- PI: Prof. PAOLO CANUTI - Dept. of Earth's
Sciences, University of Florence.
Abstract - This report presents an overview of the experiments
carried out on the Montespertoli supersite during the SIR-C/X-SAR
missions and an analysis of the results obtained by the different
groups of researchers working on this experiment. The measured
quantities taken into consideration are the full polarimetric
features at L- and C-bands and the VV polarized backscattering
coefficient s° at X- band. The
influence of frequency, polarization and incidence angle on the
response of various surface types has been investigated and statistical
relations between backscattering and soil moisture, surface roughness
and forest biomass are discussed.
1. INTRODUCTION
A research activity, which aims at a better understanding of the
information obtained from multifrequency polarimetric Synthetic
Aperture Radar to be used in hydrology has been carried out in
the framework of the SIR-C/X-SAR Project (Stofan et al. 1995).
A significant phase of this study consists of investigating the
sensitivity of radar backscattering to some important parameters
(soil moisture, surface roughness, vegetation cover and biomass)
which are of primary importance in modelling the geophysical processes
of the hydrological cycle. Although the detection of these parameters
has been the subject of many investigations, carried out in past
years with ground based and airborne sensors, only a very few
preliminary investigations have been carried out using data collected
with spaceborne sensors (e.g.: Evans et al. 1988, Touzi et al.
1992). A major problem in retrieving the hydrological parameters
is that each of them affects in a different way the radar backscattering
and separating the effects requires the use of appropriate multi-frequency
polarimetric algorithms.
Multifrequency polarimetric SAR measurements carried out during
SIR-C/X-SAR mission over the supersite of Montespertoli have been
analyzed and compared with data collected at the same frequency
and polarization with NASA/JPL AIRSAR and NASDA JERS-1 on different
dates. The sensitivity of L-band HV pol to both arboreous and
herbaceous vegetation biomass has ben confirmed; moreover, a good
separation between bare and vegetated fields has been obtained
from the ratio RL/RR at C band. A classification algorithm able
to separate different surface classes has been also carried out
using multifrequency-multipolarization images. Finally the soil
moisture and the height standard deviation of surface roughness
have been retrieved by means of the semiempirical model developed
by Oh et al. (1992) and used to validate the soil erosion model
(WEPP) on Pesa and Virginio basins.
2. EXPERIMENTAL ACTIVITIES
2.1. The experimental area
The site choosen for the experiment is the basin of the Pesa river,
located in Tuscany, on the Tirrenian side of the Northern Appennines,
on the hydrographycal left of the Arno river south of Florence.
The basin mainly consists of rounded hills in Pliocene marine
deposits, top are sometimes flat with gully edges (clay) or with
steps where alternance of gravel, sand and clay occur. Geomorphology
of Pesa basin is well related to the granulometric characteristics
of the Pliocene marine sediments and the main erosion processes
occur as landslides, sheet and rill erosion, often induced by
human activity. Yearly runoff can be estimated of 3.19 m3/sec,
maximum values can be registered in november-april, minimum values
in august-september. The climate, typically mediterranean, is
characterized by an arid period during summer and high rainfall
intensities (100 mm/h) in autumn decreasing in winter (Canuti
et al. 1992). The site is an agricultural area with few urbanization.
In the last 20 years the hillsides have been estensivley remodeled
to make vineyards accessible to mechanized farm equipment that
requires slope uniformity. Croplands are mostly represented by
vineyards, oliveyards, wheat and grass in hillside; corn, sunflower,
sorghum, alfalfa and oilseed in the flatlands. Some woodlands
are still present on the top of the hills. The Pesa river watershed
has been the object of many studies in the field of the earth
sciences and also of remote sensing techniques. On the test area
some microwave remote sensing campaigns were carried out in 1988
(AGRISCATT'88) and 1991 (MAC'91). Such activities, concerning
SAR, scatterometric and radiometric flights, were organized by
ESA (Agriscatt'88), NASA/JPL (MAC'91) and CNR/IROE of Florence.
2.2. Ground truth measurements
An extensive mapping of the two areas of Pesa and Virginio has been done in April and then up-to-dated in October. During the SIR-C flight ground truth measurements have been carried out and the following main parameters of soil and vegetation involved in the hydrological cycle have been collected and stored in a database:
gravimetric and volumetric soil moisture content at different
depths; soil roughness, i.e. standard deviation of the surface
height and correlation length; dielectric constant of soils at
L-band; soil texture; crop classification and phenological stage;
row orientation; plant density and plant height; total number
of leaves per plant; length and width of leaves; diameter of stems;
leaf area index; water content of stems, leaves and ears/flowers.
Air temperature and humidity, wind speed and direction, solar
radiation, rainfalls have been collected in one meteorological
station placed close to Cerbaia.
2.3. Calibration
Calibration activities, concerning check of absolute calibration,
inbalance factor and phase, have been performed on the area using
three thriedral corner reflectors of different sizes (one of 2.40
m by side for L-band calibration and two others of 1.80 m by side
for C- and X-band calibration) (Van Zyl 1990, Zebker et al.1991).
Some checks carried out for the first flight led to the conclusion
that all data sets of SIR-C data have been well calibrated, whereas
X-SAR data had to be calibrated. In order to guarantee a satisfactory
comparison of data collected at different frequencies and dates
on the whole dynamic range, a further check of calibration was
performed by comparing the scattering characteristics of forests
and reference bare soils whose features were controlled by ground
truth. In fact on the Pesa area three fields of bare soil were
worked with different techniques in order to achieve three types
of surface roughness with a height standard deviation of about
1 cm (almost smooth), 2 cm and 3.5 cm (quite rough), respectively.
These fields were used as cross reference between flights. Accurate
measurements of soil moisture, surface roughness and dielectric
constant have been carried out On these fields in order to test
electromagnetic models for retrieving soil roughness.
2.4. Soil erosion models
Some models able to predict soil erosion have been tested on the
area. However, the most suitable model allowing the use of remote
sensing data resulted to be the WEPP model (Water Erosion Prediction
Project) (Lane & Nearing 1989) available both in the slope
profile and basin version. To verify the reliability of the data
obtained with the model comparison between direct measurements
of infiltration and erosion carried out by means of a rain simulator
have been performed on the Pesa and Virginio areas.
3. SAR DATA
Spring and autumn are the most useful periods to study the soil
erosion processes due to the presence of bare soils showing different
surface roughnesses and relatively high values of soil moisture
content. The supersite of Montespertoli was imaged several times
at different incidence angles between 24°
and 55° by the fully polarimetric
L- and C-band SIR-C and the VV polarization X- band X-SAR during
the two SIR-C/X-SAR missions. Seven data takes were performed
in April (between the 12th and the 17th) and five in October (between
the 3 rd and the 10th). Data takes at the most important incidence
angles were in mode 16x (L- and C-bands fully polarimetric + X-band
in VV pol.), the other in 11x (L- and C-bands in HH and HV polarization
+ X-band in VV pol.). Unfortunately, with respect to the planning
of data takes, three of the most important data takes of October
are missing (fully polarimetric at incidence angles ranging between
35 and 50 degrees) and another one scheduled as 16x has been transformed
in 11x.Other SAR measurements had previously been carried out
with the airborne NASA/JPL AIRSAR during the MAC-91 campaign in
June and July 1991 and by means of ERS-1 and JERS-1 satellites
on different dates in 1992 and 1994.
Table I: PROCESSED SAR DATA OVER
MONTESPERTOLI TEST AREA
Frequency | Polarization | Theta | Ground
Resol. (m) | Dates | |
MAC'91 | C, L, P | Quad-pol. | 20°- 35° 50° | 12.2 x 6.6 | 22-29/6/91
14/7/91 |
ERS-1 | C | VV | 23° | 30 x 26.3 | 29/5/92
07/8/92 24/4/94 |
JERS-1 | L | HH | 35° | 18.3x24.6 | 24/6/92
14/4/94 |
SIR-C/
X-SAR |
|
| 20°- 60° | 25 x 20 | April-Oct.
1994 |
A summary of data analyzed in this paper is reported in Table
I. The main quantities taken into consideration are the linear
and circular co- and cross- polar components of backscattering
coefficient (s°) at L and C- band
measured with SIR-C, and the copolar backscattering measured at
X-band, VV pol. with X-SAR. In addition the available data set
has been extended including polarimetric data collected at L-
and C- band with NASA/JPL AIRSAR as well as copolar L - band (HH
pol) available from NASDA JERS-1.
Backscattering coefficients from the selected areas have been
extracted and average value of ° (in HH, VV and HV pol.)
have been computed for each fields. Approximately 100 polygons
corrispondent to selected fields have been identified. The °
values together with ground truth data concerning soil and vegetation
parameters have been stored in a database.
4. SENSITIVITY TO HYDROLOGICAL PARAMETERS
The first analysis concerned the comparison of SIR-C data with
those of AIRSAR (MAC'91) which is represented in figure 1. The
diagram shows the average value of backscattering coefficient
at L and C-band computed for the same terrain categories (forests,
olivegroves, agricultural crops and bare soil). As expected there
is a good agreement for forests which are the more stable targets
in the area. Also bare soils show similar values, whilst the difference
at C-band for agricultural crops (wheat and alfalfa) can be explained
with the different phenological state of vegetation.
Previous research has shown that, at the present state of the
art, there is a better chance of retrieving geophysical parameters
from SAR data if a classification of the observed fields has been
performed (Ferrazzoli et al. 1997). Moreover it has been pointed
out that discrimination among broad terrain classes can be better
accomplished with the use of low frequency data (P-band). On the
other hand a discrimination between bare soil and vegetation can
be achieved using C band data. Indeed, as it has already been
noted (Baronti et al. 1995), s°
at C-Band is sensitive to both small stems and leaves and, for
example, the circular and linear cross-polarized components s°RL
and s°HV allow a good
separation of bare soil from vegetated areas. Bare soil has indeed
a higher response to RL and VV (or HH) rather than to RR or HV
components. As an example, figure 2, which represents s°
RR versus s° RL,
shows that the cluster of bare soils can be separated from the
other vegetated fields where s°
RR » s°
RL. In addition SIR-C data have shown that three land
categories can be separated using C- and L-band data at VV and
HV polarization (figure 3). As a matter of fact the difference
s°HV-s°VV
has the highest value on forests and the lowest value on smooth
bare soil both at L- and C-bands, whereas on crops and ploughed
soils s°HV-s°VV
is high at C-band and low at L- band. This behavior is well in
agreement with the theoretical reasoning discussed in Ferrazzoli
and Guerriero (1995) and Baronti et al. (1995). Indeed volume
multiple scattering is generated by inclined cylinders of forest
branches and crops stems and by dielectric discontinuities due
to air bubbles in ploughed soils. On the contrary smooth bare
soil generates prevailing surface scattering ( small s°HV
).
4.1. Sensitivity to forest biomass
At the dates of the Shuttle mission only a small part of the agricultural
area was covered by crops at an early growth stage and cannot
be separated from rough bare soils. On the contrary the forests
can be well identified since they present a high and stable value
of backscattering coefficient. AIRSAR data had shown that L-band
s°HV is a suitable
indicator of crop growth whereas forest biomass is better monitored
at P-band (Baronti et al 1995, Ferrazzoli et al. 1997). However,
SIR-C data have shown that at L- band too the sensitivity of backscattering
to forest biomass is appreciable and that s°HV,
which is low for bare soils and is mainly associated to the presence
of inclined stems and large leaves, is a better indicator of wood
biomass than s°HH,
in that it changes on a wider range and with lower dispersion.
An analysis of the correlation coefficients of the relation s°
= A+B log (W) computed at L and C band for all polarizations and
at X-band for VV pol. clearly indicate that the best sensor configuration,
among those achievable from SIR-C/X-SAR, to separate dense from
thin forests is L-Band HV (or RR or 45X) and q
~ 30°- 40°
(Ferrazzoli et al.1997).
4.2. Sensitivity to Soil Moisture and Roughness
The sensitivity of the radar signal to soil moisture has been
proven in many experiments carried out over the past years although
the signal is influenced by surface roughness and vegetation cover
as well. These spurious effects are minimized for observation
at small incidence angle. However the quality of SAR images at
incidence angle q close to nadir is
affected by a relatively poor spatial resolution which makes difficult
the exact identification of field borders.
Figure 4 represents the L-band, s°HH
at q= 26°
collected with AIRSAR in 1991 and SIRC in April and October1994,
on a certain number of bare or scarcely vegetated fields. The
scattering of data appears at least partially related to surface
roughness. The square correlation coefficient is r"= 0.624.
The relation between backscattering and soil parameters can be
interpreted using the semi-empirical model developed by Oh et
al. (1992). This model relates the parameters p = s°HH/s°VV
and q = s°HV/s°VV
to soil roughness, expressed by the normalized height standard
deviation ks (k = 2p/l,
l = electromagnetic wavelength), and
to soil reflectivity Go.
These relations provide a good fit to the data when 0.1<ks<6
and 2.6<kl<19.7 (where l is the correlation length of the
surface) and allow the retrieval of ks and Go
through an iterative procedure. The model seems to be able to
satisfactorily simulate the backscattering from rough bare soils
at least in the considered range of height standard deviation
and correlation lengths. The comparison between measured and retrieved
values of ks is shown in figure 5. In the diagram, only
values of ks measured on bare soils have been considered
(Paloscia et al. 1995). Once Go
has been retrieved the correspondent soil moisture value can be
computed using the model developed by Dobson et al. (1985) which
relates soil moisture to soil reflectivity, temperature and texture.
In figure 6 the soil moisture measured on ground is compared with
the soil moisture computed from experimental s°data
through Oh and Dobson models. The results obtained with this approach
appear satisfactory in spite of some overestimation of soil moisture.
5. THE EROSION MODEL
The most suitable physically-based hydrological model allowing
the use of SAR data, seems to be the WEPP model (Water Erosion
Prediction Project, developed by USDA (Lane & Nearing 1989))
which allows a continuous simulation both in time - in that it
estimates the soil loss at time intervals - and in space - in
that it provides a punctual estimate of erosion or/and deposition
along the points of the profile.
The strategy for using SAR data in erosion models can follow this
scheme: using both electromagnetic and statistical models, some
geophysical parameters concerning soil and vegetation (such as
leaf area index, soil cover, soil moisture and roughness) can
be retrieved from SAR data, as has been partially shown in the
previous sections. On the other hand, infomation on land use and
crop classification can be obtained from SAR images as well. These
outputs can be introduced into the hydrological model along with
auxiliary data concerning climate, soil type, vegetation cover,
agricultural practices, etc., in order to obtain the soil erosion
and other parameters concerning the hydrological cycle.
In order to verify the reliability of the data obtained with the
model, a comparison with direct measurements of infiltration,
runoff and soil erosion carried out by means of a rain simulator
have been performed on Pesa and Virginio basins (Canuti et al.
1985). From these experiments a satisfactory agreement between
observed and computed soil loss has been obtained (Ballerini et
al. 1993, Paloscia et al. 1995). One of the characteristics of
this model, which makes it very interesting for the use of remote
sensing data, is the capability of reconstructing the temporal
variations of some soil and vegetation parameters starting from
a known measured value and using a priori knowledge about meteorological
conditions, soil properties and agricultural practices. The temporal
behaviour of soil roughness of two bare fields with different
surface roughnesses has been fairly well reconstructed using an
initial value retrieved from SAR data at L band throughout the
Oh model (Paloscia et al.1995). Table II shows the estimation
of the surface roughness of three bare fields. The rather good
agreement between the observed roughnesses and the values estimated
with the Oh model allowed us to use these SAR estimations as input
for the WEPP model.
# | Hstd
ground truth | Hstd
WEPP | Hstd
SAR |
1 | 1.0 | 1.40 | 1.52 |
2 | 0.9 | 1.31 | 2.03 |
3 | 1.15 | 1.35 | 1.83 |
CONCLUSIONS
From the analysis of data collected during the SIR-C/X-SAR mission
over the supersite of Montespertoli the capability of multifrequency,
multipolarization SAR images for discriminating different types
of natural surfaces has been confirmed. It has been pointed out
that discrimination between bare soil and vegetation can be achieved
using circular and linear cross-polarized components of backscattering
coefficient at C- band and that three land categories (forests,
smooth bare soils and vegetated or ploughed soils) can be separated
using C- and L-band data at VV and HV polarization. X- band data
should be useful in discriminating among similar land classes
with small structures mostly characterized by surface scattering.
In terms of the sensitivity of SAR data to the main parameters
of soil and vegetation involved in the hydrological cycle, it
has been shown that L band seems able to estimate forest biomass
as well as moisture and surface roughness of bare soils. The latter
parameters have been retrieved with a fairly good accuracy, from
experimental data by means of an empirical model.
From the analysis of erosion model, reasonable results concerning
water infiltration and overland flow and soil loss as well have
been obtained. The first attempts to introduce geophysical parameters
measured from SAR data in the model have been performed with satisfactory
results. In particular, the surface roughness of two bare soils,
retrieved by the inversion of the Oh model, has been modeled and
used by the WEPP. The results obtained are very encouraging and
suggest a continued analysis of SAR data in order to produce appropriate
input for WEPP model.
ACKNOWLEDGMENTS
This work was supported in part by the Agenzia Spaziale Italiana
REFERENCES
Ballerini P., C. Bechini, P. Canuti, S. Moretti,
S. Paloscia, S. Baronti, S. Puglisi, 1993, Analysis of multifrequency
SAR data on Montespertoli for hydrological applications. 25°
symposium, Remote Sensing and Global Change,
Graz, 4-8 April 1993, pp. .
Baronti, S., Del Frate, F., Ferrazzoli, P., and Paloscia,
S.,1993, Interpretation of Polarimetric MAC-91 data over Montespertoli
Agricultural area. Proceedings of the 25th International Symposium:
Remote Sensing and Global Change, Graz (Austria), 4-8 April
1993, pp. 290-299.
Baronti S., F.Del Frate, P.Ferrazzoli, S.Paloscia,
P.Pampaloni, G.Schiavon, 1995, SAR polarimetric features of agricultural
areas, Int. J. Remote Sensing,
Canuti, P. S.Moretti, C.Zanchi, 1985, A rain simulator
for quick field tests of infiltration and soil loss evaluation,
in Proc. of IV Int. Conference on Soil Conservation, Maracay,
Venezuela, pp. 132-144,.
Canuti P., G. D'Auria, P. Pampaloni, and D. Solimini,
1992, MAC-91 on Montespertoli: an experiment for agro-hydrology,
Proc. of the International Geoscience and Remote Sensing Symposium
- IGARSS 92, Houston, Texas, pp. 1744-1746, 1992.
Lane L. J. & M. A. Nearing (eds), Water erosion
prediction project: landscape profile documentation. NSERL
report n°2.,
National Soil Erosion Research Laboratory, Agricultural Research
Service, U.S. Department of Agricultural, Purdue University, West
Lafayette, Indiana,1989.
Dobson, M.C., Ulaby F.T. Hallikainen M. and El Rayes
M. 1985, Microwave dielectric behavior of wet soil - part II:
Four component dielectric mixing models, IEEE Trans. Geosci.
Remote Sens., 23, 35-46.
Evans, D.L., Farr, T.G., Van Zyl, J.J, and Zebker,
H.A.,1988, Radar Polarimetry: Analysis Tools and Applications.
IEEE Trans. Geosci. Remote Sens., 26, 774-789.
Ferrazzoli, P., Paloscia, S., Pampaloni, P., Schiavon,
G., Solimini, D. And Coppo, P., 1992, Sensitivity of Microwave
Measurements to Vegetation Biomass and Soil Moisture Content:
A Case Study. IEEE Trans. Geosci. Remote Sens., 30,
750-756.
Ferrazzoli, P., Paloscia, S., Pampaloni, P., Schiavon,
Sigismondi S. and Solimini, D., 1997, The potential of multifrequency
polarimetric SAR in assessing agricultural and arboreous biomass.
IEEE Trans Geosci. Remote Sens. 35, 5-17
Ferrazzoli, P. and Guerriero L., 1995, Radar Sensitivity
to Tree Geometry and Woody Volume: A Model Analysis. IEEE Trans
Geosci. Remote Sens. 33, 360-371.
Oh, Y. Sarabandi K. and Ulaby F.T., 1992, An empirical
model and an inversion technique for radrad scattering from bare
soil surfaces, IEEE Trans. Geosci. Remote Sens. 30,
370-381.
Paloscia S., P.Pampaloni, P.Ballerini, S.Moretti,
S.Sigismondi, 1995, Integration of SAR data and hydrological modeling
through multifrequency AIRSAR campaigns on Montespertoli site.
MAC Europe 1991 Final Results Workshop Proceedings, Lenggries,
Germany, 4-6 October 1994, ESA WPP-88, January 1995, pp. 65-70
Stofan E.R., Evans D.L., Schmullius C., Holt B,,
Plaut J.J., van Zyl, J.J., Wall S.D., and Way J.,1995, Overview
of the results of spaceborne imaging radar-C,X-band Synthetic
aperture radar (SIR-C/X-SAR), IEEE Trans. Geosci. Remote Sens.,
33, 817 - 828
Touzi, R., Goze, S., Le Toan, T., Lopes, A., and
Mougin, E.,1992, Polarimetric discriminators for SAR Images. IEEE
Trans. Geosci. Remote Sens. 30, 973-980.
Van Zyl, J. J., 1990, Calibration of polarimetric
radar images using only image parameters and trihedral corner
reflector responses. IEEE Trans. Geosci. Remote Sens. 28,
337-347.
Zebker, H.A., Van Zyl, J.J., Durdeen S.L and Norikane L., 1991, Calibrated imaging radar polarimetry: technique, examples and applications, IEEE Trans. Geosci. Remote Sens., 29, 942-961.
FIGURE CAPTIONS
Figure 1 - The average value of s°
at P- L- and C-band ( HH, HV and VV pol.) and X-band (VV pol.),
q = 33°-
35°,
computed for four homogeneous surface categories (forests:Y, olivegroves:O,
cultivated fields: A, and bare soils: B). a = SIRC data (April
1994), b = AIRSAR data (June 1991)
Figure 2 - s°RR
versus s°RL
at C-band. Bare soils are separated from vegetated fields. Small
captions refer to AIRSAR data and capital letters to SIR-C (labels
are the same as in previous figures)
Figure 3 - s°HV-VV
(C-band) versus s°HV-VV
(L-band).Three classes can be separated:
Forests (Y), bare soils (B) and agricultural fields (A=alfalfa,
W=wheat, P=ploughed soils)
Figure 4 - s°HH
at L-band (q
= 25°)
as a function of the gravimetric soil moisture (SMCg) of different
fields of bare soil. The line represents the best fit.
Figure 5 - The retrieved standard deviation of bare
soils (ks), computed through the Oh model, as a function of the
ground truth data
Figure 6 - The retrieved soil moisture of bare soils,
computed through the Oh model, as a function of the ground truth
data