抄録
Data assimilation, which minimizes the difference between modeled and observed statuses statistically along with objective criteria, has been applied frequently for terrestrial biosphere models in recent years. This approach enables highly accurate estimation, gap-filling of missing or sparsely distributed data, optimizing model parameters and initial state. The Markov chain Monte Carlo method is a useful data assimilation approach because of its relative simplicity in program cording. However, this approach has been applied mainly for plot-scale simulation, which does not have any lateral interaction with other distant plots and does not require a large computational load. Ensemble Kalman filter, which has advantages in the simplicity of coding and the affordability of computational load despite the unavoidable reduction in the numbers of assimilated data points and optimized parameters, has been demonstrated recently in many field sites. The adjoint (4D-VAR) method enables us to estimate the uncertainty of parameters chronologically forward and backward with moderate computational load, but we also need to find a way to avoid the problems induced by the non-linearity of terrestrial biosphere models. In addition to the expectation for high accuracy simulations, those data assimilation approaches would also offer great potential for producing useful bi-products, e.g. objective analysis data sets, and supplemental evidence of the changes in physiological and ecological responses to environmental variations by looking at the chronological variation in their responses. While much greater effort should be made to overcome the variety of operational problems in those data assimilations, greater attention has been paid to improving terrestrial biosphere modeling through multi-disciplinary research fields, including ecological, climatological and global environmental studies.
