Exploring and improving NOAA AVHRR NDVI image quality for African drylands

The accuracy of NOAA AVHRR NDVI maximum value composites can be poor due to interference from several sources, including cloud cover. The objectives of this paper are; 1. to accurately quantify noise in this imagery over Africa using geostatistics, and 2. to test four compositing techniques that may be able to reduce this noise. The nugget of the variogram model is used to compute standardized noise for five sites across Africa over 4 seasons. After removing trend and anisotropy in the NDVI sub-scenes, standardized noise estimates range from 18.5% in West Zambia to 68.2% in northern Congo. Four automated compositing methods are also tested over the West African Sahel for 13-day periods in order to improve image quality: the MVC, Maximum Value Temperature (MVT), a two-criteria algorithm that compares the two highest NDVI values for a period thereafter retaining the value with the smallest scan angle (MVCMISC), and a temperature-based algorithm similar to MVCMISC (MVTMISC). Results show that the MVT performs best for minimising cloud contamination, while the MVC is better for removing extreme scan angles. For the dual criteria algorithms, the MVTMISC performs best. The MVCMISC is better able to reduce scan angle bias for all land cover classes during the dry season, with the MVTMISC giving superior performance over the vegetative season. This work has implications for interpreting NDVI data in the context of famine early warning and developing biophysical descriptors of the African land surface at broad scales.