Search

High-resolution remote sensing imagery plays a strategic role in critical applications such as water resource management, water quality monitoring, and emergency responses to natural disasters for the quick and accurate identification and extraction of water bodies. However, traditional water body extraction methods present various challenges, particularly in the selection of image texture and characteristic features. In this study, a methodology is proposed that combines geometric features extracted from point cloud data with spectral information obtained from aerial photographs to more effectively define and extract the boundaries of water bodies. The geometric features generated from three-dimensional (3D) structure tensors are merged with the spectral information produced by the sensor system, and the well-known Random Forest (RF) classifier suitable for high-dimensional data, speed, and resistance to overfitting is used for training in water body detection. The effectiveness of the methodology developed in Matlab has been tested over four different locations in Turkey with varying topographic and vegetative characteristics. When the accuracy analysis of the detected water body boundaries is evaluated through the F-Score, the following were obtained: 85.7% for Study Area-1, 76.6% for Study Area-1 River, 93.7% for Study Area-2, 94.9% for Study Area-3, and 73.6% for Study Area-4. The study demonstrates that the presented methodology is applicable across different spatial scales and sensor types and carries potential for comprehensive uses in environmental and hydrological research. [ABSTRACT FROM AUTHOR]