Forest Vertical Information Extraction Based On P-Band SAR Tomography

Important forest vertical structure parameters such as understory topography, forest height and forest above ground biomass(AGB) are not only important forest resources parameters,but also important parameters for describing forest ecosystem. Compared to short wavelength Synthetic Aperture Radar(SAR), P-band SAR has a better penetration in forest, which has more advantages for extracting forest vertical information. However, since P-band SAR could not reflect the information of crown surface, forest height could not be inverted based on single-baseline Interferometric Synthetic Aperture Radar(In SAR) technology. In addition, the sensitivity of backscattered intensity to AGB decreases drastically for higher AGB, which is referred to as the saturation effect. Multi-dimension SAR such as Polarimetric Interferometric Synthetic Aperture Radar(Pol In SAR), multi-baseline In SAR and multi-baseline Pol In SAR make a comprehensive utilization of multiple observation data and could decsribe forest vertical structure more detailed. In view of the advantages of multi-dimension SAR, forest vertical information such as forest vertical structure profile, forest height and forest AGB were extracted with the technology of single-baseline Pol In SAR, multi-baseline In SAR tomography and multi-baseline Pol In SAR tomography. The experiments were carried out over the site of tropical forest in French Guiana. The Tropi SAR 2009 P-band multi-baseline airborne In SAR data were the data sources.The main research contents include the following aspects.(1) Forest vertical information extraction by single-baseline Pol In SAR technologyFirstly, the optimizing configuration of Pol In SAR spatial baseline was analyzed by simulation data. And the best two-track Pol In SAR data were selected from the six-track Pol In SAR data sets for the subsequent forest vertical information extraction. Then, according to the characteristics of P-band Pol In SAR data in tropical rain forest, the applicability of the Random Volume over Groud(RVo G) coherent scattering model were analyzed. And the suitable reversion algorithm to extract forest height was selected. Finally, the forest vertical structure profile was retrieved with the technology of single-baseline Polarization CoherenceTomography(PCT). The experiments showed the following results. The inverted Digital Surface Model(DSM) was underestimated and the inverted Digital Elevation Model(DEM)was overestimated with the single-baseline Pol In SAR data, which resulted in a serious underestimation of the inverted forest height with the accuracy of 59.79% and Root Mean Square Error(RMSE) of 12.06 m. The canopy top and the ground could be seperated based on the PCT method. But the resolution of the reconstructed vertical structure profile was lower.The following conclusions could be drawn from the research results. The distinguishing accuracies of the phase centers with different scattering mechanisms are different under condition of different vertical effective wave number(kz). The phase centers with different scattering mechanisms shows higher separation properties when kz varies from 0.05 to 0.15. In tropical rain forest, the elevation of the P-band SAR complex coherence phase center is higher than the actual surface and lower than the canopy top. The accuracy of the retrived forest height based on RVo G model is lower. The retrived forest vertical structure profile based on the single-baseline PCT method is difficult to accurately reflect the forest vertical structure.(2) Forest vertical information extraction by multi-baseline In SAR tomographyFirstly, three popular methods of spectral analysis, including Beamforming, Capon and Music, were analyzed using simulation data and airborne data. Then, an approach of forest height extraction using multi-baseline In SAR tomography was developed. DEM was calculated by extracting the peak of the HH polarization backscattered power in vertical direction.Similarly, DSM was calculated by means of the calibration of forest sample by extracting the peak of the HV polarization backscattered power in vertical direction. The forest height was extracted through the difference of DSM and DEM. Finally, a strategy of estimating forest AGB through multi-baseline In SAR tomography was developed. This strategy improved the forest AGB estimation model by analyzing the correlation between the values of tomographic relative reflectance in various height and forest AGB, furthermore, the tomographic features related to forest AGB were further analyzed. The results demonstrated that Beamforming performed the best imaging efficacy. The accuracy of forest height extraction was 89.16%, and RMSE was 3.59 m, which meet the requirement of forest investigation. The accuracy of forestAGB estimation was better by means of the incorporation of tomographic relative reflectance in the height of 5 m and 25 m than using the tomographic relative reflectance in the height of 5m and 25 m individually. The accuracy using the former method was 89.17%, and RMSE was46.45 t·hm-2.The following conclusions could be drawn from the research results. The DEM, DSM and forest height could be well inverted with the multi-baseline In SAR tomography method based on HH polarization and HV polarization multi-baseline In SAR data. The accuracy of forest AGB estimation could be improved by combining different layers of forest vertical structure information.(3) Forest vertical information extraction by multi-baseline Pol In SAR tomographyFirstly, three popular methods of Fully Polarimetric(FP) spectral analysis, including FP-Beamforming, FP-Capon and FP-Music, were analyzed using the airborne data. Then, an approach of forest height extraction by multi-baseline Pol In SAR tomography was developed based on SKP decomposition. DEM was calculated by extracting the peak of the enhanced ground backscattered power in vertical direction. Similarly, DSM was calculated by means of the calibration of forest sample by extracting the peak of the enhanced canopy backscattered power in vertical direction. The forest height was extracted through the difference of DSM and DEM. Finally, forest AGB eatimation based on the technology of FP tomographic estimator and the technology of SKP decomposition was studied. The feasibility of Pol In SAR tomographic features used for estimating forest AGB was studied by analyzing the correlation between the values of tomographic relative reflectance in various height and forest AGB. The results demonstrated that FP-Beamforming performed the best imaging efficacy. The accuracy of forest height extraction was 90.10%, and RMSE was 3.00 m. The correlation analysis demonstrated that the correlation between tomographic relative reflectance extracted from multi-baseline Pol In SAR and forest AGB was weak.The following conclusions could be drawn from the research results. Compared to multi-baseline In SAR tomography method, multi-baseline Pol In SAR tomography method withSKP decomposition has more advantageous to retrive forest height. The tomographic relative reflectance of the synthesised and enhanced polarization is not good for estimating forest AGB.The innovations in this paper lie in three aspects. Firstly, the robust approach of forest height extraction using multi-baseline In SAR tomography was developed. DSM was calculated by means of the calibration of forest sample by extracting the peak of the HV polarization backscattered power in vertical direction, which avoided selecting the value of power loss.Secondly, a strategy of estimating forest AGB by combining tomographic features from different layers was developed, which improved the forest AGB estimation model. Thirdly, the correlation between Pol In SAR tomographic features and forest AGB was revealed.