49
me and patience
set of templets.
ROL METHOD
airphoto map
rging of
y over large
of aerial
aracteristics is
irphoto onto the
xt, the settings
e freezed.
to basemap
can be matched
re clearly is
ield conditions,
n approach in
are sought on
basically
It is this
next Section.
ide on how to
rphotos to
itions.
a of study is
scale. (At ALS,
dered at a scale
h excellent
should be close
y to facilitate
both airphoto
llocated a number.
dinates of the
cene with the aid
n a stable base,
o factor so that
basemap scale
ting matrix of
rol points for the
in a map.
ts per airphoto
not only scale but
three control
compatibility
g airphoto points
cy between Landsat
ntograph can
points on the
ng points on the
the introduction of
re to be preferred
y O.M.I.) as they
y adjust for
pal points of
ed to the basemap,
c information
s.
at, highly
rue to scale and
orientation. The question now is how true this map
really is. In the following Section the
theoretical and empirical accuracy of map control
by Landsat is discussed.
4.3 Theoretical accuracy
ALS Landsat bulk processed photographic products
are not portrayed with a particular map projection
in mind. Apart from corrections for systematic
influences (e.g. mirror velocity) ALS also corrects
its images for panoramic distortion and earth
curvature (Dovey, 1986). Whatever the along-line
correction procedure may be, the portrayal result
can only be related to an oblique cylindrical
projection with as reference line the ground track
of the satellite. Since the maximum distance away
from this line in an MSS scene is only some ninety
km, the theoretical distortion cannot amount to
very much. As a result, ALS's bulk processed
imagery indeed provides high planimetric accuracy
over large areas as will be shown in Section 4.4.
However, in bulk processing no account is taken
of the direction in which Landsat is pointing
during image acquisition. It is simply assumed
that the satellite is pointing straight down, and
scanning at right angles to its own path. To
obtain higher accuracy than in bulk processed
imagery it is necessary to employ ground control
points with precisely known latitudes and longitudes.
This permits a yaw-pitch-roll correction to be
applied so that Landsat scenes can be produced
with a higher planimetric accuracy (Dovey, 1983).
Unfortunately, however, no precision products can
be expected for the remote areas under discussion
here, since in these areas points with known
latitude/longitude are hard to come by.
Another activity which influences accuracy is
the pinpointing of the photo points in the
Landsat image. The precision with which this can
be done is not limited to pixel size as the points
can be located to within pixel boundaries by
interpolation if the airphoto is used as guidance.
An MSS blow-up to e.g. 1/50 000 scale, of course,
does nothing to increase Landsat's inherent accuracy
but it does increase the accuracy of the cartographer
who has to locate those points within the pixels.
It should be remarked here that Landsat's spatial
resolution, contrary to popular opinion, is not so
much a function of the pixel size, but rather an
expression of the variation in the spatial
information content overlain and 'blurred' by the
pixel grid. Thus defined, Landsat's spatial
resolution is much finer than would be suggested
by its pixel dimensions.
Though it is unfeasible to quantify the method's
accuracy on the basis of the above information, it
has been possible, during the last few years in
Irian Jaya, to assess its degree of magnitude
thanks to 1 m accuracy measurements by the
topographical teams.
4.4 Empirical accuracy
In Irian Jaya Euroconsult was in a position to
test a number of times the accuracy of the method
under true field conditions. For six different
areas ranging in size between 40 000 and 80 000 ha
basemaps depicting the drainage network were
produced using the Landsat map control method. The
team also prepared a planimetric map in UTM projection
for each area, showing traverses and secondary
levelling lines (parallel to each other with 1 km
spacing), along which all encountered river and
stream intersections were indicated. The Landsat
controlled basemap and the UTM planimetric map,
both at the same scale, could now be matched by
fitting river crossing to the drainage lines.
As in this case the drainage pattern is a random
one, it follows that the intersection pattern with
parallel lines, is also random. It is remarkable
that this random field of intersection points fits
with the drainage pattern of the basemap as well
as it does. The few mismatches invariably proved
to be caused by mislocations of small streams on
the air photographs. The method of best possible
fit, used to mate intersections with streams,
showed that the mean deviation which occurred was
around 1.5 mm (when compensated for perpendicular
intersection) in random direction, at a scale of
1/20 000.
As ALS uses a process of portrayal which cannot
be defined by a particular set of mathematical
formulae, but which resembles the method as
mathematically described for an oblique Mercator
projection, and as the topographical team's results
are displayed with the assistance of the UTM
projection, a slight systematic mismatch caused by
this difference of portrayal is to be expected.
However, the absence of visible signs of systematic
variation proves that for our practical purposes
this mismatch is insignificant relative to the
position variation caused by Landsat's geometric
inaccuracies and with the uncertainties in the
positioning of the control points in the Landsat
image.
Two other facts become apparent:
* Without topographical work as a back-up, the
method is adequate for the production of
geographic maps with an accuracy requirement of
30 m, and
* the photo interpreter can always, irrespective
of tilt and scale variations, reconstruct the
location of his field data on the airphoto, thus
optimizing the quality of the interpretation.
5 DISCUSSION
5.1 Comparison
It is possible to express the mean error as a
function of control density in a slotted templet
block. Conversely, knowing the mean deviation (1.5
mm at photo scale) in the airphoto map control
method with Landsat, it is also possible to deduce
the control density that would have been necessary
had the slotted templet method been used. Of
course, this comparison is only valid if photo
tilts and scale variation are within ordinary
limits, which as we know is not the case.
Trorey (1947) has proved that the relationship
is based on the theory of errors, from which it
follows that the mean variation of image point
locations is inversely related to the square root
of the number of control points to which the
slotted templet block is laid. Or
e = kit/c)^
where e is the mean variation in mm, k a constant,
t the number of templets, and c the number of
control points. From empirical data Trorey determined
the value of the constant k as 0.16.
The theoretic control density (t/c) can now be
calculated since e=1.5 and k=0.l6. It would seem
that the 1.5 mm mean variation is comparable with
the theoretical accuracy which would have been
achieved with a control density of about 85 templets
per ground control point had the slotted templet
been used, and had airphotos been used with scale
and tilt within acceptable limits.
To achieve a result comparable to that of the
Landsat airphoto map control method when using a
set of photographs like those made available in
southern Irian Jaya, the slotted templet method