Remote sensing inversion is a new method for measuring water depth, which has a positive effect on dam operation and sedimentation study in the reservoir area. In light of the defect in statistical correlation analysis that the water depth of a remote sensing pixel represented by a single bathymetric value, this paper presents a method which takes multi point data in one pixel standing for the depth of the pixel. Taking the Haibowan reservoir in Inner Mongolia as the pilot area,the inversion factor for water depth was selected to represent the correlation between water depth and the combination of each band, and fifteen groups of dual band and five groups of multi-band inversion models were set up with different number of inversion factors in three forms of linear, quadratic and exponential equations. The better five were chosen from them,and their accuracy was tested by the sample of detection points which were not involved in the modeling,and these models were compared by the test. The results show that the optimal water depth inversion model of Haibowan reservoir is a multichannel inversion model with 12 inversion factors, and the mean absolute error of which is 0.68 m,taking 13.59 % of the average water depth. Combined the characteristics of remote sensing in short cycle and lower cost, the model can be used in a certain extent, but the inversion error of water depth might be larger in the water areas with sediment concentration and near the land. © 2018, China Water Power Press. All right reserved.
