Sensitivity analysis of a hierarchical qualitative model for sustainability assessment of cropping systems.

Sensitivity Analysis (SA) was performed on the MASC model (Multi-attribute Assessment of the Sustainability of Cropping systems), a hierarchical qualitative model built to assess the sustainability of cropping systems developed under a decision support tool called DEXi. Three approaches were tested to perform a first-order SA assuming a fixed model structure and no correlation among input variables: (i) factorial designs combined with analysis of variance (ANOVA), (ii) Monte-Carlo sampling (MC), (iii) conditional probabilities (CPs). The three tested approaches were coded under the R-statistical package in a generic way in order to be used to perform SA on every DEXi-type model. Results showed that, due to the dimension of the tree, a complete factorial design was unsuitable. MC and CPs represented efficient alternatives to perform an analysis of the overall model. CPs exploited better the hierarchical structure of the model to give exact first-order indices, while MC would be a more flexible approach for the introduction of correlations among variables. The outputs of the SA showed that there was an unbalanced distribution of the modalities of the three domains of sustainability (i.e., Economic, Social and Environmental) and that the MASC model tends to poorly discriminate the cropping systems. This may be due to the structure of the tree. Reducing the complexity of the tree using ''satellite trees'', restructuring the tree to get an equilibrated tree between the Economic, Social and Environmental dimension or increase the number of qualitative value per leaf may improve the discriminative power of the model. Such results demonstrated the interest of performing SA as a guiding tool for modellers.