Abstract: According to the characteristics of numerical GMS infrared imagery the improved statistic retrievalmethod is applied to estim the temperature and humidity. Several schemes are developed to control the quality of retrieved temperature and humidity in order to correct the errors from the image and the retrieval scheme. The variational assimilation is performed of the corrected retrieval field with the conventional observations followed by a control expenment by the MM4 model. Results show that the quality control is critical to the forecasting based onthe rationaldesigned function. The correction with the grid values is better than that of the conventional observation, and with the systematic error correction to the retrieval field included the accuracy of rainfall forecasting is improved. It is worth notice that the scheme with satellite image included can obviously improve the intensity and region forecasting of model rainfall and the prediction is more close to measured precipitation as compared with that only using the convendonal observations.