We developed an ensemble data assimilation system with a dynamical global vegetation model known as the SEIB-DGVM (Spatially Explicit Individual Base Dynamic Global Vegetation Model). The model is an individual-based model, and it includes non-Gaussian and non-linear processes. In addition, the phase space changes due to the occasional establishment and death of the plants. Therefore, we used one of the particle filter methods, the Sampling Importance Resampling (SIR), which can handle these difficulties. The experiment was performed at the Larch forest of Eastern Siberia, and the state variables and parameters of the deciduous needle leaved tree and grass were estimated.

First, an Observing System Simulation Experiment (OSSE) was performed. We simulated LAI observations from the nature run with the “true” phenological parameters: maximum photosynthesis rate and dormancy date. By assimilating the simulated LAI observations, the parameters and other dynamical variables such as Gross Primary Product (GPP), Ecosystem Respiration (RE), and Net Ecosystem CO2 Exchange (NEE) were estimated successfully. Next, we assimilated the real-world MODIS LAI data. The results showed generally good match with the FLUXNET field observation data for NEE.