Urban monitoring: new possibilities of combining high spatial resolution IKONOS images with contextual image analysis techniques

The development of high spatial resolution scanners in theory might lead to the inclusion of studies at the town quarter level or the village level. Such developments open the way to sub-regional and urban studies where the object of study has a size, smaller than a soccer field. It might enable us to monitor socio-economic components of study areas next to general land cover monitoring. In a research project on land cover change detection in Ouagadougou (Burkina Faso) an IKONOS image became available and the added value of this high spatial resolution remote sensing (RS) source could be analysed based on a comparison of the level of detail and insight in land use classification at the neighbourhood and block level.. Previous work was carried out on the basis of SPOT-XS images having a significant lower spatial resolution of 20 by 20 meters but a comparable spectral band setting. The conventional spectral-based image classification methods however exclude information captured by neighbouring pixels. City sections are often characterized by a unique spatial arrangement and size of objects such as buildings, roads and green areas. Various authors have discussed the use of so-called contextual filtering algorithms for example in vegetation and land cover classification. A disadvantage of the filtering algorithm is that the spatial resolution of the output image degrades with the size of the filtering kernel. When such a kernel is applied to a SPOT-XS image, the area under investigation covers 60×60 m., much larger than the size of the objects understudy. IKONOS with its smaller pixel size of 4×4 meters opens the way for contextual image analysis techniques at a spatial scale of interest for urban applications as the minimal kernel size of IKONOS will cover an area of approximately 12×12 m. This size much better matches the size of urban buildings and the scale of socioeconomic activities