Identification of Key Algal Species in a Large Lake Using a Multi-Method Approach

Identifying the key species contribute to potential harmful algal blooms (HABs) is essential for implementing targeted measures in HABs control. However, current approaches are limited by either low species resolution or poor spatial resolution for algal biomonitoring in large aquatic ecosystems. This study aimed to develop a novel approach to identify the key species of HABs with spatial distribution of harmful algae by integrating environmental DNA (eDNA), remote sensing, and machine learning (ML). The development of the approach was conducted in the Chinese largest freshwater lake, Poyang Lake. First, eDNA analysis revealed 606 and 104 ASVs for eukaryotic and prokaryotic algae, respectively, with Cyanophyta and Bacillariophyta being the predominant algal species in Poyang Lake during the sampling season. Subsequently, the gradient boosting tree (GBT) demonstrated the highest performance (average MAPE=11.20%) among the four ML models trained by the data from Sentinel-2 imagery, water quality, and algal eDNA. Next, the spatial genetic distribution maps of 34 algal species were generated using GBT. Finally, the contributions of algae species to algal community across different areas were identified by mapping the algal eDNA and the floating algae index (FAI) spatially. Nostocales and Stephanodiscales were identified as the key species driving FAI changes throughout Poyang Lake, with the toxic alga Nostocales exerting a greater influence in the northern lake than other species. This study demonstrated how large-scale species-level simulation of algal distributions can be achieved, which marked a significant step towards more comprehensive and refined monitoring of algal species and control of HABs.