Validation of the PESTLA model: Field test using data from a sandy soil in Schaijk (the Netherlands) 715802002

Validation of the PESTLA model: Field test using data from a sandy soil in Schaijk (the Netherlands)

[ [Validatie van het model PESTLA: toetsing met veldgegevens van een zandgrond in Schaijk.] ]

Boekhold AE, Swartjes FA, Hoogenboom FGG, van der Linden AMA

45 p in English 1993

RIVM Rapport 715802002

Toon Nederlands

English Abstract
Within the framework of the project "Validation of PESTLA" the Schaijk data set was used to analyse PESTLA model performance. The Schaijk data set contains field data on bentazon behaviour in a coarse textured humic gley soil cropped with maize. PESTLA model input parameters were derived from this data set and model simulations were performed. Comparison of model predictions with field measurements of bentazon concentrations in groundwater and contents in topsoil was done using graphical and statistical methods. The statistical method that was applied was the factor-of-f approach. This method acknowledges that both measurements and model simulations yield variable output. Variability implies that measurements and simulations are uncertain. This uncertainty is quantified and used to test the hypothesis that PESTLA model calculations are, or are not, sufficiently close to measurements. Application of the factor-of-f approach led to the conclusion that PESTLA simulated bentazon behaviour in Schaijk well. The hypothesis 'model is good' was never rejected and the hypothesis 'model is false' was rejected for almost all combinations of allowed uncertainty. The results given above were obtained with a value for bentazon half life as calculated using the standard method of linear regression on log transformed dissipation data. However, calculation of bentazon half life using non-linear regression yielded a value that was almost twice as small. PESTLA simulations using this latter value were nog in accordance with measurements anymore. This finding indicates that decisions with regard to model validity are highly sensitive to (subjective) decisions on how model input parameters are obtained.

RIVM - Bilthoven - the Netherlands - www.rivm.nl

Rijksinstituut voor Volksgezondheid en Milieu RIVM
_{( 1993-12-31 )}