Natural and anthropogenic determinants of riverine phosphorus concentration and loading variability in subtropical agricultural catchments

•Unique natural processes and anthropogenic activities induced moderate riverine P pollution in subtropical agricultural catchments.•Decreasing trend of riverine P levels was mainly caused by reduced catchment livestock density.•Landscape configuration and livestock density were major determinants of increasing riverine P.•Landscape and livestock planning need to occur simultaneously to ensure effective water quality improvements. Natural and anthropogenic determinants of riverine phosphorus (P) pollution strongly vary with different ecological and socio-economic conditions and represent a substantial challenge to aquatic environment protection and ecological restoration in agricultural catchments. Riverine water flux and total P (TP) and dissolved P (DP) concentrations were observed, and major catchment geometric, landscape and anthropogenic characteristics were investigated in one forest and ten agricultural catchments in subtropical China during 2011–2017 to identify the key determinants of the catchment riverine P levels and loadings and to develop prediction functions for these P levels and loadings. The results suggested that 39% of the collected stream water samples were determined to have TP pollution, indicating moderate riverine P pollution in the catchments. The catchment riverine P levels illustrated an increasing P trend during 2011–2016 but a decreasing P trend in 2017 through the Hodrick-Prescott (HP) filter analysis, and the decreasing trend was mainly caused by reduced catchment livestock density. Landscape configuration (cohesion index (COHESION), aggregation index (AI), mean shape index (SHMN), and patch density (PD)) and anthropogenic activities (livestock density) were identified as the major determinants of riverine P concentrations and loadings through partial least squares regression (PLSR) analysis, of which cohesion index and livestock density were the only two variables that had a variable importance for the projection (VIP) value greater than 1.0; this result confirmed the substantial effects of these variables on riverine P. The multiple linear regression prediction functions showed generally acceptable performances (R2 ≥ 0.40) for riverine P but relatively poor performance in predicting P loadings (R2 = 0.40–0.46) in comparison to predicting P concentrations (R2 = 0.51–0.54), probably due to the variations in the hydrological process in subtropical China. Therefore, the results provide useful information for identifying the na