Identification and mapping of gully using machine learning and deep learning algorithms in the loess hilly and gully region in China

Gully erosion is a critical factor contributing to global soil erosion and land degradation, posing a significant threat to ecosystem sustainability. Effective and accurate monitoring of gully erosion at large regional scales is essential for devising soil and water conservation strategies. However, current methods for rapid gully extraction are lacking in the loess hilly and gully region of the Loess Plateau. This study leverages machine learning models [random forest (RF) and support vector machine (SVM)] and deep learning models (PSPNet-R50, PSPNet-R101, SegFormer-B0, and SegFormer-B5) to extract gullies from high-resolution remote-sensing images in this region. Notably, the SVM model with a linear kernel demonstrated high performance in gully recognition, while the RF model achieved commendable results with 14 trees and a maximum tree depth of 20. Deep learning models consistently outperformed machine learning models, with all metrics exceeding 0.9407. SegFormer-B5 had the highest recall, F1-score, and mIoU values at 0.9822, 0.9811, and 0.9615, respectively. Furthermore, validation of gully morphology parameters extracted using the SegFormer-B5 model revealed root mean square error (RMSE) values for measured and predicted gully area, perimeter, length, and maximum width as 139.19 m 2 , 58.63 m, 9.10 m, and 3.71 m, respectively. The strong correlation between predicted and measured values underscores the SegFormer-B5 model’s accuracy and integrity in gully identification in the loess hilly and gully region. This research presents an accurate and efficient methodology for gully detection in the region, offering valuable academic insights and practical guidance for soil erosion monitoring and slope and gully management in the Loess Plateau.