欢迎访问林业科学,今天是

林业科学 ›› 2026, Vol. 62 ›› Issue (11): 187-196.doi: 10.11707/j.1001-7488.LYKX20260089

• 研究论文 • 上一篇    

基于可解释机器学习的林道网骨干道路建设适宜性评价

龙晨1,2,王忠伟2,*(),王擎天3   

  1. 1. 湖南交通职业技术学院运输管理学院 长沙 410132
    2. 中南林业科技大学经济管理学院 长沙 410004
    3. 中南大学交通运输工程学院 长沙 410083
  • 收稿日期:2026-02-09 修回日期:2026-09-21 出版日期:2026-11-10 发布日期:2026-11-20
  • 通讯作者: 王忠伟 E-mail:wangpmp@163.com
  • 基金资助:
    智慧物流技术湖南省重点实验室项目(2019TP1015)。

Suitability Evaluation of Backbone Road Construction in Forest Road Networks Based on Explainable Machine Learning

Chen Long1,2,Zhongwei Wang2,*(),Qingtian Wang3   

  1. 1. School of Transportation Management, Hunan Communication Polytechnic Changsha 410132
    2. School of Economics and Management, Central South University of Forestry and Technology Changsha 410004
    3. School of Traffic & Transportation Engineering, Central South University Changsha 410083
  • Received:2026-02-09 Revised:2026-09-21 Online:2026-11-10 Published:2026-11-20
  • Contact: Zhongwei Wang E-mail:wangpmp@163.com

摘要:

目的: 针对南方丘陵山地林道网等级低、通行能力差等问题,挖掘已建路网隐含的工程选线智慧,构建骨干林道建设适宜性评价模型,为林道网扩容升级提供科学依据。方法: 以我国南方8省(区)122个国家森林公园为研究对象,提取3 726.14 km已建林道数据,作为承载工程实践经验的工程案例库。将符合高标准技术规范的主干道界定为骨干道路建设适宜样本,将受地形或需求限制的低等级支路界定为建设约束样本。选取涵盖地形地貌、生态环境和人类活动3个维度的9类特征因子,采用集成学习算法XGBoost建立道路等级分级评价模型,并引入SHAP解释方法,定量解析各因子对骨干林道等级决策的全局贡献和交互补偿效应,最后以中南林业科技大学芦头实验林场为例进行实证评价。结果: 1) XGBoost模型在识别骨干林道与支线林道的建设环境差异上表现优异,AUC值达0.8271,有效表征了隐含在已建林道网中的道路等级划分经验。2) 全局重要性分析显示,土壤类型、高程、到水系距离是驱动道路等级分异的主导因子。3) SHAP交互分析量化了复杂环境下的决策边界:高程与土壤类型存在显著的耦合增强效应,高海拔区的脆弱地质稳定性是限制道路升级的决定性因子;距水系500 m范围内存在明显的河谷补偿效应。4) 利用约登指数识别出骨干林道建设适宜区最优判别阈值为0.4494。结论: 林区道路等级的划分本质上是工程效益与环境成本博弈的结果。本研究构建的林道网骨干道路建设适宜性评价模型,实现了隐性知识的显性表征,能够精准识别具有升级为骨干林道潜力的区域,研究成果可科学指导林道网的提质升级。

关键词: 林道网, 空间分异, 建设适宜性, 机器学习, 适宜性评价, 可解释人工智能, 沙普利加和解释

Abstract:

Objective: To address the low road grades and poor traffic capacity of forest road networks in the hilly and mountainous areas of southern China, this study extracts the engineering route-selection knowledge embedded in existing road networks and develops a suitability evaluation model for backbone forest road construction, thereby providing a scientific basis for the capacity expansion and upgrading of forest road networks. Method: A total of 122 national forest parks across eight provinces (regions) in southern China were targeted, and data were extracted from 3 726.14 km of existing forest roads as an engineering case library. Main roads meeting high-standard technical specifications were defined as suitability samples for backbone road construction, while low-grade branch roads limited by terrain or demand were defined as construction constraint samples. Nine characteristic factors covering three dimensions, topography and geomorphology, ecological environment, and human activities, were selected. An ensemble learning algorithm, XGBoost, was used to establish a classification evaluation model. Furthermore, the SHAP (shapley additive explanations) method was introduced to quantitatively analyze the global contribution and interactive compensation effects of various factors on decision-making for backbone forest road grades. Finally, the Lutou Experimental Forest Farm of Central South University of Forestry and Technology was used as a case study for empirical evaluation. Result: 1) The XGBoost model performed excellently in identifying environmental differences between backbone forest road and branch forest road, achieving an AUC of 0.8271, which effectively characterizes the road classification experience latent in existing networks. 2) Global importance analysis showed that soil type, elevation, and distance to water systems were the dominant factors driving road grade differentiation. 3) SHAP interaction analysis quantified decision boundaries in complex environments: there was a significant coupling enhancement effect between elevation and soil type, and geological instability at high altitudes was the decisive factor limiting road upgrading. There was a distinct valley compensation effect within 500 m of water systems. 4) The Youden index was used for identifying the optimal discrimination threshold for backbone forest road suitability areas of 0.449 4. Conclusion: The classification of forest roads is essentially the result of a trade-off between engineering benefits and environmental costs. The suitability evaluation model developed in this study achieves the explicit representation of implicit knowledge and accurately identifies areas with the potential to be upgraded into backbone forest roads. These findings provide scientific guidance for the quality improvement and upgrading of forest road networks.

Key words: forest road network, spatial differentiation, construction suitability, machine learning, suitability evaluation, explainable artificial intelligence, shapley additive explanations (SHAP)

中图分类号: