Reading landscapes: Satellite images and social science research
During the last 20 years, interest in remote sensing data has exploded. More than two thousand remote sensing papers were published in 2007, a ten-fold increase since 1987. This increased interest, however, is almost completely restricted to the natural sciences. There has been very little expansion in social science publications that use satellite imagery.
One possible reason is that the available methods for satellite image analysis are not well adapted to the needs of researchers in the social sciences. When economists, geographers or sociologist analyze a landscape they typically look for features that can say something about social structure, the type of technology used, income levels, etc. Often such features are linked to the spatial arrangement of different land-uses: size and shape of plots, size if buildings and built up areas, and how different land-uses are combined. Traditional methods, however, often fail to analyze the micro-scale variation in the spectral reflectances captured in satellite images.
The purpose of this project is to develop and test methods that use this micro-scale variation to identify objects in the satellite images that are of interest for social scientist. These methods have previously been used do identify different farming strategies in south-east Sweden, sites for traditional, summer farmsteads in the Dalarna region of Sweden, and migrant-labor dependent subsistence farms in Sodo District, Ethiopia
Bo Malmberg, Geography, Stockholm University
2008-2013
The project aim was to test whether it is possible to extract objects of interest to social scientists from satellite data. The project has shown that it is possible. For example, that it is possible to distinguish different types of farming in agriculture when spatially contextual information is utilized. The segmentations created using the project method display valuable information. One dilemma is that this new information, often need to be verified and interpreted.
During the project period, the project has focused on tests in different application areas than those covered by the application, for example, with other researchers on issues relating to damage to forests and forestry.
An important advance in the development of a method for segmentation of satellite images was made in an early phase of the project. Much time was therefore allocated to test and develop the automatic contextual segmentation technique as the project developed, and is now called WICS (Window Independent Context Segmentation). Its originality is confirmed by the fact that the project team has obtained a patent for this technology.
The project did not address the issue of engaging other scientists but in 2012 project members organized a graduate level course for PhD students in social science on remote sensing methods that was based on the research results of the project, many of these students will use remote sensing techniques and the methods the project worked out in their dissertation projects.
The project has published two articles based on the SRPC method that was presented in the application and developed this method further into an automatic contextual segmentation technique, WICS shown to produce very reliable results. PhD student Michael Nielsen has used this method successfully in two case studies, of Stockholm, and partly of the city Columbus, Ohio (USA), and in a study of a forest in Bavaria. Wästfelt has used the same technique for an analysis of landscape change in western Östergötland. This study examines the relative changes in the composition of the agricultural landscape in agriculture resulting from the deregulation of Swedish agriculture in 1991.
The project has also published a number of shorter papers that present the ideas behind the project methods, see publication list.
The projects has, as mentioned, been granted two patents in Sweden as well as internationally for the techniques developed in the project.
One question that seems central to proceed with is if there are natural contextual classes. It seems based on the project's empirical results that the contextual aspects that are used WICS segementation are so strong that it offsets some of the dilemmas that face unsupervised statistical classification.
Another issue that emerged during the project and that were not answered is what constitutes the difference between the prior method SRPC and the new WICS method. Is there a semantic level of conceptualization between the two methods? The project has worked with semantic spaces and we believe that it would be possible to describe the differences with such concepts but this requires further research.Another question is how to describe the relative dimensions of the landscape that are captured by the contextual methods used and what these dimensions imply for land use. By developing the analysis of these relative dimensions, information vital to understanding urban and landscape change can be generated.
Another important issue is how to arrive at efficient and high-quality verifications.
Presented preliminary results have been presented at the Annual meeting for American Geographers, AAG 2009, 2010, 2013, Permanent European Conference for the Study of the Rural Landscape 2012, and published in international journals.
The article Wästfelt et al 2012, confirmed that what Wästfelt found for Swedish conditions also worked in an African context.
Publication of patents demonstrate a new way to segment image information that can have a major impact.
The project's publishing strategy has been to publish in internationally recognized journals. After an invitation an article for publication in the Journal of Art History has also been written.
Publications
Wästfelt, A. The production of landscape change, Case-study over western part of Östergötland Sweden. In Prep will be submitted to Remote Sensing of Environment.
Nielsen, M. Extraction of urban areas with different functions and underlying planning ideologies using Window Independent Context Segmentation. Submitted to Computers, Environment and Urban Systems.
Nielsen, M. Ahlkvist, O. Urban area category extraction using window independent context segmentation from a SPOT 4 scene covering Columbus, Ohio. Submitted to Journal of Spatial Science.
Nielsen, M. Heurich, M, Brun, A. Malmberg, B. Automatic mapping of standing dead trees after an inscect outbreak. Submitted to Journal of Forestry.
Wästfelt, Anders. Transformations of satellite images. Submitted to Journal of Art History.
Nielsen, Michael Meinild; “Extraction of different urban area categories from satellite images using Window Independent Context Segmentation”. In: Stilla U, Gamba P, Juergens C, Maktav D (Eds) JURSE 2011 - Joint Urban Remote Sensing Event Munich, Germany, April 11-13, 2011
Wästfelt, A. Arnberg, W. 2013. Local space context measurements used to explore the relationship between land cover and land use function. International Journal of Applied Earth Observation and Geoinformation.
Wästfelt, A. Tegenu, T. Nielsen, M. Malmberg, B. 2012. Qualitative satellite image analysis: mapping the spatial distribution of farming types in Ethiopia. Applied Geography.
Ahlkvist, O. Wästfelt, A. Nielsen, M. 2012. Formalized interpretation of compound land use objects – Mapping historical summer farms from a single satellite image. Journal of Land use Science. 2012, 1–19.