耦合光谱、纹理信息的森林蓄积量估算研究

摘要

In this study,an inversion analysis of the forest growing stock volume in Linzhi County was conducted u-sing a multiple regression method.It collected Landsat 8 remote sensing images,forest inventory data(by field sur-vey organized by forest institutions on a local and national basis),and DEM from Linzhi,Tibet,China.To determine if inclusion of texture feature information into the regression would improve retrieval accuracy for our proposed inver -sion analysis model,the texture feature information in Landsat 8 was extracted using a Gray Level Co-occurrence Matrix(GLCM).Relationship between the varied bands of remote sensing images,vegetable indices,texture fea-tures,topographic factors of the Landsat,and the forest growing stock volume was analyzed and consequently inver-sion models were established according to the revealed correlation.Three inversion models for determination of for-est growing stock volume were introduced by combinations of different independent variables,namely the spectra, topographic and texture feature factors.The first multiple regression equation was organized with two variables,the spectra factor and the topographic factors;the second lied in the texture feature factor only;the last one was com-plete model with all three included.It found that the accuracy of the traditional regression model which consisted of spectra and topographic factors,was not satisfied with precision,but that of the third model established here was 80.24%,with an RMSE of 1.018,the highest among the three models.This indicated that with the introduction of texture feature information into our suggested inversion model,it increased retrieval accuracy significantly,leading to a percentage of 10.06%improvement,from 0.5843 to 0.7075 for the coefficient of total correlation.It suggested that a regression model containing three variables,the spectra,topographic and texture feature was reliable to deter-mination of the forest growing stock volume,and it would be of great significance in monitoring and management of forest resources.%本研究以Landsat 8为遥感数据源,以样地调查数据和森林资源二调数据为辅助数据对西藏林芝县的森林蓄积量进行反演研究.研究通过多元回归分析构建了林芝县森林蓄积的估算模型.为验证纹理信息的加入能否提高森林蓄积量遥感反演的精度,研究通过灰度共生矩阵提取了Landsat 8的纹理特征.在分析了森林蓄积量与遥感影像各波段、植被指数、纹理特征以及地形因子之间的相关关系后,分别以(1)光谱和地形因子、(2)纹理信息、(3)光谱因子、地形特征和纹理特征结合为自变量构建森林蓄积量的遥感估测回归模型.实验结果表明:传统的森林蓄积量反演方法得到的精度最低,而基于光谱因子、地形特征和纹理特征结合的森林蓄积量估测模型得到结果的精度最高,达到80.24%,均方根误差RMSE为1.018.研究结果证明随着纹理信息的引入,原本仅基于光谱和地形因子的森林蓄积量反演复相关系数从0.5843提高到0.7075,反演精度提高了10.06%,这说明纹理信息对森林蓄积的反演精度有提高的作用.本研究构建的基于光谱因子、地形特征和纹理特征结合的回归模型对研究区内的森林蓄积量反演具有可靠性,对于森林资源的监测和管理具有重要的意义.