Based on the GRAPES-RAFS with 3km resolution in space and 3h cycle in time, the hourly assimilation and model forecast cycle system, in which the background error correlation model is improved with superposition of five scales Gaussian components and anisotropicShorizontal correlation scales and the effect of global large-scale information to hourly analysis and forecast is examined, has been built. The numerical simulations of Gaoyou Tornado case of 12 June 2020 indicate that: (1) The hourly cycle qualities of analyses and forecasts which have assimilated more high frequent observations and allow 1h forecast as background field are obviously improved; (2) Excessive use of long-length global large-scale forecast information can reduce analysis and forecast quality; (3) Since the improved five-scale superimposed Gaussian correlation model and the anisotropic horizontal correlation scale in the 1h cycle make the background error horizontal correlation coefficient closer to the statistical results of the sample, it can more accurately simulate the mesocyclone structure of the Gaoyou tornado while the quality of model variables and precipitation forecasts is closer to the real observation.