基于机载多基线PolInSAR的森林冠层高度反演

Forest Canopy Height Inversion Based on Airborne Multi-baseline PolInSAR

以美国航空航天局(National Aeronautics and Space Administration,NASA)、欧洲航天局(European Space Agency,ESA)、加蓬航天局合作的AfriSAR项目中的Lopenp和Mondah为实验区,以L波段的多基线全极化无人驾驶飞行器合成孔径雷达(unmanned aerial vehicle synthetic aperture radar,UAVSAR)数据为数据源,分别采用高度偏差(height variance,VAR)方法、偏心率(eccentricity,ECC)方法和平均相干幅度与分离度乘积(product of average coherence magnitude and separation,PROD)方法、原点到拟合直线距离与分离度乘积(product of fitted line distance from origin and coherence separation,LINE)方法进行基线选择,并基于地面随机体散射模型(random volume on ground,RVoG)三阶段方法反演森林高度,以激光雷达获取的相对高度RH100对两个实验区的森林冠层高度反演结果进行验证。结果表明:在植被高度较小时,利用4种基线选择方法反演的森林冠层高度均比较合理,但森林高度较大时,ECC和VAR方法的低估较为严重;在Lopenp实验区中,采用VAR方法选择基线时,总体反演精度最低,R^(2)和均方根误差(root mean square error,RMSE)分别为0.36、26.27 m;采用ECC方法选择基线时,总体反演精度优于VAR方法,R^(2)和RMSE分别为0.50、13.94m;LINE方法的R^(2)和RMSE分别为0.72、8.71 m,PROD方法的R^(2)和RMSE分别为0.72、8.55 m。4种基线选择方法中,PROD和LINE方法的总体反演精度明显优于ECC和VAR方法。Mondah实验区结果的规律与Lopenp实验区的一致,说明利用多基线极化合成孔径雷达干涉测量(polarimetric interferometric synthetic aperture radar,PolInSAR)技术反演森林冠层高度时,基线选择方法对反演结果有重要影响。

In this study, Lopenp and Mondah in the collaborative AfriSAR project between National Aeronautics and Space Administration(NASA), European Space Agency(ESA) and Gabonese Space Agency are used as test areas.Taking the multi-baseline fully polarized unmanned aerial vehicle synthetic aperture radar(UAVSAR) data of L-band as the data source, we use height variance(VAR), eccentricity(ECC), product of average coherence magnitude and separation(PROD), and product of fitted line distance from origin and coherence separation(LINE) methods to select baseline, and use the three-stage method of random volume on ground(RVoG) to retrieve forest height. The forest canopy height inversion results of two test areas are validated by relative height RH100 acquired by LiDAR. The results show that the forest canopy height inversions of the four baseline selection methods are reasonable when the vegetation height is small, but the underestimation of ECC and VAR methods is more serious when the forest height is large. In Lopenp test area, the overall inversion accuracy is the lowest when using VAR method to select baseline, its R^(2) and root mean square error(RMSE) are 0. 36 and 26. 27 m, respectively. When using ECC method to select baseline, its overall inversion accuracy is better than that of VAR method, and its R^(2) and RMSE are 0. 50 and 13. 94 m, respectively. R^(2) and RMSE of for LINE method are 0. 72 and 8. 71 m,respectively. R^(2) and RMSE of PROD method are 0. 72 and 8. 55 m, respectively. Among the four baseline selection methods, PROD and LINE methods have better overall inversion accuracy than ECC and VAR methods. The regularity of the results in Mondah test area is consistent with that of Lopenp test area. It indicates that the baseline selection methods in multi-baseline PolInSAR-based forest canopy height inversion have an important influence on the inversion results.