Comparative Study of Different Machine Learning Models for Remote Sensing Bathymetry Inversion

Abstract

Water depth is an essential element of oceanographic research and marine surveying. Bathymetry inversion based on remote sensing is a time-effective, low-cost, and wide-coverage solution for shallow sea. Using WorldViewII multi-spectral remote sensing imagery and laser sounding data, Back Propagation neural network model (BP), random forest model (RF) and extreme learning machine model (ELM) were used to inverse water depth Surrounding the Chinese Ganquan island, and the inversion accuracy was compared and evaluated. The results show that among the BP, RF and ELM, the RF has the highest water depth inversion accuracy. The root mean square error (RMSE) of the check point is 0.85, the mean absolute error (MAE) is 0.60, and the mean relative error (MRE) is 3.54%. The coefficient determination R2 reaches 0.97; within the range of 0–10 m and 15–20 m water depth, the inversion of the ELM is the best; in the range of 10–15 m water depth, the RF has the better inversion effect.