Comparison of Multi-frequency SAR Land Cover Signatures for Multi-site Semi-arid Regions of Africa
Spies, B.F.1; Brown, S.C.M.1; Lamb, A.1; Balzter, H.2; Fisher, P.2
1Astrium GEO-Information Services, UNITED KINGDOM; 2University of Leicester, UNITED KINGDOM

The need for accurate land cover maps for emergency response situations is well known, but cloud cover often obscures optical satellite imagery in the tropical and sub-tropical parts of the world, particularly in the rainy season. SAR sensors offer an alternative to optical systems since they are generally not influenced by cloud cover. Although much work has been done with single-frequency SAR images, the use of multi-frequency SAR to classify land cover and land cover change has not yet been fully exploited.

This study shows the analysis and comparison of different SAR backscatter signatures (sigma-nought distributions) for distinguishable land cover types over multiple semi-arid test sites in Africa. Three sites were chosen, in Tanzania, Chad and Sudan, where existing multi-frequency data was available from the different SAR archives. Broad land cover classes from the Food and Agricultural Organisation (FAO)ís land cover classification system (LCCS) are compared. These classes are primarily vegetation, agriculture, settlements, bare areas and water. The land cover types are identified using very high resolution imagery on Google Earth. The SAR signatures analysed include a wide selection of different polarimetric modes of SAR backscatter values from ALOS PALSAR (L-band), ENVISAT ASAR (C-band) and TerraSAR-X (X-band).

The first site is located in Tanzania with data acquired during both wet and dry season for all three frequencies. Wet and dry season data are available over a second test site in Sudan, and for the dry season at a third test site in Chad. All data were acquired between 2008 and 2010. Wet and dry season data were identified and chosen by comparing acquisition times with meteorological data. The variation of SAR sigma-nought signatures for different land cover types are observed between the wet and dry seasons. In addition, the variation in these seasonal signatures are compared between the different test sites to identify whether signatures are unique to each site or whether there are common trends or signatures applicable to multiple sites. Through this comparison the use of multi-frequency SAR data for land cover classification over semi-arid regions are assessed. The analysis and comparison of the different dimensions of the SAR systems, specifically polarisation and frequency, for different land cover types is a first step towards developing a transferable classification algorithm using multi-frequency SAR over semi-arid regions. Further work will include a similar comparison of SAR signatures for forested areas, and then the development of a transferable classification algorithm using different frequency / polarimetric combinations.

This project, part of the GMES Initial Operations - Network for Earth Observation Research Training (GIONET), aims to investigate the use of multi-frequency SAR for land cover and land cover change mapping, and its application to general land cover management and emergency response situations. Different approaches of using multi-frequency, multi-polarimetric, multi-incidence angle and multi-temporal SAR datasets are investigated and some initial results are presented.