The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part Bl. Beijing 2008 
107 
dC = C - (R ), 
dX = <* >- (R >, 
dCX = C - {X ) 
where dC, dX, dCX = difference in elevation, 
C, X and R = C-, X-band elevations, and R is 
the reference elevation, 
^ ^ = spatial (buffer) averaging operator. 
Variation of difference ((7 d ) was calculated for every buffer as 
2 
a sum of the variations of the C/X-band ( <3~ C i X ) ar| d variations 
2 
of the reference elevation (<J R ): 
cr 
CiX 
+ cr: 
The pixel-based variation of the C/X-band data {CT CIX ) was 
calculated as a sum of the variation of the SRTM instrumental 
error and error caused by impenetrability in conjunction with 
C/X - pixel size (Becek, 2008): 
(tI, x = 1.55 2 +1/12 * d 2 *tan 2 (5) 
The Digital Cadastral Data Base (DCDB), one of the standard 
‘spatial’ products available in Queensland, Australia, at 
minimal cost, was used for parcel-based land cover 
classification. By means of photointerpretation of high- 
resolution 2003 aerial photography (0.15-m pixel), about 
71,500 land parcels were classified as ‘Agriculture’, ‘House’, 
‘Water’, and ‘Tree’. The ‘Tree’ parcels were also assigned 
visually assessed percentages of tree cover. Figure 1 shows 
DCDB in the background of an aerial photograph. 
Figure 4. Parcel-based classification of land cover including the 
percentage of tree cover. ‘A’ denotes ‘Agriculture’, ‘H’ - 
‘House’, ‘T35’ - 35% of ‘tree’ cover, and ‘ W’ - ‘water’. 
2.3 Method 
In the following, the data processing procedure will be 
presented. As a basic unit for calculations and data aggregations, 
a buffer with a 45-m radius centred on the centre of every C- 
band pixel was used. For every such buffer, the following data 
were available: 
a) Average X-band elevation (calculated from up to 12 
points), 
b) Average reference elevation (calculated from 13.3 (1 
-108) - in average - reference spot elevations, 
c) Type of land cover, 
d) Percentage of tree cover, 
e) Aspect calculated from C-band data, 
f) Slope calculated from C-band data, 
g) HEM value for X-band pixels, and 
h) A pixel based accuracy of C-band elevations. 
where d = 30 m or 90 m for X- or C-band pixel 
s = slope of terrain within a pixel, and 
1.55 m is an experimentally derived standard 
deviation of the instrumental errors of the 
SRTM. 
In the next step, using a reciprocal of the variance (2) as the 
weight, the weighted average differences dCw, dXw, dCXw, and 
their standard deviations were calculated for every parcel 
(belonging to one of the land cover classes). The calculations 
were done using well-known formulas: 
Buffers with missing data in any field or containing SRTM.X 
pixels with HEM>10 were omitted from further processing. 
Overall, 53,190 buffers were available for further processing. 
Data preparation also included identification of horizontal mis 
registration of both C- and X band data in relation to the 
reference DTM. This was performed by calculating the cross 
correlation plot for two perpendicular transects (N-S and E-W). 
No shift was detected at the lag level of 2.5 m. In the next step, 
the differences in elevations between SRTM and reference 
DTM for every buffer were calculated: 
g = l/o-j (5) 
where dj.jw, = weighted difference dC, dX, or dCX for i-th 
parcel 
^ ^ = spatial (parcel) averaging operator. 
The standard deviation of the average difference (Equation 4) 
can be calculated using the following formula: 
<^=<X 2 /£g (6)