Statistical Downscaling of Seasonal Forecasts of Sea Level Anomalies for U.S. Coasts

Increasing coastal inundation risk in a warming climate will require accurate and reliable seasonal forecasts of sea level anomalies at fine spatial scales. In this study, we explore statistical downscaling of monthly hindcasts from six current seasonal prediction systems to provide a high‐resolution prediction of sea level anomalies along the North American coast, including at several tide gauge stations. This involves applying a seasonally invariant downscaling operator, constructing by linearly regressing high‐resolution (1/12°) ocean reanalysis data against its coarse‐grained (1°) counterpart, to each hindcast ensemble member for the period 1982–2011. The resulting high‐resolution coastal hindcasts have significantly more deterministic skill than the original hindcasts interpolated onto the high‐resolution grid. Most of this improvement occurs during summer and fall, without impacting the seasonality of skill noted in previous studies. Analysis of the downscaling operator reveals that it boosts skill by amplifying the most predictable patterns while damping the less predictable patterns. Plain Language Summary Currently, the large computer models that form the basis of seasonal climate prediction systems produce coastal sea level forecasts spaced about 100 km apart. This is too coarse to meet the needs of U.S. coastal ocean management and services, which are becoming increasingly important as sea levels rise in a warming climate. In this study, we explored a method to provide such information on much smaller spatial scales, which better correspond to local coastal sea level measurements by tide gauges. We developed an efficient way to generate monthly sea level predictions on distances as small as 10 km apart, by applying the observed statistical relationship between sea level variations on scales of 100–1,000 km and finer‐scale coastal ocean observations to the original coarser model predictions. By testing our approach on past forecasts (“hindcasts”) from existing climate forecast systems, we found that we could improve forecasts for different local regions along both the U.S. West and East Coasts. Key Points Sea level prediction from relatively coarse operational forecasts can be enhanced to finer coastal scales using statistical downscaling Downscaling can be determined by multivariate linear regression trained from high‐resolution reanalysis and its coarse‐grained counterpart This downscaling method significantly improves skill compared to bilinea