METHOD FOR DETERMINATION OF INVARIANT RAD IOMETRIC POINTS ADAPTED TO THE SEMIARID AREAS OF WESTERN AFRICA

The follow-up of changes by multi-temporal spatial remote sensing stud ies can be carried out by successive classifications or by dynamic stu dy of the radiometric signal. In this last case, a prior stage of radi ometric rectification is required. In developing countries, determinis t methods of radiometric rectification are not operational due to lack of atmospheric optical properties automatic recorders. Ideally, empir ical methods use only data included in the image. They are founded on research of invariant radiometric points. However they are unusable in rural areas of these countries as the usual invariant radiometric poi nts (asphalt road, roof, etc.) are missing. A new approach, an automat ic selection of invariant pixels, is proposed in this paper. This is c arried out in two stages: firstly, the bare soils are selected by use of a Vegetation index, secondly, an index of brightness is used to dis tinguish the areas of extreme brightness and darkness in the selected areas. On a site of central Senegal, the quality of the algorithm is a ssessed by an identification of the sets of invariant pixels using hig h resolution aerial photography and in-situ observations. Dark pixels arise mainly on lateritics and uncultivated lands. Bright pixels are d ivided between the sandy trails and the crusted surfaces of some field s. Potential applications of multi-temporal spatial data with high res olution are suggested in terms of multi-annual vegetation studies.