Research on Benggang identification and deformation monitoring based on optical and Radar data

Abstract

Benggang erosion, a severe soil erosion landform in southern China, threatens ecological security and regional development. Conventional methods can delineate surface morphology but lack capability to characterize vertical dynamics. To address it, this study presents an integrated method of deep learning and multi-temporal InSAR based on optical and radar data, and conducts benggang identification and surface deformation monitoring in typical benggang regions of Wuhua County, Guangdong, China. The results show that a U-Net model trained on Gaofen-2 high-resolution imagery achieved accurate automated Benggang delineation with a mean IoU of 86%. Subsequently, 90 Sentinel-1 SAR scenes acquired between 2022 and 2024 were processed using SBAS-InSAR and D-InSAR techniques, with Kriging interpolation employed to generate a spatially continuous deformation field. The framework successfully resolved millimeter-scale interannual vertical surface deformation, with internal cross-validation metrics (R 2  = 0.978, RMSE = 0.544 mm, MAE = 0.313 mm) confirming high spatial consistency of the fused deformation field. The proposed framework offers a reliable and scalable technical pathway for stereoscopic Benggang monitoring and demonstrates strong potential for geohazard early-warning and ecological risk management.