Dr. Frank W. Davis
Center for Remote Sensing
and Environmental Optics
University of California
Santa Barbara, CA 93106
Co-Investigators:
John M. Melack, Univ. of Calif., Santa Barbara
(UCSB)
Biomass Modeling of the Ponderosa Pine Forests of Western North America with
SIR-C/X-SAR
for Input to Ecosystem Models
OBJECTIVES
Integrate existing forest biophysical measurements with field measurements and calibrated
aircraft
SAR
and calibrated
SIR-C/X-SAR
images to form a spatially registered data
set for model development and testing. Extend our L-band radar forest model to X-
and C-band and four polarizations.
Determine the major backscattering components from a forest using aircraft
SAR
and
SIR-C/X-SAR.
Develop and evaluate an inversion procedure through which the above-ground forest
biomass, partitioning, patchiness, and spatial and height distributions within a
stand can be estimated from
SAR
images.
PROGRESS
Our
SIR-C/X-SAR
research has focused on forest backscatter modeling and
SIR-C
data
analysis for Loblolly pine forests at the Duke Forest Supersite. In support of the
modeling and analysis,
UCSB
graduate students conducted a field study in Duke Forest
during the
SRL-1
mission. We are working in close collaboration with Eric Kasischke at
Duke University who has provided forest stand data.
Data Acquisition/Analysis
Single-look calibrated
SIR-C
data for the ten available Duke Forest scenes have been
ordered and received at
UCSB.
Four of the scenes are from the April mission, a time
of rapid leafing-out of trees and shrubs in Duke Forest, during which it rained on
three occasions. Six scenes are from October, at which time the leaves had not yet fallen,
and ground conditions were much drier than in April. X-SAR data has also been ordered
and received for nine scenes.
The following data processing has been completed for all ten
SIR-C
scenes: Sub-images
of the Durham Division of the forest (the region for which stand data are the most
complete) have been processed into sigma-0 images (L- and C-bands, polarizations).
The sub-images have been imported into Arc-Info Grid. An Arc-Info stand map (supplied
by
ERIM)
has been registered to each scene. Backscatter statistics for approximately
70 stands have been extracted and summarized for each data take. In our initial
analysis of the Duke
SIR-C
data, we examined the backscatter statistics and trends in the
data series that might be explained by variations in incidence angle or changes in
forest conditions from day to day and season to season. Other graphical analyses
done include comparisons of backscatter to stand age (and biomass) and comparisons of backscatter
from pairs of same-incidence angle data takes from
SRL-1
and
SRL-2.
An analysis
of model predictions versus observed backscatter changes resulting from soil moisture changes is in progress.
A study is under way on forest backscattering mechanisms, based on the radar target
decomposition method (Cloude, 1992; Van Zyl, 1994). This study is presently based
on
AIRSAR
data from floodplain swamp forest (Altamaha River, Georgia) and Ponderosa
pine forest (Mt. Shasta, California). It may be extended to include
SIR-C
data from Duke
Forest and Brazil.
Although the Shasta data set is not as rich as the Duke set, there are six scenes
(including one of Mode 16) that include the Shasta site. Therefore, we plan an analysis
of stand structure versus backscatter, utilizing our Shasta forest stand data. One
Mt. Shasta scene has been ordered and received at
UCSB
and a second scene was ordered
recently.
Forest Backscatter Modeling/Analysis
The Santa Barbara microwave canopy backscatter model underwent continuing evolution
during 1994-95, with several improvements in coding and function. Notably, the discontinuous
and continuous canopy models were integrated into a single program, and surface scattering models recently published by Oh et al. (1992, 1994) and Fung et al. (1992)
were incorporated. Two modeling studies have been completed since 1994. In the first,
Loblolly pine stands ranging from <1 to 14 kg/m2
were modeled and the simulated backscatter was compared to
AIRSAR
data. In the second
study, Loblolly stands ranging from 8 to 60 years of age were modeled. Sensitivity
of backscatter from these stands to variations in surface roughness and soil moisture was evaluated by means of simulation. Response of
SIR-C
backscatter to soil moisture
changes is presently being compared with the simulation results.
Field Work
During the April
SIR-C
mission, a study of the forest floor in six Loblolly stands
in Duke Forest was carried out by
UCSB
graduate students. We were assisted by local
high school students, in coordination with Eric Kasischke's soil moisture study.
We measured surface roughness, soil moisture, and L-band dielectric constant, and litter
depth, density, and volumetric moisture content. Some tree dielectric measurements
and observations of forest state were made near the time of shuttle overflights.
SIGNIFICANT RESULTS
We validated a canopy backscatter model for Loblolly pine forest stands at the Duke
Forest, North Carolina, by comparing the observed and modeled
SAR
backscatter from
the stands. Given the
SAR
backscatter data calibration uncertainty, the model made
good predictions (Wang et al. 1995) of C-HH, C-HV, L-HH, L-HV, L-VV, P-HH, and P-HV backscatter
for most of 25 stands studied. The mode overestimated C-VV backscatter for several
stands, and largely overestimated P-VV backscatter for most of the stands. Using the collected
SAR
backscatter and ground data, and the backscatter model, we studied
the influences of changes in biomass on
SAR
backscatter as a function of radar frequency
and polarization, and evaluated the feasibility of deriving the biomass from the
backscatter. This study showed that C-HH, C-HV, C-VV, L-VV, and P-VV
SAR
backscatter
may be insensitive to the biomass change. L-HH, L-HV, P-HH, and P-HV
SAR
backscatter
changed more than 5 dB as the biomass varied. This study also showed that the L-HH
and P-HH backscatter or L-HV and P-HV backscatter may be used to develop algorithms
to retrieve trunk biomass or canopy biomass of the Loblolly pine forests.
For young (< 15 years old) Loblolly pine stands at Duke Forest (North Carolina, USA),
when the ground was wet, the observed
ERS-1SAR
backscatter from short-grass fields
of 0.05 kg/m2 biomass was equal to the backscatter from the stands, and there was no significant
correlation between the backscatter and biomass (r2 = 0.19). Under dry soil conditions, the backscatter increased about 2 to 3 dB as
the biomass (Wang et al. 1994) increased from 0.05 kg/m2 to about 0.5-1.5 kg/m2, and the backscatter may be saturated near a 0.5-1.5 kg/m2 biomass level. The correlation coefficient between the backscatter and biomass was
r2 = 0.46. When the Santa Barbara microwave canopy backscatter model was applied to
simulate the
ERS-1SAR
backscatter from the stands over dry ground, modeled and observed
backscatter had similar trends with increasing biomass. For these stands, sensitivity analyses using the model showed that as the surface-soil moisture increased, the
major contributor to the total backscatter was changed from canopy volume scattering
to surface backscatter between 0.4 and about 1 kg/m2. Signal saturating at low standing biomass, and high sensitivity to soil moisture
conditions limits the value of a short-wave (C-band) and steep local incidence angle
(23 deg.) microwave sensor such as the
ERS-1SAR
for forest monitoring.
Measurements of forest floor surface roughness were made in Duke Forest Loblolly
pine stands in April 1994 in connection with the
NASA
Shuttle Imaging Radar (SIR-C)
mission. A simple, inexpensive, and field-worthy surface roughness gauge (SRG) was
built for this field study. The problem of how to define the "surface" within the layered
forest floor system was addressed. A practical approach to forest floor roughness
measurement tailored for the modeling of forest floor scattering was proposed.
We estimated rms height and correlation length of the soil surface for 56 transects. Overall
mean rms height is 2.2 cm and mean correlation length is 29 cm. Within-stand sampling
variance is large, but stand means are similar. There is no evidence to indicate
that surface roughness varies as a function of stand age or transect orientation.
The data collected can be used to parameterize forest floor surface scattering models
for Duke Forest and possibly other similar forests.
The Santa Barbara microwave canopy backscattering model was used to investigate the
sensitivity of forest backscatter to soil roughness and moisture. Surface scattering
was simulated using the empirical model of Oh et al. (1992, 1994, Salita, 1995).
Values for surface roughness and soil moisture parameters were based on the ranges measured
in Duke Forest during
SRL-1
and
SRL-2.
The simulations indicate the following:
At C-band, co-polarized backscatter was moderately sensitive to surface conditions
at 20-30 degree incidence angles for a young (8 yr.) stand and at 20 degrees for a mid-aged
(25 yr.) stand. C-HV was moderately sensitive only to the young stand at 20 degrees.
C-band was insensitive to surface conditions at shallower incidence angles and in more mature forests. L-HH was moderately sensitive in mature stands and strongly
sensitive in young and mid-aged stands in the range of surface conditions simulated.
L-VV had moderate sensitivity only for one mid-aged stand at all incidence angles.
L-HV was insensitive for all stands and angles. The insensitivity of L-HV to surface
conditions is one likely cause the for relatively strong correlations observed between
L-HV backscatter and biomass. Initial analyses of the Duke
SIR-C
data do not demonstrate the response of backscatter to soil moisture predicted by the modeling. Further
work is needed to isolate the effects of soil moisture from other factors affecting
SIR-C
backscatter.
The radar target decomposition method was verified using
SAR
images containing dihedral
corner reflectors. The method was applied to decompose multifrequency
JPLAIRSAR
backscatter from two types of forests to analyze and understand scattering mechanisms
in forested environments. For the Ponderosa pine forest (Mt. Shasta, Calif.), as
SAR
wavelength increased from C-band to P-band, scattering with an odd number of
reflections decreased, scattering with an even number of reflections increased, and
diffuse scattering showed no clear trend. For the floodplain swamp forest (Altamaha River,
Georgia), scattering with an odd number of reflections dominated at C-band. Scattering
with an even number of reflections was strong at L-band and stronger at P-band.
Diffuse scattering from both marsh and swamp forest was <20% of total scattering at C-band
and <15% at L-band and P-band. This study may be expanded to include
SIR-C
data
from Duke Forest and Brazil.
Preliminary findings based on Duke Forest
SIR-C
data analysis
1)
SIR-C
backscatter from our site is exponentially distributed as expected for dense
forest canopies that satisfy Rayleigh fading assumptions (Ulaby and Dobson, 1989).
Because the fading statistics are well described, we can place confidence bounds
around forest stand backscatter estimates. For the Duke Loblolly stands, 90% confidence
intervals are in the range of +/- 0.3 to +/- 0.7 dB, depending on stand size.
2) Sigma-0 of the forest increases as incidence angle decreases. The trend is stronger
than expected from modeling or previous data. The cause of the trend and whether
or not it is statistically significant have not yet been determined.
3) As anticipated from previous studies, the relationship between backscatter and
stand age (or biomass) is weak. Only L-HV appears to correlate directly with forest
biomass. (The indirect biomass retrieval method discussed below may overcome this
apparent limitation.)
4) In most cases, variations in backscatter between different dates (for data takes
having the same incidence angle) are on the order of 0-3 dB for individual
stands and
<2 dB for averages of the stands. We cannot be sure whether the observed variations
are due to calibration error or to known or unidentified differences in forest conditions.
The variations are of the same magnitude as stand-to-stand sigma-0 differences attributable to biomass. Within-scene relative calibration is being considered to
remove the systematic variations.
5) With the exception of one rainy morning during
SRL-1,
backscatter was not higher
for wet soil than dry soil conditions, even at small incidence angles. This apparent
contradiction to model-based expectations is under scrutiny and may lead to better
understanding of sources of forest backscatter uncertainty.
FUTURE PLANS
Biomass separability under varying conditions
Our first priority is to investigate the separability of forest stands of differing
structure and biomass under different environmental conditions for a range of incidence
angles. The main emphasis will be on Duke Loblolly stands, however, Mt. Shasta data will also be evaluated. Our approach is outlined in the following questions: Under
what conditions (incidence angle, moisture, season) do multiple regressions of
SIR-C
backscatter give the strongest predictions for total stand biomass, for crown biomass, for basal area, or for tree height? For the same conditions, how effective is the
L-HV/C-HV ratio in predicting biomass? This research will extend well into 1996.
Some of the results will be utilized in the indirect biomass retrieval study (below).
Indirect Biomass Retrieval
It has become increasingly clear over the past three years that the direct regression
approach for estimating forest biomass from backscatter is not very powerful, except
in fairly low biomass forests. The use of cross-polarization ratios, as proposed
by Ranson and Sun (1994), raises the biomass level at which the signal saturates, but
it remains to be seen how applicable the technique will be for diverse forest structural
types of moderate to high biomass. Dobson et al. (1995) speak to this issue very
plainly. In that paper the authors observe that much of the data scatter seen in regressions
of backscatter versus biomass is due to variations in forest structure. For example,
a few large trees may have a total biomass equivalent to that of many small trees but the former produce higher backscatter.
Dobson et al. advocate an indirect method for estimating biomass. In the first step,
regressions against backscatter are developed to predict crown biomass, basal area,
and height of each forest structural class. Second, estimates of trunk biomass are
derived from basal area and height using allometric equations. Finally, total biomass
for each class is the sum of crown biomass plus trunk biomass. Instead of regressions
on the aggregate forest system, the regressions are on individual structural components that are more like what microwaves actually interact with.
We plan to apply Dobson's indirect method to improve biomass estimates in the Duke
Loblolly stands. Apart from the question of extensibility of this approach to the
Loblolly forest, an equally important question needs to be addressed and will be
investigated in this study: how consistent are results from pass to pass, for varying calibration
and incidence angle, and under changeable forest conditions? The usefulness of the
indirect biomass retrieval method will depend on whether it is broadly applicable
and repeatable under different
SAR
and forest conditions.
Texture
A parallel study is planned utilizing texture measures to complement backscatter intensity
in the estimation of biomass. There is much evidence that for many forests structural
evolution accompanies biomass change over time. This is certainly true in the "old field" Loblolly stands at Duke. Furthermore, several papers demonstrate improved
discrimination of forest types in
SAR
images when texture is used to augment tone.
Differences in biomass, sensed indirectly through differences in forest structure,
may be estimated more robustly through image texture than is possible using backscatter
alone. Preliminary work has begun with a literature search and application of texture
algorithms to Mt. Shasta data. At present, we view this study's primary goal as
generating an independent channel of
SIR-C
information to feed into biomass estimation
procedures.
PUBLICATIONS
Saleta, J., 1995, Stand Discrimination in a Western Coniferous Forest Using
AIRSAR
Data
, M. A. thesis, Dept. of Geography,
UCSB.
Wang, Y., F. W. Davis, J. M. Melack, E. S. Kasischke, and N. L. Christensen, 1995,
The effects of changes in forest biomass on radar backscatter from tree canopies,
Int. J. Remote Sensing
, vol. 16, no. 3, pp. 503-513.
Wang, Y., E. S. Kasischke, J. M. Melack, F. W. Davis, and N. L. Christensen, Jr.,
1994, The effects of changes in Loblolly pine biomass and soil moisture on
ERS-1SAR
backscatter, Remote Sens. Environ.,
vol. 48, pp. 1-25.
REFERENCES
Cloude, S. R., 1992, Uniqueness of target decomposition theorems in radar polarimetry,
in Direct and Inverse Methods in Radar Polarimetry, Part I
, (ed. by W. M. Boerner et al.) Kluwer Academic Publishers, Dordrecht, The Netherlands.
Dobson, M. C., F. T Ulaby, L. E. Pierce, T. L. Sharik, K. M. Bergen, J. Kellndorfer,
J. R. Kendra, E. Li, Y. C Lin, A. Nashashibi, K. Sarabandi, and P. Siqueira, 1995,
Estimation of forest biophysical characteristics in Northern Michigan with
SIR-C/X-SAR,
IEEE Trans. Geosci. Remote Sensing
, in press.
Fung, A. K., Z. Li, and K. S. Chen, 1992, Backscattering from a randomly rough dielectric
surface, IEEE Trans. Geosci. Remote Sensing
, vol. 30, no. 2, pp. 356-369.
Oh, Y., K. Sarabandi, and F. T. Ulaby, 1992, An empirical model and an inversion technique
for radar scattering from bare soil surfaces, IEEE Trans. Geosci. Remote Sensing
, vol. 30, no. 2, pp. 370-381.
Oh, Y., K. Sarabandi, and F. T. Ulaby, 1994, An inversion algorithm for retrieving
soil moisture and surface roughness from polarimetric radar observation, Proceedings of
IGARSS
'94
, pp. 1582-1584.
Ranson, K. J., and G. Sun, 1994, Mapping biomass of a northern forest using multifrequency
SAR
data, IEEE Trans. Geosci. Remote Sensing
, vol. 32, no. 2, pp. 388-396.
Ulaby, F. T. and Dobson, M. C., 1989, Handbook of Radar Scattering Statistics for Terrain
, Artech House, Norwood, MA.
Van Zyl, J. J., 1994, Application of Cloude's target decomposition theorem to polarimetric
imaging radar data, Proceedings of the Third Spaceborne Imaging Radar Symposium
, held at Jet Propulsion Laboratory Pasadena, CA, Jan. 18-21, 1994, pp. 207-216.