il 2004
ybe the
sen to
odelled
data in
NOAA
6 data.
hey are
JAA 15
)AA 16
e grows
tisfying.
for the
ficient of
VIS and
SSes, C.9.
rest, are
1. more
2, arable
jamed as
s for the
1 class.
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B-YF. Istanbul 2004
0,8 Ie.
0,6
0,4
0,2
Coefficient of determination
12 13 17 20 21 22 23 24 25:26 27 28 29 31
Figure 3: Coefficient of determination for the different
aggregated land cover classes (left column- VIS,
right column= NIR)
In order to investigate the regression quality of either NOAA 15
and NOAA 16 sampling data, corresponding data sets were
separately introduced into the MIB. Mean values of the
obtained statistical parameters for all classes and channels are
displayed in table 2 and are compared between those obtained
by running the MIB with the combined sampling data set.
Observations from Channel Avg R° Avg se
NOAA 15 VIS 0.804 0.015
NIR 0.814 0.019
NOAA 16 VIS 0.870 0.016
NIR 0.838 0.020
NOAA 15 & 16 VIS 0.792 0.018
NIR 0.830 0.023
Table 2: Statistical results for model inversion for different
input data sets
The input data sampling set that combines NOAA 15 and
NOAA 16 observations was also introduced into the MIB in
order to compare the performance with the Roujean and the
Modified Walthall model. Table 3 displays the corresponding
results. The NTAM yields best values for the coefficient of
determination, especially in the VIS channel. However, this is
not too surprising as neither the Roujean model nor the MWM
were designed to cope with sample data encompassing the
entire growing season.
Model Channel Avg R* Avg se
Roujean VIS 0.545 0.016
NIR 0.720 0.025
Modified Walthall VIS 0.535 0.016
NIR 0.721 0.024
NTAM VIS 0.792 0.018
NIR 0.830 0.023
Table 3: Statistical results for model inversion for the three
different models
Another very interesting extension from the determination of
the correction parameters consists in a graphical visualization of
modelled reflectances for different sun sensor target geometries
(Sandmeier & Itten 1999, Dymond et al. 2001). Again, the land
cover class 17 (olive grows) is chosen for such a visualization.
As NTAM modelled reflectances depend on values for the
proxies 05, a criterion had to be established that allows to
display modelled reflectances for land cover typical ot values:
First, the mean à; value of the input sampling data was
calculated for the two channels. Then, the corresponding
standard deviation values were added and subtracted from the
mean o; values. The figures 4 to 7 show modelled reflectances
for the VIS and the NIR. The raa is the relative azimuth angle,
the vza the view zenith angle and the sza is the sun zenith angle.
The sza is set to the fixed value of 45 degrees in all graphics.
Note that no input data with a vza » 45? are introduced.
Modelling to a vza of 52.5? is done to demonstrate model
behaviour outside the range of calibration. The displayed
graphics allow to estimate seasonal and angular dependency of
the BRDF on different land covers. The hot spot is clearly
distinguishable in all graphics.
0.36
0.33
0.30
0.27
0.24
0.21
0.18
0.15
0.12
0.09
0.06
0.03
0.00
Figure 4: Modelled VIS reflectance for olive grows where
NDVI = NDVI ein - NDVldev
raa
C pup Sw S
i CUNAG quot
ye 588°
Figure 5: Modelled VIS reflectance for olive grows where
NDVI = NDVIean zh NDVI dev
43