164
Chapter 18
CSIRO Division of Atmospheric Research (DAR) using a kernel model due
to Staylor and Suttles (1986), as discussed in Cosnefroy et al. (1996).
At Tinga Tingana, the high reflectance and temporal consistency of the
target meant BRDF dominated the AVHRR variation as the Sun and view
angles changed over a one-year period. Validation sites also need good
BRDF models. Fred Prata (also from DAR) is characterizing sites at Uardry
near Hay in NSW and Amburla near Alice Springs, using innovative ground
and tower based measurements. These efforts are part of a network of
validation sites characterization and a new site in the north of Australia is
planned to be established.
At many validation sites modeled so far around the world, simple kernel
functions seem sufficient. However, for more complex land surfaces,
atmospheric and BRDF effects will need to be separated and it is not clear
whether and how consistent results will be obtained. Questions that arise in
this activity are:
Can all corrections be done with simple functions?
If not, are there a few simple “forms” for specific land cover types (i.e.,
is there a “Typology” of land cover BRDF and associated kernels)?
How does one separate atmospheric and BRDF effects?
These questions are crucial since it is one thing to characterize the
relatively simple land surface of a calibration sites, a bit harder for a
validation site and possibly very difficult for a general land surface. They are
also very pertinent at a time when people are keen to establish a consistent
and standardized set of environmental data series. BRDF effects can
dominate such series of AVHRR, Landsat and other data that are coming online (even airborne data).
5.
NDVI COMPOSITING AND CONSISTENT
AVHRR DATA TIME SERIES
In particular, there has been a very useful discussion recently concerning
AVHRR NDVI data. These data are a primary long-term data series that
many people wish to use for environmental monitoring or environmental
reporting. However, producers and users in Australia (Richard Smith,
WASTAC, DOLA, WA) are understandably worried by the greening of
deserts in winter. Li et al. (1996) have shown that the scene brightness can
account for up to 30% of the variation of NDVI for some land covers. They
have since established the effectiveness of simple kernel models for reducing
this effect—but the problem is land cover dependent.
Following on from this work, Qi et al. (1996) and Qi and Kerr (1997)
have discussed how scene brightness corrections using kernels should
Chapter 18
CSIRO Division of Atmospheric Research (DAR) using a kernel model due
to Staylor and Suttles (1986), as discussed in Cosnefroy et al. (1996).
At Tinga Tingana, the high reflectance and temporal consistency of the
target meant BRDF dominated the AVHRR variation as the Sun and view
angles changed over a one-year period. Validation sites also need good
BRDF models. Fred Prata (also from DAR) is characterizing sites at Uardry
near Hay in NSW and Amburla near Alice Springs, using innovative ground
and tower based measurements. These efforts are part of a network of
validation sites characterization and a new site in the north of Australia is
planned to be established.
At many validation sites modeled so far around the world, simple kernel
functions seem sufficient. However, for more complex land surfaces,
atmospheric and BRDF effects will need to be separated and it is not clear
whether and how consistent results will be obtained. Questions that arise in
this activity are:
Can all corrections be done with simple functions?
If not, are there a few simple “forms” for specific land cover types (i.e.,
is there a “Typology” of land cover BRDF and associated kernels)?
How does one separate atmospheric and BRDF effects?
These questions are crucial since it is one thing to characterize the
relatively simple land surface of a calibration sites, a bit harder for a
validation site and possibly very difficult for a general land surface. They are
also very pertinent at a time when people are keen to establish a consistent
and standardized set of environmental data series. BRDF effects can
dominate such series of AVHRR, Landsat and other data that are coming online (even airborne data).
5.
NDVI COMPOSITING AND CONSISTENT
AVHRR DATA TIME SERIES
In particular, there has been a very useful discussion recently concerning
AVHRR NDVI data. These data are a primary long-term data series that
many people wish to use for environmental monitoring or environmental
reporting. However, producers and users in Australia (Richard Smith,
WASTAC, DOLA, WA) are understandably worried by the greening of
deserts in winter. Li et al. (1996) have shown that the scene brightness can
account for up to 30% of the variation of NDVI for some land covers. They
have since established the effectiveness of simple kernel models for reducing
this effect—but the problem is land cover dependent.
Following on from this work, Qi et al. (1996) and Qi and Kerr (1997)
have discussed how scene brightness corrections using kernels should