Sensitivity analysis of a sensitivity analysis: we are likely overlooking the impact of distributional assumptions.

tAlthough uncertainty in input factor distributions is known to affect sensitivity analysis (SA) results, astandard procedure to quantify its impact is not available. We addressed this problem by performing a SA(generating sample of parameter distributions) of a SA (generating samples of parameter values for eachgenerated distribution) of the WARM rice model using the Sobol’ method. The sample of distributionswas generated using distributions of jackknife statistics calculated on literature values. This allowedmimicking the differences in distributions that could derive from different selection of literature sources.Despite the very low plasticity of WARM, the ranks of the two most relevant parameters was overturned in22% of the cases and, in general, differed from what achieved in earlier SAs performed on the same modelunder similar conditions. SA results were mainly affected by uncertainty in distribution of parametersinvolved in non-linear effects or interacting with others. The procedure identified parameters whoseuncertainty in distribution can alter SA results, i.e., parameters whose distributions could need to berefined.