Abstract: The Chanhe River and Bahe River in Xi'an City has abundant water resources, however, influenced by the surrounding industrialization and urbanization, the water quality is poor. In recent years, the local government has paid many efforts on the water quality improvement. In order to evaluate the control effect, by selecting the downstream section of Chanhe River and Bahe River as study area, we firstly extracted water bodies based on Sentinel-2 satellite remote sensing images using the water body index method. Then, water quality inversion model of total nitrogen (TN) and permanganate index (CODMn) was constructed using artificial neural network algorithm (ANN) and random forest method (RF), and the temporal and spatial characteristics of these two parameters were finally analyzed. The results showed that the ANN had better performance than RF in inverting water quality parameters, and the ANN had well applicability in the study area with acceptable precision. The distribution of TN and CODMn values was relatively uniform with small fluctuation however some high values existed in local area. Meanwhile, both TN and CODMn values presented a significant seasonal trend, which was closely related to the human activities along the riverbank and upstream reaches.