Automatic calibration of the Biome-BGC model with the PEST software to simulate the forest and farmland ecosystems of the Qinling Mountains in China

• The Biome-BGC model was integrated with the PEST software to achieve automatic model parameter estimation. • The new Biome-BGC-PEST package simplified the processes of model parameter estimation and sensitivity analysis. • Influential model parameters were obtained for different vegetation cover types in the Qinling Mountains of China. • The simulation performance of the Biome-BGC model was greatly improved in the Qinling Mountains. Ecological models are important tools for quantifying and evaluating the carbon and water cycles of agricultural and forest ecosystems. However, quick determination of the values of parameters of a given model remains a big challenge for most model users, especially beginners. In this study, we coupled an independent automatic parameter optimization tool of PEST (Parameter ESTimation) with the Biome-BGC model through Python programming language, and finally developed a new Biome-BGC-PEST software package for automatic model optimization. The encapsulation of the optimization process for Biome-BGC model parameters has heavily simplified model operational steps and improved model calibration efficiency. With the Biome-BGC-PEST package, sensitivity analysis and optimization of physiological and ecological parameters of the Biome-BGC model were conducted based on combined remote-sensing products of GPP (Gross primary productivity) and ET (Evapotranspiration) for the agricultural and forest ecosystems in the Qinling Mountains of China. Compared with the traditional trial-and-error methods for parameter optimization, the influential parameters estimated by the Biome-BGC-PEST package were similar, mainly including atmospheric deposition of N, symbiotic and asymbiotic fixation of N, cuticular conductance, etc. However, they were dramatically different in their sensitivity magnitudes. This was mainly because the new method greatly enhanced the efficiency of parameter optimization through allowing simultaneously tuning all of the parameters related to carbon and water fluxes. Consequently, the simulation accuracy of the Biome-BGC model was dramatically improved for the agricultural and forest ecosystems in the Qinling Mountains after parameter optimization. The R 2 (Coefficient of determination) of general GPP simulations increased from 0.67 to 0.89 and the RMSE (Root mean square error) decreased by about 37 %. Similarly, the R 2 of general ET simulations increased from 0.57 to 0.86 and the RMSE decreased by about 55 %. In conclusion, the newly established Biome-BGC-PEST package demonstrated similar or better optimization efficiency and accuracy compared to the traditional methods, which could greatly promote the application of the Biome-BGC model in relevant research of agricultural and ecological modeling.