基于梯度提升树模型的坡耕地土壤水蚀模拟与分析  

Simulation and Analysis of Hydraulic Erosion in Sloping Farmland Based on Gradient Boosting Decition Tree Mode

作　　者：李潼亮 赵梓鉴 李斌斌[2] 张风宝[1,3] 郭正 何琪琳 何庆 杨明义[1,3] LI Tongliang;ZHAO Zijian;LI Binbin;ZHANG Fengbao;GUO Zheng;HE Qilin;HE Qing;YANG Mingyi(State Key Laboratory of Soil Erosion and Dryland Farming on the Loess Plateau,College of Water and Soil Conservation Science and Engineering,Northwest A&F University,Yangling,Shaanxi 712100,China;Water and Soil Conservation Monitoring Center,Ministry of Water Resources,Beijing 100053,China;Institute of Water and Soil Conservation,Ministry of Water Resources,Chinese Academy of Sciences,Yangling,Shaanxi 712100,China;University of Chinese Academy of Sciences,Beijing 100049,China)

机构地区：[1]西北农林科技大学水土保持科学与工程学院黄土高原土壤侵蚀与旱地农业国家重点实验室,712100,陕西杨凌 [2]水利部水土保持监测中心,北京100053 [3]中国科学院水利部水土保持研究所,陕西杨凌712100 [4]中国科学院大学,北京100049

出　　处：《水土保持学报》2024年第3期54-63,共10页Journal of Soil and Water Conservation

基　　金：国家自然科学基金项目(42077071,42177338,41830758);国家重点研发计划项目(2022YFF1300805);中央高校基本科研业务费专项资金项目(2023HHZX001)。

摘　　要：[目的]针对黄土高原坡耕地土壤侵蚀过程复杂、人为干扰强烈且难以量化的特点,利用机器学习定量解析主要影响因素对坡耕地土壤水蚀的作用与贡献,模拟分析坡耕地土壤水蚀特征并探究其机理,为坡耕地土壤侵蚀的预报提供基础支撑。[方法]基于黄土高原子洲试验站坡耕地小区1959—1969年产流产沙观测数据,精细化表征其影响因子,运用梯度提升树模型对侵蚀量和径流深的变化及其影响因素的贡献进行分析。[结果]数据集中次降雨侵蚀量(0~122.72 t/km^(2))、径流深(0.02~17.20 mm)、降雨历时(2~1410 min)及平均雨强(0.02~4.63 mm)属强变异,变异系数均>1,且多数变量呈右偏态;在相同训练集和测试集划分情况下,对侵蚀量模型预测精度(R^(2)=0.81)略优于径流深模型(R^(2)=0.80),但侵蚀量模型的层数(8层)大于径流深模型(5层),表明侵蚀机理相较径流机理更为复杂;通过梯度提升树模型与SHAP算法对自变量重要性进行排序发现,影响侵蚀模型与径流模型的自变量重要性不同。[结论]受特征提取的限制,在侵蚀量与径流深较小时预测结果不理想,未来研究应当通过引入更多自变量组合方式寻找更多相关变量以提高对侵蚀事件的预测。产流和产沙的主要影响因素存在差异,降水本身特征对产流过程起主要作用,侵蚀产沙过程中主要受到降水与地形相关自变量的共同影响。基于数据驱动,为揭示黄土高原坡耕地侵蚀机理提供参考,并为区域坡耕地土壤侵蚀防治提供科学依据。[Objective]This article employs machine learning to quantitatively analyze soil water erosion in Loess Plateau slope farmland,addressing its complexity and quantification challenges due to human interference.We aim to simulate erosion characteristics,explore its mechanisms,and support erosion prediction.[Methods]Using 1959-1969 data from Zizhou Experimental Station,we characterized the influencing factors and analyzed erosion and runoff depth changes with a gradient boosting decision tree.[Results]The dataset showed significant variability in secondary rainfall erosion(0~122.72 t/km^(2)),runoff depth(0.02~17.20 mm),rainfall duration(2~1410 min),and average intensity(0.02~4.63 mm),often right-skewed.The erosion model(R 2=0.81)slightly outperformed the runoff depth model(R 2=0.80),despite its greater complexity(8 layers vs.5).Using the gradient boosting tree model and SHAP algorithm,we found differing key factors for erosion and runoff.[Conclusion]Limitations in feature extraction lead to less accurate predictions for small erosion and runoff depths.Future research should explore more independent variable combinations to enhance predictions.Main influencing factors differ between runoff and sediment production.Precipitation mainly influences runoff,while erosion and sediment production depend on precipitation and terrain-related variables.In summary,this data-driven study illuminates slope farmland erosion mechanisms on the Loess Plateau,providing a scientific basis for erosion control in the region.

关 键 词：预报模型 梯度提升树模型 坡耕地 黄土坡面 

分 类 号：S157.1[农业科学—土壤学]