63 
MULTITEMPORAL SPOT IMAGES FOR URBAN LAND-COVER CHANGE 
DETECTION OVER STOCKHOLM BETWEEN 1986 AND 2004 
Karoliina Kolehmainen and Yifang Ban 
Division of Geoinformatics, Dept, of Urban Planning & Environment, 
Royal Institute of Technology, Stockholm, Sweden 
- (kko, yifang)@infra.kth.se 
KEY WORDS: Multitemporal, SPOT, Urban, Land-cover, Change Detection 
ABSTRACT: 
The overall objective of this research is to detect new urban areas over Stockholm Region between 1986 and 2004 using 
multitemporal remote sensing. Two SPOT images acquired on 13 th of June 1986 and 29 th of July 2004 were used for changed 
detection. Three change detection methods were tested for this purpose: image differencing, principal component analysis and change 
vector analysis using normalised difference vegetation index and brightness index. The results showed that image differencing using 
the red bands and second principle components performed better in detecting new urban features than change vector analysis (the 
overall accuracies: 89%, 87% and 83% respectively & kappa: 0.77, 0.74 and 0.67). Even though overall accuracies are all above 80%, 
the kappa coefficients were much lower indicating substantial amount of omission and commission errors presented in the change 
maps. 
1. INTRODUCTION 
Remote sensing applied to urban applications is of growing 
interest. Urbanisation is a fact and it can not be stopped, the 
question is how to manage it. The migration of people and jobs 
continues, attractive regions grow and the population increases. 
Nobody can predict with certainty that the development in a 
region like Stockholm will be sustainable, and will remain so in 
the future. Therefore new efficient systems are needed for 
monitoring the changes and to help to set up a strategy to handle 
the situation. Remote sensing and change detection can offer a 
tool for this process. Satellite images provide up-to-date 
information of earth’s surface, thus can be used for analysing 
dynamics of the change and its effects. There are several 
approaches to perform change detection, including image 
algebra, transformations and classification of multitemporal 
images. The first method results in a binary change/non-change 
map, while with the classified images the information about the 
type of change can be identified. 
Lu et al. (2004) compared several change detection methods 
and found that image differencing and principal component 
analysis were to recommend when performing direct change 
detection. Liu et al. (2004) examined the mathematical 
approaches for change detection and found that principal 
component analysis was one of the most accurate methods. 
Change vector analysis, on the other hand, operates on spatially 
contiguous groups of pixels rather than on individual pixel 
isolation. It has been found to be an effective multivariate 
technique, where the type of change, in some degree, can be 
classified (Malila 1980, Johnson and Kasischke 1998). 
The increasing spatial resolution of satellite images provides 
opportunities to detect urban changes more accurately and 
SPOT images have shown their capability in this field (e.g., 
Martin and Howarth 1989). Today SPOT images are widely 
used in case studies (e.g. Zhang et al,. 2003; Weber and 
Puissant, 2003; Ferreira et al., 2004). 
The objective for this research is to detect new built-up areas, 
roads, and other new urban features over Stockholm Region 
during 1986 and 2004 using multi-date SPOT XS images. 
2. STUDY AREA AND DATA DESCRIPTION 
The study area is located in the county of Stockholm in Sweden, 
an area that covers approximately 2800 km 2 of highly 
fragmented landscape. The major land cover types are water 
bodies, forest, cultivated land, parks, roads, low and high 
density builtup areas. Stockholm region is constantly growing 
and between the mid-80ties and 2000 the population has grown 
with approximately 1 per cent each year (RUFS, 2001). The 
region has long traditions for planning towards an ecological 
sustainable society in several administrative levels. Two SPOT 
multispectral (XS) images acquired over the Stockholm on 13 th 
of June 1986 and 29 th of July 2004 were used in this study. 
3. METHODOLOGY 
1.1 Geometric and Radiometric Correction 
Accurate geometric and radiometric correction of multitemporal 
images is an important component of change detection. In this 
research, the two SPOT images were geocoded to 1:50 000 
digital topographic maps using a polynomial approach. For 
correction of the atmospheric attenuation, relative image 
normalization using regression, is applied on the 1986 SPOT 
images. Here radiometric ground control points, i.e, pseudo 
invariant features (PIFs) were selected to establish the relation 
between the corresponding images from different dates. 
2.2 Change Detection 
In general, performing detailed classification in urban areas 
with a large number of land-cover classes is a huge challenge. 
Therefore, performing direct change detection on the remotely 
sensed images is a faster and easier approach to extract the