Abstract
Early warning systems for coastal erosion and flooding are currently primarily designed for local applications, offering high-resolution, site-specific predictions. Only a few early warning systems (EWS) are used at large regional or national scales. There is also a lack of standardised indicators and thresholds, which vary widely across systems and hinder cross-regional applicability. While current EWS perform well in binary hazard detection (Yes/No hazard; 80-95% accuracy), they struggle to accurately classify intermediate hazard levels. A lack of comprehensive field datasets has impeded rigorous validation for most systems, with many assessments relying on qualitative observations. Improving the reliability of the EWS requires improving their validation against field data obtained during storms and regular updating of the topobathymetric data to include the actual pre-storm morphology. Currently, most EWS rely on outdated or synthetic morphological inputs, which increases prediction uncertainty. The computational constraints of physics-based models may prevent warnings from being issued in time and have led to the adoption of surrogate approaches that depend on robust training datasets. Furthermore, most systems focus solely on hazard detection, paying limited attention to the risk to assets or populations. Future development must prioritise stakeholder engagement and the co-design of systems that incorporate both hazard and risk assessments, in order to improve their usefulness and facilitate decision-making by end users.