Global Sensitivity Analysis for a MACRO Meta-Model for Swedish Drinking Water Abstraction Zones

In Sweden farmers are legally obliged to apply for permits for pesticide use if their land lies within a drinking water abstraction zone. The standalone modelling tool MACRO-DB 4 developed by the Swedish University of Agricultural Sciences (SLU) is available for risk assessment and decision support. MACRO-DB 4 is used by local authorities, farmers/landowners and consultants, and is based on the well-established leaching model MACRO 5.2. However, the software is costly to maintain and slow for end users. Hence, a robust meta-model of MACRO-DB 4 was developed and integrated in a web-based tool (MACRO DB Steg2 v.5) that is fast, easy to maintain, and easy to understand for stakeholders. The meta-model (implemented as an R package) is based on i) a large number of MACRO simulations for the whole agriculturally relevant area of Sweden, and ii) a trilinear interpolation tool. The simulations comprised 18 climates, 72 soils, 1 typical crop (spring cereals), 3 application seasons (spring, summer and autumn), and 150 dummy compounds consisting of a grid of normalized Freundlich coefficient Koc, degradation half-life at reference conditions DT50 and Freundlich exponent. Target variables were i) the mean leaching flux concentration over 20 years at 2 m depth (PECgw), and ii) the 20-year mean concentration in large surface water bodies (PECsw; based on pesticide inputs via drainage and baseflow) . The meta-model performs a trilinear interpolation (in the three-dimensional space of Koc, DT50 and Freundlich exponent) for log10 of PECgw or PECsw, respectively. Different crops are taken into account in the web-based tool by adjusting the pesticide interception fraction according to the BBCH stage of the crop to be modelled at the time of application. In order to identify the most important input factors for PECgw and PECsw, a variance-based Global Sensitivity Analysis (GSA) was performed for the MACRO meta-model using the Sobol’ method. This method allows to i) identify first-order (direct) and higher-order (interaction) effects for each input factor (climate, soil, application season, Koc, DT50 and Freundlich exponent), and ii) rank the input factors according to their importance.