Evaluating Current Statistical and Dynamical Forecasting Techniques for Seasonal Coastal Sea Level Prediction

The need for skillful seasonal prediction of coastal sea level anomalies (SLAs) has become increasingly evident as climate change has increased coastal flooding risks. Here, we evaluate nine current forecast systems by calculating deterministic and probabilistic skill from their retrospective forecasts (“hindcasts”) over 1995-2015, for lead times up to 6-9 months, at two United States tide gauge stations (Charleston, SC and San Diego, CA). Additionally, we assess local skill enhancement by two post-processing/downscaling techniques: an observationally based multivariate linear regression and a hybrid dynamical approach convolving sea-level sensitivity to surface forcings with predicted surface forcing variations. All these approaches face challenges stemming from whether modeled SLAs sufficiently represent observed local coastal SLA variations because of ocean model limitations and inadequacies in model initialization and ensemble spread. Some of these issues also complicate the ability of the post processing techniques to improve probabilistic skill, even though they do somewhat improve deterministic skill. In general, deterministic hindcast skill is considerably higher for San Diego than Charleston, as expected from the stronger influence of ENSO. However, ensemble spread metrics such as forecast reliability and sharpness remain low for both locations, highlighting model deficiencies in representing uncertainty. Additionally, evaluating how well any technique predicts seasonal coastal sea level variations is complicated by the forced trend component, particularly how it is estimated. Moreover, model skill is matched by a stochastically-forced multivariate linear prediction model constructed from observations, suggesting that substantial improvement remains for predicting coastal seasonal SLAs, which could also include leveraging other predicted fields, including sea level pressure and prevailing winds.