GF-3全极化SAR数据极化分解估算人工林冠层生物量

Polarimetric Decomposition Parameters for Artificial Forest Canopy Biomass Estimation Using GF-3 Fully Polarimetric SAR Data

【目的】探讨GF-3全极化SAR数据在人工林冠层生物量估算中的潜力,提出一种准确估算森林冠层生物量的方法。【方法】以内蒙古赤峰市旺业甸林场油松和华北落叶松人工林为研究对象,以GF-3全极化SAR数据为基础,结合地面实测22块样地数据,采用Freeman三分量分解、Freeman二分量分解、Yamaguchi三分量分解3种极化分解方法获得极化分解分量,分别构建各极化分解方法所对应的冠-地散射比参数(R 1、R 2和R 3),应用多元逐步回归方法建立森林冠层生物量与SAR提取参数回归模型,并运用留一法交叉检验对模型进行评价。【结果】不同极化分解方法所得分量与冠层生物量均存在较为显著的负相关关系,Freeman三分量分解体散射分量与冠层生物量的相关性(r=-0.68)高于二次散射分量(r=-0.6)和表面散射分量(r=-0.424),类似地,Freeman二分量分解体散射分量与冠层生物量的相关性(r=-0.718)高于地面散射分量(r=-0.62),而Yamaguchi三分量分解二次散射分量与冠层生物量的相关性(r=-0.743)最高,与Freeman三分量分解相比,Freeman二分量分解、Yamaguchi三分量分解的极化分解分量与冠层生物量具有更好的相关性。应用多元逐步回归方法获得的最优参数为Freeman二分量分解和Yamaguchi三分量分解对应的冠-地散射比参数R 2和R 3,建立的冠层生物量估算模型R 2=0.658,RMSE=4.943 t·hm-2;交叉验证结果表明,模型预测误差较低(ME=-0.665 t·hm-2,MAE=4.845 t·hm-2,MRE=3.33%,AMRE=23.233%,P=91.5%),且模型通过置信椭圆F检验,模型预测值与实测值一致,模拟结果较好,预测值大致分布在1∶1直线附近,模型未出现饱和点。【结论】Freeman三分量分解、Freeman二分量分解、Yamaguchi三分量分解3种极化分解方法获得的极化分解分量均与冠层生物量具有显著相关关系,极化相干矩阵旋转变换、体散射模型优化可有效提高森林区域极化分解效果,冠-地散射比参数对冠层生物量的敏感性高于任何单一极化分解分量,多种SAR极化分解参数共同使用能够较好估算冠层生物量。极化分解估算森林冠层生物量具有可行性,且不存在明显的饱和性问题。

【Objective】The object of this study was to explore the potential of GF-3 fully polarized SAR data in the estimation of artificial forest canopy biomass and to make an attempt to propose an accurate estimation method for forest canopy biomass.【Method】In this study,based on the data from 22 sample plots of Pinus tabulaeformis and Larix principis-rupprechtii artificial forest in Wangyedian Forest Farm,Chifeng,Inner Mongolia and from GF-3 fully polarimetric SAR,Freeman three-component decomposition,Freeman two-component decomposition and Yamaguchi three-component decomposition were used to obtain polarization decomposition components.The volume-to-ground scattering ratio parameters(R1,R2 and R3)corresponding to each polarization decomposition method were constructed respectively.The multiple stepwise regression method was used to establish regression models for forest canopy biomass and SAR extraction parameters,and the model was evaluated using the leave-one-out cross test.【Result】There were significant negative correlations between the components obtained by different polarization decomposition methods and the canopy biomass.The correlation between the volume scattering component of the Freeman three-component decomposition and the canopy biomass(r=-0.68)was higher than that between the double-bounce scattering component and the canopy biomass(r=-0.6),and also higher than that between the surface scattering component and the canopy biomass(r=-0.424).Similarly,the correlation between the volume scattering component of the Freeman two-component decomposition and the canopy biomass(r=-0.718)was higher than that between the ground scattering component and the canopy biomass(r=-0.62).The correlation between the double-bounce scattering component of Yamaguchi three-component decomposition and the canopy biomass was the highest(r=-0.743).Compared with Freeman three-component decomposition,the decomposition components of Freeman two-component decomposition and Yamaguchi three-component decomposition had better correlations with the canopy biomass.The optimal parameters obtained by stepwise regression were the volume-to-ground scattering ratio parameters R2 and R3 corresponding to Freeman two-component decomposition and Yamaguchi three-component decomposition,respectively.Then the canopy biomass estimation model was established(R2=0.658,RMSE=4.943 t·hm-2).The cross-validation results of the model showed that the prediction error of the model was relatively low(ME=-0.665 t·hm-2,MAE=4.845 t·hm-2,MRE=3.33%,AMRE=23.233%,P=91.5%).The model was tested by means of confidence ellipse F test,and the estimated values were consistent with the measured values.The simulation results were good,and the predicted values were generally distributed near the 1∶1 line.No saturation point appears in the model.【Conclusion】The decomposition components obtained by the three polarization decomposition methods of Freeman three-component decomposition,Freeman two-component decomposition,and Yamaguchi three-component decomposition were significantly correlated with the canopy biomass.The rotation transformation of polarization coherence and the optimization of volume scattering model could effectively improve polarization decomposition effects of forest area.The volume-to-ground scattering ratio parameters were more sensitive to the canopy biomass than each single polarization decomposition component.Collaborative use of multiple polarization decomposition components could estimate forest canopy biomass more accurately.Polarization decomposition was feasible for estimating forest canopy biomass,and there was no obvious saturation problem.