International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B4. Istanbul 2004
HOP x 100%
F A2(n — 1) (1)
where R(e) represents the confidence value in % and n is the
number of check points used in the accuracy test. Figure 2
shows realiability evolution versus the number of check points
used according the equation 1. As an inverse example, if we
wish to obtain a SD confidence value of 5%, we need about one
hundred check points. If we used 28 check points, we would
reach a 20% confidence value.
0.10 |
0100 Romer sama?
check points
Figure 2. R(e) values versus the number of check points
according the Equation 1.
Therefore, the number of check points must guarantee stability
"n error estimates. Revised research is rather heterogeneous
regarding number and accuracy of check points, and no author
has verified reliability in of these results.
Most research used a number of check points that proved
clearly insufficient for guaranteeing the validity of error results.
One article explained the use if check points from pre-existing
cartography; this procedure is not recommended, as there tends
to be no knowledge about the control map quality itself.
Methods based on GPS constitute the ideal source to obtain
these points, since they yield the coordinates with great
accuracy, and also allow to plan a spatially well-distributed
sample covering the whole area under analysis.
4.6 The DEM depuration procedure
As was indicated above, the DPW adds to each estimated
elevation datum a value for the correlation coefficient. These
values can be regarded as metadata, being estimators of the
reliability of the elevation calculated at each point. The
elevation and correlation data were exported as text files, and
then integrated into Arc View, since this GIS is not able to read
the TIN generated in Socet Set directly. The TIN was then
generated in ArcView using the points as the vertices of
triangles in a massive triangulation procedure.
This huge DEM (with no points yet eliminated) was denoted
MDE00. The other DEMs were generated by previously
eliminating those points whose correlation coefficient was less
than one of a set of threshold values. For example, MDES50 was
the result of the threshold 0.50 for the correlation coefficient
(Table 3).
For the calculation of the accuracy, we used a set of 7071
randomly distributed ground check points whose coordinates
were determined by differential GPS techniques. We then
determined the difference between these points and the
elevation values of the DEMSs, and estimated the mean error
(ME), standard deviation (SD), and root mean square error
(RMSE).
DEM name Forest No. points % points
value
MDE00 none 2 204 906 (all) 100
MDE50 0.50 1 946 805 88
MDE75 0.75 1 634 059 74
MDE80 0.80 | 457 043 66
MDE85 0.85 1 194 227 54
MDE90 0.90 810 394 36
MDE?9I 0.91 716 759 32
MDE92 0.02 617 733 28
MDE93 0.93 514 095 23
..MDE94 0.94 407 005 18
MDE95 0.95 199 745 9
Table 3. Depuration of DEM-SPOT by change in threshold
correlation values.
To ensure error reliability, we used a set of 7071 randomly
distributed check points whose coordinates were determined by
DGPS techniques. The transformation between the WGS84 and
the UTM local system was achieved by a Helmert
transformation with parameters derived from observation
measurements. These involved between 60 and 90 minutes at
five geodetic vertices around the area, with errors inferior to
0.01 m. After the geodetic frame was determined, and the GPS
processing of the check points adjusted, we were able to
calculate the difference between these points and the elevation
values of the DEM, and estimate the mean error, standard
deviation, and RMSE.
5. RESULTS
5.1 DEM-SPOT accuracy and reliability
We constructed 91 from SPOT images. Tables 2 outline the
different experimental tests. Optimal findings include:
. Erdas Imagine generates the most accurate SPOT-
DEM (7.7 m RMSE) as a TIN structure, using 14
ground control points, a 9x9 correlation window, and
using a threshold correlation value of 0.65.
. Socet Set obtains the best SPOT-DEM (8.6 m RMSE)
as a URG structure (20 m cell size), and using 13
ground control points. Socet Set. allows selection
from several matching algorithms, and the result was
more positive by using an ‘adaptive’ algorithm
instead of the specific algorithm included for SPOT
data.
A synthesis of the results is given in Table 4, which lists the
values of the mean error (ME), standard deviation (SD), ifs
confidence interval (C1=95%, 0=0.05), and RMSE.
In our case, the availability of 315 check points enabled the
error control to have a reliability of 96%. This value allows the
RMSE confidence limits to be calculated for cach DEM.
Furthermore, for a comparative analysis, we calculate error
statistics for a DEM generated from conventional cartographie
1:25.000 data.
258
Intern
Based
DEM
metho
the pi:
influer
for a c
Increa
strateg
of the
The in
carriec
Erdas
Finalh
genera
proces
Comp:
algebr
accura
of SP
differe
differe
Sourc«
SPOT
Carto;
? Mean
^ Root !
* Stand:
4 Confic
Tab
52 |
We !
hypotl
data r«
definit
and he
Carry (
withot
Figure
functi
huge
MDEt
those
thresh
of taki
It can
numbe
correl:
0.94 (;
DEM
18.5%
while |
We er
impro
the str
