Drone-Based Bathymetry Modeling for Mountainous Shallow Rivers in Taiwan Using Machine Learning

Abstract

The river cross-section elevation data are an essential parameter for river engineering. However, due to the difficulty of mountainous river cross-section surveys, the existing bathymetry investigation techniques cannot be easily applied in a narrow and shallow field. Therefore, this study aimed to establish a model suitable for mountainous river areas utilizing an unmanned aerial vehicle (UAV) equipped with a multispectral camera and machine learning-based gene-expression programming (GEP) algorithm. The obtained images were combined with a total of 171 water depth measurements (0.01–1.53 m) for bathymetry modeling. The results show that the coefficient of determination (R2) of GEP is 0.801, the mean absolute error (MAE) is 0.154 m, and root mean square error (RMSE) is 0.195 m. The model performance of GEP model has increased by 16.3% in MAE, compared to conventional simple linear regression (REG) algorithm, and also has a lower bathymetry retrieval error both in shallow (<0.4 m) and deep waters (>0.8 m). The GEP bathymetry retrieval model has a considerable degree of accuracy and could be applied to shallow rivers or near-shore areas under similar conditions of this study.