In this study, a down scaling algorithm to disaggregate the radiometer Brightness Temperature (TB) using the radar backscatter observations for SMAP (Soil ...
A DOWNSCALING ALGORITHM FOR COMBINING RADAR AND RADIOMETER OBSERVATIONS FOR SMAP SOIL MOISTURE RETRIEVAL 123 J.. 12 12 , zane heng Sh· z , 1,.". zanJze Zhao
Peng GUO
• .
1 State Key Laboratory of Remote Sensing Science, Jointly Sponsored by the Institute of Remote
Sensing and Digital Earth Chinese Academy of Sciences and Beijing Normal University ,Beijing,100101 2 Institute of Remote Sensing and Digital Earth Chinese Academy of Sciences, Beijing, 100101 3 University of Chinese Academy of Sciences, Beijing, 100049
ABSTRACT
In this study, a downscaling algorithm to disaggregate
microwave remote sensing is one of the most promising
the radiometer Brightness Temperature (TB) using the radar
techniques to monitor global near-surface soil moisture,
backscatter observations for SMAP (Soil Moisture Active
with frequent revisit and independence on the effects of
and Passive) was developed. The algorithm is based on the
clouds and solar illumination. Soil moisture retrieval using
spectral
downscaling
both
phase
and
active and passive microwave remote sensing has been
domain.
Using
the
explored for several decades, with each having distinct
information from radar measurements at fmer resolution, a
advantages [1, 2]. The passive radiometric remote sensing is
amplitude
which
information
in
combines
Fourier
new way to estimate the Fourier phase was proposed. The
very sensitivity to soil moisture, even under vegetated
algorithm
PALS
conditions while the spatial resolution is typically low
than
(�40km),which is sufficient for hydrometeorology, ecology,
datasets
has from
been
successfully
SMEX02
radiometer-only
applied
producing
inversions.
The
to
better RMSE
the
results
(Root-Mean
water resource management applications. The active radar is
Square-Error) of the downscaling Brightness Temperature
capable
are 3.26K and 6.12K for V and H polarization, respectively.
influenced
Then medium resolution soil moisture was retrieved from
structure and water content. To combine the individual
disaggregated/downscaled TB. The accuracy (RMSE) of the 3 3 downscaling soil moisture retrievals is 0.0459m /m , which
advantages of the passive and active approaches, the SMAP
is very close to SMAP science requirement of 0.04. The
development by NASA in 2008[3].The mission is targeted
results indicate that the downscaling algorithm presented in
for launch in 2014.
this
study
is
a
promising
approach
to
achieve
(Soil
fmer
of
high by
Moisture
spatial surface
Active
resolution
(�3km)
roughness,
and
Passive)
but
vegetation
was
highly canopy
selected
for
The SMAP consists of an L-band radar (1.26GHz, hh,
resolution and more accurate soil moisture retrievals for the
vv, hv polarizations) and an L-band radiometer (lAIGHz, d h h,v, and 3r and 4t Stokes parameters polarizations) that
future SMAP mission.
share a single feedhom and reflector. The deployable mesh Index
Terms-radar,
radiometer,
soil
reflector (diameter:6m) is offset from nadir and rotates
moisture,
downscaling, SMAP
about the nadir axis at 14.6 rpm, providing a conically scanning antenna beam with a surface incidence angle of 40°. SMAP will be launched into a 680km near-polar sun
1. INTRODUCTION
synchronous orbit with an eight-day repeat
cycle
and
The top few centimeters soil moisture is critical to estimate
equator crossings at 6 A.M. and 6 P.M. local time. The
the ratio between evaporation and potential evaporation at
antenna configuration yields a radiometer footprint spatial
the land surface, to compute several key variables of the
resolution at the surface of �40km and a real-aperture radar
land surface energy and water budget. In addition, surface
footprint resolution of 1-3km(over the outer 70% of the
soil moisture is the initial condition and boundary condition
swath) that provides global coverage within three days at the
to enhance weather and climate forecast skill. Soil moisture
Equator and two days at boreal latitudes(>45
is also a land state variable to determine the net carbon flux in
boreal
landscapes
and
to
develop
improved
flood
N). The
soil moisture in the top 5cm of soil with an error of no 3 3 greater than 0.04m /m at 10 km spatial resolution and 3-day
prediction and drought monitoring capability. Thus, global measurements of the soil moisture are very important to
average intervals over the global land area excluding regions
understanding the components and interactions between the
of snow and ice, frozen ground, mountainous topography,
global water, energy, and carbon cycles. Satellite-based
978-1-4799-1114-1/ 13/$31.00 ©2013 IEEE
0
baseline science mission of SMAP is to provide estimates of
open water, urban areas, and vegetation with water content
731
IGARSS 2013
no greater than 5kg/m2 (averaged over the spatial resolution
For
the same
scalar
field,
the
PSD of
unknown
scale). For the reasons mentioned above, either the SMAP
resolution can be inferred from the formula (3) which was
radiometer or the radar is difficult to individually meet the
obtained from the known coarse resolution. Then,
SMAP requirements for soil moisture spatial resolution 3 (10km) and accuracy (0.04cm /cm\
spectral amplitudes at unknown resolution (Avlln) can be
the
estimated from the formula (2), at least in the average sense. The
In this study, a downscaling algorithm that overcomes
Fourier
phase
lJ'vlln
can
be
generated
from
finer
these limitations by combining the active (radar) and passive
resolution interpolation using the bilinear, [4] i.e. determine
(radiometer) measurements to disaggregate the radiometer
phase from bilinear interpolation, denoted as DPFB. SMAP will measure the natural microwave emission in
brightness temperature was developed. The disaggregated brightness temperature with radiometer-based algorithm was
form
used to derive medium scale soil moisture to support the
backscatter
of
brightness
SMAP requirements.
increase of surface soil moisture or soil dielectric constant
(0)
temperature
and
(TB)
the
energy
of the land surface simultaneously. The
will leads to increase in radar (J and decrease in radiometer 2. THEORY OF THE ALGORITHM
Statistical
spectral
downscaling
technique
has
TB
observations, and vice-versa. Within a small region of
interest the SMAP measured been
TB and (J are expected to have a
approximately linear functional relationship[5]:
demonstrated to be able to increase the spatial resolution of integrated water vapor fields from satellite observations [4].
TBP
The principle of the spectral downscaling is the spatial field at a given resolution may be extrapolated to fine resolutions
=
a+b . a
(5)
pp
where p indicates polarization. Parameters
by properly modeling its spatial properties at any observable
a
and b are the
intercept and slop of the linear functional relationship,
scale in the frequency domain [4].
respectively. According to the distributivity property of the
For a remote sensing image, the complex Fourier spectrum (Fv) can be impressed by the spectral amplitude Av
2-D Fourier Transform (6), the Fourier phase lJ'vlln can be
and the Fourier phase lJ'v as follows:
estimated
from
radar
measurements
which
have
finer
resolution, i.e. determine phase from radar observations, denoted
(1) where s=X'
is the spatial
frequency, A is the spatial
and
1·1
linear
functional
dependence
�[a. 1; (x,y)+b· J; (x,y)] a· �[1; (x,y)] +b· �[J; (x,y)] =
indicates average and modulus operator.
(6)
where :3 indicates the inverse Fourier transform.
The mean PSD has a power law dependence on spatial
3. TEST OF ALGORITHM USING SMEX02 DATA
frequency in the Fourier frequency domain.
The PALS (Passive and Active L-band System) datasets
(3) where
The
study.
spectral density (PSD) f(Jv(s) by the following relation:
(-)
DPFR.
TBp and (J"" exhibit the higher correlation than (Jhh[], therefore the combination of TBp and (J"" was used in this
wavelength. The spectral amplitude is related to the power
where
as
between
from SMEX02 (Soil Moisture Experiment in 2002) was used
fJ is a constant which can be derived from the coarse
to
evaluate
the
performance
of
the
downscaling
algorithm. The PALS was mounted on a C-130 aircraft and h h flown over the watershed study region on June 25t , 2i , h t t and July 1S , 2nd, 5th, 6th, and 8 , 2002 during SMEX02
resolution. Spatial-spectral downscaling assumes that the whole range of spatial resolution of a scalar field can be split into a known resolution Skn and an unknown resolution
(The PALS coverage on July 1 was partial, so data was not
Sun of the spatial frequency domain. After estimating the
used in
Fourier transforms Fv for the whole range, the inverse
this
study). An extensive
datasets
of in situ
measurements including volumetric soil moisture (VSM),
Fourier transform is taken to recover the downscaled field
surface and subsurface soil temperature, soil bulk density, crop
Vd•
type
and
vegetation
water
collected during the campaign.
content(VWC)
were
The PALS radar and
radiometer have similar frequencies and incident angle to SMAP and the wet and dry soil moisture conditions is available
within
campaign period.
732
the
PALS
flight
domain
during
the
The algorithm was designed to downscale SMAP radiometer Brightness Temperature from 40km to a target resolution of lOkm; however, PALS observations have much finer spatial resolution approximately O.8km. To test the
downscaling
algorithm
using
the
PALS
data,
the
radiometer and radar measurements were gridded at �4km and �O.8km, respectively. Fig.l shows the averaged data at coarse resolution (�4km), original radiometer observations (�O.8km), the disaggregated results using DPFB algorithm
(a) DPFB
(�O.8km) and the disaggregated results using the DPFR algorithm (�O.8km). As observed in Fig.l, spatial patterns are fairly similar, and both method capture key brightness
""
temperature features. Drier and wetter regions are captured
(
well and are generally similar in both methods. However, the DPFR disaggregated results exhibit much less spurious spatial variability than DPFB, and capture well the spatial
RMSE�3.26K
structure of the original observed data. The accuracy of the
Oltslrvatlorll'\.
algorithm developed in this study (DPFR) is better, with RMSE of 3.26K and 6.47K for V and H polarization,
(b) DBFR
respectively. While the RMSE of DPFB algorithm is 3.56 K
Figure2. Plots of the observed TB and the disaggregated TB.
and 6.59K for V and H polarization, respectively. Fig.2 is
a) DPFB method, and b) DPFR method.
the plots of the disaggregated and observed brightness temperature for the DPFR and DPFB algorithm.
The
disaggregated
intermediate
product
brightness
of
the
temperature
downscaled
soil
is
an
moisture
algorithm. Any bias in disaggregated TB is removed by
06.25 .... ___
imposing
07-02 07-0S 07-