Impact of Observation Errors on All‐Sky Satellite Microwave Data Assimilation. Part I: Interplay Between Observation Error and Variational Bias Correction (VarBC)

Abstract Careful specification of observation errors is crucial for assimilating all‐sky satellite microwave radiances, although uncertainties in these errors remain substantial. Operational centers often inflate observation errors to optimize model performance. In Part I of this study, we investigate the impact of turning off an empirical inflation for all‐sky Advanced Microwave Sounding Unit‐A and Advanced Technology Microwave Sounder radiances in the Gridpoint Statistical Interpolation (GSI) data assimilation system using the Geophysical Fluid Dynamics Laboratory System for High‐resolution prediction on Earth‐to‐Local Domains model. Contrary to expectations, increasing the weight of all‐sky window channel observations exacerbated existing low‐level dry biases in analysis and degraded the subsequent forecasts. Single‐DA‐cycle experiments reveal that this paradox arises from an unintended shift in the bias estimated by variational bias correction (VarBC). This shift is driven by asymmetric sampling of cloud differences and a lenient data control strategy in VarBC, which allows samples with large background errors to influence VarBC coefficients. Our findings highlight the need for caution when calibrating the observation errors in a DA system using VarBC, as such adjustments can unintentionally alter the observation bias estimated by VarBC. The influence of observation bias changes on the analysis and forecast can be dominant. Our results raise concerns about the current VarBC data control strategy in GSI, which can make VarBC susceptible to potential model errors, especially in regions where anchor observations are limited. These motivate revised VarBC strategies to better separate model and observation biases in all‐sky radiance assimilation which will be presented in Part II.