The estimation of shallow water depth based on hyperspectral sensed dat a has been gradually matured as a reliable approach for detecting water column parameters wi th advantages of low cost and high accuracy.Semi-analytical model is widely used for many hyper spectral shallow water depth inversion models,among which maximum likelihood estimation including environmental noise and bottom intra-class variability (MILEBI) method is an important water dep th inversion method.One of the serious problems within the MILEBI method is the overestimation when the water depth is increased.Hense,an improved water depth inversion method is proposed in this paper based on a semi-analytical model.The method considers a prior distribution of depth b y including the influence of depth value on the subsurface reflectance,which may lead to an impr ovement of the accuracy for water depth inversion.The detailed solution is found by adding a re gularization term to the loss function of MILEBI method,and a new form of objective function is appli ed thereby.In this paper,a real dataset and a simulated dataset are used to evaluate the proposed me thod bench marked by the traditional approaches.The results show that in the scenario of water dep th range between 15—20 m,the mean absolute error of MILEBI is about 4.27 m,wher eas that of the proposed method is about 2.27 m,significantly better than the cl assic methods.