Process-Informed Machine Learning for Satellite-Based Prediction of Harmful Algal Blooms in the Persian Gulf

Harmful algal blooms (HABS), commonly known as red tides, represent a major ecological and economic concern in the Persian Gulf, where high temperature, elevated salinity, and intense anthropogenic pressure create favorable conditions for bloom development. This study presents a Machine Learning (ML)-based framework for the monitoring and prediction of HABS using satellite-derived biogeochemical indicators. Multi-sensor data from MODIS-Aqua and Sentinel-۳ OLCI (۲۰۱۸-۲۰۲۳) were processed to extract key environmental variables, including chlorophyll-a (Chl-a), sea surface temperature (SST), remote-sensing reflectance (Rrs), photosynthetically active radiation (PAR), and the diffuse attenuation coefficient Kd(۴۹۰). After preprocessing and labeling bloom-prone pixels, four supervised algorithms-Random Forest (RF), Support Vector Machine (SVM), Artificial Neural Network (ANN), and eXtreme Gradient Boosting (XGBoost)—were trained to classify HAB versus non-HAB conditions. Model performance was evaluated using Accuracy, F۱-score, Area Under the Curve (AUC), and Root Mean Square Error (RMSE). The results show that tree-based models outperform other approaches, with XGBoost achieving the highest Accuracy (۹۵.۴%), F۱-score (۰.۹۴), AUC (۰.۹۷), and the lowest RMSE (۱.۰۵ mg m³). These findings highlight the strong predictive capability of integrating satellite observations with advanced ML, offering a robust foundation for early-warning and continuous monitoring systems to support sustainable marine ecosystem management in the Persian Gulf.

کلیدواژه ها:

Red Tide ، Persian Gulf ، Machine Learning ، Remote Sensing ، XGBoost