Monsoonal precipitation is dominated by intraseasonal variabilities, whose skillful prediction lead time is currently less than 5 days and remains a grand challenge. Here we show that an intrinsic variability in the Indian Ocean, the Central Indian Ocean (CIO) mode, when combined with a machine learning (ML) algorithm, can produce skillful predictions of intraseasonal precipitation over the monsoon region with a lead time of over 15 days, which is close to the theoretical predictability limit. This remarkable skill improvement stems from the fact that the CIO mode is dynamically related to the intraseasonal monsoon rainfall, while the data-driven ML algorithm suppresses unwanted high-frequency noise. Using the CIO mode and the ML algorithm, the forecast system hybridizes physical fundamentals and versatility of data-driven algorithms. The identification of CIO mode and the verification of its significant contribution to intraseasonal predictions advance our understanding of the coupled monsoon system and also underscores the great potential of ML techniques in weather forecasts and climate predictions.

中文翻译：

---

利用印度洋中部模式和机器学习熟练预测印度季风季内降水


季风降水以季节内变率为主，其巧妙的预测提前期目前不到 5 天，仍然是一个巨大的挑战。在这里，我们展示了印度洋的固有变率，即中印度洋 （CIO） 模式，当与机器学习 （ML） 算法相结合时，可以对季风地区的季内降水进行巧妙的预测，提前期超过 15 天，接近理论可预测性极限。这种显着的技能改进源于 CIO 模式与季节内季风降雨动态相关，而数据驱动的 ML 算法抑制了不需要的高频噪声。使用 CIO 模式和 ML 算法，预测系统混合了物理基础知识和数据驱动算法的多功能性。CIO 模式的识别及其对季节内预测的重大贡献的验证促进了我们对耦合季风系统的理解，也强调了 ML 技术在天气预报和气候预测中的巨大潜力。