西北太平洋热带气旋频次的延伸期动力-统计预报方法和评估

介绍了西北太平洋热带气旋(TC)频次的延伸期预报方法,比较了新构建的动力-统计和统计预报模型的预测技巧,并探讨了预报误差来源及改进方向。动力-统计预报模型是基于动力模式预测的热带季节内振荡(ISO)信号及ISO-TC生成的同期统计关系来进行预报;统计预报模型则是基于TC生成的前兆ISO信号建模预报。预报评估结果显示,动力-统计混合预报模型的预报技巧高于统计预报模型,原因在于影响TC次季节变化的前兆信号并不稳定,且随着预报超前时间迅速消散,无法提供有效且稳定的可预报源;相反地,TC生成与同期的ISO背景场显著相关,动力模式对ISO(预报因子)有较好的预报能力,因此动力-统计相结合的预报方法为TC延伸期预报提供了有效途径。虽然目前动力-统计预报模型的预报技巧可达5～6周,但仍有进一步改进和提高的空间。通过对不同类型TC预报技巧检验和误差分析,研究认为年际和年代际背景场对ISO调控TC活动的影响不可忽略,且热带外ISO信号(如罗斯贝波破碎和西风急流强度等)对TC频次和轨迹也有显著影响,这些因子为TC延伸期预报提供了潜在可预报源。

A hybrid dynamic-statistical prediction model for tropical cyclone frequency over the western North Pacific and its evaluation

Prediction of tropical cyclone (TC) genesis at the extended-range to subseasonal timescale (a week to several weeks) is a gap between weather and climate predictions,which is a challenge for TC forecast.This study presents an extended-range hybrid dynamical-statistical prediction model and a statistical prediction model for TC frequency over the western North Pacific.The models are based on tropical intraseasonal oscillation signals and the TC clustering method.The fuzzy c-mean clustering method categorizes TCs over the western North Pacific into seven track patterns.Predicting anomalous TC counts in each week involves adding the observed climatological mean of weekly TC counts to obtain total genesis counts for each cluster.The probability of TC track distributions each week is derived by involving the climatology of each track probability.This model could not only predict TC number for each cluster but also the TC track distribution pattern each week.The hybrid dynamical-statistical model relies on contemporaneous statistical relationships between low-frequency variabilities and the output of the ECMWF dynamical model from the S2S dataset.The predictand is the TC genesis number over the western North Pacific during each week.Evaluation of prediction results indicates that the forecast skill of the hybrid dynamic-statistical forecast surpasses that of the statistical forecast model.The precursor signals associated with sub-seasonal TC changes dissipate rapidly,making stable forecasts challenging.In contrast,the dynamic model simulates the low-frequency background field (predictors) effectively,enhancing the hybrid model''s forecast skill.While,the current forecast skill of the hybrid dynamic-statistical forecast model extends to six weeks,further improvement is possible.Evaluation of prediction skills and error analysis of different TC clusters reveal that interannual and interdecadal variabilities of background fields on the modulations of intraseasonal oscillations on TC activity cannot be ignored.Statistical relationships between TC counts and low-frequency variabilities differ in distinct ENSO phases,suggesting potential improvement by developing forecast models based on different ENSO phases.Additionally,extratropical intraseasonal signals (e.g.,Rossby wave breaking and westerly jet intensity) significantly impact TC frequency and trajectory,which may provide more source of predictability for TC extended-range prediction.