mapping
ensk lant-
using air-
eritidskrift
r multi-
June 1978.
itizing and
MULTITEMPORAL LANDSAT DATA SETS AS A BASIS
FOR LAND USE MAPPING
by Jürg LICHTENEGGER, Department of Geography, University of Zurich and
Klaus SEIDEL, Department of Photography, Swiss Federal Institute of
Technology, Zurich
I Introduction
The analysis of a Landsat frame with respect to land use categories is un-
satisfactory: the four spectral bands used in the sensor system do not re-
cognize the differences with the necessary resolution so that all the in-
teresting categories would be distinuishable in a single picture.
Depending on the date the record is taken, the cultivations are classifiable
with different reliability: a MSS-picture taken in spring time delivers e.g.
good separation between cereals and root crops but no separation between
different classes of cereals. In early summer time one may be able to
differentiate these classes.
This leads to the simple idea of combining those four-band-pictures so that
they become 8, 12, or even more channel pictures. In this case we speak of
multispectral data sets.
11 Superposition of Landsat frames to
multitemporal data sets
We use digital methods for the analysis. In the following we will discuss,
therefore, the problems arising if one tries to superimpose Landsat frames
digitally, and we will show the resulting success using supervised classi-
fication.
Our test is a part of the "Grosses Moos", a nearly flat arable area of about
225 km? between the lakes of Neuchátel, Biel and Murten in the western part
of Switzerland. The process we are dealing with is:
l. Digital overlaying of geometrically adapted subframes
2. Supervised classification
3. Control of the classification results by ground truth
The technical problem of overlaying different Landsat frames pixel-wise
precisely is discussed in more detail inan earlier publication (1).
Landsat scenes of different overflights (the cycle-time is 18 days) do not
match geometrically. Frame center points may differ in the east-west direction
by about 20 km. For a geometrical superposition this displacement has to be
compensated for by translation of one of the frames. In a "match-point method"
one has to find sufficient identical points in both pictures. Such match-
points in a satellite-picture might be: huge flat roofs, big squares of
concrete, road-crossings, but also sharp bends of rivers. In fact, potential
sites are often smeared over several pixels or occur in neighbouring scan-
Lines, As a rule they are identifiable with a precision of only about 2 or
pixels.