidone, picloram, and triclopyr indicated that models run using "best" and "good" soil data recorded SSres values approximately 40 to 70% lower than those derived using the "limited" soil data. The GOF values for models run with LEACHM (using optimized pesticide data) indicated that for all pesticides the "best" and "good" soil data were generally similar and better performed (lower SSres) than the models run using the "limited" soil datasets (Fig. 4).

The trend of the two better soil datasets providing better results than the "limited" dataset was not apparent when using HYDRUS-2D. Models run with each of the three soil datasets provided similar SSres values (Fig. 4). This suggests that the associated parameter estimation models RETC (van Genuchten et al., 1991) and ROSETTA (Schaap et al., 2001), which use water retention curve and particle size data, respectively, to generate soil hydraulic properties, provided relatively accurate data (similar to site determined values at least) even from the varying input data.

Comparison Between Models
Analysis of variance of the GOF values for the three models indicated that for all pesticide–sample media combinations and GOF parameters there was no significant difference between the models (at p < 0.05) (Table 6). The output for each model was more dependent on the quality of the pesticide sorption and degradation data. Within the optimized data LEACHM was overall the best-performed model though there was some variation between pesticides (Fig. 4). Both GLEAMS and LEACHM performed approximately equally with hexazinone with HYDRUS-2D performing slightly better with the limited dataset. The LEACHM model generally provided the best results for procymidone, picloram, and triclopyr, while HYDRUS results were comparable for procymidone and picloram but were poor for modeling hexazinone soil and triclopyr soil solute concentrations. Results from HYDRUS-2D appeared to be more precise when using a variety of soil datasets with generally much narrower SSres value maximum to minimum ranges than GLEAMS or LEACHM. Each of the models performed on a more comparable basis when using the USDA pesticide sorption and degradation data with similar but poorer results obtained for each pesticide simulation (Fig. 2).


   DISCUSSION
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
CONCLUSIONS
 REFERENCES
 
This study provided an opportunity to assess the effects of using different quality input data on three different pesticide leaching computer models. The three main variables compared were the pesticide sorption and degradation rate/half-life of the pesticide, soil physical property data, and three different leaching models. The differing pesticide sorption and degradation parameters, Koc and half-life, had the greatest impact on the ability of model simulations to reflect field situations, providing a much greater difference in results than using the soil data of differing quality or even the selection of the model used. This finding is consistent with the observation of Garratt et al. (2002) that the influence of parameter selection is often greater than the influence of the model.

We found that simulations run with the site-optimized sorption and degradation parameters were, for all pesticide–sample media combinations, significantly more accurate than model simulations that used the USDA sorption and degradation parameters. Pang et al. (2000) noted, in reviewing the "best available literature values" selected by Wauchope et al. (1992) (where many of the values in the USDA database are derived), that these values are mostly derived from laboratory experiments using homogenous soils at room temperature. This point and the poorer performance of models run with USDA sorption and degradation parameters in our study suggests that caution should be used in interpreting simulations using this data source. For procymidone and triclopyr there was approximately an order of magnitude difference between the optimized and USDA data SSres values, while for hexazinone and picloram using the USDA pesticide characteristic data approximately doubled the SSres values. Similar patterns were reflected in the CRM and coefficient of determination GOF values. Results of simulations run using USDA derived sorption and degradation parameters indicated less contrast between the simulations run with the three different quality soil datasets, when compared to those run with optimized values. This indicates that, in this study, the accuracy of the sorption and degradation parameters impacted on the models' capacity to reflect differences in the quality of the soil property data. There was still a trend of better results using the more accurate soil data with both GLEAMS and LEACHM for each of the pesticides. This trend was not so apparent for HYDRUS-2D simulations where there was much less variation in results using different soil datasets, possibly reflecting the effect of water retention parameter estimation models used to provide required inputs of soil physical properties.

Simulations using the laboratory-determined data calculated for procymidone provided a better approximation of actual field observations than the more general USDA data, but still significantly worse than the site-optimized data. This outcome is in agreement with the findings of Dust et al. (2000) that, although important data for pesticide transformation and transport could be derived from extensive laboratory-scale experiments, these did not represent all processes that could affect pesticide fate and behavior under field conditions. All GOF parameters used in this study reflected the greater accuracy of the site-optimized leaching parameters by recording GOF values closer to the ideal values than for models using either laboratory-derived leaching parameters or those publicly available from the USDA. Figure 1 demonstrates that if the pesticide mobility and persistence parameters used are less than the "true" values, as seen in Fig. 1 with the comparison between the USDA and the optimized parameters, then the leaching will be dramatically underestimated. Conversely where externally sourced mobility and persistence data are greater than "true" values then leaching will be overestimated. This highlights the benefits of obtaining high quality site-specific pesticide leaching data, where possible. Unfortunately, however, this can only be done through inverse modeling of results obtained from field trials. These field studies are expensive and time-consuming but may prove worthwhile depending on the objectives of the study.

The LEACHM and HYDRUS-2D results for picloram were considerably closer to optimum GOF values than those derived by GLEAMS. The LEACHM and HYDRUS-2D optimized- simulation runs using each of the soil datasets provide similar results. However, when using GLEAMS there was a much greater difference between simulations run using the "best" and "good" soil property datasets than the "limited" dataset. A similar trend was apparent for both GLEAMS and LEACHM for triclopyr with the "best" and "good" soil sets providing more accurate results than the "limited" data. Once again HYDRUS-2D results were fairly similar regardless of soil group. These results indicate that in general for LEACHM and GLEAMS the "best" and "good" datasets provided similar results using either pesticide data, and when using optimized pesticide parameters these results were significantly better than the results obtained from simulation run with the "limited" dataset. This suggests that particle size and moisture release data accessed from the local reference dataset for the Horotiu soil were quite similar to those found during site characterization studies. For HYDRUS-2D there was little difference in results regardless of soil input quality, possibly showing the ability of the associated soil parameter estimation program (ROSETTA) to provide suitable water retention data from particle size distribution data.

The results obtained from our study indicated that if using LEACHM to model leaching with USDA leaching parameters, there would be little benefit in obtaining costly site-specific soil characteristic data as models run with the "limited" data performed almost as well as those run with the "best" data. This was the same for both GLEAMS and HYDRUS-2D using USDA pesticide data. If using the HYDRUS-2D model, there also appears to be little benefit in collecting soil characteristic data when using optimized and laboratory pesticide leaching data, as, for all leaching data sources, HYDRUS-2D performed similarly regardless of soil characteristic group. When using optimized data and running GLEAMS and LEACHM it appeared worthwhile to obtain soil data from either a reference set ("good" data) or site-specific data.

This assessment of cost-effective sources of data is likely to vary from site to site depending on the match of soil quality data and USDA data to the "true" values. There is potential for the "poorer" quality data to be actually closer to the "true" values than the more site-specific data (with the exception of optimized data) and as such provide a better "match" to observed data. However, cases such as these are likely to be rare and, in general, obtaining site-specific soil and pesticide sorption and degradation parameters is likely to provide the best results.

Of the three models used in this case study we observed that overall the LEACHM model consistently provided the best GOF values. While the analysis of variance of the SSres values of the three models run with hexazinone and procymidone indicated no significant difference between the "match" of the three models (with the exception of picloram models), qualitative comparisons of box plots (Fig. 1) indicate that LEACHM generally provides the lowest SSres values and HYDRUS-2D is more precise (smaller spread of data) than either GLEAMS or LEACHM. Overall, GLEAMS results were generally the least accurate, which is probably a reflection of the less complex handling of water movement and the restriction of five soil layers in the model compared to the eight layers identified at the site (Close et al., 2003a). Rekolainen et al. (2000) indicated that GLEAMS is designed to simulate chemical transfer in surface runoff and field edge and in percolation water out of the root zone. As such, GLEAMS may not be the most suitable model for smaller scale plots, such as used in these experiments. One reason for the better performance of LEACHM compared to HYDRUS-2D in this study is that HYDRUS-2D only outputs the liquid phase resident concentrations, and thus direct calibrations can be made only against soil water data. The calibration of soil concentrations was indirectly performed through calibration of soil mass in each subregion of the model domain, which can lead to incorporation of some error if the soil layers are thick. This may be a significant factor in this study as most of the observed data were from soil samples.

The results from this study indicated that for the models assessed, doing laboratory batch tests on the soils to assess leaching characteristics may lead to more accurate results than using more general USDA data, while obtaining field data provides by far the best results. A comparison of the SSres ratios (Table 7) of simulations from optimized parameters (from field data) to laboratory and USDA parameters showed that model simulations using optimized values are on average 67 to 74% more accurate than those using USDA parameters (ratios of 0.26, 0.29, and 0.33 for "best", "good", and "limited" soil groups, respectively). For procymidone, simulations using laboratory-derived parameters were approximately 70 to 88% less accurate than those using optimized parameters but 24 to 32% more accurate than those using USDA parameters.


View this table:
[in this window]
[in a new window]
  		Table 7. Summary of mean (of three models) ratios of optimized, laboratory, and USDA model derived outputs.

 

   CONCLUSIONS
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 CONCLUSIONS
REFERENCES
 
We found that the source and quality of the pesticide sorption and degradation data had a much greater influence over the ability of either of the models to simulate actual pesticide leaching than does the quality of the soil characteristic data. When using the optimized leaching data there was a difference in the simulation results between the two better soil datasets and limited dataset. This difference was not apparent when using the USDA leaching data. This reflects the smoothing effect of the generalized USDA data, which provided significantly poorer results regardless of the quality of the soil data. This also shows that with accurate sorption and degradation parameters soil data quality is likely to be important for accurate simulation results.

Overall LEACHM simulated observed data more accurately than GLEAMS regardless of the complexity of the soil data and also more accurately than HYDRUS-2D with higher quality soil property data. When using USDA data, however, there is little difference between the models. The HYDRUS-2D model performed comparatively well with sparse soil property data. One possible reason for this may be the effectiveness of the associated parameter estimation programs used to generate hydraulic properties from water retention or particle size data. We found that data quality in terms of physical soil property data had little effect on either of the models results if using easily accessible USDA pesticide data.

In this study, when using site-derived pesticide sorption and degradation parameters the more complex models (LEACHM and HYDRUS-2D) were more accurate, but when using USDA derived data all models performed more poorly and about equally. The use of site-derived sorption and degradation parameters had significantly more impact on the ability of the models to estimate field solute concentrations than the choice of the models themselves or the quality of the soil data.


   ACKNOWLEDGMENTS
 
The authors thank Paul Johnstone for carrying out some of the model simulations, and Trevor Webb and Linda Lilbourne, Landcare Research, and Mark Flintoft, ESR, for their valuable discussions and comments. The authors also thank the anonymous reviewers whose suggestions significantly improved this paper. The research was funded by contracts CO3X0303 (ESR) and CO9X0017 (Landcare Research) from the Foundation for Science, Research and Technology (New Zealand).


   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 ISI Web of Science
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 ISI Web of Science (4)
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Dann, R. L.
Right arrow Articles by Pang, L.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Dann, R. L.
Right arrow Articles by Pang, L.
GeoRef
Right arrow GeoRef Citation
Agricola
Right arrow Articles by Dann, R. L.
Right arrow Articles by Pang, L.
Related Collections
Right arrow Pesticides
Right arrow Organic Compounds
Right arrow Agricultural Pesticides

HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
The SCI Journals Agronomy Journal Crop Science
Vadose Zone Journal Journal of Plant Registrations
Journal of Natural Resources
and Life Sciences Education		 Soil Science Society of America Journal