Super typhoon Haiyan makes landfall
Super typhoon Haiyan
Subscribe here to receive special images


Lake Maggiore - Italy, 3-6 April 2000

Mapping and monitoring small ponds in dryland with the VEGETATION instrument – application to West Africa

V. Gond*, E. Bartholomé*, F. Ouattara°, A. Nonguierma+
*Joint Research Ispra, Italy
° Direction de la Météorologie Nationale, Ouagadougou, Burkina Faso
+Centre Régional AGRHYMET, Niamey, Niger

Paper (pdf file, 456 k)

Monitoring the state of small ponds is very useful in dry regions, as most of them are non-permanent and entirely constrained by the rhythm of local rainfall. These ponds determine various human activities such as watering herds, production of vegetables and other plants, and even water supply of local populations. Finally these water bodies are important for biodiversity, both of plant species and animals such as migratory birds. There is an information need at national as well as at regional levels.

On VEGETATION image colour composites water bodies and marshy vegetation show up clearly. Yet this does not mean that these features can easily be extracted, as their spectral signature may vary largely according to their specific ecological properties. In addition several other issues make the problem more difficult: atmospheric haze, bi-directional effects, and the strong N-S ecological gradient This in confirmed by the poor scores obtained when using classical image classification. To overcome this problem several procedures have been tested on both VEGETATION S1 and S10 standard products over one window centred on Burkina Faso and including parts of the neighbouring countries.

The time series extends from September till December 1999, which was a particularly humid end of rainy season in this region.Finally the most successful procedure is based on a classical photo – interpretation criterion, i. e. the local contrast. In arid lands ponds and marshes are small objects scattered in the landscape. Because of their specific nature their spectral properties are clearly different to their environment.

Thus derived channels are produced to reinforce local contrasts. These derived channels are obtained by computing regional averages on sliding windows of large size (>30 pixels) to capture the average landscape signature. Several derived channels are computed to eliminate possible confusions. Finally, simple and robust thresholds allow easy object delineation.The tests show that S1 data should be screened with a more effective cloud mask that the standard one, and that the algorithm developed in first place for S1 products also work fine on S10 products. Although validation could be carried out only in a limited manner with "historical" data and maps, it shows that commission errors are probably very few.