Synoptic-based model for reconstructing and forecasting high-frequency sea-level extremes in the Mediterranean

This paper evaluates the performance of a synoptic index-based model designed to predict extreme non-seismic sea-level oscillations at tsunami timescales (NSLOTTs) across 32 tide-gauge stations in the Mediterranean Sea, where NSLOTTs can contribute up to 50% of the total sea-level range. The model employs percentile-determined threshold exceedance criteria to define extreme NSLOTT events. A part of the time series containing half of extreme NSLOTT events is used for model training, while the rest is used for performance assessing. The baseline model integrates seven synoptic variables previously identified for a known NSLOTT hotspot and available within atmospheric reanalysis products. Various model configurations and modifications were tested to evaluate adaptability and robustness in forecasting and detecting extreme NSLOTT events. Results indicate that the model success in forecasting extreme events slightly outweighs its success in detecting observed extreme events. For stations where the baseline model performs well, this proficiency remains consistent across different configurations. However, the uncertainty in model performance is greater for these stations compared to those with poorer performance, which show minimal improvement despite configuration adjustments. Sub-basin analysis reveals that tide-gauge stations located in the eastern Adriatic Sea exhibit the best performance on average. These findings provide valuable insights for optimizing the model setup, enhancing its predictive capabilities, and improving its application in projecting extreme NSLOTT events in future climates. Ultimately, this work may contribute to coastal hazard and flooding mitigation, as well as resilience-building efforts, where extreme NSLOTT events could play a substantial role.