TOPOGRAPHIC NORMALISATION OR APPLICATION OF GEO-SPATIAL RELATIONS FOR 
IMPROVED LAND COVER CLASSIFICATION IN THE ALORA AREA (SPAIN)? 
G.F. Epema 
W.G. Wielemaker 
Department of Environmental Sciences, Wageningen Agricultural University, The Netherlands 
Commission VII, Working Group 4 
KEY WORDS: Remote, Sensing, Land, Use, Classification, Knowledge Base, Landsat 
ABSTRACT 
The Alora area (Spain) is a complex and heterogeneous environment. Classification results of land cover with four broad 
land cover classes (irrigated tree crops, rain-fed tree crops, annual crops and semi-natural vegetation) using Landsat 
Thematic Mapper data show still reasonable accuracy. Different methods are compared to enhance this classification 
further. Topographic normalization based on a DEM, using both Lambertian and non-Lambertian reflectance models did 
not improve the classification results. However, a knowledge-based post classification sorting based on relationships 
between topography, landscape, soil data and land cover considerably improved the accuracy of the land cover 
classification of the study area. This implies that a better land cover classification can be obtained when using additional 
geo-spatial data, from existing databases or during ground truth collection, than when using land cover data only. 
1. INTRODUCTION 
Knowledge about land cover and land use has become 
increasingly important. There is no doubt that remote 
sensing is the best way to obtain data over large areas 
quickly, and at an affordable price. However, the level of 
discrimination achieved by the satellite image 
classification is sometimes quite low. The discrimination 
of different land cover types by digital classification of 
satellite imagery is difficult in complex and 
heterogeneous environments like the study area (the Alora 
region, province of Malaga, Spain). In such 
environments, the digital classification of remote sensing 
satellite data offers information with a reliability, 
typically in the order of 70% (Schuiling, 1995). Despite 
great efforts to improve the spectral characterization, only 
minor improvements in classification accuracy are 
generally the result. 
This study aims at an approach to achieve high 
classification accuracy for land cover by combining 
ancillary data and remote sensing data. In this respect two 
different approaches are tested. 
One approach is used to get better results by mitigating 
the topographic effect while the other approach uses 
contextual information in a knowledge base, including 
topography, to obtain better land cover classification 
results. 
In the first approach, it will be tried to remove the 
topographic effect by modifying the surface radiance 
values recorded by the satellite sensors using the cosine 
of the angle of incidence and inclination derived from a 
digital elevation model of the area. After this procedure, 
images were again classified. 
In the second approach, also statistical relations derived 
from an existing database of the test area containing 
agronomic and other data will be examined. Most 
emphasis is however on the use of a knowledge base 
using knowledge of domain experts who worked in the 
area and study of their publications (Fresco and Guiking, 
1998, Wielemaker et al, 1996). 
The results of the different approaches will be compared 
by evaluating classification accuracy. 
2. MATERIALS AND METHODS 
The major input materials were: 
1) satellite imagery. The used image is a Landsat 
Thematic Mapper image of May 1995 with a ground 
resolution of about 30 x 30 meters. 
2) Digital Elevation Model. This model is derived from 
stereo photos and has a spatial resolution of 10 
meters, 
3) Geo-database of Alora. Within the framework of a 
training program on "sustainable land use" an 
extensive data set exists of field observations and 
interpretations related to agronomy, landscape and 
soil data. This set was used to evaluate classification 
accuracy of the different approaches. 
The three approaches of classification of this study are 
presented in figure 1: (1) a standard supervised 
342 Intemational Archives of Photogrammetry and Remote Sensing. Vol. XXXII, Part 7, Budapest, 1998 
clas 
top: 
clas 
clas 
Ap 
In 
pro 
(Ri 
of 
unk 
sel 
ger 
sin 
Foi 
ma 
dat 
cla 
pro 
lan