Optimizing Sampling Design for Chlorophyll-a Estimation in Inland Waters Using Sentinel-2 Imagery and Spatial Clustering

Abstract. Monitoring water quality in inland water bodies is critical for environmental management, yet traditional sampling methods are costly and spatially limited. Remote sensing offers a viable alternative by enabling large-scale assessment of water quality parameters such as chlorophyll-a (Chl-a) concentrations through empirical models linking spectral indices to in situ measurements. However, the accuracy of these models depends on representative sampling strategies that capture spatial and temporal variability. This study evaluates a clustering-based approach to optimize sampling site selection in the Solís Dam, Mexico, using Sentinel-2 imagery. We analyzed a one-year time series of Sentinel-2 data to compute spectral indices related to Chl-a and turbidity. Unsupervised K-means clustering was applied to stratify the reservoir into zones of distinct water quality variability, guiding the placement of 20 sampling sites. Field campaigns during dry and wet seasons (2024) provided Chl-a measurements, which were correlated with spectral indices. The Gi033BDA index showed the strongest correlation (R2 > 0.7, p < 0.01) and was used to develop a linear regression model for Chl-a estimation. Results confirmed that clustering-derived sampling points effectively represented spatial variability, though temporal mismatches (1-day lag) and samples location inaccuracies introduced minor errors. The method demonstrates how pre-stratification using remote sensing can enhance sampling efficiency while maintaining model accuracy. This approach is particularly valuable for large-scale monitoring, reducing reliance on exhaustive field campaigns. Future work should address temporal dynamics and sensor resolution trade-offs for broader applicability.