Identifying Key Taxa for Algal Blooms in a Large Aquatic Ecosystem through Machine Learning

Identifying key species responsible for excessive growth of algae communities, as reflected by the floating algae index (FAI), is crucial for developing targeted management strategies to control algal blooms (ABs). However, current approaches for algal biomonitoring in large aquatic ecosystems are limited by either low taxonomic resolution or insufficient spatial coverage. To address these limitations, this study developed a supervised machine learning (ML) approach that integrates environmental DNA metabarcoding, remote sensing, and water quality parameters to identify the key algal bloom species and map their spatial distribution. Results demonstrated that the gradient boosting tree model achieved high predictive accuracy, with a mean MAPE of 11.20% across different algal taxa. Using this model, the spatial distribution maps were generated for 34 algal taxa. Prediction accuracy was further validated by comparing model outputs with morphological survey data, revealing a significant positive correlation (Spearman's correlation coefficient 0.366-0.709, <i>p</i> < 0.05) for 75% of the species. By integrating spatial mapping of algal distributions and FAI with principal component regression, the contributions of various algae taxa to the overall community structure were quantified across different regions. <i>Nostocales</i> and <i>Stephanodiscales</i> were identified as the key taxa driving FAI variations throughout Poyang Lake, with the toxic alga <i>Nostocales</i> exerting a greater influence in the northern region compared to other species. This study presents a novel framework for large-scale species-level simulation of algal dynamics, representing a significant advance toward more precise and comprehensive monitoring and management of algal blooms.