AI-Driven Harmful Algal Bloom Prediction in Aquatic Ecosystems Using Machine Learning and Remote Sensing Data

Abstract

Harmful algal blooms (HABs) are major ecological and public-health threats in freshwater, coastal and marine ecosystems. They can reduce dissolved oxygen, release toxins, damage fisheries, affect tourism and create risks for drinking-water and aquaculture systems. Conventional HAB monitoring depends on field sampling and laboratory analysis, which are accurate but limited in spatial coverage, temporal frequency and operational response. Remote sensing provides repeated wide-area observations of aquatic colour, chlorophyll-a, turbidity and surface temperature, while water-quality sensors provide local physicochemical information. This paper proposes an AI-driven HAB prediction framework that integrates remote sensing data, in-situ water-quality measurements and meteorological variables using machine learning and deep learning algorithms. The proposed system uses spectral indices, environmental parameters and temporal observations to predict bloom risk and classify risk severity. Random Forest, Support Vector Machine, XGBoost, Convolutional Neural Network and Long Short-Term Memory models are considered, and a decision-fusion unit generates the final HAB risk probability. The methodology includes data acquisition, cloud and noise removal, spectral index computation, feature engineering, model training, time-series prediction and risk-level interpretation. The performance analysis demonstrates that the proposed fusion-based AI model improves prediction accuracy, precision, recall and F1-score compared with individual machine learning models. The proposed approach can support early warning, aquatic ecosystem management, pollution control, aquaculture safety and sustainable water-resource monitoring.