Optimizing all-sky infrared radiance assimilation with dynamic cloud-dependent error modeling

Satellite data assimilation is progressing beyond the conventional “clear-sky” approach towards the “all-sky” approach. While the former eliminates observations affected by clouds, the latter assimilates all observations including clear-sky, cloudy and precipitation conditions. The exploitation of cloud affected radiances is a promising endeavour as these observations are directly related to particularly challenging weather phenomena (e. g. convection, frontal systems, low stratus, and fog). This study focuses on the assimilation of clear and cloud affected (all-sky) radiances, from the 6.2 μm and 7.3 μm channel sensitive to water vapour in the upper troposphere using satellite data from SEVIRI, an instrument onboard Meteosat-10. The goal is to describe the improvements in short range forecasts in the high-resolution limited area Numerical Weather Prediction Model (NWP), AROME (Application of Research to Operations at MEsoscale) used at GeoSphere Austria. 3D-Var data assimilation experiments were performed to study the impact of all-sky vs clear-sky. A significant challenge is accurately representing observation errors, which are influenced by the complex and variable nature of clouds. This work implements an observation error model that dynamically adjusts error values based on cloud amount. The model addresses the increased uncertainties in cloud-dense regions by assigning higher observation errors, while clearer areas receive lower error values, in alignment with the need for spatially adaptive error characterization in all-sky conditions. Results demonstrate that the cloud-dependent error model leads to more Gaussian departures which can be expected to improve the assimilation of cloud-affected radiances, leading to better initial conditions and refined representations of atmospheric states and consequently the forecast.