Over the central United States, a large fraction of heavy precipitation and flood events have been tied to atmospheric rivers (ARs) and to two large‐scale atmospheric modes, the Pacific‐North American teleconnection pattern (PNA) and the Arctic Oscillation (AO). Here, we build on these insights to model the frequency of heavy precipitation events at 88 locations across the central United States. We use a Bayesian hierarchical modeling framework to develop and compare different models with varying degrees of complexity and nonstationary parameters that are conditioned on ARs and the prevalent climate. We show that Bayesian hierarchical models with a prior layer that allows spatially correlated parameters result in improved prediction skills over the traditional regression‐based modeling frameworks. While ARs, PNA, and AO have statistically significant relationships with the frequency of heavy precipitation events over large areas of the central United States, we find that ARs can significantly improve upon the other two covariates in the statistical modeling of extremes over the region.

中文翻译：

---

美国中部冬季强降水事件发生频率的贝叶斯层次模型：大气河流的作用

在美国中部，大部分的强降水和洪水事件都与大气河流（AR）和两个大型大气模式有关，即太平洋-北美遥相关模式（PNA）和北极涛动（AO） 。在这里，我们基于这些见解来对美国中部88个地点的强降水事件的频率进行建模。我们使用贝叶斯分层建模框架来开发和比较具有不同程度的复杂性和非平稳参数的不同模型，这些参数取决于AR和普遍的气候。我们显示，具有允许空间相关参数的先验层的贝叶斯分层模型可以提高基于传统回归模型框架的预测技巧。而AR，PNA，