Bringing Artificial Intelligence to the Coast: A Multi-Model Framework Fusing Sentinel-2 and Earth Observation Data for Near-Real-Time Beach Monitoring across 200 Sites in Cyprus

Introduction: Although artificial intelligence has reshaped much of remote sensing, relatively few systems integrate satellite observations and predictive models to support timely coastal monitoring, particularly in small island states. In Cyprus, where tourism depends heavily on coastal environments, approximately 200 beaches lack an integrated monitoring framework. One example is Posidonia oceanica: this protected Mediterranean seagrass indicates ecological value but is often mistaken for pollution, while algal blooms and bathing hazards may remain undetected.

Methods: We developed THALASSIQ, an AI framework that integrates periodically acquired Sentinel-2 imagery with more frequently updated Earth-observation and environmental data. From Sentinel-2, the framework derives the Floating Algae Index, a water-adapted NDVI, and red-edge ratios to differentiate submerged Posidonia from spectral signatures associated with green algae and cyanobacteria. Copernicus Marine and Open-Meteo provide complementary variables, including wave height, sea-surface temperature, currents, turbidity, ultraviolet radiation, and Saharan dust. These variables feed a composite Predictive Safety Index, while an XGBoost–Prophet–LSTM ensemble generates seven-day forecasts. Z-score and Mann–Kendall methods are used to identify anomalies and trends. A bilingual natural-language interface presents the available information to users.

Results: The framework was configured to provide coverage for approximately 200 beaches in Cyprus. Preliminary outputs at representative sites included sea-surface temperature near 26 °C, turbidity of 0.66 FNU, a clarity index of 88/100, and safety scores ranging from 72 to 81. These values constitute preliminary system outputs and plausibility checks rather than completed validation results. Formal validation against European Environment Agency bathing-water records and available in situ observations is ongoing.

Conclusions: THALASSIQ demonstrates a framework for combining periodically updated satellite imagery with more frequently updated environmental data to support near-real-time coastal intelligence. Formal validation is ongoing; once completed, it will allow a rigorous assessment of predictive accuracy, reliability, and transferability to other Mediterranean and island environments.