文献类型：期刊文献

中文题名：基于多源遥感数据的灌丛化草原识别技术研究

英文题名：Identification of Shrub Encroachment in Grassland by Using Multi-source Remote Sensing Data

作者：张词谦[1,2,3] 孙斌[2,3] 洪亮[1] 高志海[2,3] 王丝丝[4]

第一作者：张词谦

机构：[1]云南师范大学地理学部,昆明650500;[2]中国林业科学研究院资源信息研究所,北京100091;[3]国家林业和草原局林业遥感与信息技术重点实验室,北京100091;[4]国家遥感中心,北京100036

年份：2022

卷号：43

期号：4

起止页码：123-137

中文期刊名：航天返回与遥感

外文期刊名：Spacecraft Recovery & Remote Sensing

收录：CSTPCD;;北大核心:【北大核心2020】；CSCD:【CSCD2021_2022】；

基金：国家高分重大科技专项(21-Y30B02-9001-19/22-3);国家自然科学基金(42001386)。

语种：中文

中文关键词：灌丛化草原;多源数据;多尺度;随机森林;多元二次模型;遥感应用

外文关键词：shrub-encroached grassland;multi-source data;multi-scale;random forest;multivariate quadratic Model;remote sensing application

分类号：TP79;Q948

摘要：为探索遥感技术在灌丛化草原空间分布识别方面的应用潜力,文章采用GF-2、GF-3和GF-6等多源遥感数据,利用随机森林算法和灌丛植被覆盖度估测模型,分别从分类识别和定量提取角度对内蒙古正镶白旗开展多尺度的灌丛化草原遥感识别技术研究。结果表明:1)GF-6的纹理信息和GF-3的极化信息对提取灌丛植被覆盖度信息贡献度最高,但在不同尺度下差异明显,整体上尺度较小时纹理信息占优,尺度增大时,极化信息优势凸显;2)二元二次灌丛植被覆盖度估测模型效果最优,在多尺度下均比次优的估测模型精度提高5.95%,23.84%和20.74%;3)基于随机森林算法的草原灌丛化分类识别准确率随尺度增大而降低,整体上均低于以灌丛植被覆盖度为基础的定量监测结果。该研究的部分成果可为我国北方草原的监督管理及合理利用提供一定技术支撑。
In order to explore the application potential of remote sensing technology in the recognition of spatial distribution of shrub-encroached grassland,combing the domestic multi-source remote sensing data such as GF-2,GF-3 and GF-6 to study the remote sensing technology in identification of shrub-encroached grassland at different scales from the perspective of classification identification and quantitative extraction respectively by using random forest algorithm and scrub cover estimation model.The results showed that:1)The texture information of GF-6 and polarization information of GF-3 are most effective in the contribution to scrub cover extraction,but there are obvious differences under different scales.Overall,the texture information is superior when the window is small,and as the scale increases,the advantage of polarization information extraction becomes prominent.2)The binary quadratic model has the best effect on the estimation of grassland scrub coverage,the accuracy of the model is improved by 5.95%,23.84%and 20.74%respectively compared with that of the sub-optimal model in multi-scales.3)The accuracy of the random forest algorithm-based grassland scrub recognition gradually decreases as the scale increases,and is overall lower than the results of the quantitative monitoring based on the scrub cover.Some of the results of this study can provide some technical support for grassland supervision,management and rational utilization in northern China.