文章摘要
何 哲, 石玉玲, 李富春, 贾卓龙, 晏长根.基于LightGBM模型的甘肃省临夏县滑坡易发性评价Journal of Water Resources and Water Engineering[J].,2024,35(1):197-205
基于LightGBM模型的甘肃省临夏县滑坡易发性评价
Landslide susceptibility assessment based on LightGBM model in Linxia County, Gansu Province
  
DOI:10.11705/j.issn.1672-643X.2024.01.23
中文关键词: 滑坡  易发性评价  轻量级梯度提升机  机器学习  甘肃省临夏县
英文关键词: landslide  susceptibility evaluation  light gradient boosting machine (LightGBM)  machine learning  Linxia County of Gansu Province
基金项目:国家自然科学基金项目(42077265);甘肃省交通运输厅科技项目(2022-15)
Author NameAffiliation
HE Zhe1, SHI Yuling2, LI Fuchun3, JIA Zhuolong1, YAN Changgen1 (1.长安大学 公路学院 陕西 西安 710064 2.长安大学 地质工程与测绘学院 陕西 西安 7100543.中国二十二冶集团有限公司西北分公司 陕西 西安 710119) 
Hits: 854
Download times: 323
中文摘要:
      甘肃省临夏县地质环境复杂,滑坡灾害发育,对当地居民生产生活造成严重威胁,亦对工程建设的开展造成一定阻碍,因此,选取高效准确的机器学习方法对临夏县进行滑坡易发性评价具有重大意义。首先依据遥感影像和野外勘察资料,选取了1 718处滑坡样本,遴选了滑坡灾变的16种影响因子并建立滑坡影响因子评价体系;再结合预测精度和运行时间等指标对比了轻量级梯度提升机(LightGBM)模型与主流机器学习模型的性能;最后利用混淆矩阵分级方法进行了基于LightGBM模型的临夏县滑坡易发性评价。结果表明:临夏县重要滑坡影响因子为地表植被和地形地貌因子,其中土地覆盖为最主要影响因子;LightGBM模型预测精度高达0.931,且运行速度仅为11.7 s,既能保证高精度又极大提升了运行效率;在抽稀后的数据集上,LightGBM模型的预测表现、校准程度和分级结果均优于随机森林(RF)模型;混淆矩阵分级法的较高和高易发区内滑坡分布更为集中,在14.94%的区域内分布着86.86%的滑坡灾害点。滑坡易发性评价结果较好地反映了研究区内滑坡分布发育情况,可为当地工程建设及防灾减灾工作提供一定指导。
英文摘要:
      The complicated topographic status of Linxia County has led to the development of landslide disasters, posing a major threat to local residents’ lives and livelihoods as well as hindering the advancement of project construction. Thus, it is crucial to choose an accurate and effective machine learning technique for assessing Linxia County’s landslide susceptibility. Firstly, we selected 1 718 landslide samples based on remote sensing images and field investigation, and chose 16 influencing factors of landslide disaster to construct an evaluation system. Then the performance of the LightGBM model and the widely used mainstream machine learning models are evaluated from perspectives of prediction accuracy, running time and etc. Finally, the LightGBM model is utilized to evaluate the susceptibility of landslides in Linxia County using the confusion matrix categorization method. Results show that the main influencing factors of landslide in Linxia County are surface vegetation and topographic and geomorphic factors, among which land cover is the most influential factor. The prediction accuracy of LightGBM model can reach 0.931, and it takes merely 11.7 seconds to run, establishing an excellent performance of high precision and enhanced operational efficiency. On the extracted data set, the prediction accuracy, degree of calibration and classification results of the LightGBM model are better than those of random forest. According to the confusion matrix categorization, the landslide distribution is more concentrated in the high landslide-prone areas and extremely high landslide-prone areas, with 18.22% of the areas hosting 86.86% of landslide disaster locations. The evaluation results of landslide susceptibility are consistent with the distribution and development of landslides in the study area, which can provide some guidance for local engineering construction as well as disaster prevention and mitigation.
View Full Text   View/Add Comment  Download reader
Close
function PdfOpen(url){ var win="toolbar=no,location=no,directories=no,status=yes,menubar=yes,scrollbars=yes,resizable=yes"; window.open(url,"",win); } function openWin(url,w,h){ var win="toolbar=no,location=no,directories=no,status=no,menubar=no,scrollbars=yes,resizable=no,width=" + w + ",height=" + h; controlWindow=window.open(url,"",win); } &et=36C6DC8C7E4F18A0AB7A979107651833F11A3CAF03B3377C9D20C8577D65FAFA36DE7BCE47060BA586B0DBA058D2D443268255223CB0812A94B4960ED0E2A1DB028F0DF3B4F93C3B130CD644895FCB4A05EC0693A241863FCF25BBD79A318B74&pcid=5B3AB970F71A803DEACDC0559115BFCF0A068CD97DD29835&cid=3ECA06F115476E3F&jid=BC473CEDCB8CE70D7B12BDD8EA00FF44&yid=B80136CCD8DCBAA1&aid=5D9210679F6B7F818519A6C7A7FF4B23&vid=&iid=CA4FD0336C81A37A&sid=2BA123C6EB9D54C2&eid=EF27C460877D3C9F&fileno=20240123&flag=1&is_more=0">