Comparison of methods for the detection and extrapolation of trends in groundwater quality.

Land use changes and the intensification of agriculture since the 1950s have resulted in a deterioration of groundwater quality in many European countries. For the protection of groundwater quality, it is necessary to (1) assess the current groundwater quality status, (2) detect changes or trends in groundwater quality, (3) assess the threat of deterioration and (4) predict future changes in groundwater quality. A variety of approaches and tools can be used to detect and extrapolate trends in groundwater quality, ranging from simple linear statistics to distributed 3D groundwater contaminant transport models. In this paper we report on a comparison of four methods for the detection and extrapolation of trends in groundwater quality: (1) statistical methods, (2) groundwater dating, (3) transfer functions, and (4) deterministic modeling. Our work shows that the selection of the method should firstly be made on the basis of the specific goals of the study (only trend detection or also extrapolation), the system under study, and the available resources. For trend detection in groundwater quality in relation to diffuse agricultural contamination, a very important aspect is whether the nature of the monitoring network and groundwater body allows the collection of samples with a distinct age or produces samples with a mixture of young and old groundwater. We conclude that there is no single optimal method to detect trends in groundwater quality across widely differing catchments.

A refined lack-of-fit statistic to calibrate pesticide fate models for responsive systems.

BACKGROUND: Calibration by inverse modelling was performed with the MACRO transport and fate model using long-term (>10 years) drainflow and isoproturon (IPU) data from western France. Two lack-of-fit (LOF) indices were used to control the inverse modelling: sum of squares (SS) and an alternative statistic called the vertical-horizontal distance integrator (VHDI), which is designed to account for offsets in observed and predicted arrival times of peak IPU concentration. With these data, SS was artificially inflated because it is limited to comparison of predicted and observed IPU concentrations that are concurrent in time. The LOFs were used along with the index of agreement (d) and the correlation coefficient (r) to ascertain the fit of the calibrated models. RESULTS: Predicted arrival times of peak IPU concentration differed somewhat from observed times. All four indices indicated better model fit for the second of two validation periods when inverse modelling was controlled by VHDI rather than SS (SS = 26.4, d = 0.660, r = 0.606 and VHDI = 1.25). The VHDI statistic was markedly lower compared with the uncalibrated model (38.0) and SS calibration results (24.5). The final maximum predicted IPU concentration (44.5 microg L(-1)) for the calibration period was very similar to the observed value (44 microg L(-1)). CONCLUSION: VHDI is seen as an effective alternative to SS for calibration and validation of pesticide fate models applied to responsive systems. VHDI provided a more realistic assessment of model performance for the transient flows and short-lived concentrations observed here, and also effectively substituted for the objective function in inverse modelling.

Stationary and non-stationary autoregressive processes with external inputs for predicting trends in water quality.

An autoregressive approach for the prediction of water quality trends in systems subject to varying meteorological conditions and short observation periods is discussed. Under these conditions, the dynamics of the system can be reliably forecast, provided their internal processes are understood and characterized independently of the external inputs. A methodology based on stationary and non-stationary autoregressive processes with external inputs (ARX) is proposed to assess and predict trends in hydrosystems which are at risk of contamination by organic and inorganic pollutants, such as pesticides or nutrients. The procedures are exemplified for the transport of atrazine and its main metabolite deethylatrazine in a small agricultural catchment in France. The approach is expected to be of particular value to assess current and future trends in water quality as part of the European Water Framework Directive and Groundwater Directives.

Formation of pesticide nonextractable (bound) residues in soil: magnitude, controlling factors and reversibility.

The analysis of the coherent data on nonextractable (bound) residues (NER) from the literature and EU pesticide registration dossiers allows the identification of general trends, in spite of the large variability and heterogeneity of data. About 50% of the pesticides reviewed exhibit a low proportion of NER (less than 30% of the initial amount) while only 12% of pesticides have a proportion of NER exceeding 70%. The lowest proportion of NER was found for dinitroanilines (<20%), and the largest value was obtained for carbamates, and in particular dithiocarbamates. The presence of chemical reactive groups, such as aniline or phenol, tends to yield a larger proportion of NER. NER originating from N-heteroatomic ring were found to be lower than those from phenyl-ring structures. Among the environmental factors affecting the formation of NER, microbial activity has a direct and significant effect. Concerning the NER uptake or their bioavailability, consistent data suggest that only a small percentage of the total amounts of NER can be released. The analysis of NER formation kinetics showed that incubation experiments are often stopped too early to allow a correct evaluation of the NER maturation phase. Therefore, there is a need for longer term experiments to evaluate the tail of the NER formation kinetics. Still, the heterogeneity of the NER data between pesticides and for specific pesticides calls for great care in the interpretation of the data and their generalization.

Identification of key climatic factors regulating the transport of pesticides in leaching and to tile drains.

BACKGROUND: Key climatic factors influencing the transport of pesticides to drains and to depth were identified. Climatic characteristics such as the timing of rainfall in relation to pesticide application may be more critical than average annual temperature and rainfall. The fate of three pesticides was simulated in nine contrasting soil types for two seasons, five application dates and six synthetic weather data series using the MACRO model, and predicted cumulative pesticide loads were analysed using statistical methods. RESULTS: Classification trees and Pearson correlations indicated that simulated losses in excess of 75th percentile values (0.046 mg m(-2) for leaching, 0.042 mg m(-2) for drainage) generally occurred with large rainfall events following autumn application on clay soils, for both leaching and drainage scenarios. The amount and timing of winter rainfall were important factors, whatever the application period, and these interacted strongly with soil texture and pesticide mobility and persistence. Winter rainfall primarily influenced losses of less mobile and more persistent compounds, while short-term rainfall and temperature controlled leaching of the more mobile pesticides. CONCLUSIONS: Numerous climatic characteristics influenced pesticide loss, including the amount of precipitation as well as the timing of rainfall and extreme events in relation to application date. Information regarding the relative influence of the climatic characteristics evaluated here can support the development of a climatic zonation for European-scale risk assessment for pesticide fate.

User subjectivity in Monte Carlo modeling of pesticide exposure.

Monte Carlo techniques are increasingly used in pesticide exposure modeling to evaluate the uncertainty in predictions arising from uncertainty in input parameters and to estimate the confidence that should be assigned to the modeling results. The approach typically involves running a deterministic model repeatedly for a large number of input values sampled from statistical distributions. In the present study, six modelers made choices regarding the type and parameterization of distributions assigned to degradation and sorption data for an example pesticide, the correlation between the parameters, the tool and method used for sampling, and the number of samples generated. A leaching assessment was carried out using a single model and scenario and all data for sorption and degradation generated by the six modelers. The distributions of sampled parameters differed between the modelers, and the agreement with the measured data was variable. Large differences were found between the upper percentiles of simulated concentrations in leachate. The probability of exceeding 0.1 microg/L ranged from 0 to 35.7%. The present study demonstrated that subjective choices made in Monte Carlo modeling introduce variability into probabilistic modeling and that the results need to be interpreted with care.

Inverse modelling for estimating sorption and degradation parameters for pesticides.

The leaching model PESTRAS was used to estimate sorption and degradation values for bentazone from three lysimeter datasets using the inverse modelling package PEST. Investigations were undertaken to assess the influence on calibration results of (1) values attributed to uncertain parameters not included in the calibration, and (2) starting values supplied to the inverse modelling package. Automatic calibrations with different realistic values for the Freundlich exponent n(f) yielded different combinations of K(om) and DT50. Similarly, the supply of different starting values for K(om) and DT50 revealed that different combinations of these two parameters equally calibrated PESTRAS for two of the three lysimeters. Examination of the error surface, ie the forward running of the model for different combinations of K(om) and DT50 values, and the calculation of the goodness-of-fit to the experimental data, was found useful for identifying those instances where non-uniqueness in the calibration is likely to occur. Although the derivation of sorption and degradation values through inverse modelling is expected to offer significant benefits over laboratory determinations, care should be exercised when examining values derived through this approach. Research is needed to identify data requirements for robust estimation of sorption and degradation parameters through calibration of pesticide fate models against leaching data.

Exposure to sulfosulfuron in agricultural drainage ditches: field monitoring and scenario-based modelling.

Field monitoring and scenario-based modelling were used to assess exposure of small ditches in the UK to the herbicide sulfosulfuron following transport via field drains. A site in central England on a high pH, clay soil was treated with sulfosulfuron, and concentrations were monitored in the single drain outfall and in the receiving ditch 1 km downstream. Drainflow in the nine months following application totalled 283 mm. Pesticide lost in the first 12.5 mm of flow was 99% of the total loading to drains (0.5% of applied). Significant dilution was observed in the receiving ditch and quantifiable residues were only detected in one sample (0.06 microg litre(-1)). The MACRO model was evaluated against the field data with minimal calibration. The parameterisation over-estimated the importance of macropore flow at the site. As a consequence, the maximum concentration in drainflow (2.3 microg litre(-1)) and the total loading to drains (0.76 g) were over-estimated by factors of 2.4 and 5, respectively. MACRO was then used to simulate long-term fate of the herbicide for each of 20 environmental scenarios. Resulting estimates for concentrations of sulfosulfuron in a receiving ditch were weighted according to the prevalence of each scenario to produce a probability distribution of daily exposure.

Sources of uncertainty in pesticide fate modelling.

There is worldwide interest in the application of probabilistic approaches to pesticide fate models to account for uncertainty in exposure assessments. The first steps in conducting a probabilistic analysis of any system are: (i) to identify where the uncertainties come from; and (ii) to pinpoint those uncertainties that are likely to affect most of the predictions made. This article aims at addressing those two points within the context of exposure assessment for pesticides through a review of the different sources of uncertainty in pesticide fate modelling. The extensive listing of sources of uncertainty clearly demonstrates that pesticide fate modelling is laced with uncertainty. More importantly, the review suggests that the probabilistic approaches, which are typically being deployed to account for uncertainty in the pesticide fate modelling, such as Monte Carlo modelling, ignore a number of key sources of uncertainty, which are likely to have a significant effect on the prediction of environmental concentrations for pesticides (e.g. model error, modeller subjectivity). Future research should concentrate on quantifying the impact these uncertainties have on exposure assessments and on developing procedures that enable their integration within probabilistic assessments.

Sensitivity analyses for four pesticide leaching models.

Sensitivity analyses using a one-at-a-time approach were carried out for leaching models which have been widely used for pesticide registration in Europe (PELMO, PRZM, PESTLA and MACRO). Four scenarios were considered for simulation of the leaching of two theoretical pesticides in a sandy loam and a clay loam soil, each with a broad distribution across Europe. Input parameters were varied within bounds reflecting their uncertainty and the influence of these variations on model predictions was investigated for accumulated percolation at 1-m depth and pesticide loading in leachate. Predictions for the base-case scenarios differed between chromatographic models and the preferential flow model MACRO for which large but transient pesticide losses were predicted in the clay loam. Volumes of percolated water predicted by the four models were affected by a small number of input parameters and to a small extent only, suggesting that meteorological variables will be the main drivers of water balance predictions. In contrast to percolation, predictions for pesticide loss were found to be sensitive to a large number of input parameters and to a much greater extent. Parameters which had the largest influence on the prediction of pesticide loss were generally those related to chemical sorption (Freundlich exponent nf and distribution coefficient Kf) and degradation (either degradation rates or DT50, QTEN value). Nevertheless, a significant influence of soil properties (field capacity, bulk density or parameters defining the boundary between flow domains in MACRO) was also noted in at least one scenario for all models. Large sensitivities were reported for all models, especially PELMO and PRZM, and sensitivity was greater where only limited leaching was simulated. Uncertainty should be addressed in risk assessment procedures for crop-protection products.

Issues of replicability in Monte Carlo modeling: a case study with a pesticide leaching model.

Sensitivity and uncertainty analyses based on Monte Carlo sampling were undertaken for various numbers of runs of the pesticide leaching model (PELMO). Analyses were repeated 10 times with different seed numbers. The ranking of PELMO input parameters according to their influence on predictions for leaching was stable for the most influential parameters. For less influential parameters, the sensitivity ranking was severely influenced by the seed number used. For uncertainty analyses, probabilities of exceeding a particular concentration were significantly influenced by the seed number used in the random sampling of values for the two parameters considered, even for those cases in which 5,000 model runs were undertaken (coefficient of variation of 10 replicated analyses, 5%). A decrease in the variability of exceedance probabilities could be achieved by further increasing the number of model runs. However, this may prove to be impractical when complex deterministic models with a relatively long running time are used. Attention should be paid to replicability aspects by modelers when devising their approach to assessing the uncertainty associated with the modeling and by decision makers when examining the results of probabilistic approaches.

Calibration of pesticide leaching models: critical review and guidance for reporting.

Calibration of pesticide leaching models may be undertaken to evaluate the ability of models to simulate experimental data, to assist in their parameterisation where values for input parameters are difficult to determine experimentally, to determine values for specific model inputs (e.g. sorption and degradation parameters) and to allow extrapolations to be carried out. Although calibration of leaching models is a critical phase in the assessment of pesticide exposure, lack of guidance means that calibration procedures default to the modeller. This may result in different calibration and extrapolation results for different individuals depending on the procedures used, and thus may influence decisions regarding the placement of crop-protection products on the market. A number of issues are discussed in this paper including data requirements and assessment of data quality, the selection of a model and parameters for performing calibration, the use of automated calibration techniques as opposed to more traditional trial-and-error approaches, difficulties in the comparison of simulated and measured data, differences in calibration procedures, and the assessment of parameter values derived by calibration. Guidelines for the reporting of calibration activities within the scope of pesticide registration are proposed.

Prediction of pesticide concentrations found in rivers in the UK.

SWATCATCH is a distributed model combined with databases within a GIS as the POPPIE system to predict pesticide concentrations in rivers at the catchment outlet. The model was evaluated against a dataset of pesticide concentrations in rivers of England and Wales. More than 2000 individual analyses in each of the years 1995 and 1997 covered approximately 150 catchment-pesticide combinations drawn from 29 catchments and 16 pesticides, themselves selected to represent a range of characteristics and properties. SWATCATCH was better able to simulate maximum pesticide concentrations at any time during the year than the proportion of samples containing residues of a particular pesticide above the limit of quantification. The model simulated maximum pesticide concentrations in surface waters which were within a factor of 10 of those observed for 66-74% of catchment-pesticide-year combinations. Simulated and observed frequency of detection could not be differentiated using a Chi 2 test for 54-67% of simulations. Time series analysis for seven of the 29 catchment-pesticide combinations indicated that measured and detected series of concentrations generally followed similar patterns. The evaluation supports the intended use of the model in assisting the construction of pesticide monitoring programmes.

Sensitivity and first-step uncertainty analyses for the preferential flow model MACRO.

Sensitivity analyses for the preferential flow model MACRO were carried out using one-at-a-time and Monte Carlo sampling approaches. Four different scenarios were generated by simulating leaching to depth of two hypothetical pesticides in a sandy loam and a more structured clay loam soil. Sensitivity of the model was assessed using the predictions for accumulated water percolated at a 1-m depth and accumulated pesticide losses in percolation. Results for simulated percolation were similar for the two soils. Predictions of water volumes percolated were found to be only marginally affected by changes in input parameters and the most influential parameter was the water content defining the boundary between micropores and macropores in this dual-porosity model. In contrast, predictions of pesticide losses were found to be dependent on the scenarios considered and to be significantly affected by variations in input parameters. In most scenarios, predictions for pesticide losses by MACRO were most influenced by parameters related to sorption and degradation. Under specific circumstances, pesticide losses can be largely affected by changes in hydrological properties of the soil. Since parameters were varied within ranges that approximated their uncertainty, a first-step assessment of uncertainty for the predictions of pesticide losses was possible. Large uncertainties in the predictions were reported, although these are likely to have been overestimated by considering a large number of input parameters in the exercise. It appears desirable that a probabilistic framework accounting for uncertainty is integrated into the estimation of pesticide exposure for regulatory purposes.