Non-point source (NPS) pollution is still a serious issue in the governance of China’s water environment because the sources of pollution are scattered, non-continuous in time, and difficult to separate from point-source pollution. This review systematically organizes the progress of NPS pollution research from the beginning stage of exploratory observations to the current era of data-intensive intelligent modelling, and then presents the evidence in three analytical comparisons: research focus at different periods, operating attributes of the main watershed models, and correspondence between monitoring frequency and rainfall-driven pollution peaks. Although China’s surface-water quality has improved to some extent according to the review, the control of NPS is still limited by problems such as a lack of event-scale data, inconsistent monitoring standards, poor sharing of hydrometeorological and water-quality data, and insufficient localisation of foreign models. Although there have been recent achievements in high-resolution remote sensing, improved export coefficient refinement, SWAT-based localisation, digital twin watersheds, and hybrid AI-mechanistic models for load estimation and prediction, their application in practical governance still lacks a traceable management indicator. Future research should focus on standardised event-based monitoring, multi-source data fusion, localised and simplified model development, and transparently couple process-based models with machine-learning algorithms. The above directions will help promote better watershed management and more reasonable division of the NPS pollution control responsibility in China.
