Improving convection‐permitting all‐sky infrared radiance ensemble data assimilation through mitigating deleterious non‐Gaussian artifacts

Abstract Ensemble Kalman filters (EnKFs) with quality control (QC) are frequently used to assimilate satellite‐sensed all‐sky infrared radiance (IR) observations. EnKFs assume Gaussian forecast statistics, and QC methods frequently assume symmetric innovation statistics. That QC assumption is violated when the innovation distribution is skewed (i.e., non‐Gaussian). Unfortunately, both assumptions are violated in all‐sky IR data assimilation (DA). These violations likely generate statistical artifacts in the analysis ensemble. Such artifacts may limit the benefits of EnKFs‐based all‐sky IR DA. However, it is unclear what those artifacts are and how they limit the performance of EnKFs‐based all‐sky IR brightness temperature DA. This study addresses that knowledge gap using perfect model observing system simulation experiments (OSSEs) of EnKFs‐based all‐sky IR DA. Two artifacts resulting from those assumption violations are identified: (a) the creation of clouds in subsaturated humidity conditions and (b) a general depletion of specific humidity. The first artifact results from the EnKFs' Gaussian forecast assumption, and the latter arises from applying gross check QC on left‐skewed IR innovation distributions. As is demonstrated here, these artifacts cause severe cold temperature biases that annihilate the beneficial impacts of all‐sky IR DA. This study treats those artifacts by (a) disabling the hydrometeor analysis increments and (b) relaxing the criteria used in QC. Observing system simulation experiments demonstrate that these treatments noticeably improve the performance of all‐sky IR DA. The findings from this study not only advances the community's understanding of how non‐Gaussian statistics degrade the performance of DA but also has the potential to improve the performance of EnKFs‐based all‐sky IR DA under more realistic situations. That potential deserves future investigation.