基于梯度提升树模型的坡耕地土壤水蚀模拟与分析
DOI:
作者:
作者单位:

西北农林科技大学

作者简介:

通讯作者:

中图分类号:

基金项目:

国家自然科学基金项目(面上项目,重点项目,重大项目),国家重点基础研究发展计划(973计划)


Modeling and Analysis of Hydraulic Erosion in Slope Farmland Using Gradient Lifting Tree Model
Author:
Affiliation:

Northwest A&F University

Fund Project:

  • 摘要
  • |
  • 图/表
  • |
  • 访问统计
  • |
  • 参考文献
  • |
  • 相似文献
  • |
  • 引证文献
  • |
  • 资源附件
  • |
  • 文章评论
    摘要:

    由于人为干预强烈,坡耕地土壤侵蚀特征变化复杂多样,现有传统的侵蚀预报模型对其预测能力较差,而近年兴起的机器学习方法具备快速且高精度的预测能力。其中,梯度提升树模型(Gradient Boosting Decision Tree,GBDT)具备定量分析自变量影响力的功能,相对于其它机器学习模型具有明显的优势。本研究旨在利用黄土高原子洲地区坡耕地小区的1959-1969年间产流产沙观测数据,并精细化表征影响因子,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影响能力进行分析。数据集中次降雨侵蚀量(0-122.72 t/km2)、径流深(0.02-17.20 mm)、降雨历时(2-1410 min)及平均雨强(0.02-4.63mm)的变异系数都大于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,对次降雨土壤侵蚀量预测精度(R2=0.81)略优于径流深预测模型(R2=0.80),但次侵蚀量模型的层数(8层)大于径流深预测模型(5层),表明侵蚀机理相较径流机理更为复杂;受到特征提取的限制,在次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地侵蚀事件的预测。产流和产沙的主要影响变量存在差异,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。本研究基于数据驱动,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。

    Abstract:

    Based on the Gradient Lifting Tree Model (GBDT), a hydrological experimental dataset from the Zizhou Runoff Experimental Station in the Yellow River Basin was used to model and analyze hydraulic erosion on sloping farmland. The results showed that: 1. The coefficient of variation for secondary rainfall erosion (0-122.72 t/km2), runoff depth (0.02-17.20 mm), rainfall duration (2-1410 min), and average rainfall intensity (0.02-4.63mm) in the dataset are all greater than 1, indicating high variability. Most variables exhibit a right-skewed distribution.2. When dividing the dataset into training and testing sets, the model''s accuracy in predicting soil erosion during secondary rainfall (R2=0.81) is slightly higher than that of the runoff depth prediction model (R2=0.80). However, the number of layers in the secondary erosion model (8 layers) exceeds that of the runoff depth prediction model (5 layers), suggesting a more complex erosion mechanism compared to the runoff mechanism. 3. The prediction results are not ideal for small secondary erosion amounts and runoff depths due to limitations in feature extraction. Future research should explore additional combinations of independent variables to identify more relevant factors. 4. The main influencing variables differ between the erosion runoff and sediment production processes. Precipitation characteristics play a major role in runoff production, while erosion sediment production is mainly influenced by the combined effects of precipitation and terrain-related independent variables. This study, being data-driven, provides insights into the erosion mechanism of slope farmland in the Loess Plateau and serves as a scientific basis for sustainable regional development.

    参考文献
    相似文献
    引证文献
引用本文
分享
文章指标
  • 点击次数:
  • 下载次数:
  • HTML阅读次数:
  • 引用次数:
历史
  • 收稿日期:2023-09-05
  • 最后修改日期:2023-11-06
  • 录用日期:2023-11-16
  • 在线发布日期: 2024-04-29
  • 出版日期: