A reservoir bathymetry retrieval study using the depth invariant index substrate cluster

Abstract

In this paper, bathymetry retrieval is combined with the Depth Invariant Index (DII) substrate cluster to acquire more accurate water depth. DIIs are calculated through the selected samples that are in bright and dark pixels firstly. Then, substrates are clustered with DIIs by using the K-MEANS cluster algorithm. Last, in-situ data and Genetic Algorithm (GA) are applied to solve the models’ parameters of the Stumpf model and the Legleiter model. The feasibility of this method is investigated in the Xia Shan Reservoir, Shandong Province, China. The experimental results show that (1) When there are various bottom types in the study area, the substrates cluster before bathymetry retrieval can significantly improve the retrieval accuracy. For example, in the without cluster case, the $R^2$ values are both around 0.72 in the GF-2 image and the $R^2$ values are both 0.53 in the Sentienl-2 image, and the minimum RMSE and RRMSE values are 1.09 m and 19.36 % respectively. When substrates are clustered into two clusters and three clusters, R² values have all increased and RMSE and RRMSE values decreased. (2) Clustering substrates into more clusters may not necessarily improve retrieval accuracy. For our research area, it’s better to divide the substrate into two clusters. For the two clusters case, the bathymetry result using the Legleiter model has a higher retrieval accuracy, which RMSE is 0.76 m, R² is 0.9 and RRMSE is 11.76 %. Compared with the three clusters case, the bathymetry retrieval accuracy of the two clusters case improves more obviously.