Sensitivity of Estimated Surface CO2 Flux on Parameters Used in a Carbon Data Assimilation System

Abstract:

The estimation of surface CO₂ fluxes is important to understand the sources and sinks of near surface atmospheric CO₂. CarbonTracker, an inverse modeling system based on ensemble Kalman filter (EnKF), estimates the surface CO₂ flux using the atmospheric CO₂ concentrations as a constraint. While application of CarbonTracker for Asian region needs the optimization of EnKF parameters (e.g., ensemble size, assimilation window length, etc), appropriate parameters for Asian region have never been investigated. Therefore, in this study, several experiments using different parameters were conducted to find the optimal parameters of EnKF in CarbonTracker for Asian region. The assimilation window lengths tested are 3, 5, and 10 weeks, and the ensemble sizes tested are 100, 150, and 300. The experimental period is from 2006 to 2008.

In 2007, the optimized CO₂ flux uptake was greater than the prior CO₂ flux uptake. For each experiment, the spatial distributions of the optimized surface CO₂ flux varied seasonally. Differences between the optimized surface CO₂ fluxes of the experiments were the largest in Eurasia Boreal (EB), followed by Eurasia Temperate (ET) and Tropical Asia (TA), which is especially notable in summer. The optimized surface CO₂ flux in EB was more sensitive to the window length than the ensemble size, whereas the optimized surface CO₂ flux in ET was more sensitive to the ensemble size than the window length. In TA with few observation sites, differences between the optimized surface CO₂ fluxes of the experiments were small, which implies that determining optimal parameters in TA may be inappropriate given the current situation with few observation sites. The root mean square error (RMSE) and bias of the model CO₂ concentrations calculated by the optimized surface CO₂ flux in Asia were more sensitive to the ensemble size than the window length. More comprehensive results will be presented in the meeting.