Assimilation of radiance tendency observations from geostationary satellites in <scp>NCEP</scp> 's Global Forecast System

Abstract This study investigates an innovative approach to assimilating geostationary clear‐sky radiance observations using radiance tendencies, which represent the temporal variation of radiances. The tendencies were derived from geostationary radiance product pairs, and assimilation tests were conducted to explore this method. Data assimilation can be performed without the influence of bias in observations and models when the change in bias between observations and model predictions is sufficiently small over a given time interval. In this case, the assimilation process can focus on the change in observations during that period. For instance, the bias in surface emissivity estimation – a common issue when assimilating brightness temperature observations sensitive to land surfaces – can be effectively eliminated. Radiance tendency assimilation was demonstrated using a six‐hour cycling data assimilation system, with several scenarios explored: including only water vapor bands and/or near‐surface bands; (1) over ocean only, (2) over land only, and (3) combined land and ocean. The inclusion of water vapor bands from Advanced Baseline Imager (ABI) and Advanced Himawari Imager (AHI) increased biases and standard deviation of water‐vapor‐sensitive bands of other instruments and the bias in the moisture analysis field. When non‐water‐vapor‐sensitive ABI and AHI were assimilated, the assimilation system remained stable with reasonable adjustments made to the first‐guess fields. Reducing the thinning grid distance from 145 km to 60 km for radiance tendency observations led to better near‐surface forecast performance with ocean‐only observations compared to the combined land‐and‐ocean configuration. Mixed results were obtained for near‐surface temperature and wind vector root‐mean‐squared error (RMSE) (1000 hPa). The data assimilation system used in these experiments was not a coupled atmosphere–Earth system and therefore rejected the surface temperature increments. This limitation prevented observation information from propagating forward in time and caused inconsistencies in the boundary layer.