Unsupervised machine learning applied to Sentinel-2 time series for monitoring coastal lagoon water quality

• Sentinel-2 time series allows evaluation of water quality in a coastal lagoon. • A combination of unsupervised learning models and STL decomposition to identify spatio-temporal patterns. • Accurate zonation according to water quality degradation, with a 10 m spatial resolution. • Analysis of each zone’s patterns through its seasonality, trend, and residual components. • Clear relationship between residual values and extreme events. Artificial intelligence (AI) is transforming environmental monitoring by enabling automated, data-driven analysis of complex, high-dimensional datasets. In this study, we introduce an AI-based framework to identify the areas most affected by ecological degradation and to analyze their spatio-temporal dynamics using multispectral satellite imagery. The time series of Sentinel-2 , spanning from July 2015 to June 2024, was processed using ACOLITE atmospheric correction. Unsupervised machine learning techniques ( k -means clustering) combined with Seasonal, Trend decomposition using Loess (STL) were applied to characterize water quality patterns and temporal dynamics. The framework was developed in the Mar Menor coastal lagoon (SE Spain), where STL decomposition enabled the separation of long-term trend, seasonal variability, and residual anomalies in chlorophyll-a and turbidity, linking extreme events occurring in the lagoon with significant anomalies observed in the historical remote sensing data. This hybrid approach allowed us to differentiate regions with similar optical and ecological behavior into four well-defined zones at 10 m spatial resolution. The integration of unsupervised AI-based clustering and statistical time-series decomposition offers a powerful, scalable, and transferable methodology for automated lagoon monitoring. These results provide useful information for designing and implementing conservation and management strategies, and offer a robust analytical basis for future prevention and restoration strategies