The Study Of Data Assimilation Between Biome-BGC Ecological Process Model And Remote Sensing Observed LAI Under Three Typical Vegatation Types

Accurate estimation of water carbon flux has significant implications for terrestrial water carbon cycling,but it is also challenging.The current estimation accuracy needs further improvement.Traditional model simulations and site observations have their own advantages and disadvantages.The two must be combined.Data assimilation incorporates observations into models based on physics laws to obtain the best estimate of the model state variables as much as possible,providing an effective way for the combination of models and observations.In this study,based on the Enkaf Kalman Filter(EnKF)and Unscented Kalman Filter(UKF)data assimilation methods,the LAI(Leaf Area Index)assimilation of remote sensing observations was entered into the Biome-BGC ecological process model for evapotranspiration(ET)and Carbon flux(NEE)was simulated.In this paper,the measured data from the ground flux site were used as verification data.Based on the verification data,the two assimilation simulation results were compared with the initial simulation results of the model and the simulation results after the optimization of the parameters for analyzing.The main results are as follows:(1)Under the three vegetation types,Using the two data assimilation methods of EnKF and UKF to assimilate LAI into the Biome-BGC model,which can make the simulated LAI trajectory closer to the observed value,the simulation accuracy of the model for ET and NEE is improved,followed by deciduous broad-leaved forest,and finally grassland vegetation type.(2)Among the three vegetation types,both two data assimilation methods reach their best performance in evergreen broad-leaved forest,in this vegetation type,both the stimulated accuracy of ET and NEE are improved obviously.This is followed by deciduous broad-leaved forest,and finally grassland vegetation type(3)The EnKF assimilation method reach the best assimilative performance in the evergreen broad-leaved forest.The RMSE and NEE of the simulated ET decreased by 17.16% and 13.39% respectively,and the errors of the NEE decreased by 20.74% and 22.79% respectively.However,the performance of EnKF under the grassland vegetation type was poor and the error increased.The UKF assimilation method also perform best under the evergreen broad-leaved forest,and the overall error of the simulated variables improved the most.The UKF assimilation method performed the worst under the grassland vegetation type.(4)Under the evergreen broad-leaved forest,the two methods of EnKF and UKF get better performance in improving the accuracy of NEE simulation,and are better than those of simulated ET.In the deciduous broad-leaved forest,similarly,the two assimilation methods get better performace in improving the accuracy of NEE.However,in the grassland vegetation type,the two assimilation methods are better to improve the accuracy of ET simulation.(5)The most obvious improvement of ET simulation accuracy is based on the En KF method at the Dangxiong site in 2004.The worst effect of the improvement is based on EnKF method at the Changbaishan site in 2005.Correspondingly,the most obvious improvement in NEE simulation accuracy is based on UKF method at the Qianyanzhou site in 2005,and the worst effect of NEE assimilation was based on the EnKF assimilation treatment at the Haibei site in 2003.The data assimilation technology can establish a bridge between the model and the observation,as a result,the model simulation results can carry the observation and model information at the same time,and then improve the accuracy of the model's estimation of the ecosystem carbon-carbon flux.Meanwhile,the data assimilation technology also provide a wide expansion for remote sensing data deep-seated data mining and application.