Ensemble Forecasts for Sub-seasonal to Seasonal Rainfall over the Economic Belt of the Northern Slope of Tianshan Mountains

Abstract

The economic belt of the northern slope of Tianshan Mountains (NSTM) has important social, economic and ecological effects in Xinjiang. Thus, it is critical to improve the prediction ability of sub-seasonal to seasonal rainfall in this region. Focusing on the large terrain of Tianshan Mountains, the global high-resolution climate prediction operational system version 3 developed by China Meteorological Administration (CMA-CPSv3) is applied. The sub-seasonal to seasonal rainfall over the NSTM is predicted by using the control run, the traditional ensemble mean, and the improved optimal deterministic ensemble forecast using a probabilistic threshold (DEFPT), respectively. The DEFPT method is not intended to predict the probability of rainfall, but to forecast the occurrence (yes or no) of rainfall event in any model grid box by judging whether it exceeds a certain probabilistic threshold, and the spatial-temporal distribution characteristic is analyzed. All the predictions are evaluated by the frequency bias, equitable threat score (ETS), Hanssen and Kuipers score (HK) and anomaly correlation coefficient (ACC). The evaluation results show that the improved DEFPT method can improve the sub-seasonal to seasonal predictions of the 1-5 mm rainfall locations and persistence over the NSTM and is superior to the traditional ensemble mean and the control run. These results also indicate that it is necessary to combine numerical model prediction with proper objective ensemble prediction method, especially in regions with large terrain background and significant climatology difference. Based upon the analyses of three rainfall events (during 29 July-2 August of 2016, 7-11 June of 2017, and 8-12 July of 2020, respectively), the DEFPT method performs better in western and southern NSTM from the perspectives of rainfall locations, anomalies and persistence. However, the prediction ability is relatively low in the eastern and northern parts of NSTM, which is possibly related to the low skill of humidity prediction from each ensemble member in corresponding regions. In addition, this method can also be used in other regions by tuning the related empirical coefficients in the formula to limit forecasting biases.

Keywords

• ensemble rainfall prediction