In order to address the problem of vague responsibility quantification of water environment pollution in lake basin, which is difficult to manage and supervise accurately and scientifically, this paper adopts the Bayesian network structure and K2 algorithm to learn, and obtains the maximum number of parent nodes through the Maximum Support Tree (MWST), and then obtains the node order by the Depth-First Search Algorithm (DFS) to put forward a kind of improved MWST-DFS- algorithm that can quantify the responsibility of uncertain pollution sources in the watershed. K2 algorithm. Based on this algorithm, a Bayesian network model is constructed for the Erhai Sea as an example, and after analyzing the quantification of pollutants, it is concluded that Jiangwei station contributes more than 90% to the pollution of other stations in the watershed, and the probability that the water quality of the fourth-level dam station is less than Class II is 82%, and there are large water quality problems at the station itself, so the subsequent management process should be focused on the water quality of the Erhai Basin, which should be compared with the water quality of the fourth-level dam station and the lake inlet. In the subsequent management process, attention should be focused on the pollution sources around the Fourth Level Dam Station and the Jiangwei Station, which are hydrological stations in the lake. Compared with the traditional traceability methods, this method not only makes up for the lack of uncertainty analysis of pollution sources, but also quantifies the scientific pollution responsibility of pollution sources, which can provide a reference for the study of pollutant traceability in plateau lake basins.