127 
stands measured at L- and C-band and at 60 degrees 
and 15 degrees incidence angle and HH-polarization. 
To enlarge the accuracy the inversion was applied 
to the averaged return signal from a stand (+/- 20 
A-scans =220m). All measurements were repeated 4 
times. Some measurements have been rejected because 
of non-neglectable variations in dr or d0^(fig u re 1) 
within the measured stand. The L-band measurements 
were taken at July 18-th 1985 and the C-band measure 
ments at July 6-th 1984, both at the Roggebotzand 
site. The poplar stands were fully foliated. 
5.1 Discussion of results 
(a) C-band, 16.5 degrees inc. (figures 9a, 9b, 9c). 
The 4-level 9 meter spacing model was applied for 
the measurements in the C-band at 16.5 degrees inci 
dence angle for 3 poplar stands, with the poplar 
clones 'Robusta', 'Oxford' and 'Geneva'. These 3 
stands differ in structure. The 'Robusta' stand has 
a lower degree of crown closure and the trees are a 
few meters taller. A significant part of the radar 
return (~40%) originates (according to model assump 
tions) from the ground, the rest from layers 2 and 3 
corresponding to the tree crowns (fig. 9a). Layer 
1 corresponding to a plane beneath the crowns where 
only stems are present does not contribute signifi 
cantly. It is not apparent from these data whether 
the stems contribute through multiple reflections 
via the ground or not. (These contributions have the 
longest delay time, longer than the returns from the 
ground, but will be assigned to level 0 in this model) 
The 'Oxford' stand has a heavy understory of bee 
ches. This explains the relatively strong contributior 
of layer 1 . 
The 'Geneva' stand structure resembles the 'Oxford' 
stand structure but lacks a heavy understory. 
(b) C-band, 60 degrees inc. (figure 9d). 
At 60 degrees and for the L-band cases the returns 
from individual layers become harder to disentangle. 
Therefore the 3-level 12.5 m spacing version of the 
model was selected as a more appropriate choice. In 
these cases only the 'Robusta' parcel was analysed. 
According to the model assumptions in C-band at 
60 degrees most of the returns originate from the 
crowns, only a small amount 10-15 % originates from 
the ground or from stem-ground interactions. 
(c) L-band, 14.5 degrees inc. (figure 9e). 
In L-band at 14.5 degrees inc. ~75% of the back- 
scatter signal originates from the ground or stem- 
ground interactions. Probably the (fully foliated) 
crowns are very transparent at this frequency since 
the degree of crown closure is higher than 80%, or 
(but this is unlikely) the grassy forest floor has 
an extremely higher level of backscattering. The 
crowns contribute "25% to the backscattered power. 
(d) L-band, 61.5 degrees inc. (figure 9f). 
At 61.5 degrees both crowns (layers 1 and 2) and 
ground contribute. Even at this angle, where the path 
of propagation through the canopy is very long, 
scatterers near the ground still contribute signi 
ficantly. 
5.2 Modelling aspects. 
The radar return parameter y of the forest is di 
rectly related to the values of the coefficients 
from equation 15. This is illustrated for the 3-level 
model as 
Y total = cAz + cAl + ( 16 ^ 
in which c is a constant following from the radar 
equation (equation 13). 
The subdivision of y in a sum of contributions 
originating from horizontal levels (according 
to the assumptions of the multi-level model) can 
be related directly to the subdivision made in the 
cloud model of section 3. For the 2-layer cloud 
model 
Y total =Yl 
+ TlY2 + TlT2Y soil 
0 7) 
as a result 
Yi = cA 2 
(18a) 
liY 2 = cAl 
(18b) 
TlT2Y soil = CA ° 
(18c) 
and (section 
3) 
I1T2 “ I e q 
( 18d) 
In this way the measurement of physical parameters 
of the forest canopy other than y can contribute 
to model-making, not only for the cloud model as 
illustrated here, but in fact for all models with 
a physical basis. Note that the levels in the multi 
level model do not necessarily have to be equidis 
tant and may be adapted to the object's geometry. 
6. RETHINKING SCATTEROMETER EXPERIMENT DESIGN 
It is obvious that the DUTSCAT allows the experi 
menter to measure y values and in the case of 
forests these y values can be separated, more or 
less, in contributions from a number of horizontal 
layers. The accuracy of this subdivision can be in 
fluenced by experiment design as stated at the end 
of section 4. 
The value of y can be calculated from these 
contributions (equation 16). 
When one is only interested in the y value the 
whole procedure still has to be followed, unless the 
separation of contributions from various levels in 
the canopy can be neglected. In that case the 
'standard' pre-processing procedures can be followed, 
in which y is directly related via the radar 
equation (eq. 13) to the returned power signal. This 
procedure is far less complicated. 
One of the questions to answer now is; under which 
circumstances is standard pre-processing, where 
all scatterers are assumed to be located on a single 
plane, allowed for forests? 
In this paper a conclusive answer to this question 
will not be given. As an illustration of the impor 
tance of this matter the following simulation is 
made. 
The relative contributions, the coefficients A^, 
of the poplar stand 'Robusta' at C-band, 16.5 
degrees incidence and.HH-polarization are known 
(figure 9a). These are; level 0=43%, level 1=7%, 
level 2=48% and level 3=2%. When these numbers are 
filled out in equation 15 for the 4-level case, this 
equation can serve as a model for simulation purpo 
ses . 
In figure 10(a) the return signal is simulated for 
a 9 meter level spacing at a flying height of 323 
meter (compare with figure 7c). Figure io (b) shows 
the return signal when the level spacing is ignored. 
Since for the C-band the maximum level of power can 
be related directly to y, according to standard 
pre-processing models, it is clear from this 
example that the y value will-be under-estimated 
by approximately 2.4 dB. 
In figure 10(c) the simulated return signal for the 
9 meter level spacing at a flying height of 1800 
meter is shown. This is the maximum altitude at this 
angle of measurement for the specified operating 
range in table 2. Figure 10(d) shows the simulated 
return signal at this height of flight when the level 
spacing is ignored. The under-estimation of y is 1.3 
dB. Although, as a result of the increased height 
of flight the error decreased, it still cannot be 
ignored. 
In general the error one makes when height diffe-