Applying deep learning for boat detection and numerical modeling to assess sand mining impacts on river morphology: A case study in the Vietnamese Mekong Delta

Abstract

Effective management of riverbed sand mining is challenged by the lack of comprehensive data on sand mining volumes and their morphological impacts. This study presents an integrated framework combining deep learning, satellite imagery, and numerical modeling to monitor and assess the impacts of sand mining on river morphology in the Vietnamese Mekong Delta. A deep learning model was trained using Sentinel-1 imagery in 2023 to classify three boat types: Barge with Crane (BC), Sand Transport Boat (STB), and others. The model was then applied to detect BCs from 2014 to 2023, and the sand extraction volumes and areas were estimated. Finally, a Delft3D-FLOW model was employed to simulate the impacts of sand mining in the study period. Our deep learning model identified 386 BCs operating on the Bassac River in 2014–2023, with a total of 92.68–137.59 Mm³ of extracted sand, averaging 10.02–14.87 Mm³ annually. The numerical modeling results revealed significant riverbed incision, with a maximum annual net volume loss of −29.48 Mm³/yr and a mean erosion rate of up to −0.82 m/yr. In addition, excessive sand mining formed 23 scour holes with depths up to 11 m and incised the thalweg at rates of up to −1.18 m/yr. Sand mining maximally contributed 41.0–56.4 % of total riverbed incision during 2014–2023. These findings underscore the urgent need for improved sediment management strategies and regulatory frameworks. By providing a comprehensive assessment of sand mining impacts, this study supports the development of sustainable river management strategies in the region.