A soil color index to adjust for soil and litter noise in vegetation index imagery of arid regions

The monitoring of vegetation in arid and semi-arid regions is made difficult by the dominant and variable soil signal. Soil color and brightness differences are believed to be responsible for the Large Saharan desert "artifact" areas in AVHRR-normalized difference vegetation index (NDVI) imagery. This problem represents a lower boundary condition of vegetation detection, affecting locust monitoring in the Sahel, and detection of the \´onset of greenness\´ in vegetation index imagery. In this study, a critical analysis is made as to the internal (soil properties) and external (view-Sun geometry and atmosphere) conditions that are likely to result in high VI values over non-vegetated surfaces. Observational and simulated, bidirectional soil aid litter reflectance data are analyzed and compared with optical data from sparse shrub and grass canopies, with and without atmospheric simulations. The problem is shown to be more related to soil color (hue and chroma) and is unrelated to soil brightness. Soil color problems were found to occur, to varying extent, in the NDVI as well as the \´improved\´ NDVI variants. As a result, additional spectral bands are necessary to develop a separate soil color index, which may involve an empirical equation with the \´green\´ band or a soil-correcting mixture model.