2004 年 24 巻 3 号 p. 282-297
JERS-1 SAR images provide one of the best ways to monitor rivers in tropical forests, but digital detection of rivers from such images is made difficult because too many objects have levels of backscattering intensities similar to those of rivers. To solve this problem, we examined the appearances of the rivers in the SAR images and created feature detection models with which we performed spatial operations to evaluate the sizes of the rivers, their structural characteristics, their distribution patterns, and their association with other objects. Such operations were combined with more traditional intensity-based segmentation methods. The rivers so detected corresponded closely with those identified using JERS-1 VNIR data. We also compared the results with those obtained from SAR images observed in different seasons and found the results to be consistent with known water levels. We further applied the method to other tropical rain forests in the Amazon Basin, Congo Basin, Borneo, and New Guinea.