Abstract: Many assimilated observations can effectively improve the results of a model forecast. However, there are significant differences in the effects of various observations on the forecast. It is one of the most challenging diagnostics in numerical models to reasonably evaluate the observation contribution to the forecast. In this paper, the weather research and forecasting model’s data assimilation (WRFDA) and forecast sensitivity to observation (FSO) system was constructed in WRFDA by the method of adjoint-based FSO. Based on wind profile radar (WPR) and ground-based microwave radiometer (MWR) data obtained by the mega city project in Beijing in September 2019, the experiments on the impact of observations on the 12 h forecast of the WRF model are carried out using the WRFDA-FSO system. The contribution of wind, temperature, and humidity observations to the forecast is analyzed. The results show the following: (1) In general, the assimilated observations (MWR, WPR, Sound, Synop, and Geoamv) reduce the 12 h forecast error of the WRF model and make a positive contribution to the forecast. Among these, MWR observations have the greatest impact on the forecast, and the improvement of WPR observations on the forecast is better than that of wind field observations of sound. (2) Among the U and V observations of WPR and temperature and specific humidity observations of MWR, the positive contribution value of V observations and temperature observations to the forecast is higher, and the effect of improving the forecast is better. (3) The WPR and MWR observations, at most levels, reduce the forecast error and are a positive contribution to the forecast. The positive contribution of temperature observations is mainly below 800 hPa near the ground.