Abstract:To accurately inverse the forest height over a wider range, it is necessary to study the baseline selection method for multibaseline PolInSAR data to alleviate large errors in forest height inversion over a wider range from singlebaseline, exploring a better baseline selection method. The UAVSAR Lband multibaseline full PolInSAR data was used from five orbits obtained by JPL/NASA in Pongara, Gabon forest on February 27, 2016. Based on the maximum coherence difference (MCD) coherent optimization algorithm to make complex coherence maximum separation, the PROD method and the ECC method were improved, compared and analyzed; and verified by using the LiDAR data LVIS RH100 obtained by NASA on March 4, 2016. The error maps of the difference between the forest height and the LVIS RH100 inverted by the two baseline selection methods were plotted to analyze the results of the forest height inversion. And the density maps of the kz, canopy coherence amplitude corresponding to the two baseline selection methods and LVIS RH100 were plot to directly evaluate the difference between the ECC method and the PROD method selecting the baseline, and comparing and analyzing the pros and cons of the two baseline selection methods. Combined with these drawn graphs (forest height maps, error maps and density maps), the forest heights inverted by the two baseline selection methods were compared and analyzed. The error in low and high forest areas was large. The high forests were underestimated (the error was negative), and the forests in the low areas were overestimated (the error was positive). The underestimated or overestimation of the ECC method was greater than the PROD method, and the accuracy was inferior to the PROD method. The two methods had good consistency compared with the LVIS RH100 data. The linear equation of the ECC method and LVIS RH100 was y=0.50x+10.60 and R2=0.69. The fitted linear equation of the PROD method was y=0.63x+8.21 and R2=0.70. Validated with the LVIS RH100 data, the RMSEs of the ECC method and PROD method were 9.80m and 8.86m, respectively. The accuracy of the PROD method was improved by 9.63% than that of the ECC method. For low and high forest areas, when coherence separation or coherence amplitude between complex coherence was small, the degree of complex coherence separation and complex coherence amplitude were the main factor affecting forest height inversion. The PROD method comprehensively considered the complex coherence amplitude and coherence separation, so the PROD method was more suitable for inverting forest height than the ECC method. In mediumhigh forest areas, when the complex coherence was more dispersed and the coherence amplitude had not reached saturation, the straight line fitting effect of the coherence region was an important factor for forest height inversion. The ECC method was more suitable for inverting the forest height than the PROD method. The results demonstrated that the forest height inversed by the two baseline selection methods was consistent with LVIS RH100. The ECC method only took the linearity of the coherent region as the criterion. Because of taking into account the fitting effect and coherence amplitude of the coherent line, the forest height inverted by the PROD method was better and closer to the LVIS RH100 than the ECC method, alleviating the phenomenon that the ECC method underestimated tall forests and overestimated low forests to some extent. The PROD method was more suitable for forest height inversion in low and high forest areas, and the ECC method was more suitable for forest height inversion in moderately high forest areas.
