RING TEST WITH THE MODELS LEACHP, PRZM-2 AND VARLEACH - VARIABILITY BETWEEN MODEL USERS IN PREDICTION OF PESTICIDE LEACHING USING A STANDARD DATA SET

A ring test was carried out with three mathematical models for pestici de leaching to compare predictions from a number of modellers for a si ngle field experiment when using the same model. The exercise sought t o investigate the level of variation, if any, in model output introduc ed by user-dependent subjectivity during selection of input parameters . Five modellers were given a full description of a field experiment c arried out in the UK to determine the leaching potential of a novel pe sticide and then used the models LEACHP, PRZM-2 and VARLEACH to predic t concentrations of pesticide in soil water at 1 m depth and in soil f or a 1 m profile 220 days after application. Agreement with observed r esults was generally best for LEACHP and worst for VARLEACH, but no tw o sets of predicted results for a given model were exactly the same, e ven for the simple model VARLEACH. Differences between simulations wit h the same model were attributed to a number of input parameters which could not be derived from the experimental information provided and t hus introduced subjectivity into the modelling process. The parameters identified included dispersivity, initial soil conditions and factors determining the rate of pesticide degradation. Differences between ou tput data with the same model were of a similar order of magnitude to the variation associated with held measurements and were generally sma ller than the discrepancies between observed and predicted data. User- dependence of modelling has not previously been considered, but should be an important component in assessing model output and in evaluating the validity and use of a given programme. Model development should s eek to reduce subjectivity in selection of input parameters and improv e the guidance available to users where subjectivity cannot be elimina ted.