Quantification and attribution of errors in the simulated annual gross primary production and latent heat fluxes by two global land surface models

Abstract:

Divergence among the predictions by different global land models has not decreased over the last three assessment reports by the International Panel on Climate Change. Quantification and attribution of the uncertainties of global land surface models are important for the next phase of model improvement and development, is therefore the focus of this study. There are three sources of model uncertainties: model inputs, parameter values and model structure. Here we focus on the errors in model parameters by comparing the differences between the simulated global gross primary productivity (GPP) and latent heat flux (LE) by two global land surface models and model-data products of global GPP and LE from 1982-2005. We found that the performance of simulated annual GPP or LE by both models is most sensitive to 2 to 9 model parameters screened out by Morris method for each plant functional type (PFT). Using ensemble simulations, we applied RS-HDMR method to verify the Morris sensitivity results, and implied that about 60% of the variances of model errors in some PFTs are attributed to the sensitive parameters. We selected the combination of key parameter values that minimized the monthly errors of GPP and LE for each. Our study shows that significant improvement of model predictions can be made through parameter optimization using observations.