Abstract: The goal of time downscaling for grid precipitation forecast is to downscale the 24 h refined grid precipitation forecast with high accuracy to a finer time resolution, so as to ensure the accuracy and the consistency of total amount of refined precipitation forecast at different time intervals. At present, the single point precipitation time series of numerical model forecast is used as the proportion to downscale the forecast aimed at every one grid. In order to solve the problems in existing methods in the time downscaling procedure, such as too large rain range, too small intensity and unreasonable moving speed, the position correction, dynamic reconstruction and frequency matching algorithms are added to improve the effect of time-downscaling. Based on the evolution of ECMWF model forecasts, taking the time downscaling of 24 h grid precipitation forecast field produced by National Meteorological Centre at 20∶00 BT on 18 July 2020 as an example, the function of different algorithm steps are illustrated. Furthermore, the daily 24 h grid precipitation forecasts starting from 08∶00 BT between 1 January and 31 December 2020 are selected for batch test of time downscaling. Case analysis and batch test results show that the new added algorithm steps can improve the rationality and accuracy of hourly grid precipitation prediction. Among them, the position correction is used to correct the position from the numerical model forecast to the grid forecast, the dynamic reconstruction is used to reduce the unreasonable fluctuation of the movement and intensity of the downscaled rain belt, and the frequency matching is used to correct the problem of the larger range and weaker intensity of the downscaled rain belt. The ETS and BIAS scores of improved forecasts are improved, especially for the short-term heavy rainfall above 20 mm.