To assimilate the all‐sky infrared radiances of Himawari‐8/AHI (Advanced Himawari Imager), we investigated the statistical characteristics of a radiance simulation from a mesoscale cloud‐resolving model. Compared with AHI observations, the simulations tended to show insufficiently low brightness temperature (BT), probably because of a lack of high‐cloud prediction and insufficient cloud‐scattering in the radiative transfer (RT) calculations. For effective assimilation, quality control (QC) procedures were developed to eliminate samples in which high inhomogeneities were identified, observed BTs were very low, and simulated (background) radiances deviated substantially from observations. The probability distribution function (PDF) of observation‐minus‐background (O–B) was Gaussian for the QC‐passed samples at the humidity bands (6.2 and 6.9 µm) of the AHI. This evaluation was implemented for four representative meteorological conditions around Japan to ensure robustness. The PDF at the window band (10.4 µm), however, was not close to Gaussian, probably because of the significant negative bias even after QCs. The PDF at a low‐tropospheric temperature band (13.3 µm) was Gaussian for three out of four meteorological conditions.
The relationship of the removal of inhomogeneous scenes to the O–B statistics and BT distribution was investigated quantitatively. The stricter threshold of inhomogeneity reduced O–B difference but also decreased the percentage of low BT, resulting in the possibility of excessive removal of high‐altitude cloud samples. This relationship was enhanced for larger samples that included more diverse cloud scenes. Finally, comparison of simulations at various resolutions with two RT models revealed that the reproducibility of low BT was more affected by using a different RT model than by using higher‐resolution systems.