Riverine Bathymetry Survey Using Unmanned Aerial Vehicle-Based Multispectral Image

Abstract

River basin surveys are conducted with the aim of providing essential foundational information for the formulation of water management policies, as mandated by relevant laws and regulations. These surveys cover key investigation areas necessary for river basin management, including basic conditions, water conveyance, dimensions, environmental ecology, and more. Among the survey methods, the utilization of remote sensing data, such as drone monitoring imagery and satellite imagery, is employed for various purposes such as the safety management of hydraulic structures like dams and embankments, water quality monitoring, river terrain surveys, and assessments of changes in riverbeds. Recently, research in river basin studies has been conducted using hyperspectral imagery, which includes hundreds of spectral bands, in addition to standard RGB imagery. Hyperspectral imagery offers the advantage of high spectral resolution, making it suitable for multi-parameter assessments. However, it comes with the drawback of large initial data volumes and complex preprocessing requirements due to the abundance of spectral information. On the other hand, multispectral imagery, which collects spectral information from fewer than ten bands, is widely used, especially in agriculture and forestry. It allows for immediate monitoring of parameters like the normalized difference vegetation index (NDVI) using just two bands and facilitates the analysis of crop growth status and more. Research on bathymetric estimation using hyperspectral imagery has traditionally relied on the Optimal Band Ratio Analysis (OBRA), which utilizes band ratios highly correlated with depth to construct bathymetric maps. In this study, we applied the existing hyperspectral bathymetric estimation technique to multispectral imagery to assess the feasibility of bathymetric estimation using reduced spectral bands. We captured multispectral imagery and constructed bathymetric maps to evaluate the applicability of multispectral imagery in river basin applications. Furthermore, to overcome the limitations of traditional OBRA, we employed Gaussian mixture models for image clustering to improve the accuracy of bathymetric estimation.