This is the authors' abstract. We don't add key points for this paper.
Timely and reliable prediction of fish biodiversity across extensive river systems is pivotal for biodiversity monitoring and sustaining ecosystem stability. However, large-scale field surveys remain difficult, and broadband multispectral imagery captures habitat attributes only coarsely, limiting predictive precision. To address these challenges, Sentinel-2 multispectral observations, Gaofen-5A hyperspectral imagery, and the Google Satellite Embedding dataset were integrated with environmental DNA metabarcoding, and 7 machine learning algorithms were implemented to predict fish taxonomic diversity. A total of 149 environmental DNA samples were compiled from field campaigns and literature sources across 15 river basins in China. Habitat-informative vegetation indices were derived from pixel-level spectra, and 4 feature categories were examined, namely, multispectral variables, hyperspectral indices, deep embeddings, and fusion features combining hyperspectral information with satellite embeddings. Hyperspectral indices markedly improved prediction accuracy relative to that of Sentinel-2 multispectral predictors, achieving an R 2 of 0.80 ± 0.03 compared with 0.65 ± 0.07 for the multispectral baseline. Model performance remained stable across spectral resolutions using Gaofen-5A data. Notably, deep embedding features enabled multiple models to achieve an R 2 above 0.90, reflecting their strong capacity to integrate multisource information and characterize fish habitat conditions. Fusion of hyperspectral and embedding features produced the most accurate estimates of fish taxonomic richness with R 2 up to 0.94 ± 0.01. Overall, these outcomes demonstrate the feasibility and robustness of combining hyperspectral data with deep learned representations to predict riverine fish taxonomic richness, providing a multisource remote sensing pathway for biodiversity assessments across large spatial extents.
Answers are grounded in this paper and cite their sources.
No takes yet. Share an insight, caveat, or question.
Qi et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: