BRDF Applications in Semiarid Grassland Monitoring with the AVHRRs

---

---

4. Methods : implementation

The calibration of a surface BRDF model with satellite observations is possible in a number of ways:

1. the sensor may be designed to acquire observations at many different angles and with different solar zeniths (for example POLDER and MISR);
   
   
2. multiple sensors may be used if intercalibration is possible and differences in radiometric characteristics (i.e. band widths) can be accounted for; or
   
   
3. the revisit period may be so short that multiple overpasses over a period during which surface conditions may be assumed invariant provide an adequate sampling.

For applications using full resolution AVHRR, the second and third options are available since there are two sensors in contrasting near-polar orbits and which may be calibrated with reasonable accuracy, each providing ~two looks per location per day (depending on latitude and season). See this example of the predicted BRDF sampling provided by an AVHRR in an afternoon orbit. Note that sensors such as the ATSR-2 which are designed to view at fixed multiple view angles and which have heliosynchronous orbits do not provide optimal sampling for BRDF model inversions, since the range of view and solar angles over a short period is very limited (note also that the number of samples available from the ATSR-2 over a short timespan is much lower since the revisit period is longer).

The data used in this study were derived from four NOAA-12 (AM) overpasses and 17 NOAA-14 (PM) overpasses. Following investigations with point samples, a processing chain was built in order to demonstrate the feasibility of using a linear semiempirical BRDF model with AVHRR scenes. This takes uncalibrated channel 1 and 2 DNs and the acquisition angles (view and sun zeniths and azimuths) as inputs and outputs :

• atmospherically-corrected reflectances ("surface" reflectances) for both channels, with calibration based on the work of Rao & Chen (1996) and Loeb (1997) and atmospheric correction based on Rahman & Dedieu's SMAC algorithm (1994);
  
  
• the number of observations deemed to be uncontaminated by cloud or cloud-shadow (based on decomposition of the variation in a sequence of observations);
  
  
• the RMSE found in fitting the model to the observations (absolute error);
  
  
• the three model parameters (isotropic, geometric-optical and volume scattering); and
  
  
• the inverse matrix which is the solution to the system of linear equations comprising the inversion problem (this is required for calculation of the weight of determination, an indicator of the quality of model inversion).

The contribution of the MODIS BRDF/Albedo team (and particularly Wolfgang Lucht and Philip Lewis) in providing the AMBRALS (Algorithm for MODIS Bidirectional Reflectance Anisotropy of the Land Surface) modelling software used in this work must be acknowledged here.

When the model has been fitted to all observations, the three model parameters can be used in either BRDF adjustment using anisotropy factors ("correction"; see Wu et al 1995) or in direct forward-modelling of bidirectional reflectance to a preferred view and illumination geometry. Tests of the success of model inversion and forward-modelling may be intrinsic to the model and data (the RMSE on model fitting, the weight of determination, proportion of negative model parameters, strength of correlation between measured and modelled values...) and extrinsic (tests aginst known or expected outcomes, e.g. reduction in temporal variability over short periods, increased separability between signatures from well-defined training sites in VISNIR space).

To show the impact of BRDF on a spectral vegetation index, an NDVI series (258k) was calculated from modelled VIS and NIR reflectances in the usual way (NIR-VIS/NIR+VIS) for nadir viewing but with different solar zenith angles in the range 0 - 80 degrees. Since the model is reciprocal (i.e. reflectance is the same if the positions of the sun and sensor are exchanged), this is equivalent to having the sun overhead and changing the view zenith angle.