Page 214 - Digital Analysis of Remotely Sensed Imagery

178 Cha pte r F i v e

points of irrigation canals, roads, waterways, and intersections of
windbreaks with secondary roads. This selection represents a com-
promise among spatial distribution, site accessibility, and ease of
identification on the satellite imagery. Their ground coordinates
were also read from a topographic map.
GPS data for the urban scene were logged between April 4 and 22
of 1999. The GPS data logged on April 10 were differentially corrected
using the Pathfinder Office (version 1.01) software. The 60 loggings
contained in each GCP file were then grouped to obtain its true coor-
dinates. The total number of loggings retained in the merged GPS
files varied from 30 to 180.
The rural GPS data were collected on June 22 to 24, 1998. Not all
preselected GCPs were accessible via a vehicle due to absence of navi-
gable roads. Those inaccessible GCPs had to be abandoned. Most GCPs
were logged once, with the number of fix points varying widely from
33 to 192. A uniform interval of 5 seconds was adopted throughout
data logging during which the sky was clear. The horizontal view at
each location was unobstructed except by wind breaks at the intersec-
tion of irrigation canals. Thus, the problem of multipath (see Sec. 14.2.2
for more details) was kept to the minimum.
The obtained planimetric coordinates were used to rectify Land-
sat TM, and SPOT panchromatic (PAN) and multispectral (XL) images
based on first-order polynomials. Once the image and reference coor-
dinates of all GCPs were entered into Earth Resources Data Analysis
System (ERDAS) Imagine, rectification residuals were automatically
calculated. If the RMSE exceeded one pixel size of the image being
rectified, the GCPs with the largest residuals were sequentially
excluded from the rectification until the overall residual fell within
one pixel size for the first time. Reference coordinates were then
replaced by map coordinates or GPS coordinates averaged from a
varying number of fix points. Again, the worst GCPs were sequen-
tially removed until the overall residual became smaller than one
pixel size.

5.6.1 Impact of the Number of GCPs
The use of all 25 rural GCPs in rectifying the SPOT PAN image
resulted in an unusually low rectification accuracy of 7.13 pixels
(Table 5.4). Of these points, two were at the interface of water and
sand dunes that could have moved since the map was produced.
Their exclusion from the rectification led to a drastic reduction in the
overall residual to 2.49 pixels. This accuracy is similar to 2.65 pixels
for SPOT XL. The number of useable GCPs is reduced to 22 in rectify-
ing the TM image because GCP 11 falls outside the image area.
According to these results, there is an inverse relationship between
rectification accuracy and the number of GCPs used, regardless of the
image being rectified. Rectification accuracy is not so high for all