摘要
【目的】采用TanDEM-X单极化InSAR数据,研究基于相干系数的SINC模型森林高度估测方法,并分析5 m高分辨率的LiDAR DEM和30 m中等分辨率的SRTM DEM对模型估测精度和稳定性的影响。【方法】首先对观测的相干性进行非体散射失相干校正得到体散射失相干γVol,然后基于SINC模型将γVol的相干系数作为输入估测森林高度。以LiDAR提取的森林高度为验证数据,均匀选取150个检验样本,分别在15 m×15 m、30 m×30 m、50 m×50 m和100 m×100 m大小的样本尺度上进行精度评价,并与DSM-DEM差分法进行对比,分析2种方法的精度和适用性。【结果】5 m和30 m分辨率的参考DEM对SINC模型森林高度估测结果影响较小,随样本尺度增大其影响可逐渐忽略,当样本大小为100 m×100 m时,LiDAR DEM和SRTM DEM估测结果的R2分别为0.54、0.51,RMSE分别为2.38、2.51 m,精度分别为77.19%、75.99%;相比SINC模型法,DSM-DEM差分法在各样本尺度上的表现更好,但森林高度估测结果存在明显低估现象,必须采用森林高度实测数据进行校正,当样本大小为100 m×100 m时,R2为0.79,校正前后的RMSE分别为2.57、1.63 m,精度分别为75.44%、84.41%。【结论】基于相干系数的SINC模型法估测森林高度,以30 m空间分辨率的SRTM DEM进行地形补偿和地理编码,可以取得较好结果;虽然该方法的精度相比DSM-DEM差分法略有下降,但既不需要实测森林高度数据进行标定,也不需要输入高分辨率的DEM,具有大范围森林高度制图的潜力和更大的实际应用价值。
【Objective】Using single polarization TanDEM-X InSAR data,we studied the forest height estimation method with InSAR coherence amplitude,and analyzed the effects of different DEMs spatial resolutions(one is LiDAR DEM of 5 m resolution and the other is SRTM DEM of 30 m resolution)on forest height estimation accuracy.【Method】First,non-volumetric decorrelation was corrected from the observed coherence in order to obtain the volumetric decorrelation(γVol),and then based on the SINC model,the amplitude ofγVol was used to estimate forest height.Then forest height inversion results were validated against LiDAR CHM data and compared with the method based on phase difference.Plot sizes of 15 m×15 m,30 m×30 m,50 m×50 m and 100 m×100 m were selected to analyze the effects of spatial interpolation.【Result】The effects of DEMs in two kinds of resolutions(5 m×5 m,30 m×30 m)on inversion accuracy are small,when plot size is 100 m×100 m,the R2 are 0.54 and 0.51,the RMSEs are 2.38 m and 2.51 m,and the overall accuracies are 77.19%and 75.99%,respectively.Although the method based on phase difference has a better performance than the SINC model over different plot sizes,its results are obviously underestimated and should be calibrated with some known forest height samples.If the plot size is set as 100 m×100 m,the R2 is 0.79,the RMSEs before and after calibration are 2.57 m and 1.63 m,the overall accuracies are 75.44%and 84.41%,respectively.【Conclusion】The forest height estimation method using SINC model with the TanDEM-X InSAR coherence amplitude can obtain a good performance even if using a low spatial resolution DEM,such as SRTM DEM of 30 m resolution for terrain compensation and geo-coding.Although the accuracy of SINC model is less accurate than that of phase difference method,it requires neither plot measurements for model calibration nor high-resolution DEM for terrain compensation and geo-coding,so it has the potential for large-scale forest height mapping and higher value for practical applications.
作者
范亚雄
陈尔学
李增元
赵磊
张王菲
金玉栋
蔡丽杰
Fan Yaxiong;Chen Erxue;Li Zengyuan;Zhao Lei;Zhang Wangfei;Jin Yudong;Cai Lijie(Institute of Forest Resource Information Techniques, CAF, Beijing 100091;College of Forestry, Southwest Forestry University, Kunming 650224;Experimental Forest Farm of Wangyedian, Harqin Banner, Chifeng 024423)
出处
《林业科学》
EI
CAS
CSCD
北大核心
2020年第6期35-46,共12页
Scientia Silvae Sinicae
基金
国家重点研发计划项目“人工林资源监测关键技术研究”(2017YFD0600900)
国家重点研发计划项目“星载新体制SAR综合环境监测技术”(2017YFB0502700)。
