International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004
reported R? values are therefore valid only for the transformed
space. A scatterplot of field-based In(canopy bulk density)
measurements versus predicted values is shown in Figure 7.
Field In(Canopy Bulk Density)
4.5
T T T
-4.0 -3.5 „3.0 -2.5 -2.0
Predicted In(Canopy Bulk Density)
Figure 7. Field-based (y) versus predicted (log-transformed)
canopy bulk density measurements (x) (with 1:1 line shown).
4.4 Canopy fuel weight
The regression model for (log-transformed) canopy fuel weight
had an R? of 0.77 and an adjusted R^ of 0.75. A scatterplot of
field-based In(canopy fuel weight) measurements versus
predicted values is shown in Figure 8.
Field In(Canopy Fuel Weight)
7.5
o
T T T T T
8.0 8.5 9.0 9.5 10.0
Predicted In(Canopy Fuel Weight)
Figure 8. Field-based (y) versus predicted (log-transformed)
canopy fuel weight measurements (x) (with 1:1 line shown).
5. DISCUSSION
The results indicate that information related to three-
dimensional geometry, interferometric observables, and
backscatter characteristics of the forest canopy provided by
multi-frequency, polarimetric IFSAR systems can be used to
obtain accurate measurements of canopy fuel parameters.
Although the mature conifer forest stands within the study area
are of fairly uniform height (see Figure 5), the interferometric
observables and polarimetric backscatter data explain
variability in the vertical structural characteristics of these
forest stands, as evidenced by high correlation between field-
based and predicted canopy base heights (see Figure 6).
The use of model-based estimates for the various fuel
parameters will also introduce a significant source of variability
in the regression model. The fuel models use a number of
assumptions (i.e. uniform vertical distribution of crown fuels)
that are clearly over-simplifications of reality. More detailed
crown fuel models will be needed in the future if they are to be
used as ground truth in comparison to high-resolution remotely-
sensed estimates.
Although each predictive model contained a large number of
independent variables, the adjusted R? values were not
significantly lower than the R? for any of the models, indicating
that none of the models contained a large number of extraneous
variables working to artificially inflate this measure of fit.
Correlation matrices for the regression models also indicate that
there is little multicollinearity in these models. Admittedly, the
large number of predictor variables make it difficult to establish
a physical interpretation for these regression models. Further
work will certainly be needed to fully understand the physical
relationships between IFSAR observables and the array of
forest structural parameters.
It should also be noted that at this time, most IFSAR processing
algorithms are (perhaps not surprisingly) optimized for
extraction of bare earth elevations, not for accurate
measurement of canopy surface structure. For example, some
errors in the X-band measurement of canopy height were due to
filtering of the interferogram (Mercer, 2004). Although it is
unclear what effect this may have on the results, future work
will look at optimizing X-band processing for the forest
measurement application.
6. CONCLUSIONS
Results of this analysis indicate that IFSAR can be used to
generate accurate estimates of canopy fuel parameters
efficiently and economically over extensive areas of forest
Given the capability of IFSAR to penetrate clouds and smoke,
there may be potential for near real-time collection of fuels data
in fire-prone areas. As extensive areas of the world begin to be
covered through large-scale terrain mapping campaigns
(NextMapBritain, NextMapUSA, etc.) IFSAR data will be
increasingly available for forest mapping applications.
Once the regression analysis have been carried out and the
predictive models developed, canopy fuel parameters can be
mapped over the entire extent of the IFSAR coverage, resulting
in a GIS coverage that could be used to support fire behavior
modelling or fuels management programs.
An
20€
inte
Wo
Bei
ET
Sta
FV
Brc
Sim
Get
Ser
Bre
Sys
and
Du
Fill
pro
cla:
Tra
Get
Fin
dev
Og
Sta
Hay
inte
Ge
Hof
Mo
In:
Sen
on |
Hus
slas
on (
Imh
ram
on (
Mei
exp
Pho
Me
Anc
Can
ISP.
dim
Ore