Soil Moisture Estimation From SIR-C Measurements
D.J. Chadwick and J.R. Wang
Goddard Space Flight Center
Code 975
Greenbelt, MD 20770 USA
1. Experiment Objective.
The main objective of our study is to explore the feasibility
of estimating surface soil moisture from the multiple-frequency
and multiple-polarization measurements provided by SIR-C/X-SAR.
The approach was to first apply algorithms available in the literature
to the SIR-C data and compare the results with ground truth data
measured concurrently with the SIR-C data acquisition. Next, an
effort was made to improve on the available retrieval algorithms,
and an effort was made to develop a new algorithm which would
be more robust in areas with vegetation. The test site is located
in a well-maintained watershed near Chickasha, some 60 km southwest
of Oklahoma City, Oklahoma.
2. Summary of the Experiment.
The SIR-C mission consisted of two different shuttle flights in
April and October of 1994, each lasting about 10 days. Throughout
the second flight in October 1994 most of the soils in our test
site, the Little Washita River watershed (about 40 km by 25 km
in size), remained dry. In addition, the last three days of that
flight was devoted to interferometry and the test site was not
covered by the SIR-C measurements. As a consequence, our analysis
and research efforts have emphasized the data acquired during
the April flight. A total of twelve SIR-C data takes over the
test site, seven of them in mode 16X and remaining five in mode
11X, were acquired during the 10-day flight. Corner reflectors
were placed in five widely separated locations for SIR-C/X-SAR
calibrations. Ground truth of soil moisture was acquired within
±2 hours of the SIR-C overpasses, for a number of selected
bare and vegetated fields in the test site. Samples of vegetation
biomass and surface roughness were also obtained for several of
these fields. These ground truth data sets were processed, organized,
and reported in the Washita CD-ROM by Jackson et al (1996).
A moderate rain storm occurred in the test area just before the
first SIR-C overpass on April 11, and prior to that the soils
were fairly dry. There was no rainfall for nearly two weeks after
April 11, and measurements during April 11-18 therefore covered
a period of dry-down for the soils in the test area. The rainfall
caused the average soil moisture in the test site to increase
from about 10% to about 25% when the SIR-C passes over the area
began. The average soil moisture decreased steadily during the
April SIR-C mission to the pre-rain value of about 10%. The soil
textures in the test site vary from silty loam to fine sand, with
silty loam covering the northwestern and northeastern parts of
the watershed, fine sandy loam distributed along the Little Washita
River and upper-central part, and loamy fine sand and fine sand
spreading over the southeastern and lower-central parts of the
watershed. The percentage of sand in the soil ranges from 10%
to 90% and the percentage of clay from 5% to 30%. There are a
total of 43 rain gauges distributed fairly evenly throughout the
watershed.
Results from a preliminary analysis showed that vegetation has
a strong effect at both C and X bands, and the SIR-C data in these
bands were not ideal for soil moisture estimation. Consequently,
the L-band quad-polarized data were emphasized in the analysis
and research efforts described below.
3. Estimation of Soil Moisture.
Two different algorithms (Dubois et al., 1995; Shi et al., 1995)
developed recently for estimation of soil moisture have been applied
to the SIR-C L-band data for the Little Washita study site. The
algorithm of Dubois et al.(1995) was derived empirically based
on the ground-level scatterometer measurements of Oh et al. (1992)
and Wegmuller (1993). The algorithm of Shi et al. (1995) was based
on an approximation to the integral equation method (IEM) developed
by Fung et al. (1992). Both algorithms were derived for application
to the L-band multiple-polarized measurements within the incidence
angle range of 30_-50_. Six SIR-C passes over the test site satisfied
these conditions. Three of them on April 11, 12, and 13 (incidence
angles 28_, 42.3_, and 50.1_) were south-looking ascending passes,
and the other three on April 15, 16, and 17 (incidence angles
42.4_, 36.2_, and 30.9_) were south-looking descending passes.
The results of applying the two algorithms to these six passes
of SIR-C L-band data were reported by Shi et al. (1997) and Wang
et al. (1997). The soil moisture maps derived from both algorithms
show the trend of drying-down consistent with the ground observations.
The estimated soil moisture values (mv) were also compared with
those sampled on the ground for selected bare and vegetated fields.
Figure 1 shows a scatter plot between the mv values estimated
from the algorithm of Dubois et al.(1995) and those measured on
the ground. Different symbols are used to represent data for a
bare field, an alfalfa field, and other vegetated field to facilitate
discussion. The estimated mv values for the alfalfa field are
quite poor compared to the measured ones; five out of six times
the algorithm overestimates mv, and there is no apparent correlation
between the estimated and measured values. Results for the bare
and other vegetated fields show a good correlation between the
estimated and measured soil moisture. The standard deviation of
the comparison is about 0.059 cm3/cm3 when all the fields are
included. When the data points from the alfalfa field are excluded
from the comparison, the standard deviation improves to 0.053
cm3/cm3.
Figure 2 shows a comparison of the mv values estimated from the
algorithm of Shi et al. (1997) and those sampled from the fields.
Again, different symbols are used to denote the mv values for
the bare field, alfalfa field, and other vegetated fields. The
data points from the bare field show some correlation, but those
from the alfalfa and other vegetated fields show more scatter.
The standard deviation is found to be 0.053 cm3/cm3 when all the
fields are included in the comparison.
Both the empirically and theoretically based retrieval algorithms
are derived from ground-level radar measurements and the IEM model
only for bare soils with varying soil moisture and surface roughness
parameters. Both of them are not expected to adequately handle
the vegetated soils that comprise most of the Little Washita River
watershed. The results shown in Figures 1 and 2 conform to the
deficiency of these algorithms. An improvement to these algorithms
or a new approach is needed because most land surfaces have some
vegetation cover
4. A New Approach Under Exploration.
Radar signals from surface consist of three components: a direct
backscatter from the surface (single reflection), a component
that interacts with vegetation stalks before or after a specular
reflection from the surface (double reflection), and a diffuse
component from scattering by leaves or branches in a vegetation
canopy. It is expected that the relative contribution from these
components to the SAR signals would be different for bare and
vegetated soils; and this relative contribution may also change
with changes in mv. Theoretically, the ratio of single reflection
(90_ incidence) to double reflection (true angle of incidence
of SAR) backscatter should change with changing soil moisture
(Ulaby et al., 1981). To explore this possibility, Cloude's decomposition
method (Van Zyl, 1992) was applied to the SIR-C L-band data acquired
on April 12 and 15 and the two reflection components were calculated
for the fields with ground truth measurements. The data from these
two days were acquired at an incidence angle of 42.3_. This choice
of data sets removed incidence angle as a variable and the relative
changes in the radar backscatter components could be analyzed
with respect to soil moisture alone. Figures 3 and 4 summarize
the results from these analyses.
Figure 3 shows the change in single to double reflection ratio
as a function of change in mv for fields with ground truth measurements.
The plot clearly shows that the single to double reflection ratio
decreases in a near-linear fashion with mv. The consistent slope
suggests that regardless of vegetation, there is a direct correspondence
between the changes in the single to double reflection ratio and
mv. In Figure 4, the values of the single to double reflection
ratios are plotted as a function of mv., with each line segment
representing the change in soil moisture between 12 and 15 April.
Bare soil fields have the highest single/double reflection ratios,
with generally decreasing ratios with increasing vegetation biomass.
Figure 4 suggests that, for the single to double reflection ratio
to be useful for estimation of mv, a method has to be developed
to take into account the vegetation effect. Ranson et al. (1995)
reported a method to estimate vegetation biomass by the ratio
of backscattering coefficients at L and C bands, both at HV (horizontal
transmit and vertical receive) polarization. The method could
provide estimation of biomass up to 25 kg/m2, with a probable
error at 95% confidence interval of 2 kg/m2. As the measured biomass
values for the fields of our study area are all < 2.5 kg/m2
(Jackson et al., 1996), this probable error is quite significant.
Nevertheless, the method provides the necessary means to normalize
the vegetation effect displayed in Figure 4. Efforts to calculate
biomass using the NDVI (Normalized Difference Vegetation Index)
derived from the Landsat TM image obtained over the Little Washita
River watershed near the time of SIR-C flight lead to larger errors
in biomass estimation.
Combining the results displayed in Figure 4 and measured biomass
values, one arrives at an equation for mv as:
mv = 0.026r + 0.11b - 0.14
where r is the ratio (in dB) of single to double reflection power
at L band, VV polarization, and b is the dry biomass in kg/m2.
This algorithm is currently being tested with the SIR-C data over
our test site, and the results will be compared with those derived
from the algorithms of Dubois et al. (1996) and Shi et al. (1997).
References.
Dubois, P. C., Van Zyl, J., and Engmen, E. T. (1995), Measuring
soil moisture with imaging radar, IEEE Trans. Geosci. Remote Sens.,
33(4), 915-926.
Jackson, T., Tang, L., Hsu, A., Wood, E., O'Neill, P., and Engman,
E. (1996), Washita '94 SIR-C/X-SAR data sets, published in CD-ROM,
Goddard Space Flight Center, Greenbelt, MD 20771.
Oh, Y., Sarabandi, K., and Ulaby, F. T. (1992), An empirical model
and an inversion technique for radar scattering from bare soil
surface. IEEE Trans. Geosci. Remote Sens., 30(2), 370-381.
Ranson K.J., Saatchi, S., and Sun, G. (1995), Boreal Forest Ecosystem
Characterization with SIR-C/X-SAR, IEEE Trans. Geosci. Remote
Sens., 33(4), 867-876.
Shi, J. C., Wang, J. R., Hsu, A. Y., O'Neill, P. E., and Engman,
E. T. (1995), Estimation of bare surface soil moisture and surface
roughness parameters using L-band SAR-measurements, IGARSS'95,
Vol. I, 5-7-509.
Shi, J. C., Wang, J. R., Hsu, A. Y., O'Neill, P. E., and Engman,
E. T. (1997), Estimation of bare surface soil moisture and surface
roughness parameters using L-band SAR image data, IEEE Trans.
Geosci. Remote Sens., in press.
Ulaby, F.T., Moore, R.K., and Fung, A.K., (1981), Microwave Remote
Sensing: Active and Passive, vol. 1, Addison-Wesley pub., 456
pp.
Van Zyl, J.J. (1992), Application of Cloude's target decomposition
theorem to polarimetric imaging radar data, Soc. Photo Optical
Instr. Engin. 1748, 184-191.
Wang, J. R., Hsu, A. Y., Shi, J. C., O'Neill, P. E., and Engman, E. T. (1997), A comparison of soil moisture retrieval models using SIR-C measurements over the Little Washita River watershed, Remote Sens. Environ., in press.
Wegmuller, U. (1993), Active and passive microwave signature catalogue on bare soil (2-12 GHz), Inst. Applied Physics, Univ. Berne, Switzerland.