Sensitivity analysis of a process‐based ecosystem model: Pinpointing parameterization and structural issues

Dynamic vegetation models have been widely used for analyzing ecosystem dynamics and their interactions with climate. Their performance has been tested extensively against observations and by model intercomparison studies. In the present analysis, Lund-Potsdam-Jena General Ecosystem Simulator (LPJ-GUESS), a state-of-the-art ecosystem model, was evaluated by performing a global sensitivity analysis. The study aims at examining potential model limitations, particularly with regard to long-term applications. A detailed sensitivity analysis based on variance decomposition is presented to investigate structural model assumptions and to highlight processes and parameters that cause the highest variability in the output. First- and total-order sensitivity indices were calculated for selected parameters using Sobol's methodology. In order to elucidate the role of climate on model sensitivity, different climate forcings were used based on observations from Switzerland. The results clearly indicate a very high sensitivity of LPJ-GUESS to photosynthetic parameters. Intrinsic quantum efficiency alone is able to explain about 60% of the variability in vegetation carbon fluxes and pools for a wide range of climate forcings. Processes related to light harvesting were also found to be important together with parameters affecting forest structure (growth, establishment, and mortality). The model shows minor sensitivity to hydrological and soil texture parameters, questioning its skills in representing spatial vegetation heterogeneity at regional or watershed scales. In the light of these results, we discuss the deficiencies of LPJ-GUESS and possibly that of other, structurally similar, dynamic vegetation models and we highlight potential directions for further model improvements.