Abstract: Satellite data has been widely used as a non-conventional meteorological data in numerical weather prediction due to its global coverage and continuous, high-frequency observations.However, there are a variety of observation errors in satellite observations, including the random errors caused by the contingency of observations and systematic errors brought by the instrument itself and introduced by the Community Radiative Transfer Model, which influence the quality of satellite data to a large extent.A method of gradient information assimilation to correct the systematic errors of satellite data is proposed in this paper, which uses a gradient operator to make gradient transformation between the model variables and observation variables and realize the objective to eliminate the systematic errors.A numerical simulation of typhoon Kompasu is then conducted using the WRF (Weather Research Forecast) and WRFDA (WRF Data Assimilation system) model as well as the AIRS (Atmospheric Infrared Sounder) data.The results indicate that the method of gradient information assimilation can greatly improve the simulation of Kompasu's track, and the method can also be applied to process data of low reliability.In addition, through the assimilation diagnosis analysis, it is found that the systematic deviation of the satellite data has a great impact on numerical simulation of typhoons, and the gradient information assimilation method proposed in this paper can solve this kind of problem.