Extreme sea level variability (excluding the effects of mean sea level (MSL) and long‐period tidal cycles) at decadal to multidecadal time scales is significant along the U.S. coastlines and can modulate coastal flood risk in addition to long‐term MSL rise. Therefore, understanding the climatic drivers and ultimately predicting these low‐frequency variations are important. Extreme sea level indicators are used to represent the variations in 100‐year return water levels, estimated with a nonstationary extreme value analysis. Here, we develop prediction models in the frequency domain. Extreme sea level indicators (response) and potential predictors (traditional climate indices, sea level pressure [SLP], or sea surface temperature [SST]) are decomposed into subseries corresponding to predefined frequencies using discrete wavelet transform (DWT), and regression models are formulated for each frequency separately. In the case of traditional climate indices, subseries of climate indices that provide the highest correlation with the corresponding subseries of indicators are used in the regression models, and original indicators are reconstructed by aggregating predicted subseries. Tailored climate indices are developed for each frequency band by averaging wavelet decomposed subseries of SLP or SST from grid locations where correlations with corresponding decomposed subseries of extreme sea level indicators are highest and robust. Models with wavelet filtered climate indices reproduce the variability and general trends of the indicators. The use of tailored indices further improves the model performance in predicting extreme sea level variations. Model performance in terms of Nash‐Sutcliffe efficiency statistics varies from 0.54 to 0.93. Prediction of extreme sea level indicators using tailored indices derived from SLP and SST of initialized decadal climate model simulations is also tested to facilitate progress toward forecasting extreme sea level variations at decadal time scales.

中文翻译：

---

美国海岸线沿线极端海平面变化的可预测性

在美国海岸线上，十年到多十年的时间尺度上的极端海平面变化（不包括平均海平面（MSL）和长期潮汐周期的影响）非常重要，除了长期MSL上升外，还可以调节沿海洪水风险。因此，了解气候驱动因素并最终预测这些低频变化非常重要。极端海平面指示器用于表示100年回水水平的变化，并通过非平稳极端值分析进行估算。在这里，我们在频域中开发预测模型。使用离散小波变换（DWT）将极端海平面指示器（响应）和潜在预报器（传统气候指数，海平面压力[SLP]或海面温度[SST]）分解为与预定义频率相对应的子系列，并分别为每个频率制定回归模型。在传统气候指数的情况下，在回归模型中使用与相应指标子系列提供最高相关性的气候指数子系列，并通过汇总预测的子系列来重建原始指标。通过对来自网格位置的SLP或SST的小波分解子系列求平均，从而为每个频段制定量身定做的气候指数，其中与极端海平面指标的相应分解子系列的相关性最高且很稳健。具有小波滤波气候指数的模型再现了指标的可变性和总体趋势。定制指标的使用进一步提高了模型在预测极端海平面变化方面的性能。根据Nash-Sutcliffe效率统计数据的模型性能从0.54到0.93不等。还测试了使用源自初始化年代际气候模型模拟的SLP和SST的量身定制的指数来预测极端海平面指标，以促进在十年尺度上预测极端海平面变化的进展。