Estimation of regional surface CO_2 fluxes with GOSAT observations using two inverse modeling approaches

摘要

Inverse estimation of surface CO_2 fluxes is performed with atmospheric transport model using ground-based and GOSAT observations. The NIES-retrieved CO2 column mixing (X_(CO2)) and column averaging kernel are provided by GOSAT Level 2 product v. 2.0 and PPDF-DOAS method. Monthly mean CO_2 fluxes for 64 regions are estimated together with a global mean offset between GOSAT data and ground-based data. We used the fixed-lag Kalman filter to infer monthly fluxes for 42 sub-continental terrestrial regions and 22 oceanic basins. We estimate fluxes and compare results obtained by two inverse modeling approaches. In basic approach adopted in GOSAT Level 4 product v. 2.01, we use aggregation of the GOSAT observations into monthly mean over 5×5 degree grids, fluxes are estimated independently for each region, and NIES atmospheric transport model is used for forward simulation. In the alternative method, the model-observation misfit is estimated for each observation separately and fluxes are spatially correlated using EOF analysis of the simulated flux variability similar to geostatistical approach, while transport simulation is enhanced by coupling with a Lagrangian transport model Flexpart. Both methods use using the same set of prior fluxes and region maps. Daily net ecosystem exchange (NEE) is predicted by the Vegetation Integrative Simulator for Trace gases (VISIT) optimized to match seasonal cycle of the atmospheric CO_2. Monthly ocean-atmosphere CO_2 fluxes are produced with an ocean pCO2 data assimilation system. Biomass burning fluxes were provided by the Global Fire Emissions Database (GFED); and monthly fossil fuel CO2 emissions are estimated with ODIAC inventory. The results of analyzing one year of the GOSAT data suggest that when both GOSAT and ground-based data are used together, fluxes in tropical and other remote regions with lower associated uncertainties are obtained than in the analysis using only ground-based data. With version 2.0 of L2 X_(CO2) the fluxes appear reasonable for many regions and seasons, however there is a need for improving the L2 bias correction, data filtering and the inverse modeling method to reduce estimated flux anomalies visible in some areas. We also observe that application of spatial flux correlations with EOF-based approach reduces flux anomalies.