Coastal flooding poses a significant threat to coastal communities, adversely affecting both safety and economic stability. This threat is exacerbated by factors such as sea level rise, rapid urbanization, and inadequate coastal infrastructure, as noted in recent climate change reports. Early warning systems (EWSs) have proven to be effective tools in coastal planning and management, offering a high cost-to-benefit ratio. Recent advancements have integrated operational numerical models with machine learning techniques to develop near-real-time EWSs, leveraging data obtained from reputable databases that provide reliable hourly sea-state and sea level data. Despite these advancements, a stepwise methodology for selecting representative events, akin to wave input reduction methods used in morphological modeling, remains undeveloped. Moreover, existing methodologies often overlook the significance of compound extreme events and their potential increased occurrence under climate change projections. This research addresses these gaps by introducing a novel input schematization method that combines efficient hydrodynamic modeling with clustering algorithms. The proposed methodοlogy, implemented in the coastal area of Pyrgos, Greece, aims to select an optimal number of representative sea-state and water level combinations to develop accurate EWSs for coastal flooding risk prediction. A key innovation of this methodology is the incorporation of weights in the clustering algorithm to ensure adequate representation of extreme compound events, also taking into account projections for future climate scenarios. This approach aims to enhance the accuracy and reliability of coastal flooding EWSs, ultimately improving the resilience of coastal communities against imminent flooding threats.
