ABSTRACT Due to the insufficiencies of conventional observations and the effectiveness of bias corrections, it is still necessary to assimilate data from polar orbit satellites into a limited-area rapidly updated multiscale analysis and prediction system for Central Asia (RMAPS-CA). This study diagnoses the observation errors of the radiation brightness temperature (BT) of satellite Microwave Humidity Sounder (MHS) from multiple platforms through comparisons to the simulated BTs by the Community Radiative Transfer Model (CRTM) with soundings and a GFS input separately. The results show that the majority BIAS of the MHS BT compared with the simulated BT falls within −1.5–2.0 K and increases with an increase in the peak energy contribution height (PECH). No significant bias differences for the same channel of MHS and the same satellite are found across the different sounding stations, but obvious differences are found for the same channel from different satellites. Additionally, most of the GFS BTs are higher than the MHS BTs, which increase with the PECH. Moreover, almost all PDF distributions of the observed MHS BT bias relative to the GFS BT form a nearly Gaussian distribution, with the centre value being positive for channels 3 and 4. An assessment of the initial applications of MHS to RMAPS-CA is then conducted through two one-month retrospective runs, and forecasts initialized from analyses with and without MHS are verified against the observations. The results show that compared to the natural run for which no observations are assimilated, an improvement in the geopotential height of approximately 55.3% can be obtained by the assimilation of MHS, along with improvements of 12.8%, 4.7% and 2.3% for temperature, specific humidity (SPFH), and wind speeds below 200 hPa, respectively, for an average RMSE with initial time, which is valued at 2.9% and 3.3% for the SPFH at 2 m and wind speed at 10 m, respectively In addition, significant improvements in 24-h precipitation of 11.3%, 28.7% and 20.0% are obtained for light, medium and heavy rain, respectively, while no positive effects are obtained for rainstorm forecasts.