CRITICAL-ASSESSMENT OF VEGETATION INDEXES FROM AVHRR IN A SEMIARID ENVIRONMENT

The most frequently used vegetation index (VI), the Normalized Differe nce Vegetation Index (NDVI) and its variants introduced recently to co rrect for atmospheric and soil optical response such as Global Environ ment Monitoring Index (GEMI) and Modified Soil-Adjusted Vegetation Ind ex (MSAVI) are evaluated over a Sahelian region. The usefulness and li mitations of the various vegetation indices are discussed, with specia l attention to cloud contamination and green vegetation detection from space. The HAPEX Sahel database is used as a test case to compare the se indices in arid and semi-arid environments. Selected sites are char acterized by sparse vegetation cover and day-to-day variability in atm ospheric composition. Simulated indices values behaviour at the surfac e level shows that these VIs were all sensitive to the presence of gre en vegetation but were affected differently by changes in soil colour and brightness. We showed that GEMI is less sensitive to atmospheric v ariations than both NDVI and MSAVI since it exhibits a high atmospheri c transmissivity over its entire range for various atmospheric aerosol loadings and water vapour contents. These results were first tested o n a vegetation gradient, and secondly evaluated on a transect which en compasses various soils formations. On the vegetation gradient, it was found that GEMI computed from measurements at the top of the atmosphe re is invariable from one day to the next. On the bare soils transect, MSAVI calculated at the surface level, has shown a great insensitivit y to soil optical responses modifications, while GEMI exhibits from sp ace noticeable variability in this bright soil context. Finally, it wa s illustrated that GEMI exhibits interesting properties for cloud dete ction because of the strong decrease of its value on cloudy pixels.