Large-scale modes of climate variability, such as the El Niño-Southern Oscillation, North Atlantic Oscillation, and Indian Ocean Dipole, show significant regional correlations with seasonal weather conditions, and are routinely forecast by meteorological agencies attempting to anticipate seasonal precipitation patterns. Here, we use machine learning together with more traditional approaches to quantify how much precipitation variability can be explained by large-scale modes of variability, and to understand the degree to which these modes interact non-linearly. We find that the relationship between climate modes and precipitation is predominantly non-linear. In some regions and seasons climate modes can explain up to 80% of precipitation variability. However, variability explained is below 10% for more than half of the land surface, and only 1% of the land shows values above 50%. This outcome provides a clear rationale to limit expectations of predictability from modes of variability in all but a few select regions and seasons.

大尺度气候变率模态，如厄尔尼诺-南方涛动、北大西洋涛动和印度洋偶极子，与季节性天气条件显示出显著的区域相关性，并且通常由试图预测季节性降水模式的气象机构进行预报。在这里，我们将机器学习与更传统的方法结合使用，以量化大尺度变率模式可以解释多少降水变率，并了解这些模态非线互的程度。我们发现气候模态和降水之间的关系主要是非线性的。在某些地区和季节，气候模式可以解释高达 80% 的降水变化。然而，对于超过一半的陆地表面，解释的变异性低于 10%，只有 1% 的土地显示高于 50% 的值。这一结果为限制除少数选定地区和季节外的所有地区和季节的变异模式对可预测性的期望提供了明确的理由。