SimCLR-Enabled Wide and Deep Learning for Cyanobacterial Bloom Prediction From NASA’s PACE Hyperspectral Mission

Cyanobacterial blooms (cHABs) are harmful to aquatic ecosystems and human health. In situ sampling is labor-intensive and delayed by laboratory analyses. Remote sensing (RS) provides valuable spectral information for monitoring water bodies, but faces challenges due to the complexity of environmental factors. This letter presents a novel deep learning framework that uses self-supervised learning (SSL) with a Wide & Deep architecture to predict next week’s cHABs probability using NASA PACE mission and meteorological data. The SSL module extracts rich features from unlabeled RS data via contrastive learning and inputs them into the Wide & Deep architecture, which memorizes wide features and learns high-order abstractions from deep spectral inputs. This enables reliable generalization to unseen regions with scarce data. We evaluated our model on bloom-prone lakes: Erie (<italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">n</i> = 254: fine-tuning/validation) and Okeechobee (<italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">n</i> = 350, zero-shot testing) using binary bloom flags as ground truth. Our model achieves high predictive accuracy (0.964 ROC-AUC and 0.959 PR-AUC). A zero-shot evaluation on Lake Okeechobee shows promising generalization with 74% accuracy. Compared to existing methods, our SSL-based hybrid deep learning model offers competitive performance for scalable environmental monitoring.