.png){kind=icon}
.png){kind=icon}
Return
BCDL-C2Net: a cloud-collaborative deep learning framework for large-scale bank collapse detection in the Yangtze River
.png){kind=icon}
OA
AI
K
J
J
S
B
S
L
DOI：10.3389/fenvs.2026.1885329.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Bank collapse is a natural phenomenon that is widespread along both sides of alluvial plain rivers; leading to instability in river courses and frequent changes in erosion and sedimentation; and causing a series of economic; environmental; and social issues; which have adverse effects on sustainable development. Traditional methods of bank collapse patrol are largely limited in scope and efficiency; making it difficult to detect hazards promptly. Utilizing satellite remote sensing imagery; particularly Chinese GAOFEN and Sentinel-1 data; can provide valuable insights for monitoring and early warning systems. To address these challenges; this study proposes a cloud-collaborative deep learning framework; termed Bank Collapse Deep Learning Cloud-Collaborative Network (BCDL-C2Net); which integrates multi-source satellite remote sensing data with dual cloud platforms; namely; Google Earth Engine (GEE) and Pixel Information Expert Engine (PIE-Engine); to enable efficient and scalable bank collapse detection. Within the BCDL-C2Net framework; GEE is employed for rapid; long-term (2004–2024) and large-scale bankline change screening using the Modified Normalized Difference Water Index (MNDWI) and Otsu threshold segmentation algorithm; while PIE-Engine is utilized to implement deep learning-based refined identification and verification of typical bank-collapse events. The main findings can be summarized as follows. (1) Based on Landsat-derived bankline changes; continuous bank collapse zones longer than approximately 2 km were identified through visual interpretation. The number of continuous bank-collapse zones increased from 15 on both the left and right banks during 2004–2010 to 37 on the left bank and 34 on the right bank during 2010–2015; and then decreased markedly after 2015; with collapse-prone reaches mainly distributed around bends; branching channels; and mid-channel bars. (2) The refined identification results show that the proposed framework can identify representative collapse events with scales of approximately 100 m × 60 m at Xiaopan and 80 m × 20 m at Tianzi No.1; and has practical potential for detecting typical bank collapses larger than approximately 60 m × 20 m. (3) By coupling GEE-based large-scale screening with PIE-Engine-based refined identification using GAOFEN and Sentinel-1 imagery; BCDL-C2Net provides a hierarchical technical pathway for operational bank collapse monitoring.
Keywords:
google earth engine
Yangtze River
cloud-based deep learning
PIE-engine
satellite remote sensing monitoring
Journal
IF:
3.7
Papers:
7.9K
Citations:
2.3W
Organization
Cited Papers