There has been a growing concern on the discharge of nutrients from agricultural soils in polluting river water. A case study was conducted at the watershed level of two different agricultural land use and soil types in the eastern Hokkaido, Japan and the result was up-scaled to the regional level using ArcGIS and statistical information to predict NO₃-N concentrations by using land use analysis and N budget approaches. Water sampling was carried out from 62 and 49 sites of the main rivers and its tributaries incorporating the source to the outlet of each river system during the base flow event in August 2005 at Abashiri area (mainly agricultural farmland with Typic Dystrandepts) and Okoppe area (mainly grassland with Typic Dystrochrepts), respectively and NO₃ -N concentrations were measured. The proportions of upland agricultural land including grassland in the drainage basins were calculated by using the ArcGIS-Hydrology Modeling. The results showed that NO₃-N concentrations were significantly correlated with the proportions of upland field in the drainage basins at both the Abashiri (r=0.88, p<0.001, n=62) and Okoppe area (r=0.43, p<0.01, n=49). The regression slope was much higher for the cropland-based Abashiri area (0.025) than that of the grassland-based Okoppe area (0.009). The slopes indicated an impact intensity of upland field in polluting river water. Therefore, we defined the slope as impact factors (IF) of water quality. A comparison of these results with that of the previous study in Hokkaido indicated that the IFs were highest for intensive livestock farming (0.040), medium for mixed agriculture and livestock farming (0.020-0.030), and the lowest for grassland-based dairy cattle and horse farming areas (0.005-0.015). The results also showed that even in the areas of different soil types ranging from Typic Dystrandepts (79%), Typic Dystrochrepts (20%) to Andic Udipsamments (1%), the IFs could be well explained by the soil surplus N (r=0.92, p<0.01, n=9), which was estimated by using the N budget approach. Using the obtained regression model, the IFs for all cities, towns and villages of the Hokkaido region were estimated. The NO₃-N concentrations for all major rivers in Hokkaido were predicted by multiplying the estimated IFs by the proportion of upland fields. The predicted NO₃-N concentrations were significantly correlated (r=0.62, p<0.001, n=203) with the measured NO₃-N concentrations, reported previously. It can be concluded that estimating the proportions of upland fields in drainage basins, and calculating the soil surplus N enables us to predict river NO₃-N concentration and that its magnitude could be explained by soil surplus N regardless of soil types in the drainage basins.