The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part Bl. Beijing 2 DOS 
109 
The difference dCX for the whole AOI shown in Figure 7 
follows the Laplace distribution. The mean difference is 4.0 m, 
which means that the X-band vegetation impenetrability is less 
than the C-band one. 
12 
Tree Cover (%) 
Figure 8. Impenetrability of the C/X-band versus percentage of 
the parcel-based tree cover. 
A potentially far-reaching consequence for vegetation cover 
investigations is the relationship between SRTM bias and the 
percentage of vegetation cover of a given area. This effect is 
illustrated in Figure 8. For example, any future changes to 
vegetation cover in terms of spatial extensions (horizontal and 
vertical) can be easily identified by comparing multi-temporal 
C- or X-band impenetrability of a given area. However, it is 
necessary to note that the data for the C-band vegetation 
impenetrability shown in Figure 8 indicate about 4.8 m positive 
bias (SRTM is above the ground) for the vegetation-free parcels. 
This issue will be further investigated in the near future. 
Differences dC and dX, or elevation bias, are shown in Table 9. 
Land Cover Class 
Difference (m) 
dC± lo 
dX± la 
Agriculture 
2.96 ±3.56 
-1.46 ± 1.49 
House 
4.24 ±7.12 
0.28 ±4.18 
Tree (0-100%) 
8.29 ± 9.97 
3.69 ±6.27 
Water 
4.34 ±6.82 
0.32 ±4.54 
Table 9. SRTM C/X elevation bias for land cover classes 
‘Agriculture’ and ‘Trees’. The standard deviation varied 
between ±2.79 m for ‘Agriculture’ and ±6.88 m for ‘Trees’. 
6 
. C-X 
Exponential fit 
y = 5.113*exp(-0.004355*x) 
R2 = 0.312 
2 
0 20 40 60 80 100 
Tree Cover (%) 
Figure 10. Weighted average difference SRTM.C minus 
SRTM.X versus tree cover expressed as a percentage of the area 
of the parcel. 
The weighted average difference dCXw for ‘Tree’ was also 
calculated for every parcel as a function of the average 
percentage of tree cover. The results are shown in Figure 10. 
The data appear to follow the exponential fitting curve. The 
difference is that dCXw reaches its lowest level of about 3.3 m 
for parcels fully covered by trees. The relatively low value of 
R 2 is mainly caused by inexperienced photointerpreters who 
estimated the percentage of the tree cover. 
4. CONCLUSIONS 
A comparison of both C- and X-band SRTM elevation products 
and a high-resolution reference DTM over a large test area in 
the Gold Coast, Queensland, Australia, focused on the influence 
of the vegetation cover, which was in the leave-on state, on 
elevation bias and provided an inside look into some of the 
properties of the InSAR C- and X-band technology of elevation 
determination. There is evidence that the X-band SRTM 
penetrates the vegetation cover deeper than the C-band SRTM. 
This effectively means that the vegetation-caused error in the 
X-band SRTM is smaller than that in the C-band SRTM. 
However, this has to be further investigated due to possible 
systematic error in the C-band SRTM. 
Closer consideration of the results in Table 9, in particular for 
‘Agriculture,’ can lead to a conclusion that perhaps both C- and 
X-band SRTM datasets contain certain reference biases, e.g., C- 
band elevations are about 3 m too high, and X-band elevations 
are about 1.5 m too low. This could explain the remarks 
regarding the 4.8-m bias made in the previous paragraph. The 
negative X-band SRTM elevation bias of the order of about 2.6 
m was reported by Heipke et al., 2002. However, Ludwig et al., 
2006, have not reported such an elevation bias. 
It was also shown that there is a linear relationship between the 
percentage of land cover and the magnitude of the vegetation- 
caused bias. This property of both SRTM datasets can be 
utilised for quantitative observations of variations of vegetation 
cover. 
The level of the vegetation-caused impenetrability, e.g., about 
10 m, is responsible for an error exceeding SRTM mission error 
specification. In addition, this is valid for about 30% of the 
global landmasses covered by forests. 
Weighted average difference dCXw per Equation 4 was 
calculated for every land cover type. The results are very 
similar, varying between 3.95 m for ‘House’ and 4.43 m for 
Further investigations are needed and planned towards 
identifying the best data source of vegetation cover to be used