林业科学  2016, Vol. 52 Issue (3): 10-22 PDF
DOI: 10.11707/j.1001-7488.20160302
0

#### 文章信息

Feng Qi, Chen Erxue, Li Zengyuan, Li Lan, Zhao Lei

Forest Above-Ground Biomass Estimation Method for Rugged Terrain Based on Airborne P-Band PolSAR Data

Scientia Silvae Sinicae, 2016, 52(3): 10-22.
DOI: 10.11707/j.1001-7488.20160302

### 作者相关文章

Forest Above-Ground Biomass Estimation Method for Rugged Terrain Based on Airborne P-Band PolSAR Data
Feng Qi, Chen Erxue, Li Zengyuan, Li Lan, Zhao Lei
Key Lab. of Remote Sensing and Information Technology, State Forestry Administration Research Institute of Forest Resource Information Techniques, CAF Beijing 100091
Abstract: [Objective] To obtain an accurate estimation of forest above-ground biomass (AGB), the polynomial model integrating the terrain factors was presented based on the relationship of Synthetic Aperture Radar (SAR) response for forest AGB and terrain using the airborne P-band full Polarimetric SAR (PolSAR) data acquired by CASMSAR.[Method] Firstly, the slope map and the true forest AGB map over the study area were obtained as reference data using LiDAR data, and the forest AGB map was trained by the field AGB data. The systematical sampling was carried out based on the reference data to analyze the relationships between the backscattering intensity and the forest AGB and to analyze the changes of these relationships when the slope varied. Secondly, the local incidence angle was calculated from the LiDAR DEM and the orbit parameters of the airborne P-band SAR platform, and the polynomial model was built integrating the features of intensity, local incidence angle and look angle. Some of the sample plots were used to train the model parameters, and the others were performed as the validation samples. In order to avoid the contingency caused by sample size, more experiments were implemented with different sample size from 20m×20 m to 100 m×100 m.[Result] In the case of the plots with the size of 90 m×90 m, for the estimation model with the slope parameter (called as the second set of features) and for that without the slope parameter (called as the first set of features), the following quantitative technical targets were achieved. With the slope from 0°to 5°, the determination coefficients(R2) were 0.634 and 0.634 respectively, the root mean squared error(RMSE) were 12.07 t·hm-2 and 12.08 t·hm-2 respectively, the overall accuracies were 78.91% and 78.89% respectively. With the slope from 5°to 10°, the R2 were 0.524 and 0.523 respectively, the RMSE were 13.52 t·hm-2 and 13.97 t·hm-2 respectively, the overall accuracies were 80.57% and 80.52% respectively. With the slope above 10°, the R2 were 0.628 and 0.519 respectively, the RMSE were 13.16 t·hm-2 and 15.70 t·hm-2 respectively, the overall accuracies were 81.05% and 78.55% respectively. In addition, with the plot size increasing, the precisions of both methods were all improved. Especially, the accuracy of the estimation model with the slope parameter was higher than that without the slope parameter.[Conclusion] It was shown that the terrain had little effects on the intensity of the SAR data when the slope less than 10°, while it had a significant effect when the slope increases to more than 10°.The refined model involving local incidence angle could improve the accuracy, demonstrating the effectiveness and stability of the refined model. In addition, the accuracy would increase and tend to be stable with the scale enlarging regardless of the adopted model considered the effect of terrain or not, which revealed that the plot scale for evaluating the estimation model needed to be valued. The size of the sample plots should be considered for a reliable evaluation.
Key words: airborne SAR    P-band    PolSAR    forest above-ground biomass (AGB)    terrain

1 研究区概况及数据处理与获取 1.1 研究区概况

 图 1 研究区位置及数据覆盖范围 Fig. 1 The location of test site and the coverage area of data
1.2 地面数据获取与处理

2012年8—9月以及2013年8月，在根河森林生态站内布设66块方形样地，包括57块固定样地和9块临时样地。因调查目的不同，样地大小包括30 m×30 m，40 m×40 m和45 m×45 m，其相应的样地数量分别为35，13和18块。对固定样地每木检尺(胸径、树高、枝下高、冠幅等)，对临时样地每木胸径检尺，按照径阶抽测部分林木的结构参数，包括树高、冠幅(东西、南北)。利用陈传国等(1989)提出的异速生长方程分树种计算样地内每木AGB，并累加得到样地水平上的AGB。因样地大小不同，为了方便后续应用，将每块样地的AGB换算成单位面积的AGB。

1.3 机载LiDAR数据获取与处理

2012年8—9月，在研究区开展机载LiDAR飞行试验，以"运-5"飞机为平台，载有Leica机载激光雷达系统，获取LiDAR点云数据，激光点云密度平均为5.6个·m-2，波长为1 550 nm。

 图 2 LiDAR DEM Fig. 2 LiDAR DEM
 图 3 坡度分布 Fig. 3 Distribution map of slope
 图 4 坡度统计 Fig. 4 Histogram of slope

 $\ln W = {\rm{a}} + b \times \ln h25 + c \times \ln den70.$ (1)

 图 5 LiDAR森林AGB分布 Fig. 5 Distribution map of LiDAR forest AGB
 图 6 森林AGB统计 Fig. 6 Histogram of forest AGB

1.4 机载SAR数据获取与预处理

2013年9月13—16日，在研究区开展机载SAR飞行试验，采用"奖状Ⅱ"飞行平台，飞行高度为5 807 m，获取了P-波段PolSAR数据，飞行航向由西向东，右视方向观测，成像数据为单视复数据(SLC)，波长为0.5 m，方位向分辨率为0.625 m，距离向分辨率为0.666 m，中心入射角为55.058°。

 图 7 后向散射强度RGB影像 Fig. 7 RGB image of backscattering intensity (R:HH, G:HV, B:VV)

 $f\left( {{\theta _0}} \right) = a - b{\theta _0};$ (2)
 ${{\sigma '}_{**}} = \frac{{f\left( {{\theta _{center}}} \right)}}{{f\left( {{\theta _0}} \right)}}{{\sigma '}_{**}}.$ (3)

 图 8 后向散射强度归一化前后对比 Fig. 8 Comparison between non-normalized intensity and normalized intensity a: 归一化前后向散射强度与雷达视角关系Correlation between non-normalized intensity and radar look angle; b: 归一化后后向散射强度与雷达视角关系Correlation between normalized intensity and radar look angle.

2 研究方法 2.1 估测模型

 图 9 后面散射强度与LiDAR森林AGB的相关性 Fig. 9 Correlation between intensity and LiDAR forest AGB a，b，c means correlation between HH，HV，VV polarization intensity and LiDAR forest AGB.

 $\ln W = {a_0} + {a_1}{\sigma _{HH}} + {a_2}{\left( {{\sigma _{HH}}} \right)^2} + {b_1}{\sigma _{HV}} + {b_2}{\left( {{\sigma _{HV}}} \right)^2} + {c_1}{\sigma _{VV}} + {c_2}{\left( {{\sigma _{VV}}} \right)^2}.$ (4)

 图 10 不同坡度下后向散射强度与LiDAR森林AGB关系 Fig. 10 Correlation between intensity and LiDAR forest AGB with different slope a1，a2，a3: 坡度小于5°时各极化后向散射强度与森林AGB的关系Correlation between intensity and LiDAR forest AGB with slope below 5°; b1，b2，b3: 坡度大于5°且小于10°时各极化后向散射强度与森林AGB的关系Correlation between intensity and LiDAR forest AGB with slope above 5° and below 10°; c1，c2，c3: 坡度大于10°时各极化后向散射强度与森林AGB的关系Correlation between intensity and LiDAR forest AGB with slope above 10°.

 $cos{\theta _1} = \sin \alpha sin{\theta _0}\cos \left( {\beta - {\beta _s}} \right) + \cos \alpha + \cos {\theta _0}.$ (5)

 图 11 当地入射角 Fig. 11 Distribution map of local incidence angle

 图 12 入射角随地形变化示意 Fig. 12 Sketch of local incidence angle and slope θl:当地入射角Local incidence angle; α:当地坡度Local slope; βl:雷达方位角Radar azimuth angle; βs:坡向Local aspect; θ0:雷达视角Radar look angle.

 $\begin{array}{l} \ln W = {a_0} + {a_1}{\sigma _{HH}}\cos \left( {{\theta _1} - {\theta _0}} \right) + \\ {a_2}{\left[ {{\sigma _{HH}}\cos \left( {{\theta _1} - {\theta _0}} \right)} \right]^2} + {b_1}{\sigma _{HV}}\cos \left( {{\theta _1} - {\theta _0}} \right) + \\ {b_2}{\left[ {{\sigma _{HV}}\cos \left( {{\theta _1} - {\theta _0}} \right)} \right]^2} + {c_1}{\sigma _{VV}}\cos \left( {{\theta _1} - {\theta _0}} \right) + \\ {c_2}{\left[ {{\sigma _{VV}}\cos \left( {{\theta _1} - {\theta _0}} \right)} \right]^2}. \end{array}$ (6)

2.2 精度检验方法

 图 13 训练样本及精度检验样本分布 Fig. 13 Distribution of the training and testing sample plots
3 结果与分析 3.1 不同坡度的森林AGB估测结果精度检验与分析

 图 14 森林AGB估测结果 Fig. 14 Distribution maps of the estimated forest AGB a: LiDAR森林AGB LiDARforestAGB; b:第1组特征估测的森林AGB The estimated forest AGB based on the first set of features; c:第2组特征估测的森林AGB The estimated forest AGB based on the second set of features.
 图 15 基于全部样本的森林AGB估测结果精度评价 Fig. 15 Accuracy of the estimated forest AGB based on all testing sample plots a: 第1组特征估测精度Accuracy of the estimated forest AGB based on the first set of features; b: 第2组特征估测精度Accuracy of the estimated forest AGB based on the second set of features.

 图 16 不同坡度下的森林AGB估测结果精度评价 Fig. 16 Accuracy of the estimated forest AGB with different slopes a1,a2: 坡度小于5°时第1组特征、第2组特征估测精度Estimation accuracy based on the first, second set of features with slope below 5°; b1,b2: 坡度大于5°且小于10°时第1组特征、第2组特征估测精度Estimation accuracy based on the first, second set of features with slope above 5°and below 10°; c1,c2: 坡度大于10°时第1组特征、第2组特征估测精度Estimation accuracy based on the first, second set of features with slope above 10°.

3.2 不同尺度的森林AGB估测结果精度检验与分析

 图 17 基于所有样本的不同尺度下估测结果精度评价 Fig. 17 Accuracy of the estimated forest AGB with different plots scales based on all testing sample plots

 图 18 不同坡度下不同尺度的估测结果精度评价 Fig. 18 Accuracy of the estimated forest AGB with different plots scalesand different slopes a1，a2，a3: 坡度小于5°时估测结果的R2，Acc.，RMSE The R2，Acc.，RMSE of the estimated forest AGB with slope below 5°; b1，b2，b3: 坡度大于5°且小于10°时估测结果的R2，Acc.，RMSE The R2，Acc.，RMSE of the estimated forest AGB with slope above 5° and below 10°; c1，c2，c3: 坡度大于10°时估测结果的R2，Acc.，RMSE The R2，Acc.，RMSE of the estimated forest AGB with slope above 10°.

4 结论与讨论 4.1 结论

4.2 讨论