Satellite all-sky infrared brightness temperature (IRBT) observations are now assimilated in operational forecasting systems like the High-Resolution Rapid Refresh (HRRR) system. One plausible way to improve the impacts of assimilating all-sky IRBT in ensemble-based DA systems is to explicitly handle the non-Gaussian statistics often seen in the simulated IRBT. Ensemble-simulated IRBTs are often non-Gaussian because IRBT values can change dramatically depending on the presence or absence of clouds. Separate treatment of ensemble members with clouds and without clouds could potentially address this non-Gaussianity.

In a recent publication, we proposed handling the clear and cloudy members separately using an efficient bi-Gaussian extension of the classic ensemble Kalman Filter (EnKF). Our Bi-Gaussian EnKF (BGEnKF) differs from the original EnKF in two ways. First, our BGEnKF employs two separate linear updates: one to update clear members using the statistics of clear members, and another to update cloudy members using the statistics of cloudy members. In contrast, the EnKF only applies a single linear update to all members using the statistics of all members. Second, our BGEnKF also updates the proportion of clear and cloudy members to reflect the analysis ensemble’s likelihood in the two types of cloud scenes. For instance, the assimilation of a cloudy IRBT observation often causes the BGEnKF to increase the number of cloudy members and decrease the number of clear members. This update to the proportion of clear and cloudy members is not explicitly performed by the EnKF.

In this talk, we will use observing system simulation experiments (OSSEs) over the equatorial Indian Ocean to compare the impacts of assimilating all-sky IRBT using the EnKF and the BGEnKF. IRBT observations from the Meteosat Visible and Infrared Imager (MVIRI) infrared window channel will be assimilated.