eDNA and AI Identification Reveal Complementary Signals in Phytoplankton Monitoring

Phytoplankton regulates aquatic energy transfer and biogeochemical cycling, yet accurately characterizing community composition remains challenging because different monitoring methods capture complementary but incomplete signals. Here we present a comparative integration study combining environmental DNA (eDNA) metabarcoding, AI-assisted identification using YOLOv7 model, and manual identification to evaluate their complementary strengths in phytoplankton monitoring. Applied to a strongly regulated river-reservoir continuum in the Three Gorges region, a site-based field survey integrated eDNA, YOLO, manual identification and environmental covariates. Results support two hypotheses. First, the methods showed a division of strength. eDNA expanded taxonomic coverage and improved detection of rare taxa, whereas YOLO and manual identification provided abundance information for dominant morphotypes. Second, combining methods produced complementarity gains by increasing taxonomic among methods, and improving cross-method characterization of community structure and environment-community relationships. The integrated assessment identified two sites with elevated phytoplankton-related risk based on low diversity and high biomass, but this result should be interpreted as preliminary spatial screening rather than validated bloom early warning. Overall, our results show that molecular and morphology-based methods provide complementary ecological information, while broader application will require regional calibration, expanded taxonomic training coverage, and temporal validation.