Radiometric analysis of SPOT-VEGETATION images for burnt area detection in Northern Australia

Abstract

Radiometric analysis of SPOT-VEGETATION (VGT) images acquired over Australia was carried out as a basis for the development of an algorithm to map burnt areas in woodland savannas. We analysed the variability of daily ground reflectance and its relationship with illumination and viewing geometry. Finding that the geometrical effects can be parameterised by the phase angle (angle between the illumination and the viewing directions) and the viewing zenith angle (VZA), we fit a simple linear model to the observations. The results show that about 60–70% of the variability in the daily reflectance is caused by geometrical effects. The residual 30–40% of the variability is probably due to changes in vegetation condition, such as senescence, and residual atmospheric contamination. We tested temporal compositing as a practical method of reducing the variability in the reflectance whilst retaining the burnt area signal. We inspected the radiometric and geometrical effects of four different compositing criteria and showed that minimum near infrared (NIR) is the most appropriate for burnt area mapping over the study area. In order to analyse the sensitivity of the VGT spectral bands and derived indices to changes induced by fire, we extracted burnt area spectral signatures for different vegetation types. The persistence of the burnt signal, as observed with each band and index, was analysed. Among the bands, NIR is shown to be the most sensitive to fire occurrence. There is a clear drop in the reflectance immediately after the fire and it remains very low during subsequent weeks. On the other hand, the burnt signal in the short-wave infrared (SWIR) band is showed to be strongly dependent on the vegetation cover type and on the age of the burnt area. Among the indices, the Global Environment Monitoring Index (GEMI) is identified as the most suitable for detecting changes induced by fire on the vegetation cover.