The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part Bl. Beijing 2008 
from 
sets, 
The 
data 
DOQQ 
are (x, y) 
- 12 points 
in 
and 
The 
their 
common window size and area to create difference images 
from which comparative accuracy statistics were generated. 
Figure 6 illustrates the subset area used for comparing 
Cartosat-1 DEM against the NED data. 
The masking step was applied only to the ASTER DEM to 
resolve cloud prevalence issues. The geoid was adjusted in 
both the Cartosat and ASTER DEMs, produced using ENVI 
software, to account for differences stemming from their 
ellipsoidal reference base. Once a consistent set of DEMs 
(resampled to appropriate pixels sizes) from Cartosat-1, 
SRTM, and ASTER were generated, they were subtracted 
from the NED source to derive difference images. Basic 
statistics including mean, standard deviation, minimum and 
maximum were also calculated. Figure 7 is one example of 
the difference image, produced from subtracting the 
Cartosat-1 October 2005 DEM from the NED. The main 
reason why the difference image looks like a shaded relief 
image is because of a spatial offset in the y direction. White 
areas in the difference image depict positive differences, 
while black ones show negative differences. 
Figure 7: NED minus Oct 2005 Cartosat-1 DEM difference 
image (Mean = -16.973, Std. Dev. = 12.257) 
To assess the horizontal accuracy (x, y), the Cartosat-1- 
derived orthorectified images were compared with the 
DOQQ data. This was done on the basis of a dozen ground 
control points (GCP), whose positional accuracy was 
evaluated on both the orthorectified and DOQQ data sets. 
GCPs were selected to provide an optimal coverage across 
the image (Figure 8). 
5 m USGS DOQQ 5 m Cartosat Orthoimage 
Figure 8: Typical ground control points distribution 
In addition to the process described above using the ENVI 
software module, Cartosat-1 DEMs and orthorectified images 
also were generated using PCI Ortho Engine. Results from 
both are included in Table 2. 
5. Analysis and Results 
The following analysis and results are based on Cartosat-1 
DEMs that were generated in 2006 using PCI software and in 
2008 using an improved version of the ENVI DEM module. 
The most accurate Cartosat-1 DEMs that emerged from these 
studies were the absolute DEMs produced by the PCI 
software. Mean vertical error for those DEMs was slightly 
less than 4 meters, and the standard deviations were less than 
10 meters (Table 2). In addition, the horizontal accuracies of 
the Cartosat-1 DEMs were approximately 5 meters x-y, 
which were the best horizontal accuracies achieved. Relative 
DEMs are produced without any GCPs and only use the 
instrument-derived ephemeris data. The relative DEMs 
produced from ASTER data were only slightly less accurate 
than the absolute Cartosat-1 DEM produced by PCI software. 
But these relative DEMs were significantly more accurate 
than the relative Cartosat-1 DEMs produced from either PCI 
or ENVI software. Even the absolute Cartosat-1 DEMs 
produced from the ENVI software were less accurate in a 
vertical sense than the relative DEMs produced from ASTER 
data. However, the horizontal accuracies were better than 
those for the relative DEMs produced from ASTER data. 
The most accurate DEMs evaluated were those produced 
from SRTM data, at least in terms of vertical accuracy. 
However, the absolute Cartosat-1 DEMs produced by PCI 
software were only slightly less accurate than the SRTM 
DEMs. The accuracy measured in this study is relative to the 
NED reference, which has an error as well. So the reported 
accuracies are not absolute vertical accuracies to the ground. 
Drum Ms. DEM 
X* 
Y* 
r* 
DMt/SoRwar« Sourer 
RMSEx 
RMSEy 
M»»rt 
Std. D«v. 
Cartosat-1 (Oct. 2005) Rel. PCI 
104 343 
137.701 
-45.235 
23.617 
Cartosat-1 (Nov. 2005) Rel. PCI 
29.573 
145.916 
-16.409 
25.854 
Cartosat-1 (Oct. 2005 Abs. PCI 
5495 
5.490 
3.916 
9.576 
Cartosat-1 (Nov. 2005) Abs. PCI 
3.275 
5.939 
-3.549 
5.232 
Cartosat-1 (Oct. 2005) Abs. ENVI 
2819 
7220 
-16.973 
12.257 
Cartosat-1 (Nov. 2005) Abs. ENVI 
3 376 
13.182 
-19.570 
19.920 
ASTER (Ail. 2000) Rel. Silcast 
23 260 
19 280 
-7.295 
10.014 
ASTER (Mar. 2004) Rel. Silcast 
18 290 
19 240 
2.195 
8.460 
SRTM 
N/A 
N/A 
-2.196 
2.955 
Error in meters 
*DEM-derived Orthorectified GCPs minus DOQQ 
**NED minus DEM 
Table 2: Drum Mts. DEMs: Accuracy assessment