Bathymetric information for shallow coastal/lake areas is essential for hydrological engineering applications such as sedimentary processes and coastal studies. Remotely sensed imagery is considered a time-effective, low-cost, and wide-coverage solution for bathymetric measurements. This study assesses the performance of three proposed empirical models for bathymetry calculations in three different areas: Alexandria port, Egypt, as an example of a low-turbidity deep water area with silt-sand bottom cover and a depth range of 10.5 m; the Lake Nubia entrance zone, Sudan, which is a highly turbid, unstable, clay bottom area with water depths to 6 m; and Shiraho, Ishigaki Island, Japan, a coral reef area with varied depths ranging up to 14 m. The proposed models are the ensemble regression tree-fitting algorithm using bagging (BAG), ensemble regression tree-fitting algorithm of least squares boosting (LSB), and support vector regression algorithm (SVR). Data from Landsat 8 and Spot 6 satellite images were used to assess the performance of the proposed models. The three models were used to obtain bathymetric maps using the reflectance of green, red, blue/red, and green/red band ratios. The results were compared with corresponding results yielded by two conventional empirical methods, the neural network (NN) and the Lyzenga generalised linear model (GLM). Compared with echosounder data, BAG, LSB, and SVR results demonstrate higher accuracy ranges from 0.04 to 0.35 m more than Lyzenga GLM. The BAG algorithm, producing the most accurate results, proved to be the preferable algorithm for bathymetry calculation. |
