Great Victoria Desert: Remotely Sensed Data and Data Analysis
For this research, Landsat Multi-Spectral Scanner (MSS) covering the period from 1979 to 1994 and two Thematic Mapper (TM) scenes were obtained from the Australia Centre for Remote Sensing (ACRES). In addition to the multispectral data, in 1993 CSR was also one of the first AIRSAR acquisitions using JPL's airborne polarametric radar.
To the right, subset of 1981 MSS scene over the testsite. The Yamarna basin is the leaf-shaped feature at the top of the image, and the firescars show up a bright white. Notice the unique shapes of the firescars.
The initial step in the study was to register all of the imagery to the Universal Tranverse Mercator (UTM) coordinate system. A second order transformation with an allowable
rms error of <0.5 was used with a nearest neighbor resampling.
Fire Extraction
After all of the images are registered, the fires from each year were extracted. Recently burned areas in the Great Victoria Desert are readily distinguishable in Landsat MSS imagery. Ash is highly adsorptive in the near-infrared bands whereas vegetation is highly reflective (Channey, 1994). Even years following a burn , the difference between the burned and non-burned vegetation differs enough to define sharp fire boundaries. The change detection was done by subtracting the previous year's brightness values from the current year's brightness values. From this resultant layer, a pyramid segmentation (Acton, 1994) method was used to accurately
delineate the fire boundaries. From this process, a map of each year's fires was created.
To see the fire statistics for each year, click here.
To see the firemap for all years, click here.
Radiometric Correction
Radiometric correction of satellite imagery is a difficult, yet necessary step in understanding temporal changes of vegetation. The brightness value which is recorded at the satellite sensor is often in need of correction due to sensor degradation and atmospheric attenuation. The MSS data in this study was atmospherically corrected based on a darkest pixel improvement method (Chavez, 1988) and then converted to surface reflectance. The computed reflectance values of known vegetation types were compared to those obtained from CSIRO using a laboratory spectrometer. The computed reflectance values compared reasonably close to the spectrometer results. These variations are expected since the ground response is not
included in the CSIRO responses and the amount of pixel mixing in MSS data is fairly significant.
AIRSAR coverage
In October of 1993, JPL flew their Airborne SAR (AIRSAR) sensor over a target site known as Red Sands. They flew two passes over the testsite in two different modes, 20Mhz fully polarametric radar and 40Mhz Topsar Radar. The Southern strip, which was flown at 20Mhz, has a ground resolution of approximately 9 meters and has HH, VV, and HV polarizations of the C, L, and P bands. The northern strip, flown in Topsar Mode, has a ground resolution of approximately 4 meters and has HH, VV, and HV for the L and P band and the VV polarization for the C band.
The images look quite different in comparison to multispectral data since this is an active sensor rather than a passive sensor. It was hoped that the different wavelengths of the radar would be able to penetrate through the sand of Great Victoria Desert, but the flight was acquired during a heavy thunderstorm, so the amount of soil penetration is thought to be minimal. The AIRSAR is useful as a ground truth tool for gathering
signature files for MSS data, in that individual Marble Gum trees and other vegetation classes such as Mulga, thryptomene, and spinifex are easily seen in the AIRSAR data.
|
Last Modified: Wed Apr 14, 1999
CSR/TSGC Team Web
|
