南水北调与水利科技（中英文）2025，Vol.23Issue(2)：260-266,7.DOI:10.13476/j.cnki.nsbdqk.2025.0033
基于DEM的黄土塬区地貌形态类型划分方法
Automatic classification method of landform morphology in loess tableland based on DEM
摘要
Abstract
Accurate delineation of geomorphic patterns establishes a crucial basis for research into soil erosion,simulating sediment and water transport models,and preventing and controlling soil degradation in the loess plateau.The region's terrain is complex,and the various processes involved in soil and water interactions require precise geomorphological classification methods.Traditional approaches predominantly rely on image segmentation techniques based on geomorphological principles and deep learning algorithms.However,both methods present challenges,particularly the difficulty in directly classifying geomorphic feature surfaces.The extraction of such surfaces often depends on identifying geomorphic feature lines,such as shoulder-lines,which further complicates the process by increasing the complexity of the surface acquisition. To tackle these challenges,a systematic analysis of the geomorphic features typical of the loess tableland was conducted,leading to the development of a novel automated classification method based on digital elevation model(DEM)data.This method leverages DEM to generate data on drainage networks,slope gradients,and terrain curvature.By making use of the catchment relationship,the method traverses the topographic data and applies slope information to classify the geomorphology of the loess plateau into three distinct types:tableland,slope,and gully.Misclassifications are subsequently corrected using fuzzy terrain curvature adjustment,ensuring both fine-scale classification accuracy and the preservation of overall terrain continuity and consistency.This approach minimizes the impact of local noise and errors that could lead to classification deviations in conventional methods. The effectiveness of the proposed classification method was validated using a case study in the Nanxiaohegou watershed of Qingyang City,Gansu Province.The method was employed to classify the geomorphological features of the area based on DEM data with a resolution of 12.5 meters.The manual classifications derived from remote sensing images as compared to the automatic classification algorithm demonstrated a high degree of accuracy,achieving an overall classification accuracy of 88.4％,with a Kappa coefficient of 0.88.The intersection-over-union values for the various landform types were 0.899 for tableland,0.731 for slope,and 0.378 for gully,indicating superior performance relative to similar algorithms,although further refinement is needed for the gully classification. The proposed method offers a simplified yet highly effective means of classifying landforms,using catchment relationships and slope mutations to avoid the need for complex line feature extraction prior to surface feature identification.The results demonstrate the robustness of the method in the typical loess plateau regions,with its ability to accurately identify tableland,slope,and gully types.This approach not only enhances the precision of landform classification but also provides a reliable technical foundation for applications in soil and water conservation as well as sediment and water transport modeling.The method's potential to improve geomorphological analysis highlights its value in advancing both research and practical efforts in the field of landscape and erosion management.关键词
地貌形态/自动分类/DEM/黄土塬区/数字地形分析Key words
landform morphology/automatic classification/DEM/loess tableland/digital terrain analysis分类
建筑与水利引用本文复制引用
尤际州,周祖昊,刘佳嘉,王坤,于新哲..基于DEM的黄土塬区地貌形态类型划分方法[J].南水北调与水利科技（中英文）,2025,23(2):260-266,7.基金项目
国家自然科学基金黄河水科学研究联合基金项目(U2243601) （U2243601）
"十三五"国家重点研发计划项目(2016YFC0402405) （2016YFC0402405）
