Evaluation of global ensemble prediction models for forecasting medium to heavy precipitations

In this study, 24-h forecasts of CMA, ECCC, ECMWF, KMA, NCEP and UKMO models were extracted from the TIGGE database and evaluated over selected stations in Iran within the 2010-2018 period. Daily forecast data were interpolated using the inverse distance method to the location of selected synoptic stations, while the frequency bias were further corrected using quantile mapping. The accumulation interval for precipitation was 24 h. Moreover, both raw and frequency-bias-corrected data were evaluated in deterministic and dichotomous modes at precipitation depth thresholds of 5, 15, 25, 35, and 45 mm. Forecasts were further partitioned and evaluated in different seasons in selected stations located in the margin of Alborz and Zagros Mountains. Best performing precipitation models did well over October-December season along the Alborz Mountains and over April-June along the Zagros Mountains. Results indicated that the forecasts were quite improved especially in high threshold after the frequency bias correction. Compared to other models, the ECMWF model provided best results in most stations. In contrast, NCEP performed poorly and could not provide a reliable medium to heavy precipitation forecasts. In general, none of the forecast models provided accurate estimates for precipitation depths over the 35-mm threshold. In fact, at higher precipitation thresholds, all models underpredicted the number of precipitation events.

Automated summary

This study evaluated 24-h forecasts of various meteorological models over selected stations in Iran, with ECMWF performing best after frequency bias correction and NCEP showing poor performance. None of the models accurately estimated precipitation depths over the 35-mm threshold, with all models underpredicting precipitation events at higher thresholds.