Development of Forest Information Measurement System by Multi-Wavelength and Multi-Polarization High-Resolution Synthetic Aperture Radar

这项研究的目的是开发从由日本宇宙航空研究开发机构(日本宇宙航空研究开发机构)和国家信息和通信技术研究所(NICT)联合开发的高分辨率多波长多极化合成孔径雷达(PI-SAR)中提取森林信息的技术。测试地点是北海道Tomakomai的针叶林。首先，对2002年11月获取的PI-SAR图像与同期地面测量的森林数据和2003年8月的森林数据进行回归分析。利用传统的雷达散射截面(RCS)技术，发现在所有极化条件下，X波段RCS与森林参数之间没有显著的相关性。然而，L波段的RCS随着森林生物量的增加而增加，最高可达40吨/公顷左右。这些趋势与几位研究人员已经报道的类似。接下来，因为高Re…更多的分辨率SAR图像显示了森林的结构，寻找图像纹理与森林信息之间的关系。结果表明，图像幅度服从K分布的概率密度函数，并且交叉极化图像的K分布的序参数与森林生物量之间存在着很强的相关性。此外，还发现有序参数随着生物量的增加而增加，最高可达100吨/公顷左右，远远超出了传统RCS方法的饱和极限。根据回归曲线，估计了未知森林的生物量，并与2005-2006年地面测量的生物量进行了比较。经比较，模型的准确率为86%。最后，利用地面测量的所有生物量数据对回归模型进行更新。该模型被认为对整个北海道地区平地针叶林生物量的估算是有效的，并且将传统的RCS技术与本研究开发的纹理分析相结合，可以将估算森林生物量的精度提高到更高的水平。较少

The purpose of this research is to develop techniques to extract forest information from the high-resolution multi-wavelength and multi-polarization synthetic aperture radar, Pi-SAR, developed jointly by the Japan Aerospace Exploration Agency (JAXA) and the National Institute of Information and Communications Technology (NICT). The test site is the coniferous forests in Tomakomai, Hokkaido. First, regression analyses are carried out between the Pi-SAR images acquired in November 2002, and the forest data measured simultaneously on the ground and the forest data measured in August 2003. Using the conventional technique that utilizes the radar cross section (RCS), it is found that there is no significant correlation between the X-band RCS and forest parameters at all polarizations. The L-band RCS, however, is found to increase with increasing forest biomass up to approximately 40 tons/ha. These trends are similar to those already reported by several researchers. Next, because the high-re … More solution SAR images appear to show the structures of the forests, the relation between the image texture and forest information is sought. As a result, the image amplitudes are found to obey the K-distributed probability density function; and that strong correlation exists between the order parameter of the K-distribution in the cross-polarized images and the forest biomass. Further, it is found that the order parameter increases with increasing biomass up to around 100 tons/ha which is well beyond the saturation limit of the conventional RCS method. From the regression curve, the biomass values of unknown forests is estimated and compared with those measured on the ground in 2005-2006. The comparison yields the model accuracy of 86%. Finally, the regression model is updated using all biomass data measured on the ground. This model is considered to be effective for estimating the biomass of coniferous forests on flat ground in the entire areas of Hokkaido; and the accuracy of estimating the forest biomass can be improved to much higher levels by combining the conventional RCS technique and the texture analysis developed in this study. Less