the
AM,
ich),
r as
test
data
ight
heat
nner
TIR
sion
The
rtho
s of
‘the
uted
s as
IAPRS & SIS, Vol.34, Part 7, “Resource and Environmental Monitoring”, Hyderabad, India,2002
many as possible input parameters of the applied physical
models were measured, such as total dry matter and total water
content of samples, leaf area index and mean leaf angle. The
wet samples corresponding to an 0.25m^ area were weighed,
oven dried and weighed again to assess total dry matter and
total water content. Leaf area index and mean leaf angle were
measured with the Licor LAI-2000 analyser. Combining leaf
area index with total dry and wet matter, specific dry matter and
specific water content are derived. From the mean leaf angle
and a shape parameter the leaf angle distribution LAD is derived
based on an ellipsoidal distribution. Using the Licor LA1-2000,
ears, stems and leaves of winter wheat plants cannot be
separated for the assessment of leaf area index and mean leaf
angle. Thus, all measurements are made without separation of
different plant components. The chlorophyll content has been
estimated qualitatively considering the visual appearance of the
leaves. Most measurements are repeated to assess accuracy
properties.
4.2 Estimation of vegetation parameters at the test sites
The estimation of vegetation parameters was conducted using
the described models and Daedalus multispectral scanner data.
The measurement sites within the fields are arbitrarily divided
in ground control points and validation points.
X Ground control point
O Validation point
T Mass point
100 meters
Figure 2. Data acquisition with Daedalus ATM scanner and
ground truth measurements on June 28" 2000.
Overview about the ground control and validation
points at the left and mass points with disturbed
vegetation (black regions) at the right.
The ground truth measurements at the ground control points are
an essential part of our model, whereas the measurements at the
validation points are used to prove the accuracy of the inversion
process. No measurements are made at mass points, which are
used to estimate unknown vegetation parameters at any position
within the field. To reduce computing time mass points are
chosen in a grid of 70x10 pixels. Figure 2. illustrates the
distribution of the different types of points. In this special case
seven ground control points are chosen. Disturbed pixels and
pixels near the wheel tracks have been eliminated. Thus; mass
points lying on the eliminated pixels have been excluded from
the inversion process. The Daedalus multispectral scanner data
have been smoothed with a mean filter of mask size 5x5.
Approximate values for the vegetation parameters at mass
points are estimated using simulated annealing. After the least-
squares adjustment, the resulting maps of vegetation parameters
are calculated by interpolating between the estimated vegetation
parameters (v. Figure 3.).
specific dry matter g/cm2 specific water content
0.04 d
14 0.035
# 0.03
0.025
Figure 3. Maps of estimated vegetation parameters, which are
results of the model inversion. The model inversion
has been conducted at validation and mass points.
4.3 Accuracies
Our goal is to derive a strategy for the use of ground control
points. From a practical view, the number of necessary ground
control points should be low to reduce required ground truth
measurements. On the other hand, the robustness of the
inversion process and attained accuracies of the estimated
vegetation parameters should be high. In figure 4 two kinds of
accuracies of the estimated leaf area index for four
combinations of ground control points are illustrated. The
theoretical standard deviation, which is derived from the least-
squares adjustment, corresponds quite well with the empirical
deviation at the validation points with a tendency of higher
empirical deviations for all combinations. The empirical
deviation is the difference between measured and estimated
vegetation parameters. ',
Q Ground control point
0.15 Theoretical accuracies
0.15 | Empirical accuracies
1
Ground control points
Figure 4. Maps of leaf area index estimated with Daedalus
scanner data of June 27" 2001. For four
combinations of different ground control points the
empirical and theoretical deviations of the leaf area
index at the validation points are calculated.
Simulation studies point out the influence of ground control and
mass point constellations on the accuracies. Figure 5. shows the
relation between the number of ground control resp. mass points
and the theoretical accuracies of the vegetation parameters and
