Geoscience Reference
In-Depth Information
Figure 8.3
Percentage of land-cover change by Thematic Mapper (TM) image (all causes).
Clearing is a sensitive issue politically in Australia because of its impact on biodiversity
conservation, greenhouse gas emissions, and land and water salinization. In most states, farmers
are now required to apply for a permit before clearing vegetation. Because this study represented
the first major operational use of remotely sensed data in Australia, there was substantial interest
in the reliability of the LC change estimates.
Traditionally, accuracy assessment of data sets derived from satellite imagery involves compar-
ing them against an independent or reference source of information assumed to be correct (i.e.,
aerial photographs). However, no suitable data were available for our 288
-ha study area. The
limited existing aerial photography had already been used in the quality assurance process, which
required that every change pixel identified be checked against another data source (Kitchin and
Barson, 1998). Additionally, for most of the 156 TM scene pairs, LC change was a rare event; over
96% of the scenes had less than 3% of their area affected by change (Figure 8.3).
A methodology that did not require a reference data set and could be applied to change data
produced using a variety of approaches to image processing and radiometric calibration (Table 8.1)
had been developed by Lowell (2001) to evaluate the LC change maps produced for the ALCC
project by the state agencies. This chapter reports on the application of Lowell's method and on
the reliability of the estimates of change produced for the Australian ALCC project.
¥
10
6
Land-Cover Change Detection Methods Used for Each State
Table 8.1
State
Method
NSW
Unsupervised classification of combined 1991 and 1995 images
NT
Band 5 subtraction of 1990 and 1995 images
QLD
Thresholding of band 2, 5, and NDVI difference images
SA
Unsupervised classification (150 classes) of combined 1990 and 1995 images
TAS
Thresholding of NDVI difference data
VIC
Unsupervised classification of combined 1990 and 1995 images to create woody,
nonwoody, woody increase, and woody decrease
WA
Combined 1990 and 1995 images and carried out canonical variant analysis
based on biogeographic regions to identify suitable indices and bands to classify
land-cover change
