THE PRECISION OF THE NDVI DERIVED FROM AVHRR OBSERVATIONS

Vegetation studies using NOAA-AVHRR data hove tended to focus on the u se of the normalized difference vegetation index (NDVI). This unitless index is computed using near-infrared and red reflectances, and thus has both an accuracy and precision. This article reports on a formal s tatistical framework for assessing the precision. of the NDVI derived from NOAA-AVHRR observations. The framework is based on the ''best pos sible'' precision concept, which assumes that signal quantization is t he only source of observational error. While the radiance resolution o f a spectral observation is essentially fixed by the instrument charac teristics, the reflectance resolution is the radiance resolution divid ed by the cosine of the solar zenith angle. Using typical solar zenith angles for AVHRR image acquisitions over Australia, +/- 0.01 NDVI uni ts is typically with ''best possible'' precision attainable in the NDV I, although this degrades significantly over dark targets, and at larg e solar zenith angles. Transforming the computed NDVI into a single by te for disk storage results in little or no boss of precision. The fra mework developed in this article can be adapted to estimate the ''best possible'' precision of other vegetation indices derived using data f rom other remote sensing satellites.