Sentinel 3 OLCI and Machine Learning for Cyanobacteria Bloom Detection Over Small Inland Water Target

High nutrient input agricultural practices and nutrient enrichment have been identified as the main factors driving cyanobacterial harmful algal blooms (cyanoHABs) formation in Uruguay. Current agricultural practices are already inflicting significant harm on aquatic ecosystems and human well-being, with future forecasts indicating a worsening of these trends. Thus, real-time detection of cyanoHABs in inland freshwater ecosystems is imperative for mitigating potential threats. Conventional cyanobacteria detection models often entail intricate procedures, necessitating the integration of biophysical, chemical, or on-site DNA sequencing measurements. Leveraging satellite imagery offers a cost-effective means to pinpoint cyanoHAB occurrences across extensive spatial and temporal scales. Within the European Space Agency's (ESA) satellite fleet, the Sentinel 3 satellites equipped with the Ocean and Land Color Instrument (OLCI) present a notable resource. In this study, we employed the cyanoHAB detection algorithm, Maximum peak-height (MPH), on OLCI data from the Laguna del Sauce lagoon to establish a baseline detection accuracy of 85%. The success of MPH's tree-like algorithm encouraged us to use machine learning classification algorithms based on decision trees to improve detection accuracy. The XGBoost classification model outperformed the other models by achieving an accuracy of 92%.