摘要
【目的】探究Landsat 8多光谱影像结合地统计学方法估算乔木林地上碳储量的可行性和适用性,为应用Landsat 8多光谱影像结合地统计学方法估算区域森林参数提供参考。【方法】以浙江省内的一景Landsat 8多光谱影像覆盖的范围为研究区,以乔木林地上碳储量为研究对象。通过外业调查获取专项调查数据,并采用生物量转换因子和树种含碳率参数,计算得到乔木林地上碳储量数据。基于Landsat 8多光谱影像和DEM数据,提取植被指数、纹理特征、主成分变换因子、缨帽变换因子和地形因子,采用皮尔森相关系数法和方差膨胀因子法对这些因子进行优选,生成用于建模的自变量集。分别采用稳健估计和协同克里格插值法构建乔木林地上碳储量模型,并对所构建的模型精度进行对比分析。【结果】本实验所提取的因子经皮尔森相关系数法筛选后,得到22个自变量因子,经方差膨胀因子法优选后,共有7个自变量因子(比值植被指数、非线性植被指数、海拔、第2波段的平均值纹理、第4波段的相关性纹理、第7波段的平均值纹理、第7波段的方差纹理)用于建模。协同克里格插值法构建模型的决定系数(R2)为0.45、均方根误差(RMSE)为9.88 t·hm-2、平均绝对偏差(MAE)为7.75 t·hm-2、总预报偏差的相对误差(RE)为0.24%,其拟合精度优于稳健估计法(R2=0.42,RMSE=10.15 t·hm-2,MAE=8.03 t·hm-2,RE=0.27%)。本文所采用的皮尔森相关系数法结合方差膨胀因子法可有效地考虑变量间的相关性及共线性问题,可以在一定程度上提高所构建模型的稳定性,所采用的协同克里格插值法考虑了变量的空间分布特征,与传统的统计模型相比具有较好的应用优势。【结论】本研究为应用Landsat 8多光谱影像结合协同克里格插值法快速估算森林碳储量及其他森林参数提供了新的途径。
【Objective】The feasibility and applicability of Landsat 8 multispectral images combined with geo-statistics was studied in estimating the aboveground carbon storage of arbor forest,to provide a reference for the application of Landsat 8 multispectral images combined with geo-statistics in estimating regional forest parameters.【Method】Taking a Landsat 8 image in Zhejiang province as the research area,and taking the aboveground carbon storage of arbor forest as the research object.Based on the special survey data obtained from the field survey,the aboveground carbon storage of arbor forest was calculated by the biomass conversion factor and the carbon contents of the tree species.The vegetation indices,texture features,principal components,tasselled cap transformations and topographic factors were extracted from a Landsat 8 image and DEM data,meanwhile,the optical selection of these factors were carried out by Pearson correlation coefficient and Variance Inflation Factor(VIF)and the preferred factors were used to construct the models.Then,we estimated the aboveground carbon storage of arbor forest using Co-kriging method and robust regression and compared the accuracy of the two estimation methods.【Result】22 factors were obtained by selecting the factors by Pearson correlation coefficient,and 7 factors(Ratio Vegetation Index,Non-linear index,altitude,Mean texture of the 2nd band,Correlation texture of the 4th band,Mean texture of the 7th band,Variance texture of the 7th band)were obtained for constructing models after the selection by VIF method.The determination coefficient(R2),root mean square error(RMSE),mean absolute error(MAE)and relative error(RE)between estimated and measured values of model based on Co-kriging method were 0.45,9.88 t·hm-2,7.75 t·hm-2,and 0.24%.The R2,RMSE,MAE and RE values of model based on robust regression were 0.42,10.15 t·hm-2,8.03 t·hm-2,and 0.27%.So the carbon storage model constructed based on Co-kriging method was better than the robust regression,and the higher estimation accuracy was also acquired.Pearson correlation coefficient and VIF nethod adopted in this study can effectively consider the correlations and collinearities of variables,and can improve the stability of the constructed models to a certain extent.The Co-kriging method can consider the spatial distribution characteristics of factors and has better application advantages than traditional statistical models.【Conclusion】This study can provide a reference for the rapid estimation of forest carbon storage and other forest parameters using the Co-kriging method and Landsat 8 multispectral images.
作者
邱新彩
郑冬梅
王海宾
安天宇
许等平
郝月兰
彭道黎
QIU Xincai;ZHENG Dongmei;WANG Haibin;AN Tianyu;XU Dengping;HAO Yuelan;PENG Daoli(College of Forestry,Beijing Forestry University,Beijing 100083,China;Academy of Inventory and Planning,National Forestry and Grassland Administration,Beijing 100714,China;Planning&Design Institute of Forest Products Industry,National Forestry and Grassland Administration,Beijing 100010,China)
出处
《中南林业科技大学学报》
CAS
CSCD
北大核心
2020年第11期138-146,共9页
Journal of Central South University of Forestry & Technology
基金
国家林业和草原局948项目(2015-4-32)
国家重点林业工程监测技术示范推广项目([2015]02号)。
