Sensitivity analysis of a radionuclide transfer model describing contaminated vegetation in Fukushima prefecture, using Morris and Sobol' - Application of sensitivity analysis on a radionuclides transfer model in the environment describing weeds contamination in Fukushima Prefecture, using Morris method and Sobol' indices indices

The increasing spatial and temporal complexity of models demands methods capable of ranking the influence of their large numbers of parameters. This question specifically arises in assessment studies on the consequences of the Fukushima accident. Sensitivity analysis aims at measuring the influence of input variability on the output response. Generally, two main approaches are distinguished (Saltelli, 2001, Iooss, 2011): - Screening approach, less expensive in computation time and allowing to identify non influential parameters; - Measures of importance, introducing finer quantitative indices. In this category, there are regression-based methods, assuming a linear or monotonic response (Pearson coefficient, Spearman coefficient), and variance-based methods, without assumptions on the model but requiring an increasingly prohibitive number of evaluations when the number of parameters increases. These approaches are available in various statistical programs (notably R) but are still poorly integrated in modelling platforms of radioecological risk assessment. This work aimed at illustrating the benefits of sensitivity analysis in the course of radioecological risk assessments This study used two complementary state-of-art global sensitivity analysis methods: - The screening method of Morris (Morris, 1991; Campolongo et al., 2007) based on limited model evaluations with a one-at-a-time (OAT) design; - The variance-based Sobol' sensitivity analysis (Saltelli, 2002) based a large number of model evaluations in the parameter space with a quasi-random sampling (Owen, 2003). Sensitivity analyses were applied on a dynamic Soil-Plant Deposition Model (Gonze et al., submitted to this conference) predicting foliar concentration in weeds after atmospheric radionuclide fallout. The Soil-Plant Deposition Model considers two foliage pools and a root pool, and describes foliar biomass growth with a Verhulst model. The developed semi-analytic formulation of foliar concentration, derived from an ordinary differential equations system, involved ten parameters. After Sobol' analysis, foliar concentration (at times ranging from 0 to 8 months after deposition) was essentially sensitive to drop-off initial concentration, weathering rate and biomass growth rate, with strong interactions between them. This study stressed that sensitivity analysis is an essential part of assessments, by identifying the non-influential parameters, for which default values can be used, and pointing out critical parameters, for which site-specific calibration (Nicoulaud-Gouin et al., submitted to this conference) is recommended and their temporal and spatial variability should be included in the model structure. It motivated the development of an integrated toolbox allowing end users to conduct sensitivity analyses on a wide range of models, from analytic to more complex ones, including those developed in the Symbiose platform (Gonze et al, 2010). (authors)