Reconstructing multi-decadal total phosphorus export from Chinese coastal rivers using Landsat and machine learning

Riverine phosphorus transport plays a pivotal role in regulating coastal biogeochemical cycles and food security, yet high-resolution spatio-temporal monitoring remains challenging in China. In this study, a satellite-based framework is presented to reconstruct total phosphorus (TP) fluxes from 92 Chinese coastal rivers over the period 1984-2018. The national TP flux is updated to 192 ± 37 × 10³ t P/yr, indicating that budgets based solely on major rivers potentially underestimate TP exports, as previously understudied small rivers account for ∼34% of the total national TP flux. The examined rivers show widespread decreasing trends in TP fluxes, with 62% exhibiting significant declines and only 13% showing increases (p < 0.05). Furthermore, the national TP flux exhibits a clear regime shift around 1998. From 1984 to 1998, TP flux increased significantly (6.3 × 10³ t/yr, p < 0.05), whereas it decreased significantly from 1998 to 2013 (slope = -5.9 × 10³ t/yr, p < 0.05), followed by stronger interannual variability during 2013-2018. An explainable machine learning attribution analysis further reveals that climate-driven hydrologic variability and shifting anthropogenic factors are important contributors associated with variations in China's coastal TP export, with reservoir-related variables showing strong associations with decreasing TP fluxes. This study presents a practical remote-sensing approach for spatio-temporal reconstruction of riverine TP fluxes and provides an observational benchmark to assess how changing TP delivery may influence coastal water quality and eutrophication risk across China's marginal seas.