Abstract
Coastal communities, together with their resources and various ecosystems, are often susceptible to natural disasters such as hurricanes, tropical storms, and sea level rise (SLR). Wetlands and marsh vegetation can lessen the impact of storm surges and coastal floods. However, the increasing rate of sea level rise may lead to the destruction and retreat of these natural ecosystems. This study utilizes a combination of machine learning (ML) for coastal storm surge modeling and an assessment of its negative impact on land use and landcover (LULC). The objective is to evaluate the extent of flooding and the resulting loss of natural habitat caused by the landfall of two significant hurricanes along the southwest coast of Florida. The ML models were trained to compute the enduring capacity for recurrent floods caused by hurricanes resembling recent ones, specifically hurricanes Irma (2017) and Ian (2022). The models were trained and validated using gridded data obtained from Sentinel-1 and Landsat 8 OLI. The NOAA SLR Digital Elevation Model data was utilized as an input parameter for machine learning-based hydrodynamic modeling. It has been utilized to forecast the highest storm surge height and the maximum water level resulting from wind stress and wave setup, as well as to demarcate the impacted habitats in the Florida coastal ecosystem. The ML-based modeling methodologies used to evaluate the effects of storm surges and flooding on coastal ecosystems can be enhanced by incorporating additional essential sea state variables into the model. In this study, we advanced in this area by adapting the model framework to incorporate alterations in landforms, in addition to changes in the oceanic environment (such as changes in shape and structure and the development of urban areas along the coast). The model accurately predicts the extent of land affected by surges caused by hurricanes Irma (3413 ha) and Ian (3754 ha), as well as their negative effects on land use and land surface elevation. The findings indicate that the movement of storm surge and subsequent flooding and SLR inland was characterized by a high level of dynamism and was strongly impacted by geomorphology and landscape characteristics. The most accurate machine learning model achieves an AUC score of 0.99 when forecasting the occurrence of storm surge and its movement towards land. Presently, we present a novel simulation framework that incorporates modifications to the terrain in order to enhance the accuracy of machine learning-based hydrodynamic storm surge prediction models for upcoming scenarios of sea level rise and storm conditions.
