The purpose of this paper is to evaluate JERS-1 SAR data for determining vegetation types in arid regions. First, a noise speckle filter was applied to the original JERS-1 SAR image using a Map Filter with an adaptive 7×7 window. Multitemporal SAR images were registered with the JERS-1 OPS image using a secondorder polynomial function. The accuracy of registration was within 0.5 pixel RMS error. Second, a small part of the study area was extracted from the fullscene image for further analysis. The NRCS values of each extracted image were computed with the known Calibration Factor for the NASDA supplied JERS-1 SAR data. In order to smooth SAR imagery, focal operation was used. This operation provides a mean filter function for computing the values to replace noisy pixels in an image.
The resultant image was smoother than prior to filtering. Therefore, pixel values were smoothed to some extent with surrounding pixels. Their tones were thus somewhat simplified after filtering. Following this color composite, an image based on three scenes above was generated to identify the training samples (signatures) . Multitemporal color composite image was evaluated for vegetation type discrimination using supervised and unsupervised classification algorithms. As a result, two thematic layers were generated. Finally, the accurately classified classes from both of the classification results were gathered to obtain the final classification result. A test site along the Tarim River in the Tarim Basin, China, was selected for this purpose.
View full abstract