Sensors which scan at large angles relative to the surface such as the AVHRR present special problems in interpreting their output because important changes in view zenith as well as changes in the direction of illumination across the scanline and from season to season result in different radiance values at the sensor for the same surface, irrespective of whether the orbit is sun-synchronous. For any quantitative use of VIS and NIR data from the AVHRR, surface BRDF must be taken into account in addition to the effects of changing atmospheric attenuation and scattering with varying path length; see Corrections for Bi-directional Effects at the Canada Centre for Remote Sensing and the papers by Wu et al (1995) and Li et al (1995) listed in the references section. The problem is often cached in terms of "scan angle effects" or "solar angle effects" but it is the two underlying reasons for the observed phenomema which must be accounted for : the processes governing surface BRDF and changing path length.
The impact of surface BRDF on AVHRR applications can be visualised in a number of ways, including spurious deviations in a series of observations, obvious joins between mosaicked images and lack of separability of classes in VISNIR space (if classes composed of distinct grassland community types cannot be resolved with space observations, there is little hope of reliably obtaining useful biophysical information). Studies of the BRDF of vegetated land surfaces from ground, air and space show that even limited changes in viewing and illumination geometry cause important variations in reflectance estimates; tilting a field radiometer by only 40 degrees to view in the backscattering and forward-scattering directions close to the principal plane can result in deviations of more than 20% in near-infrared reflectance factor estimates. See this ground-based example of a bidirectional reflectance distribution from radiometry carried out in the Inner Mongolia grasslands in August 1996. More evidence (if required!) is here on Scott Goetz' pages .
In spite of this evidence, attempts continue to be made to use data from the AVHRR in environmental monitoring and other applications with no explicit correction for BRDF effects. Attempts made to correlate NDVI from AVHRR reflectances composited over periods of less than a month with environmental or ecological indicators such as rainfall or pasture degradation always lead to an unsatisfactory result; see Eklundh, 1998 and Bastin et al, 1995. The former found that the weakest relation between a temporal series of AVHRR NDVI and rainfall occurs at 10-day compositing resolution, with a median R-squared correlation coefficient of only 0.10. It is significant that extending the compositing period to one month or simply averaging the NDVI series produced increasingly better relationships; cf. Zhu and Yang (1996). While the effects of changing view and illumination are sometimes mentioned in studies using AVHRR, the reference is usually made in passing and other disturbance factors are sought to try to explain the anomalies apparent.
Even the large AVHRR reprocessing programmes of the NOAA/NASA Pathfinder Land and IGBP/EDC/USGS Global 1km AVHRR Land Project which aim to provide consistently-processed global AVHRR reflectances and brightness temperatures over long periods do not account for the effects of BRDF explicitly; instead attempts are made to reduce angular dependency by the restriction of viewing zeniths to within 42 degrees of nadir and by adopting a maximum-value compositing (MVC) technique with the NDVI (which is less dependent on BRDF than the channel reflectances). Other methods including temporal smoothing techniques have been used in crop monitoring programmes in Europe (for example, as part of the JRC Space Applications Institute's MARS programme). For global biosphere modelling, Sellers et al (1992) developed the FASIR-NDVI (Fourier-Adjusted, Solar zenith angle correction, interpolation and reconstruction) method, used in the International Satellite Land Surface Climotology Project. However, these techniques are not adequate and do not really stand up to scrutiny; see Zhu and Yang, Roujean et al, 1992b, Gallo & Huang (1998), and an important paper by Cihlar et al (1994) on alternative compositing scenarios. Indeed, a number of studies have found that maximum-value compositing with NDVI preferentially selects off-nadir observations rather than near-nadir ones provided that an adequate atmospheric correction is applied, contradicting the most frequently-cited reason for adopting MVC (Holben, 1986) (more citations here).
The reason for this situation is that until recently (1992) there was no adequate and operationally feasible means of describing or explaining the BRDF of vegetated surfaces using sensors such as the AVHRR.
...but why bother with the AVHRR? Isn't it
a meteorological sensor?