Decoding the transport thresholds of emerging contaminants in watersheds using explainable machine learning.

Understanding watershed emerging contaminants (ECs) transport is vital for pollution control but challenging due to complex land-climate interactions and limited models. This study collected 517 seasonal water samples from the Huangshui River (2020-2024) and quantified microplastics (MPs), antibiotics, heavy metals, and water quality indicators. A novel machine learning (ML-SHAP) framework was developed to model ECs transport (train R² = 0.94, test R² = 0.65), integrating multiscale land use (200, 500, 1000, 2000 m riparian buffers), landscape metrics (Patch Density (PD), Largest Patch Index (LPI), Contiguity Index Mean (CONTIG-MN)), and 11 climate variables. Overall, the water quality and heavy metals complied with Class III and Class I standards (GB3838-2002), respectively. However, MPs (1831 items/L) and antibiotics (55.33 ng/L) posed significant threats to regional water security. MPs transport was enhanced in fragmented urban land (PD > 1 in 2000-m buffer) and highly connected cropland (LPI > 50 in 500-m buffer), whereas antibiotic transport intensified in cropland with low landscape connectivity (LPI < 50 in 1000-m buffer). Notably, forest (cover > 45 % in 1000-m buffer) and grassland (CONTIG-MN > 0.5 in 500-m buffer) effectively mitigated ECs transport. Therefore, enhancing riparian forest and grassland connectivity while reducing urban fragmentation within a 2000 m buffer could substantially mitigate the transport of ECs. MPs transport increased under heavy rainfall (>6 mm) and low wind speeds (<1.2 m/s), while antibiotic concentrations rose under strong winds (>2 m/s), low rainfall (<2 mm) and weak solar radiation (<1.7 × 10⁷ J/m²). Climate warming under SSP585 increased MPs by 10.90 items/L and antibiotics by 0.007 ng/L per decade. Low-emission SSP245 with 40 % riparian reforestation reduced pollutants. These findings provide new mechanistic insights into ECs transport and offer a novel model for watershed ECs management.