In: Wagner W., Székely, B. (eds.): ISPRS TC VII Symposium - 100 Years ISPRS, Vienna, Austria, July 5-7, 2010, IAPRS, Vol. XXXVIII, Part 7B 
343 
with different levels of adjacent illumination sources. For those 
conditions various other reflectance quantities are needed 
(Martonchik et al, 2000). 
t 
Nadir 
Back 
ward 
WWltlll J J/////X 
WVUU I « J I////// 
\\\%V% I I I I 4///✓// 
> I * » *///✓✓/ 
%%%*•*** 
» » « 4 
iff« 
IMI IMWW 
9 91 immv 
Figure 1. Illustration of the view-illumination geometry of trees 
in the focal plane of a nadir looking aerial camera. The 
flying direction is upwards and Sun is 26° left of it. The 
white rectangles depict the nadir and backward viewing 
in the ADS40 line sensor, which we used in this study. 
Depending on the view and illumination directions, the pixel 
can view, in the extreme case of hot-spot geometry, shadow-free 
targets. When a tree crown is back-lit, the pixels sample mostly 
shadowed targets or forward-scattering canopy elements. The 
diffuse light incident at a tree consists of the light scattering in 
the atmosphere, but also of light scattered by the adjacent trees 
and the background. Taller neighboring trees are both reflectors 
that contribute to the incident light, but they also attenuate 
hemispherical diffuse light. Multiple scattering by trees contri 
butes to the total radiance towards the sensor. 
The spectra of a tree in an aerial image are measured from a 
sample of pixels that are geometrically linked to the tree. If the 
geometry of the canopy is known, it is possible to sample the 
crown for the Sun-lit and shaded parts (Korpela, 2004; Larsen 
2007).. 
There are many sources of inter- and intratree reflectance variat 
ion. The varying phenological and physiological status affects 
reflectance. Epiphytic lichens, flowers, and cones constitute 
sources of variation. The structure of branches, shoots, and 
needles, and the whole branching pattern and crown shape vary 
between and inside individuals, and the crown structure changes 
with age. The functioning, structure, and the environment all in 
terrelate in a tree. Structural differences explain largely the va 
riation in reflectance and anisotropy. 
The scale of observation has important implications. In aerial 
images, a crown can be sampled by hundreds of pixels, while in 
satellite images several crowns fill a pixel. In sub-meter pixels, 
the scale is at the level of branches and shoots. Since bidirectio 
nal effects have their origin in sub-pixel shadow casting, it is 
possible to observe branch-level anisotropy. Scale is linked with 
sensor altitude and the medium. The majority of the atmosphe 
ric effects occur below the 3-6 km altitude, which stresses the 
importance of atmospheric modeling in airborne images if target 
reflectance data is strived for. 
The introduction of digital sensors, direct sensor orientation, 
and image post processing systems, have all altered photogram- 
metric practices. In forest applications, this development has 
been covered by the expansion of LiDAR. Digital sensors are 
relatively calibrated to a uniform response or absolutely calibra 
ted to produce at-sensor radiance (ASR) data. Among photo- 
grammetric sensors, absolute calibration exists for the ADS40 
line camera. It measures ASR in 4 bands in two directions. A 
laboratory calibration is applied throughout a radiometry chain 
that also includes radiometric correction methods, which are 
implemented in the Leica XPro software (Beisl et al., 2008). 
The performance of ADS40 in tree species classification in 
Finland was simulated in Heikkinen et al. (2010). An additional 
band at 710-725 nm provided the best improvement. With the 
original 4 bands, the simulated sp. classification accuracy was 
75-79%, while it was up to 85-88%, using the fifth band at the 
red-edge. 
The aspects of utilizing reflectance anisotropy or calibrated 
images in high-resolution RS forest applications are largely 
unexplored. Studies have indicated that anisotropy might provi 
de additional information for sp. classification (Deering et al., 
1999). Line sensors offer fewer viewing directions, but owing to 
their view-geometry (Fig. 1), they sample the reflectance aniso 
tropy in ID. This may facilitate the interpretation compared to 
frame images. Our overall objective was to explore the ADS40 
line camera for tree sp. classification of Scandinavian forests, 
where airborne line sensors or airborne calibrated reflectance 
data have not been tested thus far. Our three detailed objectives 
were as follows. 
1. Implement an ADS40 sensor model. 
2. Develop a method by which Sun-lit, shaded, camera- 
visible, and occluded parts of crowns can be determined 
to enable an extraction of image features in different illu 
mination classes (Korpela, 2004; Larsen, 2007). 
3. Examine the anisotropy and variation of spectral image 
features in radiometrically corrected images to study the 
performance potential of these data for tree species dis 
crimination.. 
2. MATERIALS AND METHODS 
2.1 Study area and reference trees 
The experiments were carried out in Hyytiálá, southern Finland 
(61°50'N, 24°20'E). The study area extends 2x6 km and com 
prises protected and commercial forests. We used reference 
trees measured in 2005-2009 in 121, 0.04-1.8-ha plots. The 
age of trees was 15-150 years, and a total of 15687 reference 
trees were formed by image- or LiDAR-visible trees (visual in 
terpretation). Merely 3.8% of the trees had a relative height of 
below 0.5. Tree maps were used to derive a proximity class for 
each tree, which describes the dominant species among the ad 
jacent trees. 
2.2 ADS40-SH52 and LiDAR data 
The ADS40 flight was carried out on August 23, 2008 at 10-12 
local time (7-9 GMT) in 15 strips. Solar elevation was 27°-37° 
and there were few clouds during the campaign at 700 m agl. 
Reflectance targets and in-situ radiometric measurements were 
carried out simultaneously. The strips were flown at 1,2, 3, and 
4 km altitudes and multispectral data was recorded for the nadir 
(N00A) and backward (B16A) view directions. At 1 and 2 km 
only N00A or B16A view was active. Strips were flown mostly 
in S-N direction, but also in E-W. We used discrete-return 
LiDAR data from 2006-2008 for the tasks of tree crown mo 
delling and occlusion determination (section 2.6).