The benefit of neighborhood methods for the assimilation of cloud‐affected radiances evaluated in observing‐system simulation experiments

Abstract Deficiencies in the representation of hydrometeors and subgrid‐scale motions in numerical weather prediction models cause systematic differences between satellite observations and their model equivalents, which pose a significant challenge for the assimilation of cloud‐affected radiances. In this study, we compare methods for mitigating the impact of systematic errors in the context of satellite data assimilation. Specifically, we consider thinning, superobbing, a novel technique referred to as cloud‐fraction assimilation, and their multiscale variants. To measure their respective benefits, we apply these methods to the assimilation of 7.3 m infrared and 0.6 m visible satellite radiances in an observing system simulation experiment (OSSE). The OSSE is based on the Weather Research and Forecasting model and the ensemble adjustment Kalman filter. A 250‐m resolution simulation of weakly organized deep moist convection serves as nature run, and assimilation experiments are conducted with a 2 km resolution. Our results show that superobbing and cloud‐fraction assimilation clearly outperformed thinning. The benefit of cloud‐fraction assimilation varied over time, which suggests that the parameters of the assimilation algorithm can be further optimized for this technique. However, the impact of cloud‐fraction assimilation was remarkably large, considering its reduced information content compared with radiance fields. Cloud fractions may therefore be a viable approach to mitigate the effect of complex biases and structural errors in real‐world model simulations.