基于EL-InVEST模型的玛多地震对高寒湿地面积及生境质量影响研究  

Effects of Maduo Earthquake on Alpine Wetland Area and Habitat Quality Based on EL-InVEST Model

作　　者：顾天江 杜凯[1,2,3] 毛旭锋 金鑫[1,2] 于红妍 唐文家[5] 吴艺 刘泽碧 GU Tianjiang;DU Kai;MAO Xufeng;JIN Xin;YU Hongyan;TANG Wenjia;WU Yi;LIU Zebi(Key Laboratory of Surface Process and Ecological Conservation of Qinghai-Tibet Plateau/Key Laboratory of Physical Geography and Environmental Process of Qinghai Province,Xining 810008,P.R.China;School of Geography Science,Qinghai Normal University,Xining 810008,P.R.China;Qinghai South of Qilian Mountain Forest Ecosystem Observation and Research Station,Huzhu 810500,P.R.China;Qinghai Qilian Mountain National Park Qinghai Service Guarantee Center,Xining 810008,P.R.China;Qinghai Provincial Department of Ecology and Environment,Xining 810008,P.R.China)

机构地区：[1]青藏高原地表过程与生态保育教育部重点实验室/青海省自然地理与环境过程重点实验室,青海西宁810008 [2]青海师范大学地理科学学院,青海西宁810008 [3]青海祁连山南坡森林生态系统国家定位观测研究站,青海互助810500 [4]青海祁连山国家公园青海服务保障中心,青海西宁810008 [5]青海省生态环境厅,青海西宁810008

出　　处：《生态环境学报》2025年第2期209-221,共13页Ecology and Environmental Sciences

基　　金：青海省自然科学基金青年项目(2024-ZJ-960);青海师范大学中青年科研基金项目(KJQN2022002)。

摘　　要：为准确评估2021年玛多地震对高寒湿地的影响,选取震中6度烈度带为研究区,依托Google Earth Engine(GEE)平台,融合Sentinel-2、Sentinel-1以及SRTM1数据,构建原始光谱、植被指数、水体指数、红边指数、纹理特征、地形、雷达特征7个遥感特征集,利用Stacking算法构建集成学习模型实现震前、震后的高寒湿地分类;结合InVEST模型实现地震前后生境质量的定量评估。结果表明,1)集成学习分类精度优于支持向量机和随机森林,特征优选方案下分类精度最高(92.741%,Kappa系数为0.902),较支持向量机和随机森林有所提高。2)在原始光谱的基础上加入纹理特征,湖泊湿地和河流湿地的分类精度分别可达0.987、0.933,加入地形特征后沼泽湿地分类精度可达0.857,纹理特征中的方差和相关性以及地形特征中的坡度、高程对高寒湿地分类效果影响显著。3)地震造成湿地面积减小,其中沼泽湿地减少43.088 km^(2),变化率为-0.254%;河流湿地减少31.654 km^(2),变化率为-3.522%;湖泊湿地面积下降4.971 km^(2),变化率为-0.303%。而裸地面积增加了36.160 km^(2)。4)生境质量均值从震前0.523下降到震后0.482,高等级生境质量面积占比从7.399%下降到5.993%,而低等级生境质量面积占比从2.191%上升到7.658%。该研究显示地震对高寒湿地产生负面影响,研究结果可为后续灾后生态恢复与管理提供依据。On May 22,2021,a 7.4-magnitude earthquake struck Maduo County,Guoluo Tibetan Autonomous Prefecture,Qinghai Province,China,causing damages to natural landscapes,such as wetlands,rivers,and lakes.Thus,this study aimed to accurately assess and quantify the impact of earthquakes on the alpine wetlands in the region.To address the research aim,we selected the 6-degree intensity zone of the Maduo earthquake as the research area,using the Google Earth Engine(GEE)as the platform,which was combined with Sentinel-2,Sentinel-1,and SRTM1 to extract research data.Remote sensing features,seven experimental schemes including the original spectrum,and vegetation index,original spectrum and water index,original spectrum and red edge index,original spectrum and texture feature,original spectrum and terrain feature,original spectrum and radar feature,and feature selection of random forest importance ranking.Finally,the habitat quality of the study area before and after the earthquake was systematically evaluated based on the classified alpine wetland data combined with the InVEST model.The influence of remote sensing features on alpine wetland classification was also explored.Changes in the alpine wetlands before and after the earthquake were quantified.Changes in habitat quality in the study area before and after the earthquake were quantitatively analyzed.The results showed that:1)the integrated learning model constructed based on the stacking algorithm was superior to the stochastic forest in terms of classification accuracy,and the stochastic forest was superior to the support vector machine in terms of classification accuracy.When ensemble learning was used for classification,the classification accuracy of the alpine wetlands reached 92.741%,and the kappa coefficient was 0.902.Compared to a single support vector machine or random forest model,the classification accuracy was improved to a certain extent.2)After adding textural features to the original spectrum,the classification accuracy of the lake wetland and river wetland re

关 键 词：遥感分类 Google Earth Engine 地震 高寒湿地 特征优选 哨兵2号 

分 类 号：X14[环境科学与工程—环境科学] TP79[自动化与计算机技术—检测技术与自动化装置]