Assessing Meteo-HySEA Performance for Adriatic Meteotsunami Events

TL;DR

The paper evaluates Meteo-HySEA, a GPU-accelerated meteotsunami model, against the CPU-based AdriSC-ADCIRC system in the Adriatic Sea using three documented events (2014, 2017, 2020). Both models are forced by WRF downscaled ERA reanalyses, with validation against tide gauges and microbarographs to assess timing, amplitude, and periods of meteotsunamis; results show forcing quality largely governs realism, with Meteo-HySEA capturing timing and spatial patterns but often producing longer periods. A major advantage of Meteo-HySEA is its computational efficiency, enabling rapid high-resolution, multi-grid simulations including inundation for potential operational early warning. The study highlights the need for systematic observational validation and enhanced atmospheric forcing (ensembles, data assimilation) to reliably forecast harbor seiches and inundation, while confirming the potential of GPU-accelerated meteotsunami modeling for real-time hazard assessment.

Abstract

Meteotsunamis are atmospherically driven sea-level oscillations that can trigger hazardous coastal flooding, particularly in resonant bays. This study assesses the GPU-based Meteo-HySEA model for meteotsunami simulation in the Adriatic Sea, benchmarking its performance against the CPU-based AdriSC-ADCIRC system. Three documented events (2014, 2017, 2020) were simulated using WRF downscaling of ERA reanalyses and validated with tide-gauge and microbarograph observations. Both models are limited by the underestimation of mesoscale pressure disturbances in the atmospheric forcing. Meteo-HySEA generally reproduces the timing and spatial variability of sea-level oscillations and often yields larger amplitudes than ADCIRC, but it tends to overestimate dominant wave periods, particularly in enclosed basins. Differences in oscillation persistence underscore the need for further validation against high-resolution tide-gauge data to assess whether Meteo-HySEA captures harbor seiches more realistically or ADCIRC better represents physical energy dissipation. Crucially, GPU acceleration provides order-of-magnitude gains in computational efficiency, enabling rapid high-resolution, multi-grid simulations including inundation, and thus offering strong potential for operational early warning.

Figures (19)

• Figure 1: Location and topo-bathymetry of the Adriatic basin with a zoom on the study area including the Vela Luka, Stari Grad and Vrboska harbors. Observational network based on 1-min data at microbarograph locations (pink circles) and at tide-gauge locations in the Vela Luka and Stari Grad harbors as well as 5-min data from various pressure sensors (white dots).
• Figure 2: Adriatic Sea and Coast (AdriSC) modeling suite setup including the domains of the different models used (WRF, ROMS and ADCIRC) with a zoom on the ADCIRC model mesh at the locations of interest: Vela Luka, Stari Grad and Vrboska harbors.
• Figure 3: Meteo-HySEA nested grid system setup. A five-level nested grid system is shown, where the outlines indicate the extent of each grid. The spatial resolutions range from 1 km (blue), 250 m (red), 60 m (green), 30 m (black), and 7 m (magenta). The bottom images display zoomed views of the highest-resolution grids covering the areas of Stari Grad, Vrboska, and Vela Luka.
• Figure 4: Meteotsunami events of June 25 and 26 2014. Spatial distribution of the maximum pressure disturbances for (a) the WRF-ERAI and (b) WRF-ERA5 simulations. In panel (c), observed and modeled filtered atmospheric pressure at the Vis, Vela Luka and Vrboska MESSI microbarograph locations.
• Figure 5: Meteotsunami events from June 28 to July 02 2017. Spatial distribution of the maximum pressure disturbances for (a) the WRF-ERAI and (b) WRF-ERA5 simulations. Panel (c) compares observed and modeled filtered atmospheric pressure at the Vela Luka, Stari Grad and Vrboska MESSI microbarograph locations.