Storm surges and coastal inundation during extreme events in the Mediterranean Sea: the IANOS Medicane

Abstract

The IANOS Medicane was one of the most severe storms that have formed in the Mediterranean Sea with Category 2 Hurricane characteristics. The storm induced a significant increase in sea-level elevation along its pathway and caused storm surges at the central Ionian Sea with consequent impacts on coastal regions of the Ionian Islands and western Greece. An integrated approach, based on hydrodynamic ocean simulations, coupled to meteorological and coastal flooding simulations, is used in combination with field and satellite observations to analyze the marine weather conditions, the storm surge characteristics, and the coastal inundation characteristics due to the impact of IANOS Medicane in September 2020. The evolution of the Medicane and the respective storm surge in the ocean have been successfully recorded by the met-ocean simulations, part of an active public-access operational forecast system. Both wind and atmospheric pressure patterns affected the storm surge variability over the Ionian Sea. The direct intrusion of the Medicane from the central Mediterranean Sea toward the Ionian Sea formed storm surges over several coastal areas, even before the storm’s landfall, due to the accompanying onshore currents. Storm surges in the order of 30 cm generated extensive flooding over lowland coastal areas, as confirmed by both satellite (Normalized Difference Water Index, NDWI) and numerical (coastal inundation modeling) data. Satellite-derived and simulated estimations showed that, in specific coastal regions, the run-up of seawater extended up to 200 m inland, depending on the hydraulic connectivity between the lowland areas, which determined the inundation extents during the storm surge.

Appendices

Appendix A

CoastFLOOD is a 2-D horizontal, mass balance, coastal inundation model, which is based on the concepts of the established LISFLOOD-FP model for coastal (and river) plain flooding (Bates et al. 2005, 2010; Hunter et al. 2005). The robustness of LISFLOOD-FP model’s approach has been thoroughly validated and broadly applied in floodplain areas (Horritt and Bates 2002; Neal et al. 2011; Seenath et al. 2016). We hereby combine this approach with a wet/dry cell assignment technique for flood fronts over steep slopes (Brufau et al 2004; Castro et al. 2005). Therefore, the model considers the wetting and drying of separate cells in the integration process, yet all computations are set to finish when the highest flooded area is achieved, viz. it is not allowed to simulate the retreat of seawater toward the sea after the storm surge starts to attenuate on the coastal boundary. The flood routing module makes use of very fine spatial resolution (dx = 2 m; see Sect. 2.2) computational domains based on raster grids (Horritt and Bates 2001).

CoastFLOOD has a rather simplistic finite difference hydraulic flow module of very high resolution, based on a raster grid, particularly fitted to reproduce the 2-D water expansion due to surge-induced seawater setup causing coastal inundation (Bates and De Roo 2000; Horritt and Bates 2001; Makris et al. 2020; Skoulikaris et al. 2021). A meridional-to-zonal direction decomposition of the inundation flow components allows the 1-D flow equations for seawater overland propagation to be solved separately for each front of a typical 2-D raster grid cell (Bates et al. 2005; Bradbrook et al. 2004; Hunter et al. 2005). This provides an easy 2-D solution representation in each horizontal direction (zonal and meridional, i.e., x- and y-direction, respectively) and final integration of the flow (Bates et al. 2010). The continuity equation corresponds to the mass conservation principle for the calculation of local water depth (free-surface height) in each gridded cell (Bates et al. 2005). The flood flow route is then estimated by a simplistic quad-tree search algorithm for downstream cells and saved in an updated matrix every time step by a dry/wet cell storage code, based on the difference of hydraulic head between neighboring cells (Hunter et al. 2006). The flow rate in each cell and direction are derived from a Manning’s law approach using the water surface elevation above land level incorporating bottom friction. The continuity and momentum equations for the calculation of change in the volume of flow, following Neal et al. (2011), are described in detail by Skoulikaris et al. (2021).

CoastFLOOD model also incorporates a “static-level” inundation module operating in “Bathtub” mode, i.e., tracing and marking the flood-prone lowland cells with ground elevation z below a predefined threshold (e.g., the storm surge water level) on the computational raster grid (Didier et al. 2019; Schmid et al. 2014; Yin et al. 2017). The Bathtub technique is too simplistic in terms of hydraulic processes and may lead to non-physical overestimations of coastal flood extents (Didier et al. 2015; Ramirez et al. 2016; Vousdoukas et al. 2016). Thus, an enhanced bathtub approach with hydraulic connectivity (“Bathtub HC” mode), i.e., allowed water flow in adjacent cardinal and diagonal directions (“eight-side rule”), is also available to constrict implausible overestimation of possibly inundated areas by coastal seawater masses (Karamouz and Fereshtehpour 2019; West et al. 2018; Williams and Lück-Vogel 2020). This method neglects bottom friction due to floodplain terrain roughness and permeability, time integration for the entire duration of the storm surge event, and water flow height and velocity that affect the overland flood extension from the coastline. However, it performs better than a mere bathtub approach and therefore provides more conservative inundation results with less unrealistically detached flooded areas.

The potentially flooded coastal land area is precalculated, as the hydraulic-connectivity bathtub module runs in CoastFLOOD’s initial phase of simulation before it commences the full-scale numerical solution by the spatial- and time-stepping algorithm (Bates and De Roo 2000; Bates et al. 2005, 2010; Skoulikaris et al. 2021). This eventually helps with deriving the coverage percentage of inundated areas (by number of wet cells) that are numerically estimated by the flood flow model compared to the potentially flooded areas of the raster grid (Makris et al. 2020; Skoulikaris et al. 2021). Typically, inconsistencies in the digital elevation model (DEM) may induce inaccuracies in the derivation of flooded areas. In the present case study, the implemented grid is of very high resolution (dx = 2 m), taking into account the most significant topographic details in the coastal zone, such as engineered urban infrastructure, buildings, ports, roads, and natural formations in the floodplain, farmlands, beach slopes, dunes, emerged barriers, hillocks, rural geodetic peculiarities (Murdukhayeva et al. 2013; Kahl et al. 2022).

Appendix B

The Normalized Difference Water Index (NDWI) is computed based on Sentinel-2 Band3 and Band8 bands of the ocean color images:

$$NDWI = \frac{{\rm Band}3 - {\rm Band}8}{{{\rm Band}3 + {\rm Band}8}}$$

(1)

where Band3 and Band8 are the visible green light and the near-infrared radiation of the spectrum, respectively. The estimation of the coastal flooded area is based on the derived NDWIs of two images, the one before (15 September) and the second after (20 September) the IANOS passage; the difference of the two NDWIs provides information about the inundated area. Specifically, after calculating NDWI for each of the two images, the magnitude of the difference in NDWI for each 10-m pixel (NDWI value after the storm minus the NDWI value before the storm) was calculated. A positive difference is interpreted as an increase in soil moisture due to the storm indicating a prior presence of water on that pixel. In order to conservatively identify the areas that received large amounts of water due to the storm, it is considered that pixels with difference values greater than 0.1 correspond to wet soil (“wet” areas that were potentially flooded during the storm). The next step was to filter the NDWI values calculated for 15 and 20 September with a threshold of zero (all cells with positive NDWI values are identified as cells full of water) and by this to identify the actual water areas before and after the storm. The difference between the two corresponds to the areas that were still flooded on 20 September due to the storm (“flooded” areas). It is worth pointing out that the areas identified as flooded due to the storm had increased NDWI difference values greater than 0.5 in several cases, which confirms the result. It is noted that although the method is not able to distinguish the source of the flood (e.g., storm surge, drainage runoff, precipitation), it provides useful information about the inundation levels of lowland areas close to the coastline.