College of Geology Engineering and Geomatics, Chang'an University, Xi'an 710054, China.
College of Geology Engineering and Geomatics, Chang'an University, Xi'an 710054, China; Key Laboratory of Western China's Mineral Resources and Geological Engineering, Ministry of Education, No. 126 Yanta Road, Xian 710054, China,.
Sci Total Environ. 2021 Aug 20;783:147040. doi: 10.1016/j.scitotenv.2021.147040. Epub 2021 Apr 14.
Gully head erosion significantly contributes to land degradation, and affects gully dynamics on the Loess Plateau of China. Modeling with a gully head erosion susceptibility map (GHEM) is an essential step toward appropriate mitigation measures. This study evaluates the effectiveness of two varieties of advanced data mining techniques-a bivariate statistical model (certainty factor (CF)) and a machine learning model (random forest (RF)) for the accurate mapping of gully head erosion susceptibility taking the Dongzhi Loess Tableland of China as an example at a regional scale. A database comprising 11 geographic and environmental parameters was extracted with 415 spatially distributed gully heads, of which 70% (291) were selected for model training and 30% (124) were used for validation. An accuracy evaluation using the area under the curve (AUC) value revealed that the CF model was 84.1% accurate, while the AUC value of the RF model map was 88.8% accurate. According to the RF method used to assess the relative significance of predictor variables, the most significant factors influencing the spatial distribution of the GHEM were the slope angle, slope aspect, topographic wetness index, and slope length. The GHEM can ultimately aid in decision making with respect to soil planning and management and sustainable development of the study area, which can be applied to other similar loess regions.
沟头侵蚀严重导致土地退化，并影响中国黄土高原的沟谷动态。通过沟头侵蚀敏感性图（GHEM）进行建模是采取适当缓解措施的重要步骤。本研究以中国董志塬黄土台地为例，在区域尺度上，评估了两种先进数据挖掘技术的有效性-二元统计模型（确定性因子（CF））和机器学习模型（随机森林（RF）），用于准确绘制沟头侵蚀敏感性图。该数据库包含 11 个地理和环境参数，提取了 415 个空间分布的沟头，其中 70%（291）用于模型训练，30%（124）用于验证。使用曲线下面积（AUC）值进行准确性评估表明，CF 模型的准确率为 84.1%，而 RF 模型图的 AUC 值的准确率为 88.8%。根据用于评估预测变量相对重要性的 RF 方法，影响 GHEM 空间分布的最重要因素是坡度角、坡度方向、地形湿度指数和坡度长度。GHEM 最终可以帮助制定与土壤规划和管理以及研究区可持续发展有关的决策，可应用于其他类似的黄土地区。
