yed many prominent 
i to known surface 
wed differences in 
j Gum site quality 1 
n flooded) appearinc 
r (i.e. vegetation on 
¡d the colour lighter, 
adar appeared dark, 
tigation of bispectral 
ind Landsat images, 
le range of radar 
ses was developed 
e cover classes 
i >Dark 
¡3 
Swamp 
Agriculture 
water 
ark as they acted as 
jiving antennae. The 
so the density of the 
te quality (Forestry 
e forest type (i.e. the 
the radar response. 
5), who found that 
increasing age (or 
(during the needle 
iund that for L-Band 
correlated with tree 
Des of pine forest in 
id higher backscatter 
phenomena may be 
i subject to an 80% 
iis aspect is to be 
verall classification 
-B data classification 
ng accuracy was nol 
r the supervised 
[¡cation accuracy 
Table 5 - Mapping accuracies for the supervised classification of the 
combined Landsat and SIR-B data. 
Class A 
VV 
F 
I 
II 
III 
cr 
“T - 
TM 
Agriculture (AJ 27 
27 
0 
0 
o 
o 
Swamp (S) 3 
11 
1 
7 
1 
6 
29 
18 
62 
38 
Water (W) 
1 
28 
1 
30 
2 
7 
93 
Box (B) 
23 
1 
1 
2 
2 
29 
6 
21 
79 
Red Gum SQ1 (I) 
4 
1 
19 
6 
1 
31 
12 
39 
61 
Red Gum SQ2 (II) 
2 
2 
2 
21 
1 
2 
30 
9 
30 
30 
Red Gum SQ3 (III) 
2 
12 
1 
7 
7 
1 
30 
23 
77 
23 
Total No. of 
Pixels (T)30 
20 
30 
44 
24 
36 
10 
12 
209 
No. Comissions 3 
9 
2 
21 
5 
15 
3 
12 
% Comissions 10 
45 
7 
48 
21 
42 
30 
100 
U = Unclassified pixels 
OM = number of omissions 
P = percentage of omissions 
CA = class mapping accuracy 
statistical testing of thematic map accuracy. Rem. Sens, of the 
„ Environ. 7:3-14. 
Wu S. (1984). Analysis of synthetic aperture radar data acquired ovei 
a variety of land covers. IEEE Trans, on Geosci. and Rem. Sens. 
GE-22(6):550-5578. 
There was some qualitative’evidence to suggest that the 
remote sensing data was more accurate than some, sections of the 
site quality and vegetation maps used for ground truthing and 
mapping accuracy assesment. A more detailed ground truthing 
exercise is needed to evaluate whether some misclassified pixels ar< 
actually correctly classified, and in fact it is the ground truth data 
which is inaccurate. 
Some research has been undertaken to determine the optimal 
combination of wavelength, polarization, resolution and look angle 
for agricultural applications (De Loor, 1974; Ulaby, 1975; Brakke el 
al., 1981; Dobson et al., 1983), though much still needs to be done in 
forestry. Reliable models to describe radar backscatter from forests 
also need to be developed. 
4. Conclusions 
The highest overall classification accuracy of 65% was obtained 
with co-registered Landsat MSS and SIR-B radar data. SIR-B provides 
additional information for delineation of forest types and site 
quality classes for the Riverina forests of Australia, though the 
amount of extra information is limited. Stand structure appeared the 
main factor affecting radar backscatter from forests. 
Acknowledgements 
Mr T. Lee provided assistance in running the static average 
filter which he developed on the Dipix system, at the Centre tor 
Remote Sensing, Unversity of New South Wales. Ms L. Bischof 
provided valuable advice on the operation (and peculiarities) of the 
Dipix system. 
We are also grateful to The Forestry Commission of N.S.W. and 
the Department of Conservation, Forests and Lands, Victoria, who 
made available ground truth maps and reports of the study area. 
Literature cited 
55.1% 
502% 
56.4% 
best overall result, 
on which increased 
itributed information 
SS. The Landsat and 
¡dual class mapping 
sky and Sherk, 1975) 
) site quality classes 
data, so these result! 
ever, Benning et al. 
>tic forest types into 
''Jew Zealand., with 
i 58%; a poor result 
, (1980) used a four 
and estimated the 
l was 6 m. Guidon et 
>l airborne MSS and a 
;t terrain, and showed 
vith the MSS imagery 
results. The airborne 
sat MSS, due to the 
anner. 
st three principal 
ie Landsat MSS data 
e matrix produced by 
it 84.2% of the total 
was due to the radar 
mponents, generally 
Benning V.M., Ching N.P., Bennetts R.L., Ellis P.J., Beach D.W. 
(1981jNew Zealand land use cover and forestry mapping from a 
satellite. Proc. Second Aust. Rem. Sens. Conference, Canberra. 
p2.1.1-2.1.5. 
Brakke T.W., Kanemasu E.T., Steiner J.L., Ulaby F.T., Wilson E. (1981). 
Microwave radar response to canopy moisture, leaf area index and 
dry weight of wheat corn. Rem. Sens of Environ. 11:207-220. 
De Loor G.P., Jurriens A.A., and Gravesteign H. (1974). The radar 
backscatter from selected agricultural crops. IEEE Trans, on 
Geoscience Workshops GE-12(2):70-77. 
Dobson M., Ulaby F.T., Moezzel S. (1983). A simulation study of the 
effects of land cover and crop type on sensing soil moisture from 
an orbital C-band radar. Int. Geoscience and Rem. Sens. Symp., Vol 
1., TA-1-3. 
Forestry Commision of N.S.W. (unpub). Murray Management Plan 
(Draft). Forestry Commission of N.S.W., P.O. Box 2667, G.P.O., 
Sydney, N.S.W. 2001, Australia. 
Fung A.K. and Ulaby F.T. (1983). Matter-energy interaction in the 
microwave region. In Manual of Remote Sensing, Vol. 1, Chapt. 
4:115-164. 
Guidon B., Gentle M.R., O'Callaghan J.F., Briggs I.C., Dreiven G. (1980). 
Integration of MSS and SAR data of forested regions in 
mountainous terrain. Proc. 14th Int. Symp. on Rem. Sens, of Environ. 
Vol. 3, pp. 1673-1690. 
Hoekman D.H. (1985). Radar backscattering of forest stands. Int. J. 
Rem. Sens. 6(2):325-343 
Inkster D.R., Lowry R.T., and Thompson M.D. (1980). Optimal radar 
resolution studies for land use and forestry applications. Proc. 
14th. Int. Symp. on Rem. Sens, of Environ. Vol. 2, pp. 865-882. 
Kalensky z. and Sherk L.R. (1975). Accuracy of forest mapping from 
Landsat CCTs. Proc. 10th. Int. Symp. on Rem. Sens, of the Environ. 
Vol. II pp. 1159. 
Ulaby F.T. (1975). Radar response to vegetation. IEEE Trans. Antenna. 
Propagation., AP-23(1):36-45 
Van Genderen J.L., Lock B.F., Vass P.A. (1978). Remote sensing 
519