Abstract Global climate models (GCMs) produce informative seasonal forecasts of global precipitation months ahead of the occurrence for hydrological forecasting. Meanwhile, the skill of GCM forecasts varies by location and initialization time. In this paper, we investigate the anomaly correlation, which indicates the correspondence between forecasts and observations, for 10 sets of global precipitation forecasts in the North American Multi-Model Ensemble (NMME) project. We propose to use principal component analysis to characterize the variation of anomaly correlation. We identify the existence of spatial and temporal patterns at the global scale. The spatial pattern reveals that high (low) anomaly correlation at one initialization time coincides with high (low) anomaly correlation at other initialization times. In other words, for a grid cell, the anomaly correlation at different initialization times tends to be similarly high, or low. It is observed that some of the regions where grid cells are with overall high anomaly correlation tend to exhibit tele-connections with global climate drivers. On the other hand, the temporal pattern suggests that the anomaly correlation tends to improve with initialization time. This pattern is attributable to data assimilation that bases forecasts at a later initialization time on more global observations and simulations. Generally, the two patterns are effective and explain 50% to 70% of the variation of anomaly correlation for the 10 sets of NMME forecasts. The projections of anomaly correlation vectors onto the two patterns help illustrate where and when the NMME precipitation forecasts are skillful.

中文翻译：

---

NMME 全球降水季节预报的预测性能：时空视角

摘要 全球气候模型 (GCM) 在水文预报发生之前的几个月就产生了全球降水量的信息性季节预报。同时，GCM 预测的技巧因位置和初始化时间而异。在本文中，我们研究了北美多模式集合 (NMME) 项目中 10 组全球降水预报的异常相关性，表明预报与观测之间的对应关系。我们建议使用主成分分析来表征异常相关性的变化。我们确定了全球范围内空间和时间模式的存在。空间模式表明，一个初始化时间的高（低）异常相关性与其他初始化时间的高（低）异常相关性一致。换句话说，对于网格单元，不同初始化时间的异常相关性往往同样高或低。观察到，一些网格单元整体异常相关性高的区域往往表现出与全球气候驱动因素的遥相关。另一方面，时间模式表明异常相关性倾向于随着初始化时间而改善。这种模式归因于数据同化，该数据同化在以后的初始化时间基于更多的全球观测和模拟进行预测。一般来说，这两种模式是有效的，并且解释了 10 组 NMME 预测的异常相关性变化的 50% 到 70%。异常相关向量在这两种模式上的投影有助于说明 NMME 降水预测在何时何地是熟练的。 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .