A New Approach to Estimate Forest Parameters Using Dual-Baseline Pol-InSAR Data
Lu Bai(1,1), Wen Hong(1), Fang Cao(1) and Zhou Yongsheng(1)
(1) Institute of Electronics, CAS, Lab 1, No. 19 BeiSiHuan XiLu, 100190, Beijing, China
Many algorithms based on DoA technique are proposed to separate dominant scattering centers from the observation. For instance, the MUSIC and CAPON algorithm obtains dominant scattering centers via minimizing the effect of noise subspace. ESPRIT algorithm and its improved algorithm indicate efficient approach to estimate the interferometric phase of each dominant scattering center. However, these algorithms identify the signal subspace based on the scattering information. They do not explore the interferometric coherence information and make efforts to suppress the errors in interferometric phase. In this paper, we combine the coherence optimization to obtain the signal subspace. In order to estimate and suppress the phase noise, dual-baseline PolInSAR data are used in the proposed algorithm.
According to the assumption, the signal subspace has high coherence between these observations while the noise subspace is less correlated. Therefore, we identify the signal subspace using coherence optimization instead and they are estimated from dual-baseline data. Furthermore, we introduce the assumption that the scattering centers stay at constant location and the estimated interferometric phase contain phase errors. To obtain the expected phase of dominant scattering centers, we explore dual-baseline PolInSAR data to estimate the phase error. After the removal of flat earth phase, the expected phases of dominant scattering centers are related to the baseline and the phase errors in two baseline are not correlated. Thus, we uses the difference in the interferometric phase estimated from the two baseline data to estimate their phase errors. After the suppression for phase noise, we can obtain the accurate estimation for the dominant scattering centers.
The experimental data are acquired by ESAR system in 2003. The data covers 21 forest test sites in Traunstein, Germany as shown in Fig. 1. The grey color indicates the average forest height their. To validate the performance of proposed algorithm, its estimation for the forest height is compared with those given by the ESPRIT algorithm using single baseline data.