Abstract:Based on the precipitation forecast dataset from the European Centre for MediumRange Weather Forecasts (ECMWF) and the daily accumulated precipitation of 2403 national surface meteorological observation stations across China, the calibration on daily precipitation by means of the categorized rainfall regression and the further calibration by means of the frequencymatching method are conducted. The results show that compared with the bilinear method, the categorized rainfall regression is more effective in decreasing the forecast biases, and improves the correlation coefficient with the observed data and the equitable threat score. The forecasts of different thresholds become more accurate after applying the frequencymatching method, with the smaller area deviation. The false alarm rate of light rain and the missing rate of heavy rain are also both reduced. Improvements of the forecasts by the categorized rainfall regression and the frequencymatching method differ in initialized times, rainfall thresholds and lead times. After the calibration of the categorized rainfall regression, the forecast initialized at 20:00 is improved with a larger magnitude than that at 08:00. The improvement of the forecast is relatively limited for rainfall thresholds of 01 mm and 50 mm, but significant for thresholds of 5 mm, 10 mm and 15 mm. Additionally, the amplitude of the improvement increases slightly over the lead time. The improvement induced by the frequencymatching method is greater for precipitation forecasts initialized at specific times that show worse performances.