Reduction of nutrient loads has long been a management focus of watershed restoration efforts in many regions, including the Chesapeake Bay. To better understand the regional patterns and drivers of total nitrogen (TN) and total phosphorus (TP) load trends across the Bay watershed, we analyzed TN and TP load data from the Chesapeake Bay Non-Tidal Network (NTN) stations, respectively, using advanced machine learning approaches1,2. Cluster analysis revealed regional patterns of short-term trends and categorized the NTN stations into distinct clusters. In addition, random forest models identified regional drivers of the trend clusters by quantifying the effects of major nutrient sources and watershed characteristics (i.e., land use, geology, physiography). Results showed evidence that improved agricultural nutrient management has resulted in declines in agricultural nonpoint sources, which in turn contributed to water-quality improvement in the period of analysis. Results also showed that water-quality improvements are less likely to occur in the Coastal Plain areas, reflecting the effect of legacy nutrients. To provide spatially explicit information for the entire watershed including unmonitored areas, the developed models were used to predict TN and TP trend clusters at the fine scale of river segments, which are more relevant to watershed planning. Across the Bay watershed, about two thirds of the river segments had declines in TN, TP, or both. Despite such progress, continued nutrient reductions, especially from agricultural nonpoint sources, are imperative in order to achieve the nutrient reduction goals. Overall, this work demonstrates that advanced machine learning approaches can help better understand the regional patterns and drivers of nutrient trends in large monitoring networks, providing insights directly applicable to watershed management and planning.