REPRESENTATION OF VEGETATION BY CONTINENTAL DATA SETS DERIVED FROM NOAA-AVHRR DATA

Coarse resolution satellite data provide a unique means by which to examine the spatial characteristics of surface phenomena at a wide range of spatial scales. In this study images of the Normalized Difference Vegetation Index (NDVI) with a resolution of 8 km derived from the National Oceanographic and Atmospheric Administration (NOAA) Advanced Very High Resolution Radiometer (AVHRR) are studied at a variety of spatial scales ranging from 8 to 512 km. Scale variance analysis is used to examine the spatial characteristics of the NDVI up to the scale comparable with that used for global modelling. This technique identifies the scales at which spatial variation is taking place and the relative magnitude of the variation. Both annual and seasonal images of Africa from 1987 are examined. The analysis revealed substantial differences within the continent in the scale at which spatial variation takes place. Commonly, for the annual image there is an increase in spatial variation with coarsening spatial resolution, although certain areas of complex surface conditions show markedly different patterns. There are substantial changes in the spatial characteristics of the NDVI with time. Analysis of monthly maximum value composites for September and February revealed different responses in scale variance as a function of spatial resolution. These spatial differences were most marked for areas where vegetation possesses strong seasonality. Interpretation of these results leads us to believe that different factors appear to be controlling the spatial variation of the NDVI at different scales. Averaging at coarse grid cell sizes of 512 km as a means of representing surface conditions results in varied success in representing the NDVI. Averaging areas of transition and surface heterogeneity may result in a substantial over-simplification of surface conditions. Consideration needs to be given to the spatial characteristics of areas and their temporal variability if satellite-derived data are going to be validly applied to global models.