AntiBloom: A Novel Deep Learning-Powered Device to Predict Harmful Algal Blooms

Abstract

Harmful Algal Blooms (HABs) affect biosystems, leading to Unusual Marine Mortality Events (UMEs). Predicting algal bloom events can be costly and challenging. In this study, we developed a method for predicting algal blooms using an Artificial Neural Network (ANN) with Keras and Sklearn. We also designed a buoy that can collect real-time data on wind speed, UV index, pH, water temperature, and salinity to calculate real-time predictions of algal bloom. With a low cost of $150, the buoy was equipped with imaging and notification capabilities using Amazon Web Services (AWS), enabling relevant parties to be informed of harmful conditions. The ANN was trained using data from the National Centers for Environmental Information, which provided water quality data sets for Lake Erie. The model demonstrated 96.34% accuracy in predicting elevated levels of chlorophyll-a, a common marker for detecting algae. The buoy and its algorithm significantly improve over current methods of detecting elevated chlorophyll-a levels, affirming their potential to be mass-produced and usable by local authorities.