Mapping Informal Settlements Through High Resolution Imagery
- Paper number
IAC-06-B1.5.06
- Author
Mr. Mattia Stasolla, University of Pavia, Italy
- Year
2006
- Abstract
Informal settlements are gaining interest in the remote sensing community, due to the fact that they are human settlements with large extensions, experiencing rapid changes and mixed land uses, which make them very difficult to be automatically extracted from remotely sensed imagery. However, the availability of high resolution optical and (in the near future) SAR data may improve the mapping and monitoring requirements of administrations and NGOs. Thus, it is really timely to understand to what extent current or new algorithms may be adapted or devised to extract land cover and land use maps for such areas. The approach is the natural improvement of current monitoring techniques, based on ground or aerial surveys. According to the European initiative “Global Monitoring for Environment and Security” (GMES), particular attention should be paid to urban areas, and ESA is funding phase 2 of the GMES Urban Service (GUS) project. This works extends the use of GUS legend to informal settlements, providing an efficient procedure aimed at their mapping through high and very high resolution data. The procedure is dependent on the ground resolution of the input image, and has been tested using SPOT multispectral and panchromatic images, as well as Quickbird pan-sharpened data. The first step of the procedure is always the use of a trained classifier (a fuzzy ARTMAP neural network) for land cover mapping. Very basic land cover classes, in accordance with the AFRICOVER available information, are considered in this step, and, according to our results, even coarser data, like radar images from the ASAR sensor on board of the ENVISAT-1 satellite, may be enough to this task. Higher spatial resolution (down to a few meters) is required however for the following phase, which performs a land-cover to land-use reclassification based on the analysis of the neighborhood of each pixel. Alternatively, if ground or spectral resolution is not sufficient, texture information may be exploited to discriminate between different classes. Finally, for very high resolution, a different procedure has been evaluated: after classification, classes related to artificial surfaces like shelters are extracted and objects are reclassified according to their size. After any of the resolution-driven procedures a map according to Level 4 of GUS legend was obtained, and achieved overall accuracies are around the 90% value. This research was carried out on two locations in Sudan, i.e. Al-Fashir and Nyala, thanks to the data made available for the Joint Research Center.
- Abstract document
- Manuscript document
IAC-06-B1.5.06.pdf (🔒 authorized access only).
To get the manuscript, please contact IAF Secretariat.