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


Lake Maggiore - Italy, 3-6 April 2000

Classifying land cover types with VEGETATION data in dryland:
A case study in Burkina Faso

V. Gond, E. Bartholomé
Joint Research Ispra, Italy

Land cover mapping is an essential component of environmental assessment. In the framework of the international conventions each signatory country must prepare reports with reliable figures for a series of items, including surfaces allocated to various land-cover types. This is for instance the case for the Framework Convention on Climate Change where potential natural carbon pools and sinks (e. g. forest surfaces) should be properly assessed. The principle applies also to the other conventions. The need is thus clearly defined: land cover mapping should provide accurate figures at the national level, and the location accuracy should be high enough to help year-to-year monitoring and to guarantee credibility to the information provided. The objective of the present study is to assess the capacities of VEGETATION data to provide the best possible detail both spatially and thematically.

To carry out the study S1 as well as S10 products were used, together with derived products, such as the NDVI and what we call here the NDWI (Normalised Difference Water content Index, after Gao 1996): (IR-MIR)/(IR+MIR).

The time-series analysis shows that these 2 indices give slightly different information regarding season duration. In addition there are differences in signal intensities that can possibly be interpreted either as differential atmospheric effect on the indices, or as a specific environmental information.

In dry regions, land-cover mapping can more easily than elsewhere be carried out with cloud-free single-date images at various stages of vegetation development. Furthermore, in these environments vegetation types can be discriminated through soil type spectral properties rather than through biomass amount. In single date images bi-directional effects can be minimised both by using band combinations and local contrast techniques.

As simple classifiers, such as the ISODATA algorithm, applied either on time series or on multispectral data sets do not lead to very satisfactory results compared to what can be photo-interpreted from the images, alternative procedures were evaluated. This is especially the case for instance for linear features such as valley bottoms

A series of examples are analysed in various environmental conditions and compared to reference material such as vegetation and soil maps, and high resolution satellite data.