基于Landsat8的深圳市森林碳储量遥感反演研究

摘要

With Landsat 8 remote sensing images acquired in 2014 as datum source,remote sensing retrieval of forest carbon storage in Shenzhen was conducted and the spatial distribution was analyzed.A total of 168 sample plots were selected by a stratified random sampling procedure.A total of 31 vegetation indices were extracted from the images and used as independent variables,and urban forest carbon storage from field sampling plots was a dependent variable.Multivariate-stepwise regression model,Logistic regression model and radical basis function (RBF) neutral network model were developed to estimate forest carbon storage of the study area.The results showed that:the estimation accuracy of RBF neutral network model was the highest with the greatest determination coefficient and smallest root mean square error (RMSE) of 0.829 and 9.131 t · hm-2,respectively;the determination coefficient and RMSE of Logistic regression model were 0.723 and 11.821 t · hm-2,taking the second place.Multi-stepwise regression model had the lowest estimation accuracy with the determination coefficient and RMSE of 0.662 and 12.870 t · hm-2.Therefore,the relationship between urban forest carbon and image derived spectral variables could be modeled and described better by RBF neutral network model.The spatial distribution of forest carbon storage of the study area was characterized by larger estimates in the southeast coast and smaller estimates in the development zones of the mid-west city,being consistent with the actual spatial patterns of forests.%以2014年Landsat 8遥感影像为数据源,研究了深圳市森林碳储量遥感反演模型的构建及其空间分布情况,对城市生态系统碳循环研究具有重要意义.采用分层随机抽样的方式布设168个样地,结合外业样地数据,从遥感影像中提取31个植被指数作为自变量,分别构建了多元线性回归模型、Logistic回归模型和Radical Basis Function(RBF)径向基函数神经网络模型,进而估算该地区的森林碳储量并比较分析.结果表明,RBF神经网络模型的估算精度最高,决定系数最大且均方根误差最小,分别为0.829 t· hm-2和9.131 t· hm-2;Logistic回归模型估算精度次之,决定系数和均方根误差分别为0.523 t· hm-2和11.821 t·hm-2;多元线性回归模型估算精度最低,决定系数最小,均方根误差最大,分别为0.438t· hm-2和12.870 t·hm-2.可见,RBF神经网络模型能更好地模拟森林碳储量与各个因子之间的关系.研究区森林碳储量的空间分布特点表现为东南沿海部分碳储量大,中西部城市经济开发区碳储量小,与实际森林分布基本一致.