Assessing the sensitivity of a land surface model through qualitative and quantitative tests

Abstract:

Land surface models (LSMs) represent land-atmosphere fluxes in global climate models and are therefore key for projecting the future of the global carbon cycle. LSMs are highly non-linear with many parameters that are difficult to constrain. Understanding the sensitivity of model outputs to the uncertainty of model parameter values is essential for assessing the robustness of model results, identifying model deficiencies, and for systematic model tuning. This presentation will evaluate the sensitivity of the Canadian Land Surface Scheme Including Biogeochemical Cycles (CLASSIC) when forced with local conditions from a selection of eddy covariance measurement sites. As a first step, we employ a qualitative test (Morris method) that identifies the most influential parameters. Focusing on a subset of parameters, we then implement a quantitative test (Sobol method) that shows the relative importance of the most influential parameters. The results are then used for identifying model deficiencies and systematic model tuning.