Unveiling environmental indicators of algal blooms using interpretable AI

Le résumé fourni par la source

Harmful algal blooms represent a significant environmental challenge with substantial ecological and economic impacts worldwide. Traditional predictive approaches often fail to capture the complex, nonlinear relationships between environmental drivers and algal proliferation. This research introduces the Algae-Net, a neural network model designed to accurately predict algal density and species co-occurrence patterns, which based on a high-resolution environmental monitoring dataset collected over four years (2022–2025, 31 consecutive months) in a river basin in Sichuan Province, China. The model achieved an average R 2 of 0.9778 for algal density prediction and algal coexistence prediction performance (micro-AUC: 0.8904) on test datasets. Through gradient-based attribution methods, we identified total nitrogen, temperature, and conductivity as the primary drivers of algal blooms, simultaneously analyzing their interactions. Algae species-specific environmental preferences were also elucidated, revealing distinct ecological niches. Notably, we found that increased conductivity mitigated the synergistic impact of temperature and nutrients on bloom development. TN concentration below 1.0 mg L −1 should be strictly maintained during high-temperature (>25°C) periods to prevent bloom formation under moderate electric conductivity. Our approach transcends traditional modeling by providing mechanistic insights into algal dynamics, offering water resource managers a robust tool for early bloom warning and targeted algae species intervention strategies.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.