Integrating a model with remote sensing observations by a data assimilation approach to improve the model simulation accuracy of carbon flux and evapotranspiration at two flux sites

Model simulation and in situ observations are often used to research water and carbon cycles in terrestrial ecosystems, but each of these methods has its own advantages and limitations. Combining these two methods could improve the accuracy of quantifying the dynamics of the water and carbon fluxes of an ecosystem. Data assimilation is an effective means of integrating modeling with in situ observation. In this study, the ensemble Kalman filter (EnKF) and the unscented Kalman filter (UKF) algorithms were used to assimilate remotely sensed leaf area index (LAI) data with the Biome-BGC model to simulate water and carbon fluxes at the Harvard Forest Environmental Monitoring Site (EMS) and the Dinghushan site. After MODIS LAI data from 2000–2004 were assimilated into the improved Biome-BGC model using the EnKF algorithm at the Harvard Forest site, the R2 between the simulated and observed results for NEE and evapotranspiration increased by 7.8% and 4.7%, respectively. In addition, the sum of the absolute error (SAE) and the root mean square error (RMSE) of NEE decreased by an average of 21.9% and 26.3%, and the SAE and RMSE of evapotranspiration decreased by 24.5% and 25.5%, respectively. MODIS LAI data of 2003 were assimilated into the Biome-BGC model for the Dinghushan site, and the R2 values between the simulated and observed results for NEE and evapotranspiration were increased by 6.7% and 17.3%, respectively. In addition, the SAE values of NEE and ET were decreased by 11.3% and 30.7%, respectively, and the RMSE values of NEE and ET decreased by 10.1% and 30.9%, respectively. These results demonstrate that the accuracy of carbon and water flux simulations can be effectively improved when remotely sensed LAI data are properly integrated with ecosystem models through a data assimilation approach.