A Multimodel Carbon Assimilation System Using a Modified Ensemble Kalman Filter and Bayesian Model Averaging Scheme

Monday, 15 December 2014
Shupeng Zhang1, Xiaogu Zheng1, Zhuoqi Chen1, Jingming Chen2, Guocan Wu1 and Xue Yi1, (1)Beijing Normal University, Beijing, China, (2)University of Toronto, Toronto, ON, Canada
Atmospheric CO2 abundance data can be used to constrain surface carbon fluxes and evaluate prediction skills of ecosystem models. In this study a multimodel carbon assimilation system is developed for assimilating atmospheric CO2 abundance data into three ecosystem models and exploiting the diversity of prediction skills of these models. The assimilation approach is based on a modified ensemble Kalman filter (EnKF) which estimates the inflation factor of the forecast error with a maximum likelihood function. The Bayesian model averaging scheme infers best predictions of ecosystem carbon fluxes by weighting individual predictions based on their probabilistic likelihood measurements. The proposed system was used to estimate the terrestrial ecosystem carbon fluxes from 2000 to 2008 and evaluate ecosystem models in different areas of the globe and at different times.