Much of the focus on future sea level rise concerns the forced trend associated with anthropogenic warming, but on shorter time scales, internal climate variability has the potential to exacerbate the…

Much of the focus on future sea level rise concerns the forced trend associated with anthropogenic warming, but on shorter time scales, internal climate variability has the potential to exacerbate the regional impacts of this long-term trend. The aim of the proposed work is to explore the predictability of regional sea levels (with uncertainties) due to large-scale climate variability at subseasonal-through-seasonal (S2S; defined here as approximately 2 weeks through 9 months) lead times. We will do this through analysis of tide gauge observations and climate model simulations and by leveraging recent, state-of-the-art XAI approaches for climate prediction and uncertainty quantification developed by the PIS. Since predicting variability of coastal sea levels and climate patterns is notoriously difficult, we do not expect to make accurate predictions all of the time. Thus, a significant focus of this work will be identifying when prediction skill exists, and when IT does, exploring its predictability sources. The proposed work is organized into four distinct activities. Activity I aims to quantify relationships between large-scale climate patterns and regional sea level extremes at S2S lead times using tide gauge observations. Activity II then focuses on developing and applying state-of-the-art explainable AI (XAI) approaches to predict coastal sea levels within CESM2, with an emphasis on uncertainty quantification. Activity III then utilizes these explainable AI algorithms to diagnose the large-scale sources of predictability and how they may change in the coming decades. Finally, activity IV aims to compare sources of predictability across large ensemble climate models and the tide gauge observations. The proposed project addresses the NOAA CPO coastal inundation opportunity by developing and exploring state-of-the-art advances in machine learning to identify and leverage sources of predictability of regional coastal sea level. By focusing on sources of predictability and large-scale climate drivers, both in observations (tide gauge data) and climate models (e.g. CESM2) this work aims to ?to improve the understanding and modeling of coupled...drivers/contributions to mean and extreme total water levels? [solicitation]. Machine learning methods have been identified by the NOAA-DOE PPP workshop report as ?innovative ways of observational analysis in support of process understanding, analysis and correction of model biases...? And ?an area that requires additional focused research?. The proposed focus on predicting societally-relevant climate hazards also serves to advance the science needed to support the NOAA strategic plan of building a climate ready nation. The proposed work meets NOAA readiness level 1 and/or 2 in that the general machine learning approach has demonstrated proof-of-concept success by the PIS, and the CO-PI has recently demonstrated success using the MJO and enso as coastal sea level predictors.