1.State Key Laboratory of Soil and Water Conservation and Desertification Control,College of Soil and;Water Conservation Science and Engineering (Institute of Soil and Water Conservation),Northwest A&F University,Yangling,Shaanxi 712100,China
2.Institute of Soil and Water Conservation,Chinese Academy of;Sciences and Ministry of Water Resources,Yangling,Shaanxi 712100,China
Objective This study aims to construct a gully erosion susceptibility assessment model in order to provide an important scientific basis for early warning of gully erosion occurrence and its precise management in the black soil region of Northeast China. Methods Kedong County in Heilongjiang Province, a typical thick-layer black soil region, was selected as the study area. Erosion gullies were extracted based on 0.4 m high-resolution remote sensing images. Redundant variables were eliminated using the variance inflation factor (VIF). Four machine learning models covering linear, nonlinear, and ensemble learning architectures-logistic regression (LR), support vector machine (SVM), random forest (RF), and extreme gradient boosting (XGBoost)-were compared. The SHAP (Shapley Additive exPlanations) interpretability model was introduced to quantify the contribution of each influencing factor to gully erosion susceptibility. Results 1) XGBoost demonstrated the best predictive capability for gully erosion susceptibility (training set AUC=0.81, test set AUC=0.75), followed by the SVM and RF (test set AUC=0.74), while LR showed the weakest predictive capability (AUC=0.67). 2) The gully erosion susceptibility predicted by the four models all exhibited a spatial distribution pattern of "low in the periphery and high in the center" at the county scale, but there were significant differences in the susceptibility levels. The RF model predicted the largest proportion of high gully erosion susceptibility, accounting for 45.0%. The gully erosion susceptibility predicted by the SVM and XGBoost models in the study area was predominantly moderate and very low levels, accounting for 56.2% and 59.2%, respectively. In contrast, the LR prediction results indicated that low gully erosion susceptibility was dominant, covering 39.4% of the study area. 3) SHAP analysis revealed that the digital elevation model (DEM) was the dominant factor influencing gully erosion susceptibility (SHAP value=0.55), exhibiting a significant positive effect. The stream power index (SPI) had the second strongest positive driving effect on gully erosion susceptibility (SHAP value=0.30), while the normalized difference vegetation index (NDVI) showed a clear negative effect on gully erosion susceptibility (SHAP value=0.37). Conclusion In the thick-layer black soil region of Kedong County, significant differences are observed in the gully erosion susceptibility levels evaluated by the four machine learning models. The integration of the XGBoost model with the SHAP method enhances the interpretability of machine learning models in predicting gully erosion susceptibility. Moreover, gully erosion susceptibility is jointly driven by the coupled effects of "terrain-dominated control, vegetation regulation, and confluence hydrodynamics triggering". These findings deepen the understanding of gully erosion occurrence and provide theoretical guidance for optimizing gully erosion control measures in black soil regions.
ROC曲线越靠近左上角,代表模型在保持高真正率的同时拥有低假正率,即综合分类性能越优。作为ROC曲线的量化指标,曲线下面积(area under curve, AUC)被用于直接衡量模型的分类精度。AUC的取值在[0,1],其值越接近1,表明模型性能越好。本研究参考国际通用标准[27]:AUC>0.9时,模型精度极高;0.7<AUC<0.9时,模型精度较高;0.5<AUC<0.7时,模型精度较低。在本研究中,将综合运用混淆矩阵、准确率及ROC曲线与AUC值,全面评估和比较各模型的沟蚀预测能力。
ValentinC, PoesenJ, LiYong. Gully erosion: Impacts, factors and control[J].Catena,2005,63(2/3):132-153.
[2]
PoesenJ, NachtergaeleJ, VerstraetenG, et al. Gully erosion and environmental change: Importance and research needs[J].Catena,2003,50(2/3/4):91-133.
[3]
NyssenJ, PoesenJ, MoeyersonsJ, et al. Dynamics of soil erosion rates and controlling factors in the northern Ethiopian Highlands-towards a sediment budget[J].Earth Surface Processes and Landforms,2008,33(5):695-711.
[4]
GordonL M, BennettS J, AlonsoC V, et al. Modeling long-term soil losses on agricultural fields due to ephemeral gully erosion[J].Journal of Soil and Water Conservation,2008,63(4):173-181.
[5]
YangJiuchun, ZhangShuwen, ChangLiping, et al. Gully erosion regionalization of black soil area in northeastern China[J].Chinese Geographical Science,2017,27(1):78-87.
LiShan. The Ministry of Water Resources organized the investigation of erosion gully in the black soil area of Northeast China (excerpt)[J].Hydro Science and Cold Zone Engineering,2023,6(8):eF0003.
[8]
DongYifan, WuYongqiu, QinWei, et al. The gully erosion rates in the black soil region of northeastern China: Induced by different processes and indicated by different indexes[J].Catena,2019,182:e104146.
LiShanshan, WangChunmei, TangJie, et al. Development processes from ephemeral to permanent gullies and their critical terrain thresholds in a typical watershed of the mollisol region, Northeast China[J].Geomatics and Information Science of Wuhan University,2026,51(2):345-356.
LiuCongtan, FanHaoming. Research advances and prospects on gully erosion susceptibility assessment based on statistical modeling[J].Transactions of the Chinese Society of Agricultural Engineering,2024,40(4):29-40.
[13]
HitouriS, MeriameM, Sk AjimA, et al. Gully erosion mapping susceptibility in a Mediterranean environment: A hybrid decision-making model[J].International Soil and Water Conservation Research,2024,12(2):279-297.
[14]
GaoRuilu, GaoMaofang, YaoShuihong, et al. Gully erosion susceptibility mapping considering seasonal variations of NDVI using a machine learning approach in the mollisol region of China[J].Soil and Tillage Research,2025,245:e106322.
[15]
LiuCongtan, FanHaoming, JiangYanyan, et al. Gully erosion susceptibility assessment based on machine learning: A case study of watersheds in Tuquan County in the black soil region of Northeast China[J].Catena,2023,222:e106798.
[16]
LiJianjun, ChenYulan, JiaoJuying, et al. Gully erosion susceptibility maps and influence factor analysis in the Lhasa River basin on the Tibetan Plateau, based on machine learning algorithms[J].Catena,2024,235:e107695.
[17]
ShapleyL S. 17. a value for n-person games[M]∥Contributions to the Theory of Games (AM-28), Volume II. Princeton: Princeton University Press, 1953:307-318.
[18]
LundbergS M, LeeS I. A unified approach to interpreting model predictions[C]∥Proceedings of the 31st International Conference on Neural Information Processing Systems.2017:4768-4777.
[19]
白建宏.东北黑土区侵蚀沟分级初探[J].中国水土保持,2017(10):41-42.
[20]
BaiJianhong. Preliminary study on classification of erosion gully in black soil area of Northeast China[J].Soil and Water Conservation in China,2017(10):41-42.
[21]
ArabameriA, SahaS, RoyJ, et al. Landslide susceptibility evaluation and management using different machine learning methods in the gallicash river watershed, Iran[J].Remote Sensing,2020,12(3):e475.
[22]
HosmerD W Jr, LemeshowS, SturdivantR X. Applied logistic regression[M].Hoboken:Wiley,2013.
[23]
PourghasemiH R, YousefiS, KornejadyA, et al. Performance assessment of individual and ensemble data-mining techniques for gully erosion modeling[J].Science of the Total Environment,2017,609:764-775.
[24]
BreimanL. Random forests[J].Machine Learning,2001,45(1):5-32.
[25]
HoT K. The random subspace method for constructing decision forests[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,1998,20(8):832-844.
PumainD, KanevskiM, PozdnoukhovA, et al. 2009,Machine learning for spatial environmental data: Theory, applications and software, EPFL Press, 2009,377 p[J].Cybergeo,2010:371-377.
[28]
KanevskiM, TimoninV, PozdnukhovA. Machine learning for spatial environmental data: Theory, applications, and software[M].EPFL press,2009.
[29]
FriedmanJ H. Greedy function approximation: A gradient boosting machine[J].The Annals of Statistics,2001,29(5):1189-1232.
[30]
LiSijin, XiongLiyang, HuGuanghui, et al. Extracting check dam areas from high-resolution imagery based on the integration of object-based image analysis and deep learning[J].Land Degradation and Development,2021,32(7):2303-2317.
[31]
ArabameriA, ChenWei, LocheM, et al. Comparison of machine learning models for gully erosion susceptibility mapping[J].Geoscience Frontiers,2020,11(5):1609-1620.
[32]
GarosiY, SheklabadiM, ConoscentiC, et al. Assessing the performance of GIS- based machine learning models with different accuracy measures for determining susceptibility to gully erosion[J].Science of the Total Environment,2019,664:1117-1132.
[33]
HuangDonghao, ZhaoXinrui, YinZhe, et al. Utilizing geodetectors to identify conditioning factors for gully erosion risk in the black soil region of Northeast China[J].International Soil and Water Conservation Research,2024,12(4):808-827.
[34]
LiuXin, WangDichen, GuoMingming, et al. Determination of future gully erosion risk and its spatially quantitative interpretability of driving factors in regional scale using machine learning algorithms[J].Geoderma,2025,459:e117396.
LiJianjun, ChenYulan, JiaoJuying, et al. Detecting gully occurrence risks using multivariate nonlinear spatial modeling in the Lhasa River basin of China[J].Transactions of the Chinese Society of Agricultural Engineering,2022,38:73-82.
[37]
LiuCongtan, FanHaoming, WangYixuan. Gully erosion susceptibility assessment using three machine learning models in the black soil region of Northeast China[J].Catena,2024,245:e108275.
[38]
SørensenR, ZinkoU, SeibertJ. On the calculation of the topographic wetness index: Evaluation of different methods based on field observations[J].Hydrology and Earth System Sciences,2006,10(1):101-112.