Increasing riverine phosphorus (P) levels in headwaters due to expanded and intensified human activities are worldwide concerns, because P is a well-known limiting nutrient for freshwater eutrophication. Here we adopt the conceptual framework of the SPAtially Referenced Regressions On Watershed attributes (SPARROW) model to describe total phosphorus (TP) sources and transport in a headwater watershed undergoing rapid agricultural expansion in the upper Taihu Lake Basin, China. Our models, which include variables for land cover, river length, runoff depth, and pond density, explain 94% of the spatio-temporal variability in TP loads. Agricultural lands contribute the largest percentage (61%) of the TP loads delivered downstream, followed by forestland (21%) and urban land (18%). Future agricultural expansion to 15% of the total basin area is possible, which could lead to a 50% increase in TP loads. According to our analysis, an average of 24% of the total P export from the watershed landscape was intercepted in ponds. The exported amount was subsequently retained by tributaries and along the mainstem river, accounting for 14% and 43% of their inflowing loads, respectively. The remaining ∼6 tons yr of TP was eventually transported into Tianmu Lake, in Southeastern China. The model identified several sub-catchments as hotspots of TP loss and thus logical sites for targeted management. Our study underscores the significance of agricultural expansion as a factor that can exacerbate headwater TP pollution, highlighting the importance of landscapes to buffer TP losses from sensitive hilly catchments. This also points to a need for an integrated management strategy that considers the spatial-varying P sources and associated transport of TP in precious headwater resources.