In recent years, the vulnerability of coastal regions has increased significantly due to the effects of climate change. Measures must be taken to protect these coastal regions, which are disproportionately affected by extreme weather events and other damaging factors, and to increase their resilience. In this study, we propose a conceptual patch-based extension to the unsupervised Local Outlier Factor (LOF) anomaly detection algorithm to enable hotspot detection in Earth observation data. We validate our approach on Synthetic Aperture Radar (SAR) data using both synthetic and real-world anomalies and demonstrate that these methods outperform an autoencoder and a temporal Reed-Xiaoli (RX) approach, which are widely used for anomaly detection. Additionally, we generate coastal hotspot maps that identify areas requiring greater protection against extreme weather events and other hazards. These maps allow us to provide recommendations to decision-makers and governance bodies. • Extension of LOF algorithm, enabling localization of anomalies in SAR data. • Validation of these unsupervised ML algorithms with synthetic and real-world anomalies. • Generation of coastal hotspot maps to improve coastal protection.
Koslow et al. (Thu,) studied this question.