Spatial variability and future susceptibility of thaw hazards on hillslopes across the Qinghai-Tibet Plateau
Jiahui Yang, Yanchen Gao, Qingbai Wu, Miles Dyck, Francis Zvomuya, Xinrui Fang, Jia Zhou, Quojin Qi, Hailong He
Global and Planetary Change, 266: 105647 (2026)
论文解读
青藏高原多年冻土区山坡解冻危害广泛分布，对区域景观稳定性和基础设施安全构成威胁。本研究建立了基于文献资料的山坡解冻危害数据集，并通过目视解译识别非解冻危害样本。在此基础上，选取18个环境预测因子，利用9种机器学习和深度学习模型评估青藏高原多年冻土区山坡解冻危害的空间易感性，并结合SHAP方法识别关键预测因子。结果表明，NDVI、降水季节性、年均降水量和年均地温是影响山坡解冻危害易感性的关键环境因子。随机森林、Bagging、XGBoost和Gradient Boosting模型均表现出较高的预测性能，AUC均超过0.95。基于这4种高性能模型预测结果的统计集成均值显示，高易感性和极高易感性区域约占多年冻土区面积的5.53%，主要分布于青藏高原中部和东北部；低易感性和极低易感性区域占89.39%，主要分布于青藏高原南部。未来情景预测表明，低易感性和极低易感性区域略有扩大，而高易感性和极高易感性区域有所减少。本研究为青藏高原山坡解冻危害清单编制、寒区地质灾害制图以及北半球寒区风险评估提供了科学依据。
研究要点
- ✓建立了更加完善的山坡解冻危害综合评价体系
- ✓基于9种机器学习和深度学习模型开展多年冻土区山坡解冻危害易感性评估
- ✓利用18个环境预测因子和SHAP方法识别山坡解冻危害的关键驱动因素
- ✓随机森林、Bagging、XGBoost和Gradient Boosting模型均取得较高预测性能，AUC超过0.95
- ✓高易感性和极高易感性区域约占多年冻土区面积的5.53%
- ✓低易感性和极低易感性区域占多年冻土区面积的89.39%
引用格式
Jiahui Yang, Yanchen Gao, Qingbai Wu, Miles Dyck, Francis Zvomuya, Xinrui Fang, Jia Zhou, Quojin Qi, Hailong He (2026). Spatial variability and future susceptibility of thaw hazards on hillslopes across the Qinghai-Tibet Plateau. Global and Planetary Change, 266, 105647. https://doi.org/10.1016/j.gloplacha.2026.105647
@article{Yang2026,
author = {Jiahui Yang, Yanchen Gao, Qingbai Wu, Miles Dyck, Francis Zvomuya, Xinrui Fang, Jia Zhou, Quojin Qi, Hailong He},
title = {Spatial variability and future susceptibility of thaw hazards on hillslopes across the Qinghai-Tibet Plateau},
journal = {Global and Planetary Change},
year = {2026},
volume = {266},
pages = {105647},
doi = {10.1016/j.gloplacha.2026.105647},
}