AVHRR bidirectional reflectance effects and compositing

Abstract

The purpose of this study was to assess the magnitude of bidirectional reflectance distribution (BRD) effects in AVHRR Channels 1 and 2 over land, their dependence on land cover, and the possibility of correcting for these effects as part of the AVHRR compositing process. The study was carried out in central Canada using cloud free full (1 km) resolution AVHRR images over 20 days near the peak green period in 1988. All images were coregistered and resampled. Uniform sample sites representing cropland, deciduous and coniferous forest, and wetlands were chosen using Landsat images. The effect of atmospheric corrections and the performance of three vegetation indices were also assessed. The results showed that BRD effects in the AVHRR data depend on land cover type, spectral band, the viewing geometry, and the degree of atmospheric correction. Atmospheric corrections increased the value of NDVI but reduced its variation with satellite zenith angle. BRD dependence in AVHRR Channels 1 and 2 could be modeled quite accurately, but spectral band- and land cover-specific model parameters were required for a good fit with the measurements. It is concluded that in AVHRR compositing based on the maximum NDVI value, the most-cloud-free pixel selection should precede atmospheric corrections and the correction for BRD effects. BRD correction models should be derived using data with similar spectral and spatial characteristics, preferably from the same sensor.