Water conservation in forest ecosystems has ecological functions such as regulating climate and maintaining ecological water balance. As an alpine region, the Qinghai-Tibet Plateau cannot manually observe water conservation on the spot due to its high altitude and harsh environment. In order to better obtain water conservation in alpine regions, remote sensing technology is introduced, and the value of water conservation in a specific area is obtained through remote sensing inversion. This study takes Nyingchi Bayi District as the study area. The study area uses four types of vegetation: Nyingchi Spruce, Alpine Quercus, Alpine Pine and Snow Rhododendron as the main tree species. Remote sensing images cannot directly obtain water conservation information, but the value of water conservation can be inverted by constructing a quantitative relationship between vegetation leaf spectral information and water conservation. To study the quantitative relationship between different vegetations and water conservation, collect 1 000 leaf samples and water conservation data from 10 sampling points for each vegetation, use ASD spectrometer to obtain hyperspectral data, select fitting parameters through correlation, and build a regression model of water conservation. The Sentinel-2 remote sensing image was used to invert the water conservation distribution of vegetation in the study area, and the inversion results were verified. The results show that the reflectance spectra of the four types of vegetation leave all show similar regularities. The difference is not obvious in the visible light band, and there are four obvious water absorption bands in the near-infrared to the mid-infrared band (700~1 400 nm), and the reflectivity in the red to near-infrared band highest. The spectral reflectance showed the order of Alpine Quercus>Alpine Pine>Lingzhi Spruce≈Snow Rhododendron. Through experiments, the vegetation canopy interception, litter water holding capacity and soil water content are obtained. The sum of the three represents the water conservation capacity of the vegetation, and the relationship between the spectral characteristics of the vegetation and the water conservation capacity is analyzed. Moreover, through the Pearson coefficient to evaluate the quantitative relationship between band parameters and water conservation, it is determined that the four parameters R₅₄₀, R_{1 950}, NDWI and NDVI are significantly related to water conservation. Based on the above parameters and the water conservation of the four types of vegetation, a regression model of water conservation was constructed. The vegetation water conservation in the study area was inverted through the model to verify the simulation accuracy. The overall inversion accuracy R² is greater than 0.7, and the RMSE is less than 10. It shows that the prediction model has a good inversion effect, and the model can effectively estimate the water conservation of the forest ecosystem.