Multimodel assessment of CMEMS storm-surge forecasts under record-breaking Gloria storm

On January 19th-24th 2020, the Western Spanish Mediterranean (WM) coast was hit by the storm Gloria, one of the most extreme meteorological events ever recorded in the region. A strong North-South atmospheric pressure gradient, linked to a high atmospheric pressure system centred over the British Islands (1050hPa), favoured outstanding easterly winds across the WM. Several buoys moored along the Iberian Mediterranean coast beat their record of significant wave height (reaching 8.44 m at Valencia buoy) and a wind-driven storm-surge locally beat the record along the Valencia coastline.

Operational storm-surge forecasts were provided by different services at the WM area. Models presented both commonalities and differences, due to their intrinsic features (physics, resolution, forcing data, assimilation scheme, etc). A way to synthetise all these model outcomes, is by building an ensemble that integrates all of them. Ensemble techniques, such as Bayesian Model Average (BMA), not only generate combined forecasts; but also may compute confidence intervals, that are specially suitable when the ensemble members diverge.

Since 2018, the Puertos del Estado (PdE) ENSURF (ENsemble SURge Forecast) system delivers probabilistic forecasts at WM tidal stations, by combining in a BMA: (i) near-real time tide gauge data and (ii) forecasts from the PdE Nivmar system and the CMEMS MED-MFC and IBI-MFC services. Consequently, this contribution aims to assess the performance of these storm-surge forecasts under storm Gloria at two levels: (i) individually and (ii) integrated within the ENSURF system. Each forecast solution has been analysed at several tidal stations, and no single model outperforms at all tidal stations and synoptic conditions. Then, it is confirmed that the probabilistic forecast gives significant added value with respect to existing operational systems.

At an individual level, on those tidal stations in which the surge was mainly wave and wind-driven, MED-MFC performed better, with emphasis on the growth-phase of the surge. IBI-MFC showed good skill on those stations with wind-driven surges, and those mean sea level pressure-driven (MLSP) surges in which the Atlantic-Med water-mass exchanges are important. Finally, Nivmar exhibited good performance on MSLP-driven surges.

At the integrated level, the ENSURF forecast presents lower bias and RMS, plus higher correlation than most of its ensemble members. Despite these error metrics, though, further work is also needed on the BMA for estimating the peak of the storm-surge event. The results for this contribution, then, may serve to plan forthcoming improvements in the current coastal sea-level forecast systems.