Tinglong Zhang 1,2,3, Rui Sun3, Changhui Peng 1
1Laboratory of Global Change and Ecosystems Forecasting, College of Forestry,
Northwest A&F University,
2College of Resources and Environmental Sciences, Northwest A&F University,
3School of Geography and Remote Sensing Sciences, Beijing Normal University

Ecological process model can simulate terrestrial ecosystem carbon cycle process and carbon flux. It is a very useful work on carbon cycle research. However, model simulation has its own limitations, there are always some difference between simulation results and real situation, how to improve the accuracy of model to simulate vegetation productivity? one way is to improve and optimize the model, even propose new models; Another way is model combined with observations, using data assimilation methods, observation was joined into model while model is running, By these, model trajectory was adjusted and model simulated error was corrected, simulation results was made to carry information from both model and observations . Thus, it can effectively reduce error and improve accuracy of model simulating.
In this paper, simulated annealing algorithm was used to optimize physiological and ecological parameters of Biome-BGC ecosystem model. Optimization results show that: Using the optimized parameters, model simulation results was closer to actual observations. Parameter optimization can effectively reduce uncertainty of model simulation.
When using Biome-BGC model to simulate water, carbon flux at Harvard Forest, a problem was found that some years in the vegetation growing season, vegetation gross primary productivity (GPP) will suddenly drop to a very low level, and continued more for a long time, but not so the actual flux observation. Biome-BGC model was improved and adjusted at Harvard Forest area. After adjustment, the model simulated accuracy of water, carbon flux are greatly improved.
Using ensemble Kalman filtering (ENKF) assimilation algorithm, observation LAI (whether field measured or remote sensing inversion) were assimilated into model, can effectively improve the accuracy of model simulated results (NEE and evapotranspiration) and reduce errors at Harvard forest EMS flux station and Dinghushan flux stations.