:ati*ns. H*w- 
! information 
'est*. It is 
¡ateilite data 
lappe r and 
: information. 
Symposium on Remote Sensing for Resources Development and Environmental Management / Enschede / August 1986 
Classification of the Riverina Forests of south east Australia 
using co-registered Landsat MSS and SIR-B radar data 
i postgraduate 
»graphy uni- 
! author is 
Director, 
on Systems 
; valuable 
thankful to 
mission in 
it, Government 
for their 
•P. 
Revised 
.a, New Delhi, 
>82. A compa- 
.cation pro- 
ir an area of 
Basilicata, 
f of Environ, 
>• Automatic 
sensed data. 
>s, Division 
‘each to world 
■ s. Rep. No. 
Stockholm. 
il rain forest. 
I. 
[ues for image 
in remote 
•ess. 
A.K.Skidmore, P.W.Woodgate & J.A.Richards 
Centre for Remote Sensing, University of New South Wales, Kensington, Australia 
Abstract: The Riverina forests of south east Australia have been extensively managed for 150 years as a 
productive source of railway sleepers and sawn timber. This study was the first Australian forestry 
application to evaluate the use of SIR-B radar (co-registered with Landsat MSS data) for mapping forest types 
and site quality classes. The techniques used for radar speckle reduction, registration of images and 
classification or cover classes are discussed. Results show that the classification accuracy was superior 
when the two data sources were used in combination rather than individually. 
1. Introduction 
The objective of the study was to map forest types and site 
qualities of the Riverina forests using SIR-B imagery as a data 
source by itself and in combination with Landsat MSS data. 
The study area straddles the Murray River, which is the bordei 
between the States of New South Wales and Victoria in Australia 
(see Figure 1). 
Figure 1 - Location of the study area 
About 15,000 years ago, a 10m high fault line developed in a 
north-south direction across the course of the Murray River, 
effectively damming the river and causing the river to diverge into 
two arms, to the north and south. The huge triangular sedimentary 
delta that formed was subject to periodic flood inundation caused by 
the high winter/spring precipitation in the headwaters of the Murray 
River. This flood plain is now dominated by virtually pure 
monospecific stands of River Red Gum (Eucalyptus camaldulensis), 
due to this species' unique ability to withstand periodic flooding, in 
lower lying areas of the delta, shallow lakes and swamps are in 
various stages of silting up. Aeolian sand hills rise up to 12m above 
the flood plain and support tree species including Yellow Box 
(Eucalyptus melliodora) and Grey Box (Eucalyptus microcarpa). 
However, local variation in topography is generally less than 2m, anc 
shadowing effects can be considered minimal or non-existant. 
Interestingly, at the time of the SIR-B overpass, the forest complex 
was experiencing an 80% flood of the total forest area. The Landsat 
image was recorded about a month later when the flood had just 
receded. 
Three site quality classes have been defined for River Red Gum 
stands (Table 1). Site quality is the actual (or potential maximum 
average) height of trees in a forest stand, and is also an indication o 
the stand density. Stand density is a measure of stand basal area (i.e 
cross sectional area of tree stems at 1.3m per unit area) or stocking 
Table 1 also details the other major cover types in the study area. 
This present forest structure has been modified by man's 
activities. The Aboriginal population regularly burnt the forest to 
maintain an open woodland condition of veteran trees, which 
enhanced the value of the forest for hunting. From the 1840's, 
European man used the forest for grazing runs and for timber. Curren 
logging is on a selection basis, with some overmature trees being 
removed during logging to improve regeneration. Stands are 
uneven-aged and very variable in tree size and distribution as a 
result of this history. However, stand density (basal area or 
stocking) is correlated to site quality (see Table 1), with red gum 
(site quality 1) being the densest forest. 
The study area was selected because the Centre for Remote 
Sensing at the University of New South Wales had acquired a clear 
Table 1 - Major cover types in thè Riverina Forests 
Land cover Typical location of 
type occurence 
Definition and description of 
land cover type 
River Red 
Gum Site 
Quality 1 
Frequently flooded e.g. 
river bends. Areas 
with good access to 
subterranean water. 
Dominant tree height (or pot- 
ential tree height) of 31-45 m. 
Higher stand density (70 
m^/ha).Heavily stocked regen 
eration. Ground cover of leaf 
litter or grass. 
River Red 
Gum Site 
Quality 2 
Intermediate levels of 
the floodplain. Depth to 
watertable 6-9 m. 
Dominant heights of 21-31 m. 
Increasing number of woody 
understorey plants. Moderate 
stand density. 
River Red 
Gum Site 
Quality 3 
Higher levels of flood- 
plain. Depth to water- 
table > 9 m. Infrequent 
floods of short duration 
Poor stand development. Open 
savannah woodland of <21 m in 
height. Woody understorey 
species more pronounced. 
Yellow 
and Grey 
Box 
Irregularly flooded and 
flood-free areas 
Stands vary in dominant height 
from 6-30 m. Grass component 
in understorey. 
Swamps 
and Giant 
Tussock 
Rushland 
Watercourse and 
semipermanent swamps 
Tussock grass formation of 
2-3 m. 
SIR-B radar image of the Riverina forests from the flight of Space 
Shuttle Challenger in October 1984. Comprehensive forest type and 
site quality maps already existed for the area, and were made 
available by the Forestry Commission of N.S.W. and the Department o 
Conservation, Forests and Lands of Victoria. A unique opportunity 
thus occured to generate forest type maps using SIR-B imaging rada 
combined with Landsat MSS data of approximately the same dates. 
Details of the Landsat MSS and SIR-B radar are described in Table 2. 
Table 2 - Description of the Landsat MSS and Sir-B radar 
Landsat MSS 
SlR-B Radar 
Source 
Landsat-4 
Space Shuttle 
Acquisition date 
17 November 1984 
13 October 1984 
Acquisition time 
0930 
0100 
Pixel resolution 
79 x 56 m 
12.5 m 
Wavelength 
5 x 10-5 cm - 
1.1 x 10-5 cm 
23.5 cm 
Incidence angle 
orthogonal 
32.7 - 39.3 degrees 
Two obstacles had to be overcome to meet the objectives of 
the study. The first was radar speckle, which is an unavoidable 
product of the illumination of a surface by coherent monochromatic 
radiation. Despite the fact that SIR-B imagery was produced by 
averaging four independent looks, further speckle reduction was 
necessary prior to classification to prevent aberrant speckled pixels 
from causing misclassification. Secondly, spatial resolution