The modeling toolkit BUVARD (BUffer strip for runoff Attenuation and pesticides Retention Design tool) consists in several steps like analyzing the watershed and its characteristics (soil, climate, cultural practices), and running dynamical models, in particular the mechanistic model VFSMOD. At the end this toolkit delivers the optimal VFS width considering the needed filter efficiency (for example, 70% of runoff reduction).The present study aims at making possible to simplify this very complete method, to determine its sensitivity to input parameters in new climatic and agronomic conditions by reducing the computational cost of the modeling toolkit. A metamodel (or surrogate model) developed on local conditions would allow to perform GSA with low cost yet ensuring it is based on physics. It would help users understanding the most important processes in the VFS they want to design and would increase the operational scope of the modeling toolkit. We performed a much smaller sampling of input parameters using the Latin Hypercube Sampling method optimized with a maximin criteria. The metamodel is then be based on a Kriging approach with a very good quality of prediction. It is a promising tool to perform uncertainty analysis and sensitivity analysis at alow cost.