For decades the marine ecological models have sustained progressive developments and been subjected to an increasing degree of complexity in their processes, forcings and parameterization. In parallel, the validation techniques have evolved from visual to statistical comparisons, allowing fair estimates of the bias and correlations between model results and reference data. Still, it is difficult to estimate in advance what will be the uncertainty attached to any model prediction because of the complexity of the ecological models and the non linearity of their response to a change. Also, it is not trivial to determine the uncertainty of the model response due to one specific forcing, especially when this forcing is variable in time and space. The uncertainty in an ecological model response is somewhat linked to the model sensitivity to a perturbation. Since the non-linear model response to a perturbation may vary in a wide range of possibilities, we chose to base our assessment on the probability theory by applying a “light” Monte-Carlo experiment. It consists in a reduced number of randomly-perturbed simulations where knowledge of the system allows narrowing the range of perturbations. We applied the light Monte-Carlo experiment on a biogeochemical model in the English Channel and the southern North Sea (3D-MIRO&CO). The uncertainty on modelled chlorophyll a prediction was studied as a response, first, to random wind perturbations and, second, to random phytoplankton autolysis values. Statistical and probabilistic quantification of the results is being presented for the Belgian coastal and offshore zones.
Located in
Library
/
RBINS Staff Publications
