Robust machine learning methods for climate research

将神经网络等机器学习方法引入气候研究，有助于更好地理解厄尔尼诺-南方涛动(ENSO)等气候现象的非线性特征。不幸的是，北极涛动(AO)、太平洋-北美(PNA)遥相关、热带外ENSO效应和太平洋年代际振荡(PDO)等热带外气候现象，由于其信噪比较低和/或数据记录相对较短，给当前一代机器学习方法带来了极大的困难。提出了新的稳健方法，这些方法应该会显着改善对这些气候现象的非线性分析(在某些情况下还包括预测)。这些稳健的方法不仅有助于分析观测数据，而且有助于分析大气环流模式(GCM)的气候模拟，从而使我们不仅能够了解气候系统的非线性行为，而且还可以了解在全球变暖情景下非线性行为是如何变化的。

Machine learning methods such as neural networks have been introduced to climate research, leading to a better understanding of the nonlinear properties of climate phenomena, such as the El Niño-Southern Oscillation (ENSO). Unfortunately, extra-tropical climate phenomena such the Arctic Oscillation (AO), the Pacific-North American (PNA) teleconnection, extra-tropical ENSO effects, and the Pacific Decadal Oscillation (PDO), with their weak signal-to-noise ratio and/or relatively short data records, pose great difficulties for the current generation of machine learning methods. New robust methods are proposed which should significantly improve on the nonlinear analysis (and in some cases prediction) of these climate phenomena. These robust methods will be useful in analyzing not only observed data, but also the general circulation model (GCM) climate simulations, thereby allowing us to understand not only the nonlinear behaviour of our climate system, but also how the nonlinear behaviour changes under global warming scenarios.