AI-driven forecasting of harmful algal blooms in Persian Gulf and Gulf of Oman using remote sensing

Abstract

This study develops an artificial intelligence (AI) model to forecast harmful algal blooms (HABs) in the Persian Gulf and Gulf of Oman using freely available remote sensing data, including chlorophyll-a (Chl-a), sea surface temperature (SST), salinity, and wind. The model introduces novel features such as spatial and temporal standard deviations of Chl-a concentration and a derived gradient feature. Correlation analysis indicated that these features enhance predictive capability. A multi-layer artificial neural network (ANN) was trained using a 66%/34% data split for training and testing, achieving 88.7% accuracy in binary classification (bloom/non-bloom) with an area under the ROC curve (AUC) of 90.1%. Overfitting was mitigated by monitoring training and validation loss, both of which consistently decreased over epochs, confirming robust model generalization. The use of standard deviation in SST and salinity highlights their influence on bloom dynamics, providing key insights into algal bloom drivers. The focus on freely available data enables stakeholders to better manage the environmental challenges posed by HABs.