The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B7. Beijing 2008 
15 
are helpful to complement the spectral measurements for a 
complete and robust characterization of vegetation canopies 
and heterogeneity of a coniferous canopy based on its degree of 
reflectance anisotropy as observed by the multi-angular 
imaging spectrometer CHRIS. This study was performed on the 
Swiss National Park (SNP) study site using a four-angle 
CHRIS/PROBA data set that was acquired over the SNP on 17 
February 2004 (sun zenith: 59.7°, azimuth: 165.4°) under cloud 
free conditions. The data set was subsequently geometrically 
and atmospherically corrected following the approach described 
in the preceding chapter. 
CHRIS/PROBA observations were investigated and related to 
surface structure using the parametric Rahman-Pinty-Verstraete 
(RPV) model (Rahman et al., 1993), which simulates the 
anisotropy of a surface reflectance as a function of four 
parameters. It has been shown that the RPV Minnaert function 
parameter k, describing the degree of anisotropy, is related to 
canopy structure and subpixel heterogeneity (Pinty et al., 2002; 
Widlowski et al., 2004). Hence, the Minnaert function 
parameter k, which quantifies the overall shape of the surface 
Bidirectional Reflectance Factor (BRF) is of particular interest. 
Based on the k values, the anisotropy of the observed HDRF 
can be classified into a bell- (A>1) or bowl-shaped (&<1) pattern. 
Bell-shaped BRFs are associated to heterogeneous canopies of 
medium density over a bright background. The inversion of the 
RPV model against the multi-angular data over a subset of the 
pre-processed CHRIS/PROBA scene provided spatial fields of 
the RPV model parameters describing the anisotropy of the 
observed surface reflectance. The performance of the inversion 
was affected by errors due to sloping terrain (i.e., topography). 
Thus, the subsequent interpretation of the retrieved model 
parameters was restricted to areas with slopes of up to 10° and 
inversion uncertainties below 10%. For those conditions 
measured multi-angular data were fitted well by simulated BRF 
based on the retrieved RPV parameter sets. 
This case study demonstrated the successful inversion of the 
parametric RPV model against the independent information 
source of multi-angular CHRIS/PROBA observations. The RPV 
inversion allowed discriminating between different surface 
types based on their inherent anisotropy. Results showed the 
potential to distinguish within a forest stand between closed 
canopies and ones of medium density, thus delivering 
quantitative surface structural information. 
3.2 Canopy Biochemistry Estimation 
Knowledge about plant biochemistry is important for a range of 
environmental applications (Asner and Vitousek, 2005; Curran, 
2001; Ustin et al., 2004). 
The objective of the second case study was the investigation of 
directional CHRIS/PROBA data for an improved estimation of 
foliar nitrogen concentration (C N ) and water content (C w )- We 
investigated a) whether the added information in remotely 
sensed multi-angular data can improve C N and C w estimates 
and b) whether certain sensor viewing angles emerge to be 
beneficial for estimating C N and C w . The study was performed 
on the forests of the Swiss Plateau study site Vordemwald 
(VOR) as described above. In July 2004, an extensive field data 
campaign took place covering 15 subplots that were chosen 
according to their species composition. At each subplot 3-10 
tree crowns were determined for foliar sampling. The trunk 
position of each tree was measured with a Trimble GeoXT GPS 
receiver. A complete CHRIS/PROBA scene (five viewing 
angles) acquired on 1 July 2006 over the VOR study site was 
geometrically and atmospherically corrected following the pre 
processing methodology described earlier in this paper. In total, 
spectra of 60 field-sampled crowns were extracted from the five 
CHRIS/PROBA images by using the geographical trunk 
positions (vector data) of the sampled trees to locate the crown 
pixels in the images (Gorodetzky, 2005). CHRIS/PROBA data 
acquisition was two years after field data collection but during 
the same phenological period (July). We assumed a stable C N 
and C w level during July and only small inter-annual variability 
for nitrogen concentration and for leaf water status due to 
similar climatic conditions in the years 2004 and 2006. 
Multiple linear regression analysis was applied to fit models 
between the dependent variables (C N and C w ) and all possible 
viewing angle combinations of four spectral data sets: SPEC, 
BNC, CRDR, NBDI. SPEC includes original reflectance values; 
BNC includes band depths normalized to the waveband at the 
center of the absorption feature (Curran et al., 2001; Kokaly and 
Clark, 1999); CRDR includes continuum-removed derivative 
reflectance (Mutanga^t al., 2004; Tsai and Philpot, 1998) and 
NBDI includes normalized band depth index values (Mutanga 
et al., 2004). To limit the number of spectral wavebands used in 
the regression models, this study employed a statistical variable 
selection method, namely an enumerative branch-and-bound 
(B&B) search procedure (Miller, 2002). 
To summarize the findings of this case study, the following 
conclusions can be drawn: 1) additional information contained 
in multi-angular data improved regression models for C N and 
C w estimates and lowered cross-validated RMSEs considerably, 
2) strongest effects upon R 2 for each data set (SPEC, BNC, 
CRDR, NBDI) yielded from models with the same number of 
viewing angles can be achieved when adding a second and third 
viewing angle (Figure 3 and Figure 4), 3) monodirectional 
models developed on backward scattering viewing directions 
were generally superior to models based on forward scattering 
data and 4) untransformed reflectance data (SPEC) often 
outperformed continuum-removed data when using only one 
viewing zenith angle (Huber et al., 2008). 
Figure 3: Model coefficient of determination (R 2 ) of nitrogen 
concentration regressed on the data sets SPEC, BNC, CRDR 
and NBDI, respectively. R 2 represents the mean of all models 
with the same number of viewing zenith angles involved. For 
instance R 2 of one angle stands for the mean of five 
monodirectional models (± 36°, ±55° and nadir).