Zooplankton release substantial amounts of dissolved organic matter (DOM), yet their contribution to the molecular composition of natural aquatic DOM pools remains poorly constrained. Here we integrate zooplankton-derived DOM molecular formula from incubations with machine learning models to predict zooplankton contribution toward aquatic DOM composition. Among five tested models, random forest (RF) outperformed linear and other nonlinear approaches, achieving high predictive accuracy (R² = 0.815 for both zooplankton‑ and non‑zooplankton‑derived classes), indicating that DOM source attribution is governed by nonlinear molecular feature interactions. RF‑predicted zooplankton‑derived molecules exhibited lower molecular mass and aromaticity (AIₘₒd, DBE), higher H/C ratios, and enrichment in protein‑ and lipid‑like compounds, consistent with a more labile DOM molecules. SHAP feature importance identified S/C ratio, m/z, phosphorus content, DBE, and O/C as key discriminators, confirming that the model captures mechanistic biochemical signatures. Applying the trained model to 16 water‑column samples spanning an estuarine–coastal gradient revealed that zooplankton‑derived molecules consistently comprised 16.5–20.1% (mean 18.3%) of total assigned molecular intensity. Remarkably, the predicted zooplankton‑derived DOM proportion exhibited a clear day‑night reversal in vertical distribution, surface enrichment at night (mean surface–bottom difference +1.45%, p = 0.044) and bottom enrichment during the day (–1.01%, p = 0.020), consistent with the diel vertical migration behaviour of zooplankton. Together, these findings demonstrate that zooplankton imprint a persistent and behaviourally dynamic molecular signature on estuarine coastal DOM pools. Our scalable machine‑learning framework provides a semiquantitative foundation for tracing zooplankton‑derived DOM in aquatic systems, enabling future assessments of how zooplankton contribute to shape DOM molecular fingerprint.
 

暂无