Impact of Model Resolution in ACOS-xCO2 Observation Assimilation

Assimilation techniques are generally employed in two major applications, to fill the gaps in sparsely sampled observations or to improve the uncertainties in a model system. Both applications rely on estimating optimal states that minimize the total deviation between the observation and model forecast. In order to compute the deviation, the model forecast is transformed into observation space via an observation operator which represents sampling geometry, sounding profile, measurement sensitivity, and processing error of an observing system. Ideally, the observation space transformation implies the spatial and temporal resolution of the transformed model forecast matches those of the observation. However, current global chemistry-transport models (CTM) cannot deliver the required observation resolution of ACOS-XCO2 whose foorprint diameter ranges between 10 km and 50 km and sampling frequency is less than a minute. The quality of the assimilation degrades as the resolution mismatch increases, but the degrade level varies depending on the surface emssion property of the region, observation period, and assimilation duration. In this study, we present the impact of model resolution in ACOS-xCO2 observation assimilation based on two types of statistical analyses: 1) sample-level comparison between the bias-corrected V3.3 ACOS-xCO2 observations and the model forecasts in three spatial/temporal resolutions (2°x2.5° in 15 min, 1°x1.25° in 5 min, and 0.5°x0.625° in 2.5 min) over one year period (2010); 2) Evaluation of the model runs before and after assimilating V3.3 ACOS-xCO2 observations in the above three spatial/temporal resolutions compared to the HIPPO and TCCON data sets. The CO2 flux inventory developed by the NASA's Carbon Monitoring System-Flux (CMS-Flux) task and GEOS-5/MERRA meteorology fields developed by the NASA's Global Modeling and Assimilatoin Office (GMAO) were employed for the study.