Multi-Reservoir Water Quality Mapping from Remote Sensing Using Spatial Regression

Quantitative prediction of toxin-producing Aphanizomenon cyanobacteria in freshwaters using Sentinel-2 satellite imagery

This study aimed to develop an empirical model to predict the spatial distribution of Aphanizomenon using the Ridiyagama reservoir in Sri Lanka with a dual-model strategy. In December 2020, a bloom was detected with a high density of Aphanizomenon and chlorophyll-a concentration. We generated a set of algorithms using in situ chlorophyll-a data with surface reflectance of Sentinel-2 bands on the same day using linear regression analysis. The in situ chlorophyll-a concentration was better regressed to the reflectance ratio of (1 + R₆₆₅)/(1-R₇₀₅) derived from B4 and B5 bands of Sentinel-2 with high reliability (R² = 0.81, p < 0.001). The second regression model was developed to predict Aphanizomenon cell density using chlorophyll-a as the proxy and the relationship was strong and significant (R² = 0.75, p<0.001). Coupling the former regression models, an empirical model was derived to predict Aphanizomenon cell density in the same reservoir with high reliability (R² = 0.71, p<0.001). Furthermore, the predicted and observed spatial distribution of Aphanizomenon was fairly agreed. Our results highlight that the present empirical model has a high capability for an accurate prediction of Aphanizomenon cell density and their spatial distribution in freshwaters, which helps in the management of toxic algal blooms and associated health impacts.