中央氣象局全球預報模式對不同BSISO相位下臺灣梅雨季降雨特徵之預報能力評析：以2016－2017年為例

By comparing the rainfall and circulation forecasted by Central Weather Bureau Global Forecast System (i.e. CWB/GFS) with the observational data over East Asia, this study evaluates the capability of CWB/GFS in forecasting the impact of two different types of boreal summer intraseasonal oscillations (named as BSISO1 with a 30-to-60-day oscillation period and BSISO2 with a 10-to-30-day oscillation period) on the Mei-yu season (May and June) rainfall over Taiwan during 2016-2017. For the model forecast, we focus on the analysis of lead times from day-1 to day-15 (denoted as LT1~15). For the observational data, we use rain gauge observation, GPM (Global Precipitation Measurement) IMERG (Integrated Multi-satellite Retrieval for GPM) precipitation data, and NCEP (National Centers for Environmental Prediction) Reanalysis version 2. Our analyses of observational data show that the characteristics of Mei-yu season rainfall in Taiwan during 2016-2017 are under the modulation of atmospheric circulation changes over East Asia related to BSISOs. In general, heavy rainfall events tend to occur over Taiwan at phase 7-1 of BSISO1 and at phase 4-6 of BSISO2, when the signal of large-scale convective zone propagates into the areas nearby Taiwan, Southeast China and Okinawa. By comparing these observational features with the features forecasted by CWB/GFS, it is noted that CWB/GFS can depict the relationship between the occurrence timing of heavy rainfall (＞30mm/day) in Taiwan and the phases of BSISOs (i.e. heavy rainfall mostly occurred at phase 7-1 of BSISO1 and at phase 4-6 of BSISO2). Also, based on the analysis of spatial correlation (Scorr) and root-mean-square-error (RMSE) between CWB/GFS and observation, we note that LT1~5 has the largest Scorr and smallest RMSE (i.e. the best forecasting performance), LT11~15 has the smallest Scorr and largest RMSE (i.e. the worst forecasting performance), whereas the forecasting performance of LT6~10 is between them. However, quantitatively, the CWB/GFS rainfall forecasts tend to ＂underestimate the heavy Mei-yu rainfall events＂ and ＂overestimate the weak Mei-yu rainfall events＂ in Taiwan; this feature is found in all of the forecast results of LT1~5, LT6~10 and LT11~15. Further examination on ＂Why CWB/GFS is able to capture the relationship between the occurrence timing of heavy rainfall in Taiwan and the phases of BSISOs?＂, our results show that is related to the good skill of CWB/GFS in forecasting the propagating features of rainfall and circulation over the areas nearby Taiwan, under the modulation of BSISOs. These results help a better understanding of the values of applying CWB/GFS in rainfall forecast over Taiwan. Notably, this is a pilot study, which uses 2016-2017 CWB/GFS forecast data as an example, to examine the capability of CWB/GFS in forecasting the rainfall changes in Taiwan under the modulation of BSISOs. Further research work is proposed to examine whether above results are suitable for other years when more CWB/GFS forecast data are available in the future.