In the granite regions of southern China, benggang poses a substantial threat to the ecological environment due to significant soil erosion. This phenomenon also imposes constraints on economic development, necessitating substantial investments in restoration efforts in recent decades. Despite these efforts, there remains a notable gap in comprehensive risk assessment that integrates both the erosion risk and disaster risk associated with benggang. This study focuses on a representative benggang area in Wuhua County, Guangdong province, employing transformer methods and high-resolution imagery to map the spatial pattern of the benggang. The integrated risk of benggang was assessed by combining soil-erosion risk and disaster risk, and cultivated land, residential land, and water bodies were identified as key disaster-affected entities. The machine-learning Segformer model demonstrated high precision, achieving an Intersection over Union (IoU) of 93.17% and an accuracy (Acc) of 96.73%. While the number of large benggang is relatively small, it constitutes the largest area proportion (65.10%); the number of small benggang is more significant (62.40%) despite a smaller area proportion. Prioritization for benggang management is categorized into high, medium, and low priority, accounting for 17.98%, 48.34%, and 33.69%, respectively. These priorities cover areas of 30.27%, 42.40%, and 27.33%, respectively. The findings of this study, which offer benggang management priorities, align with the nature-based solutions approach. Emphasizing the importance of considering costs and benefits comprehensively when formulating treatment plans, this approach contributes to sustainable solutions for addressing the challenges posed by benggang.