Calibration of QPF Using the Optimal Percentile Method

Compared to a QPF from a deterministic model, an ensemble based QPF provides extensive information, such as forecast uncertainty and probability of heavy rain. Calibration techniques of QPF are required to improve forecast reliability and accuracy. The techniques in operation commonly include the frequency-matching method (FMM) (Zhu et al., 2015) and the probability match method (PMM) (Elbert 2001). This study proposes a new calibration method of QPF, named as the optimal percentile method (OPM), which is applied to obtain the most likely QPF outcome with assistance of the FMM and the PMM. Firstly, based on the QPF and the corresponding observation in the same time period and the same forecast lead-time, the OPM is used to calculate the threat score (TS) and the Bias score (BS) for the different ensemble percentiles, select the percentile which can obtain the highest TS from the specified range of BS, and then apply this optimal percentile value into QPF. Secondly, the percentile forecast is further calibrated by the FMM, in order to eliminate the discontinuity of QPF. Finally, the spatial distribution of QPF is adjusted with the probability match method, in order to further reduce the position prediction error of the QPF.

In the context of 2011-2014 ECMWF ensemble based PQPF and corresponding in-site observation over China, the optimal percentiles for different forecast hours are calculated and then applied to the QPF of 2015 summer rainfall. The result shows that TS scores for the heavy rain of 12-36 hour forecast (≥50 mm / day) can be improved from 0.166 (best TS from deterministic forecast) to 0.206, which is also higher than the FMM and PMM alone. The improvement is significant for medium-range forecast of all thresholds, with TS increased by 47.4% for 132-156 hours lead-time for heavy rain. Additionally, the OPM has reduced the amount of false alarm as appropriate. The cases show that the OPM can improve the QPF in the following aspects: 1) to enhance the precipitation amount effectively, or to expand the heavy precipitation area when the uncertainty is large.; 2) to capture the possible occurrence of heavy rainfall events, and indicate the most likely location (especially for medium and long range forecasts); 3) to predict the strong precipitation area that is missing from the deterministic forecast; and 4) to provide valuable correction for heavy rainfall location.