In: Wagner W., Székely, B. (eds.): ISPRS TC VII Symposium - 100 Years ISPRS, Vienna, Austria, July 5-7, 2010, IAPRS, Voi. XXXVIII, Part 7B
ASSESSMENT OF THE IMPACT OF UNCERTAINTY ON MODELED SOIL SURFACE
ROUGHNESS ON SAR-RETRIEVED SOIL MOISTURE UNCERTAINTY
E. De Keyser a *, H. Lievens b , H. Vemieuwe a , J. ÄIvarez-Mozos c , B. De Baets a , N.E.C. Verhoest b
a Department of Applied Mathematics, Biometrics and Process Control, Ghent University, Coupure links 653, B-9000 Gent, Belgium
b Laboratory of Hydrology and Water Management, Ghent University, Coupure links 653, B-9000 Gent, Belgium
c Department of Projects and Rural Engineering, Public University of Navarre, Spain
* Eva.DeKeyser@UGent.be
KEY WORDS: Soil moisture retrieval, SAR, Uncertainty assessment, Linear regression, Soil surface roughness
ABSTRACT:
Soil moisture retrieval from SAR images using semi-empirical or physically-based backscatter models requires surface roughness pa
rameters, generally obtained by means of in situ measurements. However, measured roughness parameters often result in inaccurate
soil moisture contents. Furthermore, when these retrieved soil moisture contents need to be used in data assimilation schemes, it is
important to also assess the retrieval uncertainty. In this paper, a regression-based method is developed that allows for the parameteriza
tion of roughness by means of a probability distribution. This distribution is further propagated through an inverse backscatter model in
order to obtain probability distributions of soil moisture content. About 70% of the obtained distributions are skewed and non-normal
and it is furthermore shown that their interquartile range differs with respect to soil moisture conditions. Comparison of soil moisture
measurements with the retrieved median values results in a root mean square error of approximately 3.5 vol%.
1 INTRODUCTION
Soil moisture is a key variable in various earth science disciplines
such as hydrology, meteorology and agriculture. The models that
are mostly used in these disciplines generally require spatially
distributed soil moisture as an input. As the microwave backscat-
tered signal from a bare soil surface is partly influenced by the
soil moisture content, radar remote sensing can be used to meet
these high spatial resolution requirements. Currently, only ac
tive microwave sensors, of which the Synthetic Aperture Radar
(SAR) is the most common imaging configuration, are able to
capture small-scale soil mositure patterns.
Several backscatter models exist that calculate the backscattered
signal, given soil moisture, soil surface roughness and incidence
angle, polarization and wavelength of the radar signal. Soil sur
face roughness refers to the unevenness of the earth’s surface due
to natural processes or human activities, and is generally statisti
cally described by the root mean square (rms) height, the correla
tion length and an autocorrelation function (Ulaby et al., 1982a).
Unfortunately, soil surface roughness parameters are difficult to
measure as several experiments have shown that roughness pa
rameterization depends on profile length (Callens et ah, 2006;
Davidson et ah, 2000; Ogilvy, 1988; Oh and Kay, 1998) and the
measurement technique (Mattia et ah, 2003a), meaning that dif
ferent roughness parameter values can be obtained for the same
surface. These problems occur because natural surfaces behave
as a self-affine fractal surface (Shephard and Campbell, 1999;
Dierking, 1999; Shepard et ah, 2001), while most of the backscat
ter models assume a stationary random surface.
Amongst the various methods that exist to overcome this param
eterization problem, Su et ah (1997) suggested the use of an ef
fective roughness parameter, which is estimated by means of re
motely sensed data in combination with soil moisture measure
ments. This parameter then replaces the in situ roughness mea
surements for soil moisture retrieval from successive SAR acqui
sitions. This concept is applied successfully in different studies
(Verhoest et ah, 2000; Baghdadi et ah, 2002,2004,2006; Rahman
et ah, 2007; Alvarez-Mozos et ah, 2008) and will also be used in
this study.
SAR retrieved soil moisture maps are often used in hydrological
models or in data assimilation schemes. For the latter applica
tions, the Ensemble Kalman filter (Evensen, 2006) is frequently
used to assimilate remotely sensed hydrologic information (Re-
ichle, 2008). This method relies on the value of the observed vari
able and assumes a normal distribution, for which the mean value
and the variance of the observed variable are required. There
fore, retrieval algorithms should provide not only soil moisture
content, but also a quantification of its uncertainty.
The research questions to be answered in this study are:
1. How can the uncertainty on effective soil surface roughness
be quantified?
2. How does this uncertainty influence the uncertainty on re
trieved soil moisture?
For this purpose, all other sources of uncertainty were ignored,
such as uncertainty on the backscattered signal, uncertainty in
duced by vegetation cover or by the backscatter model.
2 METHODOLOGY
The methodology used in this study is based on a relationship that
was found between effective roughness parameters and backscat
ter coefficients. A linear regression model was used to model this
relationship (Lievens et al., 2010) and can furthermore be used
to quantify the uncertainty on the modeled soil roughness as a
probability distribution. Using a Monte Carlo method to prop
agate this probability distribution through an inverse backscatter
model, a probability distribution for soil moisture content is ob
tained.
