345 
SOME PROBLEMS ASSOCIATED WITH 
LARGE AREA MAPPING FROM LANDSAT 
R. Wright and N.K. Hubbard 
Department of Geography 
Aberdeen University 
St. Mary's, High Street 
Old Aberdeen, AB9 2UF, 
Scotland (U.K.) 
BIOGRAPHICAL SKETCHES 
Robert Wright received his B.Sc. degree in Geography with honors from 
Glasgow University, his B.Sc. in Photogrammetric Engineering from the 
International Institute for Aerial Survey and Earth Sciences (ITC) in 
the Netherlands and his M.S. in Remote Sensing from the University of 
Michigan. He is a lecturer at Aberdeen University where he is respon 
sible for teaching and research in remote sensing, photogrammetry and 
land surveying. He is principal investigator on a project funded by 
the Natural Environment Research Council to develop and test a method 
for producing a land cover map of Scotland from LANDSAT data and is 
also an experimenter in the European SAR-580 synthetic aperture radar 
experiment. 
Neil K. Hubbard is a research assistant in the Geography Department at 
Aberdeen University working on the N.E.R.C.-funded project mentioned 
above. He received his B.A. degree in Geography with honors from 
Nottingham University and his M.Sc. in Environmental Technology from 
Imperial College, London. He has spent three summer vacations working 
for the aerial photography unit of the Ministry of Agriculture in 
Cambridge and is currently registered for a Ph.D. degree at Aberdeen 
University. 
ABSTRACT 
A project is described which has the aim of producing a color printed 
map of land cover types of mainland Scotland (about 75,000 km 2 ) using 
digital data from seven LANDSAT scenes. Initial attempts to classify 
the data involved a 'supervised' approach using an advanced image data 
processing system (the Plessey IDP3000). When class parameters derived 
from this system were extrapolated from 'training' areas to classify a 
complete scene, unacceptable errors in classification appeared. The 
main cause of these poor results is the high variability of Scottish 
topography and land use over short distances and the variable atmos 
pheric haze present on apparently cloud-free overpass days. To over 
come these difficulties, a less automated classification has been 
adopted, sub-dividing each scene into blocks of similar topography, 
land use and atmospheric conditions. A preliminary color map of one 
scene has been produced from the IDP3000 classification and this scene 
has been reclassified by the modified approach using nine sub-scenes. 
Statistical testing of part of this classification indicates than an 
overall accuracy of 87.5 per cent can be achieved at the 95 per cent 
confidence limit. 
INTRODUCTION 
Since the advent of the LANDSAT series of satellites in 1972, a great 
deal of effort has been expended on digital classification of land