Abstract: Non-point source (NPS) pollution has posed a serious risk in the watershed. It is often required for the driving mechanisms and response patterns of the NPS pollutants. In this study, the quantitative analysis was conducted on their load uncertainty associated with the NPS pollution. 74 representative catchments were located in the Mintuo and Jialing River basins, the tributaries of the upper Yangtze River. The dataset was collected from the long-term monitored hydrologic and water quality data, geospatial data of the underlying surfaces, and pollution source inventories. An uncertainty quantitative framework was presented to integrate α-shape envelope algorithms with the weighted least squares regression. The tail noise was controlled in the Delaunay triangulation network of the NPS dataset, according to the characteristic threshold. The boundary geometric sensitivity was simultaneously regulated to capture the spatial patterns and potential uncertainty regions within the dataset. NPS load variability was systematically characterized using three key parameters, i.e., the slope of the upper boundary envelope (the maximum potential loads), the slope of the lower boundary envelope (the baseline transport capacity), and the envelope angle (the range of load fluctuations). The variation range of the NPS loads was also quantified to explore the impacts of the different hydrological and underlying surface conditions on the uncertainty of NPS loads. The results revealed that there were consistent right-skewed log-normal distributions across all pollutants, hydrological-driven extremes for the ammonia nitrogen (NH₃-N), permanganate index (COD_Mn), and total phosphorus (TP) tied to <5% of storm events, and source-dominated total nitrogen (TN) maxima unaffected by rainfall intensity. A statistically significant positive correlation was observed between runoff volume and both the magnitude and variability of NPS (NPS) pollutant loads of NH₃-N, COD_Mn, TP, and TN, indicating the increasing load quantities and variation ranges under elevated runoff conditions. The variability of the upper boundary envelope angles for NH₃-N, COD_Mn, and TP NPS loads exceeded that of the lower boundary envelope angles. The range of the maximum loads for the four pollutants significantly surpassed the degree of variation in the average loads. NH₃-N and TP NPS loads exhibited higher sensitivity to changes in runoff. The uncertainty of NPS loads for the four pollutants demonstrated the significant interannual variability. The angles of both the lower and upper boundary envelopes for the four pollutants shared a clear increasing trend as the rainfall increased. Whereas the envelope angles decreased with the increasing rainfall. The increasing information content of the underlying surface led to a substantial rise in the uncertainty of the NPS loads. NH₃-N exhibited no significant correlation with the influencing factors, such as the river network density, vegetation coverage, and watershed area, while the uncertainty of the COD_Mn, TN, and TP NPS loads decreased with the increasing natural vegetation coverage. The uncertainty in the COD_Mn and TN loads was predominantly influenced by the underlying surface conditions, whereas the TP uncertainty exhibited the dual dependence on both runoff processes and surface conditions. The non-parametric Alpha Shapes algorithm was applied to quantify the flow-dependent variation in the NPS pollution load using multi-source datasets. There was some influence of the hydrological and underlying surface factors on the load uncertainty. The finding can also provide a scientific basis to identify the key sources of uncertainty in the basin and watershed.