Prediction of harmful algal blooms in the Persian Gulf using remote sensing and artificial intelligence modeling

Harmful algal blooms (HABs) represent a significant environmental and socio-economic threat across the Persian Gulf region, impacting marine ecosystems, public health, and coastal economies. This study shows an advanced predictive pipeline that uses satellite remote sensing data with ensemble Artificial intelligence (AI) modeling to forecast HAB events along the coastlines of Bandar Abbas, Qeshm, and Hormuz. Key environmental variables, including chlorophyll-a concentration, sea surface temperature (SST), and remote sensing reflectance (R rs ) at wavelengths of 412, 443, 488, 513, and 555 nm, were extracted from MODIS-Aqua imagery, providing a comprehensive depiction of the spatial and temporal variability in the marine environment. We employed a species distribution modeling approach that integrates an ensemble of five machine learning (ML) algorithms—Random Forest (RF), Boosted Regression Trees (BRT), Support Vector Machines (SVM), Multi-Layer Perceptron (MLP), and Radial Basis Function (RBF)—to mitigate the limitations of single-model predictions and enhance forecast reliability. Our modeling framework utilized 1809 confirmed HAB presence observations alongside 13,396 systematically generated pseudo-absence points, with model performance validated through bootstrapping and cross-validation over 713 daily prediction intervals. The ensemble model, formulated via AUC-weighted aggregation of individual predictions, achieved a robust average Area Under the Curve (AUC) of 0.95 and a peak True Skill Statistic (TSS) of 0.85 . Specifically, a case study on November 23, 2008, yielded a sensitivity of 96.67 % and specificity of 74.37 %, highlighting the model's proficiency in correctly identifying HAB events. Variable importance plots pinned SST and certain Rrs bands (particularly at 443 and 555 nm) as key predictors, which concurs with established drivers of algal growth. Further, this coupled method not only yields high-resolution spatial and temporal forecasts of bloom events but also refined insight into environmental mechanisms underlying HAB dynamics, informing effective coastal governance and policy making. Collectively, these findings illustrate the promise of combining remote sensing data with ensemble AI methods to create effective early-warning systems and inform targeted management practices for reducing impacts of HABs in the Persian Gulf. • Advanced Ensemble AI Modeling: The study integrates five machine learning algorithms—Random Forest, Boosted Regression Trees, SVM, Multi-Layer Perceptron, and RBF Networks—into an ensemble framework, significantly improving prediction reliability for harmful algal blooms (HABs). • High Model Performance: The ensemble model achieved a strong average AUC of 0.877 and a peak TSS of 0.7105, with a case study showing 96.67 % sensitivity, demonstrating high accuracy in detecting HAB events. • Key Environmental Predictors Identified: Sea surface temperature (SST) and remote sensing reflectance at 443 nm and 513 nm were identified as the most influential variables in predicting HAB occurrences. • Extensive Dataset and Validation: The model utilized 1894 HAB presence records and 14,276 pseudo-absence points, validated across 835 days using bootstrapping and cross-validation techniques for robustness. • Impactful Applications for Coastal Management: This approach offers high-resolution, spatiotemporal HAB forecasts for key regions like Bandar Abbas, Qeshm, and Hormuz, providing valuable insights for early-warning systems and informed environmental policy-making.