Abstract
Using Earth system models for seasonal sea‐level prediction remains challenging due to model biases and initialization shocks. Here we present a hybrid dynamical approach for seasonal sea‐level prediction to alleviate some of these issues. The approach is based on convolving atmospheric forcings with sea‐level sensitivities to these forcings. The sensitivities are pre‐computed by the adjoint model of the Estimating Circulation and Climate of the Ocean (ECCO) system. The forcings are a concatenation of ECCO forcings before prediction initialization and a 10‐member predicted atmospheric forcing ensemble from the Community Climate System Model version 4 (CCSM4) after initialization, with offline forcing bias corrections applied using the observationally‐constrained ECCO seasonal forcing climatology. As a pilot study, we conducted 12‐month hindcasts from 1995 to 2016 in Charleston (United States East Coast). Our approach avoids drifts in CCSM4 sea‐level predictions and beats seasonal climatology and damped persistence as predictors up to a 6‐month lead time. The prediction skill comes from two factors: (a) ECCO forcings prior to prediction initialization influence sea level after initialization through delayed oceanic adjustments (e.g., coastally‐trapped waves, open‐ocean Rossby waves, and advection of steric anomalies) leading to skillful predictions beyond 2 months after initialization, and (b) the 10‐member CCSM4 ensemble forcing predictions have relatively good skill at 1–2 months lead times. Our method is computationally efficient for operational sea‐level prediction at specific locations and can attribute sea‐level prediction skill and uncertainty to specific forcings or forcing from particular regions, thereby providing useful information to seasonal prediction centers for improving their prediction systems.
Plain Language Summary
Predicting sea level a few months ahead can help coastal communities to prepare for elevated flood risks. This is becoming more and more important because of the increase in floods due to rising sea levels. However, seasonal prediction models often struggle with predicting sea level. Here we present a new approach that combines predictions of atmospheric forcings, such as wind, heat fluxes, and precipitation with pre‐computed maps that show how sea level responds to changes in these forcings to predict sea‐level changes up to a few months ahead. We have tested this new approach for Charleston, South Carolina, and we find that our approach shows a promising prediction skill on lead times up to about 6 months. At short lead times, the relatively good skill of sea‐level prediction comes from the fidelity of the predicted atmospheric forcings. At lead times beyond 2 months, the observed forcings prior to prediction times enhance the sea‐level prediction skill because they cause delayed adjustment of sea level. Here, the ocean carries memory from past atmospheric forcing to influence future sea level.
Key Points
We developed a novel way to predict sea level (SL) by convolving its sensitivities to forcings with observed and coupled‐model predicted forcings
A pilot project applying the method for seasonal hindcasts of Charleston SL shows positive prediction skill up to 6‐month lead time
Sea‐level hindcasts using observed but not coupled‐model predicted forcings have even better skill for 2–6 months of lead times
