Revisit the Observation Errors for All-Sky Microwave Radiance Observations

Although many all-sky satellite microwave radiance observations have already been assimilated in operational models, there remains large uncertainty in their observation errors. For example, in GFS, the observation errors for all-sky AMSU-A and ATMS channels are assigned values much larger than their estimated errors (e.g., can be 2-3 times), even though the estimates already account for the representation errors. The additional inflation of the observation errors is empirical and requires rigorous tuning (hereafter, empirical inflation). While the empirical inflation is a practical approach to address other inappropriate assumptions within DA systems, it comes at the expense of under-utilizing these observations.

In this work, we will closely examine the impact of the observation errors for all-sky microwave observations on the analysis and forecast using the Geophysical Fluid Dynamics Laboratory (GFDL) System for High-resolution prediction on Earth-to-Local Domains (SHiELD) model. SHiELD is coupled with the GSI-based hybrid 4DEnVar, which is similar to the operational GFS. We found that turning off the empirical inflation leads to mixed results in analysis and forecasts in SHiELD. In particular, turning off the empirical inflation exacerbates the near-surface cold and dry biases in SHiELD. We found that this is due to an interaction between the empirical inflation and the variational bias correction (varBC). We will investigate the model and observational biases by refining the existing varBC. Based on the refined varBC, we will also examine the effect of replacing the empirical inflation with an objective and non-parametric observation error method, the evolving-Gaussian method, on the performance of DA. Our final goal is to find a way to better extract the information from the all-sky radiances observations.