Premium
Parameter optimization for carbon and water fluxes in two global land surface models based on surrogate modelling
Author(s) -
Li Jianduo,
Duan Qingyun,
Wang YingPing,
Gong Wei,
Gan Yanjun,
Wang Chen
Publication year - 2018
Publication title -
international journal of climatology
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.58
H-Index - 166
eISSN - 1097-0088
pISSN - 0899-8418
DOI - 10.1002/joc.5428
Subject(s) - environmental science , biosphere , climatology , climate model , global optimization , meteorology , scale (ratio) , computer science , econometrics , climate change , mathematics , algorithm , geology , ecology , biology , oceanography , physics , quantum mechanics
ABSTRACT Errors are quite large in the simulated carbon and water fluxes obtained by global models used for the Fifth Assessment Report of the Intergovernmental Panel on Climate Change, and reducing those errors is important for improving our confidence about these models and their projections. Errors in model parameter values are a major cause of those large modelling errors but can be significantly reduced if model parameter values are optimized. While parameter optimizations have been carried out at local sites or regional scales, parameter optimizations have been rarely conducted at the global scale because of the high computing costs required to optimize a large (>100) number of model parameters. In this study, we used an adaptive surrogate modelling based optimization (ASMO) method to maximize the match between simulated monthly global gross primary production (GPP) and latent heat flux (LE) derived by two global land surface models (LSMs) and the model‐data products for global GPP and LE from the 1982–2008 period generated by the Max Plank Institute. The ASMO method only required a few hundred model runs to find the optimal values of all optimized parameters for the two global LSMs [the Australian Community Atmosphere‐Biosphere‐Land Exchange (CABLE) and joint UK land environment simulator (JULES)]. Our results show that up to 65% of the model errors can be reduced by parameter optimization for most of the plant functional types (PFTs) and that the model performances of CABLE and JULES are significantly improved at 72 and 93% of the land points, respectively. At last, we discuss the limitations of this work and recommend that parameter optimization based on surrogate modelling using various observational data sets and acceptable prior information of uncertainties in model structure and observations should be considered as a key step in improving the performance of global LSMs or model intercomparisons.