基于数据同化的哈佛森林地区水、碳通量模拟

模型模拟和站点观测是陆地生态系统水、碳循环研究最主要的两种手段，但各有优势和不足，若二者相互结合，则能更准确地反映生态系统水、碳通量的动态变化.数据同化为模型与观测结合提供了一条有效的途径.本文采用哈佛森林环境监测站相关数据，利用集合卡曼滤波同化算法，将实测叶面积指数(LAI)和遥感LAI同化进入Biome BGC模型中，对该地区水、碳通量进行模拟.结果表明：与未同化模拟相比，将1998、1999和2006年实测LAI数据同化后，模型模拟碳通量(NEE)与通量观测NEE的决定系数(R²)平均提升8.4%；蒸散发(ET)的R²平均提升10.6%；NEE的绝对误差和(SAE)和均方根误差(RMSE)平均下降17.7%和21.2%，ET的SAE和RMSE平均下降26.8%和28.3%.将2000-2004年MODIS LAI 产品与模型同化后，NEE、ET模拟值与观测值间的R²分别提升7.8%和4.7%；NEE的SAE和 RMSE分别下降21.9%和26.3%，ET的SAE和 RMSE分别下降24.5%和25.5%.无论实测LAI还是遥感观测LAI，同化进入模型都能不同程度地提高水碳通量的模拟精度.

Simulation of water and carbon fluxes in Harvard forest area based on data assimilation method.

Model simulation and in situ observation are the two most important means in studying the water and carbon cycles of terrestrial ecosystems, but have their own advantages and shortcomings. To combine these two means would help to reflect the dynamic changes of ecosystem water and carbon fluxes more accurately. Data assimilation provides an effective way to integrate the model simulation and in situ observation. Based on the observation data from the Harvard Forest Environmental Monitoring Site (EMS), and by using ensemble Kalman Filter algorithm, this paper assimilated the field measured LAI and remote sensing LAI into the Biome-BGC model to simulate the water and carbon fluxes in Harvard forest area. As compared with the original model simulated without data assimilation, the improved Biome BGC model with the assimilation of the field measured LAI in 1998, 1999, and 2006 increased the coefficient of determination R2 between model simulation and flux observation for the net ecosystem exchange (NEE) and evapotranspiration by 8.4% and 10.6%, decreased the sum of absolute error (SAE) and root mean square error (RMSE) of NEE by 17.7% and 21.2%, and decreased the SAE and RMSE of the evapotranspiration by 26.8% and 28.3%, respectively. After assimilated the MODIS LAI products of 2000-2004 into the improved Biome-BGC model, the R2 between simulated and observed results of NEE and evapotranspiration was increased by 7.8% and 4.7%, the SAE and RMSE of NEE were decreased by 21.9% and 26.3%, and the SAE and RMSE of evapotranspiration were decreased by 24.5% and 25.5%, respectively. It was suggested that the simulation accuracy of ecosystem water and carbon fluxes could be effectively improved if the field measured LAI or remote sensing LAI was integrated into the model.