次季节-季节预报模式对华南前后汛期降水的预报技巧评估

Capability of Subseasonal-to-Seasonal Prediction Models in Forecasting Precipitation in South China During Rainy Seasons

关键词:  次季节-季节预报模式  华南  汛期  降水  预报技巧

Key words:  subseasonal-to-seasonal prediction models  South China  rainy season  precipitation  prediction skill

摘要:

利用次季节-季节(S2S)预报计划多模式回报数据集和国家气象观测站逐日降水数据，通过确定性指标评估了 CMA、ECMWF、NCEP、JMA、UKMO 这 5家 S2S 模式对华南前后汛期降水的预报技巧，并分析了不同类型降水事件的可预报性。各模式在表征降水强度与变率方面存在一定局限性，普遍高估(低估)华南西北部(东南部)降水强度，低估全区降水变率。在时间相关技巧方面，前后汛期大多模式可提前1～2候熟练预测(相关系数通过了 0.05 的显著性检验)大部分地区降水。根据平均方差技巧，模式可预报性较低，多数模式对大部分区域降水的熟练预报(平均方差技巧大于 0)仅在第 1候。ECMWF模式显示了最熟练、稳健的预报能力，基于时间相关与平均方差技巧所揭示的可预报性上限为 2～3候。在短预报时效(1候)下，后汛期模式预报技巧较高于前汛期，但随预报时效的增加，模式预报能力较前汛期下降更迅速。在 S2S 时间尺度，降水异常事件通常具有比平均事件更高的可预报性。

Abstract:

This study evaluated the deterministic skills of subseasonal-to-seasonal (S2S) prediction models, including the CMA, ECMWF, NCEP, JMA, and UKMO models, in predicting precipitation during the rainy seasons in South China. It analyzed the predictability of various types of precipitation events utilizing the S2S prediction multi-model reforecast dataset and daily precipitation data from national meteorological stations. The results revealed limitations in these models’ability to predict precipitation intensity and variability. Generally, these models tended to overestimate precipitation intensity in the northwestern part of South China while underestimating it in the southeastern part. Moreover, these models underestimated the overall precipitation variability in the region. In terms of temporal correlation, most models showed useful skills, with correlation coefficients statistically significant at the 95% confidence level, in predicting precipitation for most areas during the rainy seasons, with a lead time of 1-2 pentads. However, most of these models exhibited low predictability in accurately capturing the anomalous precipitation patterns (measured by mean square skill), and their useful forecast lead time (mean square skill > 0) was only 1 pentad for most regions. Among the models, ECMWF demonstrated the highest level of prediction capability, exhibiting strong and consistent performance as indicated by the temporal correlation and mean square skill, with a predictability upper limit of 2 to 3 pentads. For forecasts with a short lead time (1 pentad), these models exhibited higher prediction skills for the second rainy season. However, as the lead time increased, the models' skills declined more rapidly during the second rainy season. Anomalous precipitation events generally exhibited higher predictability than average events at the S2S timescale.