应用全极化合成孔径雷达数据构建多变量估算森林地上生物量模型  

Constructing a Multivariate Estimation Forest Above-Ground Biomass Model Using Full-Polarization SAR Data

作　　者：贾康 刘媛媛 范文义 Jia Kang;Liu Yuanyuan;Fan Wenyi(China Coal Aerial Surveying and Remote Sensing Group Co.,Ltd.,Xi’an 710199,P.R.China;Heilongjiang Province Natural Resources Rights and Interests Investigation and Monitoring Institute;Key Laboratory of Sustainable Forest Ecosystem Management,Ministry of Education(Northeast Forestry University))

机构地区：[1]中煤航测遥感集团有限公司,西安710199 [2]黑龙江省自然资源权益调查监测院 [3]森林生态系统可持续经营教育部重点实验室(东北林业大学)

出　　处：《东北林业大学学报》2024年第1期61-66,102,共7页Journal of Northeast Forestry University

基　　金：国家自然科学基金项目(31971654);民用航天技术预先研究项目(D040114)。

摘　　要：以河北塞罕坝机械林场为研究对象,Alos-2全极化SAR数据以及实测样地数据为基础,采用多种方法进行目标分解,计算多极化通道的后向散射系数,通过皮尔逊相关性检验,筛选显著相关的参数,构建线性及非线性偏最小二乘(PLS)生物量估算模型,并对模型进行检验。结果表明:线性PLS模型均方根误差为21.75 t·hm^(-2),决定系数(R^(2))为0.6185,平均相对偏差绝对值为31.1%,平均偏差为-1.844 t·hm^(-2);非线性PLS模型均方根误差为18.218 t·hm^(-2),决定系数(R^(2))为0.6969,平均相对偏差绝对值为23.86%,平均偏差为0.0001 t·hm^(-2)。对模型贡献最大的参数为非负特征值分解的奇次散射参数、VH极化后向散射系数、HV极化后向散射系数、特征向量分解参数l1和主要散射机制平均参数。非线性方法在模型拟合和预测中都表现出更好的效果,全极化SAR目标分解参数以及交叉极化下的后向散射系数,在估算森林地上生物量时没有饱和点出现。Taking the Saihanba Mechanical Forest Farm in Hebei as the research object,using Alos-2 fully polarization SAR data and field measurements as the basis.Multiple methods were employed for target decomposition and calculation of the backscattering coefficients of the multiple polarization channels.Significant correlation parameters were screened through Pearson correlation analysis to construct both linear and nonlinear partial least squares(PLS)biomass estimation models,which were then validated.The results showed that the root mean square error(RMSE)of the linear PLS model was 21.75 t·hm^(-2),with a coefficient of determination(R^(2))of 0.6185,an average absolute relative deviation of 31.1%,and an average deviation of -1.844 t·hm^(-2).The RMSE of the nonlinear PLS model was 18.218 t·hm^(-2),with an R^(2) of 0.6969,an average absolute relative deviation of 23.86%,and an average deviation of 0.0001 t·hm^(-2).The parameters that contributed the most to the models were the odd-order scattering parameters from non-negative eigenvalue decomposition,VH polarization backscattering coefficient,HV polarization backscattering coefficient,l1 parameter from eigenvalue decomposition,and average parameter lambda of the main scattering mechanism.The nonlinear method performed better in both model fitting and prediction.The fully polarization SAR target decomposition parameters and backscattering coefficients in cross-polarization did not exhibit a saturation point when estimating forest aboveground biomass.

关 键 词：偏最小二乘法 全极化SAR 极化分解 森林地上生物量 

分 类 号：S757[农业科学—森林经理学]