Temporal Stability of Soil PDF

Review and Analysis
Karl Vanderlinden
Harry Vereecken
Horst Hardelauf
Michael Herbst
Gonzalo Mar nez
Michael H. Cosh
Yakov A. Pachepsky*
This paper reviews the current knowledge base on the temporal stability
of soil water contents, highlights the
interac ons of the temporal stability controls, and proposes research
avenues which can advance mul ple
applica ons that the temporal stability of soil water contents currently
has in environmental monitoring,
modeling, and management.
K. Vanderlinden, IFAPA, Centro Las TorresTomejil, 41200 Alcal del Rio, Spain. Harry
Vereecken, H. Hardelauf, and M. Herbst,
Agrosphere (IBG-3), Ins tute of Bio- and
Geosciences, Forschungszentrum Jlich
GmbH, 52428 Jlich, Germany. G. Marnez, Dep. of Agronomy, Univ. of Cordoba,
14071, Cordoba, Spain. M.H. Cosh, USDAARS, Hydrology and Remote Sensing Lab.,
Beltsville, MD, 20705, USA. Y.A. Pachepsky,
USDA-ARS-EMFSL, Beltsville, MD, 20705, USA.
K. Vanderlinden and G. Mar nez, Currently
at USDA-ARS-EMFSL, Beltsville, MD, 20705
USA. Received 23 Nov. 2011. *Corresponding
author (Yakov.pachepsky@ars.usda.gov).
Vadose Zone J.
doi:10.2136/vzj2011.0178
Received 23 Nov. 2011.
Soil Science Society of America

5585 Guilford Rd., Madison, WI 53711 USA.
All rights reserved. No part of this periodical may

be reproduced or transmi ed in any form or by any
means, electronic or mechanical, including photocopying, recording, or any informa on storage
and retrieval system, without permission in wri ng
from the publisher.
Temporal Stability of Soil

Water Contents: A Review of Data
and Analyses
Temporal stability (TS) of soil water content (SWC) has been observed throughout a wide
range of spa al and temporal scales. Yet, the evidence with respect to the controlling
factors on TS SWC remains contradictory or nonexistent. The objec ve of this work was
to develop the first comprehensive review of methodologies to evaluate TS SWC and to
present and analyze an inventory of published data. Sta s cal analysis of mean rela ve
dierence (MRD) data and associated standard devia ons (SDRD) from 157 graphs in 37
publica ons showed a trend for the standard devia on of MRD (SDMRD) to increase with
scale, as expected. The MRD followed generally the Gaussian distribu on with R2 ranging
from 0.841 to 0.998. No rela onship between SDMRD and R2 was observed. The smallest
R2 values were mostly found for nega vely skewed and platykur c MRD distribu ons. A
new sta s cal model for temporally stable SWC fields was proposed. The analysis of the
published data on seven measurement-, terrain-, and climate-related poten ally controlling factors of TS SWC suggested intertwined eects of controlling factors rather than single
dominant factors. This calls for a focused research eort on the interac ons and eects of
measurement design, topography, soil, vegeta on and climate on TS SWC. Research avenues are proposed which will lead to a be er understanding of the TS phenomenon and
ul mately to the iden fica on of the underlying mechanisms.
Abbrevia ons: EOF, empirical orthogonal func on; MRD, mean rela ve dierence; KMRD, kurtosis of MRD;
MABE, mean absolute bias error; MRD, mean rela ve dierence; MRD, spa al mean of MRD; RD, rela ve
dierence; SDMRD, standard devia on of MRD; SDMRD2, variance of MRD; SDRD, standard devia on of
RD; SDRD spa al mean of SDRD; SMRD, skewness of MRD; SWC, soil water content; TS, temporal stability
The concept of TS in soils was first introduced by Vachaud et al. (1985). They observed
that at specific locations within a field, the field averaged soil water storage was preserved
in time. In addition, they found that locations could be spotted where soil was consistently
wetter or consistently dryer than average across the field leading to a persistent bias. The TS
SWC has also been reported both at scales finer than the field, e.g., within a plot (Rolston
et al., 1991; Pachepsky et al., 2005; Herbst et al., 2009), and at scales coarser than the
watershed, e. g. regional scale (Martnez-Fernndez and Ceballos, 2003). Temporal stability is often called time stability, and other terms such as rank stability or order stability
have been proposed and discussed (Chen, 2006).
Information on TS SWC has found multiple applications in environmental monitoring,
modeling, and management. It has been shown to be useful, for example, in locating time
stable sites viewed as the most representative locations (Vachaud et al., 1985; Kachanoski
and de Jong, 1988; Rolston et al., 1991) and in up- and downscaling soil water contents
(Cosh et al., 2004; Jacobs et al., 2004; Guber et al., 2008; de Rosnay et al., 2009). The TS
SWC was instrumental in developing methods to infi ll missing data from malfunctioning probes (Pachepsky et al., 2005; Dumedah and Coulibaly, 2011). Data on TS found
direct use in hydrologic modeling (Brocca et al., 2009; Heathman et al., 2009) and in
SWC monitoring data assimilation in soil water flow modeling (Pan et al., 2012). Field
management zones were delineated based on TS SWC (da Silva et al., 2001; Starr, 2005).
It was hypothesized that TS SWC information can be useful in improving hydrologic
modeling by accounting for differences in antecedent soil water contents (Gmez-Plaza et
al., 2001; Zehe et al., 2010; Minet et al., 2011), to design adequate treatment experiments
with replicated plots (Kamgar et al., 1993; Reichardt et al., 1993; Cassel et al., 2000; Rocha
et al., 2005) and to design sensor networks and optimize the number of sensors (Mohanty
and Skaggs, 2001; Heathman et al., 2009).
www.VadoseZoneJournal.org
Currently the number of publications on TS SWC exhibits accelerated growth. Yet, the basic questions about TS SWC and its
controls remain unanswered. Moreover, the evidence found in literature with respect to TS SWC controls remains contradictory.
The practically important assumption that an increase in spatial
scale affects the time stability remains untested. An inventory of
existing examples of the TS SWC analysis seems to be in order
to begin answering these questions in a systematic manner. Such
inventory, based on available studies in literature, was the purpose
of this work. Specific objectives were (i) to analyze available literature to identify the key controls affecting TS SWC, (ii) to evaluate
the effect of measurement techniques on TS SWC, (iii) to evaluate
time and scale effects on TS SWC, and (iv) to suggest an identifiable spatiotemporal model of the TS SWC.
6 Quan
fying TS SWC
Two groups of methods have been proposed in literature to

characterize the TS SWC. The fi rst group of methods uses all
observations made during the observation period. These methods
include mean relative difference (MRD) and time independent
spatial patterns (EOF analysis). The second group of methods uses
pairs of observation times. It includes the temporal persistence
regression, the Spearman rank correlation, and the Pearson correlation. In addition, in this section, methods to select representative
locations are revised and a Gaussian approximation of the distribution of MRD is proposed.
Mean Rela ve Dierences

The use of MRD (Vachaud et al., 1985) is currently by far the most
often applied technique to research the TS SWC. The relative difference RDij between the SWC ij at observation location i at the time
j and the spatial average SWC at the same time j is defined as
RDij =
ij
[1]
The MRD for the location i becomes then

MRDi =
1
Nt
j= N t
RDij ,
[2]
j =1
where Nt is the number of observation times. The standard

deviation SDRDi of the set RDi,1, RDi,2 ,, RDi, Nt of relative
differences at the location i over the observation period is usually
computed along with MRDi as
SDRDi =
1 j= N t
(RDij MRDi )2 .
N t 1 j =1
[3]
The value of SDRDi serves as one of the measures of the TS

SWC (Vachaud et al., 1985; Pachepsky et al., 2005), while ij
at points with MRDi 0 is considered to be representative for
j throughout time. To simplify notations, the subscript i will

be omitted where possible. It is understood that both MRD and
SDRD are location-specific and as such are functions of the spatial variables.
Pa ern Analysis with EOF

Empirical orthogonal function (EOF) analysis, or principal component analysis (PCA) is a widely applied statistical method for
analyzing large multidimensional datasets and for searching patterns in them. Th is method has been applied to SWC datasets
to extract dominant patterns (Yoo and Kim, 2004; Perry and
Niemann, 2007), which are similar to MRD patterns. EOF analysis partitions the observed variation into a series of time-invariant
spatial patterns (EOFs) that can be multiplied by temporal varying
(but spatially constant) coefficients and summed to reconstruct
observed soil moisture patterns. EOFs can be mapped and these
maps can be compared with maps of various soil, landscape, and
land use properties in search of similarities (Jawson and Niemann,
2007; Perry and Niemann, 2007; Korres et al., 2010; Ibrahim and
Huggins, 2011).
Temporal Persistence: Regression and

Correla on
Regression across the studied area between water contents at two
different observation times was suggested as a means of quantifying the TS SWC (Kachanoski and de Jong, 1988). The regression
equation was
ij2 = a j1 j2 ij1 + b j1 j2 .
[4]
A close linear regression relationship between ij1 and ij2 was

interpreted as the manifestation of the temporal persistence. This
type of TS is weaker than the TS described earlier on, corresponding to time-independent MRD and small SDRD, because both
intercept and slope of the regression may change as times j1 and j2
change. This looser definition of TS appears to be useful in characterization of observations in which the spatially stable pattern is
different over different periods within the total period of observations. da Silva et al. (2001) used the slopes in Eq. [4] as metrics of
the TS. It was suggested to distinguish types of temporal persistence by the significance of the differences between the regression
slope and one, and by the significance of the difference between the
intercept of the regression equation and zero (Grant et al., 2004).
Kachanoski and de Jong (1988) noted that temporal transformations other than the linear transform Eq. [4] could be employed to
characterize the time stability. The use of the Pearson correlation
coefficient between measurements at two observation times (Cosh
et al., 2004) is closely related to the temporal persistence concept of
Kachanoski and de Jong (1988). Also Kamgar et al. (1993) found
a similar correlation between measurements at different observations days, across depth. Another related approach is the one of
Biswas and Si (2011a, 2011b) who used wavelet coherency analysis
to analyze the scale-dependent persistence of SWC across depth

and the TS in time.
Spearman Rank Correla on

The non-parametric Spearman rank correlation coefficient rs was
also proposed to quantify the TS SWC by comparing SWCs at
two different observation times (Vachaud et al., 1985). This correlation coefficient between SWC measured at j1 and j2 observation
times is computed as
i= N
rS ( j1 , j2 ) = 1
6 i=1 s [ R(i , j1 ) R(i , j2 )]
( N s 1) N s ( N s +1)
[5]
with Ns the number of spatial observation locations, i = 1,2,,

Ns, R(i,j) the rank of ij at location i and observation time j. The
closer rs is to one, the more temporally stable are the SWC patterns.
When introducing rs, Vachaud et al. (1985) noted that this method
cannot help in the selection of the positions of measurement locations, and that this test may be questionable if differences between
measured values are smaller than experimental uncertainties themselves. The latter can be the case in situations in which either the
probability density function is very uniform or the experimental
determinations are very crude.
One application of this coefficient is to characterize the memory
in spatial SWC patterns by comparing rs values as the difference
between j2 and j1 (and therefore time between the measurements
j1 and j2) gradually increases (Rolston et al., 1991).
Representa ve Loca ons

Representative locations are usually defined as the locations where
measured soil water contents either are close to the average water
contents or can be easily transformed to obtain such averages. The
term catchment average soil moisture monitoring (CASMM) sites
was proposed for such locations at the watershed scale (Grayson
and Western, 1998).
Several methods were proposed to define the representative location. The simplest method is to use the location with the MRD
closest to zero (Vachaud et al., 1985). However, this approach
ignores the fact that MRD are actually statistics with inherent
errors characterized by SDRD. Therefore several of them may
be indistinguishable. A modification of the above approach is to
use the location where the highest TS is observed, i.e., the lowest
SDRD, and to provide a constant offset which is used to acquire a
mean SWC value across the observation area (Starks et al., 2006;
Heathman et al., 2009). Guber et al. (2008) and Schneider et al.
(2008) suggested looking for locations with smallest MRD and
SDRD. de Rosnay et al. (2009) proposed to use the polar coordinates with MRD values as radii and SDRD as the angular
coordinate to combine TS information about different locations.
Jacobs et al. (2004) proposed to combine MRD and SDRD in a
single value of a so called root mean square error,
RMSE i = MRDi 2 + SDRD2i
[6]
and select the location with smallest RMSE. The mean absolute
bias error, or MABE (Hu et al., 2010b) was also proposed in
the form
MABE i =
j= N t
RDij MRDi
j =1
1 + MRDi
[7]
The location with the minimum value of MABE was suggested to

be the most representative.
Landscape position and soil texture have also been considered as
indicators for selection of the representative location. Jacobs et al.
(2004) analyzed daily surface soil moisture measurements using
time stability analysis in function of soil texture, clay, sand and
topography within four fields in the Walnut Creek watershed.
They showed that locations with a mild slope (0.9 to 1.7%) consistently exhibit time stable features with MRD close to zero. Hilltop
and steep slopes underestimated consistently the field-average
SWC. The stable locations in terms of the RMSE were found to
have a moderate to moderately high clay content as compared to
the field average. Grayson and Western (1998) expected that sites
that represent field means should be found in field neutral locations, defi ned by slope and aspect. The representative locations
were located in areas reflecting average topography characteristics,
in terms of elevation and slope in works of Thierfelder et al. (2003)
and Brocca et al. (2009). Teuling et al. (2006) found for three
different datasets that topographic attributes were not useful in
identifying representative sites for the spatial average SWC.
Finding a single location to estimate average water contents at
several depths simultaneously proved to be difficult. Tallon and
Si (2003) were able to find only one site to be representative, with
small MRD and SDRD, for two separate depths. Guber et al.
(2008) showed that the best representative locations, those with
the smallest SDRD, were different for five depths in the soil profile.
Field sites considered temporally stable, with MRD close to zero,
for the surface soil moisture were not stable for the profi le soil
moisture in the study of Heathman et al. (2009). Hu et al. (2010a;
2010b) found that one site can be representative of five soil depths
and four soil layers, respectively.
The effect of the temporal frequency of SWC measurements on the
selection of the representative location has only been researched to
a small extent. Brocca et al. (2010) found, by randomly selecting
(100 times) between 5 and 35 sampling days from the data set, that
with 12 sampling days the representative points, with MRD closest
to zero and smallest SDRD, were correctly identified in 90% of the
cases. Also considering the 12 sampling days, characterized by the
wettest and driest conditions, the representative points could be
retrieved. Martnez-Fernndez and Ceballos (2005) found under
Mediterranean conditions that approximately 1 yr of measurements was required to determine the representative point, with
MRD closest to zero and smallest SDRD, both at extents of 600
m and 40 km.
Gaussian Approxima on of the Distribu on

of MRD
The use of the Gaussian distribution could make it possible to apply
powerful parametrical statistical techniques and tests in TS SWC
analysis. Inspection of published dependences of MRD on the
location number (see next section) showed that these dependences
are often symmetrical and resemble the Gaussian cumulative function. Therefore, the normal probability distribution function,
MRD MRD
1
i
p (MRDi ) = 1 + erf
2 SDMRD2
2
[8]
could be fitted to the experimental dependencies of MRD on i

if the probability p(MRDi) will be defined as [1/2 + (i 1)]/Ns.
Parameters in Eq. [8] are the mean MRD, MRD, and the variance SDMRD2 .
The fitted MRD and SDMRD can be compared to the sample mean,
MRD =
1
Ns
i= N s
MRDi ,
[9]
i =1
and the sample standard deviation,

SDMRD =
1 i= N s
(MRDi MRD
N s 1 i=1
[10]
respectively. Here, Ns is the number of locations and sample mean

and standard deviation are defined over the whole observation period.
6 Experimental Data on TS SWC

Summary of the Experimental Data
Data Set and Descrip ve Sta s cs

Descriptive statistics and measurement-related parameters of
published MRD data are provided as a spreadsheet in the online
supplementary material section. An effort was made to have this
information as complete as possible. Mean relative difference
values and associated SDRD were digitized from 157 graphs in
37 publications. For each case, descriptive statistics were directly
calculated, including the sample MRD and SDMRD, their
Gaussian fitted analogs and the corresponding coefficient of determination (R 2) of the Gaussian fit; the coefficient of kurtosis of
MRD, KMRD; the coefficient of skewness of MRD, SMRD; and
the sample mean and standard deviation of the SDRD, SDRD,
and SDSDRD, respectively. In addition, for each dataset SWC
observation-related parameters were recorded, such as the maximum distance between two sampling locations (extent); the
average separation distance between sampling locations, in case

of irregularly distributed points, and for regular measurement
grids, the grid spacing (spacing); the number of observation times
(Nt), the number of observation points (Ns), the duration of the
observation period (period), the distance between the soil surface
and the center of the measured layer (depth), and the thickness of
the measured layer (thickness), among other parameters that are
provided in the spreadsheet.
Table 1 presents the descriptive statistics of MRD and SDRD,
and the measurement-related parameters for the entire dataset. In
some cases published MRD graphs have not been plotted according to the traditional Vachaud et al. (1985) method as shown in
Eq. [1] and Eq. [2]. Therefore, data corresponding to 19 graphs
from four publications (Martnez-Fernndez and Ceballos, 2003;
2005; Bosch et al., 2006; Cosh et al., 2006) were excluded from
the analysis.
The histograms of measurement-related characteristics are given in
Fig. 1. Overall, the dataset is strongly skewed towards the remote
sensing studies with short measurement periods, mainly surface
SWC measurements, and thin soil layers. To account throughout
the analysis for the large heterogeneity in measurement depth and
thickness, data sets were separated into surface, subsurface and
profi le measurements. Surface measurements correspond to SWC
observations in the top 0.2 m, found generally for remote sensing
studies, while measurements in the top 0.3 m and deeper are considered profi le measurements. Subsurface measurements are those
made at specific depths, with varying thickness.
Non-zero MRD
Several datasets showed MRD values substantially different
from zero. This contradicts the definition of MRD, because
the sum of the MRD is equal to zero at any observation time,
and therefore also MRD has to be zero [Eq. 12 and Eq. 24].
Small deviations of MRD from zero could appear during
the digitizing process due to the limited quality of the graphs.
About 11, 20, and 29% of the cases had |MRD| > 0.05, 0.02,
and 0.01, respectively.
Assuming that no errors have been made in the computation of
MRD in the original papers, the only explanation for non-zero
MRD can be a difference between numbers of observations used
on different observation times due to sensor malfunctioning or other
possible causes of data loss. The effect of the incomplete observation
on MRD is illustrated with the synthetic example of Table 2. The
ideal TS SWC is assumed for three locations. The top part of the
table summarizes the case when the data are complete, for which
MRD = 0. The bottom part of the table shows the case when two
measurementstime 2 at location (1) and time (3) at location 3
are absent. The absence of two observations leads to the non-zero
MRD = 0.025.
Table 1. Descriptive statistics of the available data on temporal stability of soil water content (TS SWC), (see also supplementary information) presented in terms of mean relative differences (MRD), standard deviation of relative difference (SDRD), and measurement-related parameters. Digitized
data from 19 graphs, corresponding to four different papers, were excluded from the analysis.
N
Min.
Max.
Mean
SD
Skewness
Mean relative difference and standard deviation of relative difference

Fitted MRD
138
0.166
0.098
0.022
0.032
1.403
Sample MRD
138
0.052
0.097
0.000147
0.014668
1.828
Fitted SDMRD
138
0.013
0.866
0.213
0.151
0.876
Sample SDMRD
138
0.012
0.778
0.217
0.148
0.744
R2
138
0.841
0.998
0.969
0.030
2.043
KMRD
138
2.009
7.313
0.482
1.149
3.258
SMRD
138
0.948
2.438
0.253
0.536
0.853
SDRD
137
0.010
0.650
0.120
0.090
2.181
SDSDRD
137
0.004
0.387
0.062
0.058
2.656
44000
4869
10006
2.3
Measurement-related parameters
Extent (m)
138
Spacing (m)
131
.5
10000.0
1318.0
3070.9
2.3
Nt (days)
136
87840
3078
14993
5.5
Ns
138
210
45
38
1.4
Period (days)
137
11
1292
234.5
277.1
1.9
Depth (m)
138
0.025
3.500
0.414
0.618
2.9
Th ickness (m)
135
0.040
1.50
0.36
0.36
1.3
R 2:
MRD: spatial mean of MRD, SDMRD: standard deviation of MRD,

coefficient of determination of the Gaussian fit, KMRD: kurtosis of MRD, SMRD:
skewness of MRD, SDRD: spatial mean of SDRD, SDSDRD: standard deviation of SDRD, Extent: maximum distance between two sampling locations,
Spacing: average separation distance between sampling locations, in case of irregularly distributed points, and for regular measurement grids it is the grid spacing, Nt : number of observation times, Ns: number of observation points, Period: duration of observation period, Depth: distance between the soil surface and
the center of the measured layer, Th ickness: thickness of measured layer.
Fig. 1. Overview of experimental designs in studies of temporal stability of soil water content (TS SWC). Frequency distributions are given for (a) spatial extents of observations defined as is the maximum distance between two sampling locations, (b) the number of sampling locations Ns, (c) duration
of observation periods, (d) depths of the center of the layer in which soil water content (SWC) or soil water storage were recorded, (e) thickness of the
layer in which SWC or soil water storage were recorded, and (f ) position of the measurement layer in soil profiles. Surface: from 0 to 0.2 m, Profile:
from 0 to 0.3 m and deeper, Subsurface: at specific depths, with varying thickness.
Table 2. Synthetic example of the effect of the incomplete measurement set on the relative difference (RD), the mean relative difference (MRD), and
the average mean relative difference MRD.
Measured SWC (cm3cm3)
Time
Loc. 1
Loc. 2
RD
Loc. 3
Avg. across Loc. 1 to 3
Loc. 1
Loc. 2
Loc. 3
MRD
Complete measurements
1
0.250
0.310
0.190
0.250
0.000
0.240
0.240
0.280
0.340
0.220
0.280
0.000
0.214
0.214
0.310
0.370
0.250
0.310
0.000
0.194
0.194
0.160
0.220
0.100
0.160
0.000
0.375
0.375
MRD
0.000
0.667
0.667
0.250
0.000
0.240
0.240
Partly absent measurements (marked as ND)

1
0.250
0.310
0.190
ND
0.340
0.220
0.280
ND
0.214
0.214
0.310
0.370
ND
0.340
0.088
0.088
ND
0.160
0.220
0.100
0.160
0.000
0.375
0.375
MRD
0.029
0.229
0.276
0.025
Figure 2a shows that non-zero MRD are

most likely to occur in cases when SDMRD are
relatively large, and that SWC measurements
for the entire soil profi le result more often in
MRD values close to zero, than measurements at specific depths. That may happen in
part because the profi le data are obtained by
integrating measurements at several depths,
and therefore in absence of measurements at
some depths one still is able to obtain the value
for the whole profi le.
Fig. 2. Relationships between (a) the spatial average of the mean relative difference (MRD),
The standard deviation of SDRD, SDSDRD,
sample MRD, and the standard deviation of the MRD, sample SDMRD, and (b) between
is related to the spatially averaged SDRD,
the spatial mean of the standard deviation of the relative differences, SDRD, and the standard deviation of the RD standard deviations, SDSDRD, distinguishing between surface (from
SDRD, as shown in Fig. 2b. On average,
0 to 0.2 m), profile (from 0 to 0.3 m and deeper), and subsurface (at specific depths, with varythere is a linear dependence between the
ing thickness) SWC measurements.
two values: the less stable the RD, i.e., the
greater the SDRD, the more temporal variability of the RDs is expected. Soil water
content measurements averaged over the entire soil profile show
Gaussian Fit
the smallest SDRD and SDSDRD, indicating that RDs for
The Gaussian distribution function, Eq. [8], was fitted to the data,
this type of measurements are more stable in time and that
providing a fitted MRD and SDMRD, and R2 . Results are sumSDRD is more stable in space than for SWC measurements at
marized in Table 1. Overall, the fit was mostly satisfactory, with
specific depths.
R2 values larger than 0.990, 0.985, 0.980, 0.970, and 0.950 in 22,
36, 49, 64, and 83% of all cases, respectively. Figure 3a shows the
Standard deviation of RD appeared to be sensitive to extent.
negatively skewed R2 distribution. The highest R2 values (>0.97)
The average standard error of SDRD for the extent ranges
were unrelated to sample SDMRD (Fig. 3b), indicating that the
<200, 2002000, and >2000 m were 0.086 0.010, 0.129
quality of the Gaussian fit is independent from the magnitude of
0.010, and 0.127 0.023. Interestingly, a similar dependence
the spatial variability of MRD. The smallest R 2 values tended to
can be traced for the SDSDRD, i.e., standard deviations of the
be associated with small (<0.1) and intermediate (0.30.4) sample
SDRD. The average values of SDSDRD within the scale ranges
SDMRD values. Th is often corresponds to the absence of large
< 200 m, 2002000 m, and >2000 m were 0.032, 0.066, and
absolute values of MRD at the ends of the sample distribution
0.073, respectively.
MRD curves.
The accuracy of fitting Eq. [8] to

observed MRD demonstrated a clear
dependence on scale. The average
values of determination coefficients
R 2 in the extent ranges <200 m, 200
2000 m, and >2000 were 0.983, 0.972,
and 0.951, respectively, and were significantly different at p = 0.05. The
minimum values of R 2 were 0.903,
0.890, and 0.840, in the same ranges
of extent, respectively. Mean relative
difference tends to be more normally
distributed in space at smaller scales
than at larger scales. This implies
that SWCs tend to have distributions
closer to normal at the smaller scales
compared to the larger scales. Most
probably, many other factors come into
play at larger scales that do not occur
at smaller scales: climate variability,
catchment-based processes, more pronounced topographic effects.
Fig. 3. (a) Histogram of R2 for the Gaussian fit of the mean relative difference (MRD) data and (b)
relationship between R2 and the sample standard deviation of the mean relative difference (SDMRD).
Dispersion in the overall linear relationship between sample and fitted

SDMRD increased with increasing
SDMRD (Fig. 4a), with minimal dispersion for SDMRD <0.15 (Fig. 4b).
This behavior was independent from
measurement depth (graphs not shown).
To evaluate the effect of non-zero
MRD on the Gaussian fit, the oneFig. 4. Relationship between the sample and fitted standard deviation of the mean relative difference
(SDMRD), (a) for the complete SDMRD data range and (b) for SDMRD <0.15.
parameter version of Eq. [8] was also
fitted, imposing MRD = 0. An F-test
at the 0.05 probability level was used
SDMRD < 0.15 (blue dots in Fig. 5d) again showed no relationship
to decide whether the two fits performed equally well, which was
between R2 and SMRD.
true for 32% of the cases.
Positively skewed distributions find their origin in a small number
Figure 5 shows the relationships between R2 , KMRD, and SMRD.
of permanently-wetter-than-average observation points. Soil water
2
The smallest R values correspond to cases with probability density
regimes at these wet points are presumably non-locally controlled
functions that are flatter and have longer tails (negative KMRD)
by topography related attributes (Grayson et al., 1997). In many
than the Gaussian function (Fig. 5a). Negative KMRD values were
studies, observation points were not located randomly within the
encountered in 79% of the datasets. Most of the cases for which
fields or catchments. Their position within the fields or catchments
the one- and two- parameter Gaussian fits perform equally well
was often chosen along transects and in accordance to extreme
(red dots in Fig. 5a) could be included in this category, while for
topographical attributes, such as the highest or lowest point, on
data sets with SDMRD <0.15 (blue dots in Fig. 5b) no relationridges or in drainage lines, so that the SWC regimes of these prefership was found between R2 and KMRD. About 61% of the cases
entially located points can be significantly different from the other
could be considered to have moderately asymmetrical distributions,
locations (e.g. permanently flooded or wetter points near the drain0.5 > SMRD > 0.5, while 6% showed SMRD < 0.5 and 33%
age lines or near catchment outlets). However, also highly variable
SMRD > 0.5. The smallest R 2 were found for negatively skewed
physical soil properties in combination with heterogeneous soil
distributions (SMRD < 0), while positively skewed distributions
profi les could cause permanently wetter-than-average observa2
(SMRD > 0) generally showed a higher R (Fig. 5c). Datasets with
tion points (e.g., deficient drainage conditions). A small number
of positive extreme MRD values did not

seem to affect negatively the R 2 of the
Gaussian fit, but will result generally in
significant differences between the oneand two-parameter Gaussian fits and in
positively biased estimates of the fitted
MRD. Cases for which the one- and
two-parameter Gaussian models performed equally well showed mostly a
negative SMRD.
Negatively skewed distributions can
originate from a small number of permanently dryer-than-average observation
points. Soil water dynamics at drier
points might be dominated by local
controls (Grayson et al., 1997), mainly
soil related, such as positively skewed
log-normal hydraulic conductivity pdfs,
or locations with dominant preferential flowalthough they could also be
associated with preferential location
according to topographical attributes,
such as south facing slopes (Tallon and
Si, 2003). At least in this dataset the
occurrence of SMRD < 0 is much less
frequent than SMRD > 0, and leads in
most cases to a strong degradation of
the quality of the Gaussian fit. Figures
5e and 5f show that for most of the cases
where SMRD < 0, also KMRD < 0.
This condition corresponds also to most
of the cases for which the one- and twoparameter Gaussian fits perform equally
well. Further research is required to confirm the controlling factors behind the
Gaussian behavior of MRD and possible
deviations from it.
Spearman Rank and

Pearson Correla on
Spearman rank correlation coefficients
Fig. 5. Relationships between coefficient of determination (R2), coefficient of kurtosis (a) and (b), and
are used to measure the strength of TS
coefficient of skewness (c) and (d) of MRD (KMRD and SMRD, respectively). To represent KMRD
SWC and enable the comparison of TS
data on a logarithmic scale, an offset equal to 3 was used. Red dots represent cases for which the one- and
SWC across depth, at different sites or
two-parameter Gaussian fits performed equally well. Blue dots represent cases for which SDMRD <0.15.
for different measurement periods. For
three plots in central Italy, Brocca et al.
(2009) found rs values ranging from 0.14 to 0.62, from 0.20 to
largest rs for the sandy loam soil with gently rolling topography
0.65, and from 0.30 to 0.86, for measurement periods of 61, 89,
and rage land cover (LW03), intermediate rs for the silty loam
and 244 d, respectively. They recall that the highest values were
soil with gently rolling topography and range land cover (LW13),
obtained for the site with the steepest slope, in agreement with
and the smallest rs for the silty loam soil with flat topography
findings by Mohanty and Skaggs (2001), who found for periods of
and a split winter wheat/grass cover (LW21). Figure 6 shows
less than 1 mo at three fields in the Little Washita watershed the
the time dependency of the rs for these three cases, which can be
Pearson correlation coefficients were used (Lin, 2006; Cosh et

al., 2006; 2008; Herbst et al., 2009; Heathman et al., 2009) to
characterize the temporal persistence by the strength of linear
relationships between SWC at different observation times
(Kachanoski and de Jong, 1988).
Factors Aec ng TS SWC

Means, frequency, and locations of data collection as well as sitespecific soil, landscape, and weather properties have been shown to
affect the TS metrics. Knowing these effects is beneficial both for
selection of the representative sampling locations and for projecting the TS for locations that have not been sampled yet.
Sensors
Fig. 6. Time dependency of the Spearman rank correlation coefficient
(rs) for three fields in the Little Washita Watershed (Mohanty and
Skaggs, 2001).
considered representative for situations with high, intermediate,

and low rs found in other TS SWC studies. Coppola et al. (2011)
found rs ranging from 0.53 to 0.99 during a measurement period
of 2 mo. Schneider et al. (2008) found higher rs between consecutive measurement days for sites with homogeneous vegetation as a
consequence of heavy grazing, than for non- or less heavily-grazed
sites with a denser and more heterogeneous vegetation. Similar
statements regarding the negative effect of growing vegetation on
rs were also made by Gmez-Plaza et al. (2000), who found along
three transects in a catchment rs ranging from 0.5 to 0.95. Hu et
al. (2009) found, excluding the topsoil layer an increasing mean
rs with depth, from 0.85 to 0.90, at 40 and 80 cm, respectively.
Biswas and Si (2011b) found interannual cycles in TS SWC as
expressed by rs. The rs were highest (>0.97) when comparing observations from the same seasons, even from different years, while rs
was smaller (<0.90) for observations from different seasons.
In general, rs is higher for profi le SWC than for surface or topsoil
SWC measurements. Vachaud et al. (1985) reported rs values for
1-m soil layers at two sites, ranging from 0.79 to 0.99 and from 0.66
to 0.78, respectively. Sampling dates with the highest SWC showed
the highest rs. Martnez-Fernndez and Ceballos (2003) found
in their regional study for 1-m soil profi le measurements rs ranging from 0.57 to 1, observing the smallest values during transition
periods from dry to wet states. Grant et al. (2004) found rs > 0.90
for consecutive days throughout a 2 yr period for a top layer of 0.75
m. The rs found by Rolston et al. (1991) for 1.5-m soil profi les in
an almond orchard ranged from 0.32 to 0.99, and those obtained
by Comegna and Basile (1994) for 0.9-m soil profi les ranged from
0.25 to 0.81. Kamgar et al. (1993) found increasing rs with depth
for three 1-m soil layers, ranging from 0.12 to 0.91 in the top layer
and from 0.87 to 1 in the deepest layer. Also Cassel et al. (2000)
found increasing rs with depth.
In the complete data set, the Theta-probe was used in 45 cases,

mainly in remote sensing studies, followed by TDR (32 cases) and
Neutron probe (35 cases). The Multisensor capacitance probes
and Hydra-probes were used in 16 and 5 cases, respectively, while
gravimetric SWC measurements were used in 3 cases, and a Polar
Scanning Radiometer (PSR) and ECH2O probes both in 1 case.
Figure 7 compares statistical distributions of sample SDMRD
among measurement methods. The Theta-probe measurements
show the smallest scatter in SDMRD, probably because this
technique was mostly used in remote sensing studies where rather
homogeneous field conditions are targeted. Distributions for the
Neutron probe and the Multisensor capacitance probe, where
remarkably similar, with similar means (p = 0.94).
The effect of the measurement method in identifying the presence of TS SWC was first demonstrated by Kirda and Reichardt
Fig. 7. Distribution of the sample standard deviation of the mean

relative differences (SDMRD) obtained with different sensors (G:
gravimetric, NP: neutron probe, MCP: multi-sensor capacitance
probe, TP: Theta-probe, HP: Hydra-probe, PSR: Polar Scanning
Radiometer). The caps at the end of each box indicate the minimum
and maximum, the lower and upper quartiles delimit the box, and the
line in the center of the box is the median.
(1992). They compared rs derived from SWC measurements using

tensiometers, neutron probes, and resistance blocks. They observed
TS SWC using neutron probe measurements but not with two
other sensors. The authors suggested that the lack of TS SWC for
tensiometer and resistance block measurements was due to the
extreme textural variability at the experimental site. The soil at
this site stems from alluvial deposition and consists of a mixture of
sand and gravel, and compacted packs of clay. The neutron gauge
measurements provided average water content integrated over a relatively large sphere of influence compared to the values obtained
from tensiometers and resistance blocks and therefore, the influence of soil variability on the TS SWC was minimal. Reichardt et
al. (1997) pointed out that at least part of the TS SWC (expressed
in terms of MRD) may be an artifact due to inadequate neutron
probe calibrations as a consequence of varying local soil properties. Hu et al. (2009) compared the effects of different neutron
probe calibration procedures on TS SWC in a small catchment.
Location-specific and a catchment-wide calibration resulted in
a similar ranking of MRD values. All the calibration equations
provided almost identical spatial mean SWC estimates, while differences appeared among the corresponding standard deviations.
An interaction of soil properties and sensor properties seems to be
able to affect the observation of TS SWC.
Depth and Layer Thickness

An overview of the statistical distributions of SDMRD for
three classessurface, subsurface, and the whole profileis
shown in Fig. 8. The MRD values derived from subsurface SWC
Fig. 8. Box and Whisker plots for the sample standard deviation of
the mean relative differences (SDMRD) according to surface (from
0 to 0.2 m), profile (from 0 to 0.3 m and deeper), and subsurface (at
specific depths, with varying thickness) soil water content (SWC)
measurements. The caps at the end of each box indicate the minimum
and maximum, the lower and upper quartiles delimit the box, and the
line in the center of the box is the median. Black dots are outliers, more
than 1.5 times the inter-quartile range larger than the upper quartile.
Different letters represent different averages at p < 0.05.
measurements generally show a larger SDMRD than values

obtained from surface measurements. The difference of averages
across these two groups was significant at P < 0.001. The profi le
water storage had the average SDMRD very close to that of surface
SWC. The difference between the average values was not significant at the 0.05 significance level.
Large or small SDMRD can occur across all depths (Fig. 9), while
intermediate SDMRD values, roughly between 0.1 and 0.3, were
mainly found for depths <0.5 m. Also for individual studies with
multi-depth MRD data, both increasing (e.g., Cassel et al., 2000;
Tallon and Si, 2003; Lin, 2006; De Lannoy et al., 2007; Guber et
al., 2008; Heathman et al., 2009; Hu et al., 2010b) and decreasing
(e.g. Kamgar et al., 1993; Gmez-Plaza et al., 2001; Starks et al.,
2006; Choi and Jacobs, 2011; Martnez et al., 2010) SDMRD with
depth was found. Vegetation and agricultural activity (e.g., tillage) could explain lower SDMRD in surface soils. Pachepsky et al.
(2005) who used the data from a non-vegetated site observed only a
weak decrease of SDMRD with depth. Also the large-extent study
of Martnez-Fernndez and Ceballos (2003) for mostly sandy soils
did not observe dependencies of the SDMRD on depth.
Spa al Scale
Data were grouped into three extent categories: <200m, 2002000
m and >2000 m. Figure 10 shows box-and-whisker plots of sample
SDMRD for the three categories. Excluding the outlier for the
200- to 2000-m category (Fig. 10), the averages standard errors
of SDMRD were 0.07 0.01, 0.24 0.01, and 0.28 0.02, respectively, being the average of the first category significantly different
(p < 0.001) from the others. This result could be expected since for
Fig. 9. Relationship between the sample standard deviation of the

mean relative differences (SDMRD) and the depth to the center of
the measured soil layer.
Fig. 10. Box and Whisker plots of sample standard deviation of the
mean relative differences (SDMRD) for three categories of extent
(<200, 2002000, >2000 m). The caps at the end of each box indicate
the minimum and maximum, the lower and upper quartiles delimit the
box, and the line in the center of the box is the median. Black dots are
outliers, more than 1.5 times the inter-quartile range larger than the
upper quartile. Different letters represent different averages at p < 0.001.
an increasing extent of the study area, additional factors of spatial

variation of SWC (e.g. convective rainfall with partial coverage
of the study area) can manifest itself and may therefore affect the
presence of TS SWC. A substantial body of the literature on spatial variability of SWC points in this direction [e.g., Famiglietti
et al. (2008), and references therein]. We note that it is not known
whether TS SWC can be maintained under increasing extent of
the study area. The hypothesis that an increase in spatial scale at
a specific site leads to an increased SDMRD could be tested with
data from studies with different extent which were
performed at two USDA research watersheds
Walnut Creek and Little Washitaduring the so
called SMEX and SGP campaigns when the satellite images, airborne remote sensing, and ground soil
water content monitoring were collocated in time.
data from Choi and Jacobs (2011), measured at 14 moments in

time during SMEX05 with a Theta-probe at 8 cm depth (depth
interval of 6 cm), at 3 points within each of the 10 studied fields
within the watershed. Spacing within and between the fields was
approximately 100 m and 2 km, respectively, with a total extent
of approximately 17.5 km. It is worth noting that the Cosh et
al. (2004) measurements came from long-term installed equipment, while the Choi and Jacobs (2011) data were obtained using
a destructive approach and are therefore technically not repeatable. Although non-co-located observations in time could increase
the SDMRD, at least in this case the effect seems to be minimal.
The similarity in SDMRD obtained in both studies, for different extents, may indicate that SDMRD reaches a maximum at a
certain threshold extent, beyond which it remains constant. The
similar SDMRD also indicate, at least for this extent magnitude,
that SDMRD is not affected by replicating local sampling. The
multi-depth MRD data (Fig. 11, points 4a4e) from Choi and
Jacobs (2011) show also that SDMRD decreased with depth, reaching approximately 60% of the topsoil variability at 30 cm depth.
The SDMRD of the top 31-cm soil profi le (Fig. 11, point 4f) corresponds well with the SDMRD at a depth of 18 cm (Fig. 11, point
4d). Surface SWC measurements with a Theta-probe were reported
by Joshi et al. (2011) for SMEX02 and SMEX05 campaigns in field
WC11 (Fig. 11, points 5a and 5b, respectively) and the tile-drained
WC12 field (Fig. 11, points 5c and 5d, respectively). At both fields
similar SDMRD were obtained for both sampling campaigns. The
smaller SDMRD for the WC12 field may be a consequence of the
tile-drainage system. Teuling and Troch (2005) found that soil
drainage during wet periods destroyed SWC variability created by
spatially variable transpiration. The SDMRD for field WC11 was
smaller than the SDMRD obtained from Choi and Jacobs (2007)
at the same field (Fig. 11, point 3a), but with almost four times less
observation points. The spread obtained for the large extent Polar
Scanning Radiometer (Fig. 11, point 5e) was lower than the spread
Walnut Creek Watershed

Temporal stability of SWC has been intensively
studied in the Walnut Creek catchment and surroundings, Iowa, during SMEX02 and SMEX05.
Figure 11 shows the sample SDMRD versus extent.
Cosh et al. (2004) provided MRD data for 12 measurement points with a spacing and extent of 2 and
24 km, respectively, and a measurement period of
52 d. Soil water content was measured with Hydraprobes at a depth of 5 cm (with a depth-interval of
4 cm) during SMEX02. A similar SDMRD value
was obtained (Fig. 11, point 4b) for the MRD
Fig. 11. Sample standard deviation of mean relative difference (SDMRD) versus extent
for the Walnut Creek watershed, distinguishing between surface (from 0 to 0.2 m), profile
(from 0 to 0.3 m and deeper), and subsurface (at specific depths, with varying thickness)
soil water content (SWC) measurements. Data from Cosh et al., 2004 (1); Jacobs et al.,
2004 (2a2d); Choi and Jacobs, 2007 (3a3j); Choi and Jacobs, 2011 (4a4f ); and Joshi
et al., 2011 (5a5e).
obtained from the other data sets, as a consequence of

the measurement method and its support.
Surface soil MRD data were also provided by Jacobs et
al. (2004) for four fields (WC11, WC12, WC13, and
WC14, points 2a-2d, respectively, in Fig. 11), which
were densely measured using a Theta-probe during
SMEX02. The SDMRD for field WC11 was similar
to the spread found for the spatially dense Joshi et
al. (2011) data (Fig. 11, 5a and 5b), but for the tiledrained field W12 the spread was higher. The spread
for field WC13 was highest and close to the values
obtained from Choi and Jacobs (2007) (Fig. 11, point
3f). Overall, the Walnut Creek data, coming from a
total of 26 MRD graphs from five publications, seem
not to support the increasing SDMRD with increasing extent found in Fig. 10 for the complete dataset.
Th is could be a consequence of the rather homogeneous low relief topography, poor surface drainage
and homogenous clay content (23 to 31%) that characterizes the watershed (Choi and Jacobs, 2011).
Fig. 12. Sample standard deviation of mean relative difference (SDMRD) versus extent
for the Little Washita watershed, distinguishing between surface (from 0 to 0.2 m),
profile (from 0 to 0.3 m and deeper), and subsurface (at specific depths, with varying
thickness) soil water content (SWC) measurements. Data from Mohanty and Skaggs,
2001 (1a1c); Starks et al., 2006 (2a2j); Heathman et al., 2009 (3a3k); Joshi et al.,
2011 (4a4d).
Li le Washita Watershed
Data on the TS SWC for the Little Washita Watershed are
summarized in Fig. 12. The field and catchment data show an
increasing SDMRD with extent. At 600 m an average SDMRD
of 0.15 was observed whereas at the scale of 35 km the average
value was 0.32. Important variation in the SDMRD was observed
at each of the observation scales (Fig. 12). Mohanty and Skaggs
(2001) presented the TS analysis of surface Theta-probe-measured
SWC at three fields during SGP97, using a regular measurement
grid of 49 points with a spacing and extent of 100 and 600 m,
respectively. The SDMRD for the flat LW13 field (Fig. 12, point
1b) was smaller than the SDMRD for fields LW03 and LW21 (Fig.
12, points 1a and 1c), which showed both a rolling topography and
identical SDMRD. For soil surface measurements at fields LW12,
LW13, LW21, and LW45, during the extremely wet remote sensing Cloud and Land Surface Interaction Campaign (Heathman
et al., 2009), substantially lower SDMRD were obtained (Fig. 12,
points 3a3d). In each field eight points were monitored, with an
approximate spacing of 200 m and an extent of 600 m.
Soil water content was measured by Starks et al. (2006) at 15-cm
depth-intervals down to 60 cm using TDR probes at eight sites,
with an average spacing and extent of approximately 10 and 21 km,
respectively. During SGP97 measurements were repeated 29 times
and during SMEX03 six times. Standard deviation of MRD was
highly consistent between both sampling campaigns and across
depth. The first two topsoil layers (Fig. 12, points 2a and 2e; and
2b and 2f) showed similar, but higher SDMRD than compared
to the deeper layers (Fig. 12, points 2c and 2 g; and 2d and 2h).
Considering the entire profi le (60 cm), almost identical, SDMRD
were obtained for both campaigns (Fig. 12, points 2i and 2j).
Joshi et al. (2011) analyzed the same Theta-probe measured soil surface data as Mohanty and Skaggs (2001), for fields LW03, LW13,
and LW21 (Fig. 12, points 4a, 4b, and 4c, respectively). For field
LW21 also MRD data from the CLASIC2007 experiment were
provided (Fig. 12, point 4d). Except for field LW03, for which the
SDMRD was twice as large as for the other fields, similar SDMRD
were obtained for both studies. Overall, for this watershed, published MRD data showed an increasing SDMRD with extent.
Spa al Density
The spatial density, defi ned as the ratio Ns/Extent, was used to
summarize the data with respect to spacing. Figure 13 demonstrates that SDMRD is most variable at intermediate densities,
roughly from 0.02 to 0.3 m1, and is generally higher for smaller
densities. Standard deviation of MRD appears to be generally less
than 0.2 for increasing density from 0.3 m 1. Datasets with these
larger densities were obtained with a spacing and extent ranging
from 0.5 to 22 m, and from 3 to 164 m, respectively, with Ns ranging from 10 to 108. We hypothesize that for these high densities
and fine scale studies, the effects of non-local controls, including
rainfall, are possibly minimized, and that non-local effects are the
dominant controlling factor for the smallest densities, leading to
increasing SDMRD with decreasing density. At intermediate densities local and non-local controls interplay, leading to a wide range
of possible SDMRD values.
Both the extent and the resolution of the sampling scheme had
an impact on the mean of the distribution of SWC in the work of
Petrone et al. (2004). In terms of the distribution of soil moisture,
the extent of sampling within a grid was not as significant as the
density, or spacing, of the measurements in this work. We note that
it is not known how the spatial configuration of sampling locations
affects the manifestation of TS SWC. The spatial configuration
that vegetation controls affected the time stability of soil moisture in the 0 to 20 cm layer making
it difficult to estimate soil moisture values from
earlier measurements using the time persistence
concept as expressed in Eq. [4]. Manfreda and
Rodrguez-Iturbe (2006) found that the length of
the measurement method affected the variance of
the long-term spatial mean daily SWC, especially for
shallow rooted vegetation. The SDMRD might be
seen as the TS equivalent of this variability.
Soil Proper es
Soil properties are commonly implicated in the existence and extent of TS SWC since the ability of soil
to retain and to transmit water is the obvious reason
for differences in SWC. The effect of soil properties
Fig. 13. Dependence of the sample standard deviation of the mean relative difference
and local terrain attributes on SWC patterns was
(SDMRD) on the spatial density of measurements [number of spatial observations (Ns)/
referred to by Grayson et al. (1997) as local controls.
Extent], distinguishing between surface (from 0 to 0.2 m), profile (from 0 to 0.3 m and
They considered the dry SWC state to be dominated
deeper), and subsurface (at specific depths, with varying thickness) soil water content
(SWC) measurements.
by vertical fluxes, which depend mainly on local soil
properties, while lateral fluxes are minimized due to
was shown to be an important factor in characterizing the spatial
the low hydraulic conductivity. Soil texture was sugvariability of SWC (Manfreda and Rodrguez-Iturbe, 2006).
gested to affect the TS by Vachaud et al. (1985). In their work, two
locations that always showed the largest and smallest water storage
Temporal Scale
had a total percentage of particle size smaller than 20 m by weight
Both measurement frequency and duration of the observation
in the first meter of 60 and 49%, respectively. Field water retention
period can be considered as components of the temporal scale to
was strongly related to the silt+clay content in their work. Better
define the TS. Most campaigns included in the dataset were shortstability, i.e., higher rs, was observed in the sandy loam soils than
term for the purposes of remote sensing validation studies. Several
in silt loam soil in the work of Mohanty and Skaggs (2001). Soil
studies had high temporal frequency provided by sensor networks.
type, as characterized by bulk density, clay and sand content, was
responsible for nearly 50% of spatial variability of MRD in the
Degradation of TS SWC with time was demonstrated by Rolston
work of Cosh et al. (2008). Soil thickness was shown to affect RD
et al. (1991). In their work, single sampling locations identified
and to be a factor of TS SWC by Zhu and Lin (2011).
during 1 yr gave estimates of mean storage during the following
year with some increase in error. However, use of the same sampling
Relatively low variations in soil texture and structure across study
locations for more than 2 yr increased the error in storage estimates
areas was suggested as the reason for the absence of significant difin this work. On the other hand, Schneider et al. (2008) found
ferences between MRD in different locations across the site with
time-stable locations with a low deviation from mean field SWC
volcanic soil (Comegna and Basile, 1994). However, measurements
and low standard deviation for research sites in China. Although
were done under the developing barley crop, and root activity might
the time stability characteristics of some points varied between
decrease the TS manifestation as suggested by Cassel et al. (2000).
years, the selected points were appropriate to predict mean SWC
of the sites for multiple years. The authors question the feasibility
Vegeta on
of finding temporally stable locations with low SDRD when time
Land cover and crop development can affect the TS SWC. Root
series longer than 2 yr are considered. In a large-scale study, Cosh
activity was suggested as the factor weakening the TS as measured
et al. (2006) found for half-hour SWC measurements that TS was
by the Spearman rank correlation coefficient (Kamgar et al., 1993;
maintained throughout the different seasons of a 21-mo measureCassel et al., 2000). Shallow root activity or no root activity was
ment period. In another large-scale study, Martnez-Fernndez
observed in most of the cases when the depth effect on TS was not
and Ceballos (2003) found that TS SWC was maintained, i.e.,
detected. In the work of Pachepsky et al. (2005), a similar degree
points maintain their rank in the MRD curve, throughout a 3-yr
of TS SWC was observed at different depths in a loamy soil withmonitoring period.
out vegetation. Temporal persistence was found to depend on
grazing management and the related plant cover at the field scale
TS SWC can be changed during the growing period due to the
by Schneider et al. (2008). At larger scales, land cover or vegetaeffect of root activity. Gmez-Plaza et al. (2001) used rs to show
tion do not necessarily affect the spatial distribution of soil water
content (Venkatesh et al., 2011). Existing feedbacks between TS

SWC and the TS in vegetation patterns complicate assigning a
definite cause-effect relationships between vegetation and SWC
variability (Canton et al., 2004; Ruiz-Sinoga et al., 2011). Teuling
and Troch (2005) found vegetation to increase the SWC variability during the growing season under unstressed conditions, due
to heterogeneous transpiration, but to decrease it as soon as the
SWC drops below a threshold value and transpiration becomes
supply-limited and soil-controlled. In addition, SWC variability
was found to be reduced by drainage after rainfall.
Topography
Topography as a control has been analyzed in the majority of
works on TS SWC. Results have been inconclusive partly because
different topographic descriptors along with different designs of
measurement campaigns were used in different landscapes. The
differences in topographic position were shown to impose the TS
SWC (Tomer and Anderson, 1995; Biswas and Si, 2011c). GmezPlaza et al. (2000) studied the TS SWC at three different transects
with slopes varying between 33 and 41%. They showed that topographic effects or local topography were the mean causes of TS
SWC in a sense that the bias of each location with respect to the
mean SWC value was maintained. Th is can be explained by the
fact that dry and wet positions at the top and bottom of the slope,
respectively, remain stable in time. No substantial effect of topographic variables on TS was found in relatively flat areas (Kaleita
et al., 2004).
during the growing season, especially at 0.1- and 0.4-m depths,

while during the non-growing season and at 0.8-m depth these
influences became less significant but terrain attributes became
more prominent. However, at the scale of the landform unit, topography dominated over soil properties in the steep sloping landform
units (>8% slope), regardless of soil depth; whereas soil properties
dominated over topography (especially during drier growing seasons) in relatively flat landform units (<8% slope). At the smaller
spatial scales (plot and slope transect scales), soil properties exert a
first order control on the soil moisture variation. However, in the
areas with less soil variability, the influence of terrain attributes
on SWC variation increased. The authors concluded that interplay
of terrain, soil, and crop and their impacts on soil moisture variability are complex and dynamic across the agricultural landscape,
the degree of which is a function of spatial scale, soil depth, and
season. The best time-stable features were found at mild slopes
with moderate to moderately high clay content as compared to the
field average (2830% clay) in the work of Jacobs et al. (2004). This
corresponds to the findings of Grayson and Western (1998) who
found the representative points for the average SWC to be located
near the mid-slopes and in areas that have topographic aspect close
to the catchment average.
Climate and Seasonality
Topographic effects on time stability appear to be scale dependent.

Kachanoski and de Jong (1988) demonstrated that the spatial pattern of the change in soil water storage during the recharge period,
for scales less than 40 m and especially for scales less than 30 m, is
significantly related to the spatial pattern of surface curvature. The
spatial pattern of surface curvature may therefore be responsible
for the breakdown in time stability for the recharge period at scales
less than 40 m that these authors observed. de Rosnay et al. (2009)
observed the loss of the relationship between TS SWC, i.e., MRD,
and topography at scales smaller than 80 m.
The effect of climate and seasonality on TS SWC has to our knowledge not directly been analyzed. Therefore we use SWC dynamics
here as a proxy for climate and seasonality. Earlier studies showed
better time stability for the drier sites in terms of the SDRD and
the RMSE, although this was not confirmed by Choi and Jacobs
(2007). Several studies found that TS SWC is lower during dry
periods, based on a comparison of Spearman rank correlation
for wet and dry periods (Kachanoski and de Jong, 1988; GmezPlaza et al., 2000). Martnez-Fernndez and Ceballos (2003) used
data of the REMEDHUS network to analyze the effect of wet
and dry conditions on the temporal persistence of soil moisture.
They found that dry points are more time-stable than wet points
in terms of SDRD. During wet and dry periods, similar Spearman
rank correlation coefficients between subsequent measurements
were obtained, indicating similar temporal persistence. Drying
periods typically preserve temporal persistence whereas transitional periods from consistently dry to consistently wet periods
may break it (Kachanoski and de Jong 1988). Soil moisture studies
are generally performed in areas where soils are not overwetted,
so the average of the surface is dry which gives a short interval for
moisture at the low end, and more of an interval from the average
to the saturation stage. Overall, this leads to the conclusion that
SWC itself is one of the factors controlling its TS.
The intertwined influences of scale, topography, soil properties,

and vegetation development on TS can be expected. Zhu and Lin
(2011) found that at the farm scale, both soil and terrain influenced
soil moisture variation regardless of season and soil depth. The
influences of crop and soil on soil moisture variation were observed
The effects of weather, topography, soil properties, and vegetation on TS SWC are probably intertwined and complementary.
Choi and Jacobs (2007) analyzed the effect of soil properties on
SWC variability and noted that the principal component analysis
Topography was described as a non-local control of soil water

dynamics by Grayson et al. (1997), and claimed to be the dominant
factor during wet SWC states, when lateral water flux shapes the
spatial SWC distribution. Using non-local topographic attributes
such as the upslope drainage area and the distance from drainage
channel along with local attributes, such as slope, elevation, and
aspect explained a substantial part of the variation in MRD in
work of Brocca et al. (2009).
Rela ve Role of Controls
demonstrates that rainfall and topography explain surface soil

moisture variability changes as soils dry, while soil parameters control the maximum relative variability. Overall, no clear dominant
controls can be identified that are consistent throughout literature.
Partly, the reason for that is that data are not available to do so,
time series are often limited or only few locations are available.
However, there may be a fundamental issue of control interactions that has not been so far substantially researched. The idea
that dominant SWC controls, and the way they interact, change
as SWC switches between more stable wet and dry SWC states
(Grayson et al., 1997; Teuling and Troch, 2005) might be a good
starting point to do so.
Z(u , t ) = M (u , t ) + R(u , t )
[11]
Z(u , t ) = m(u , t ) + R(u , t ) ,
[12]
The deterministic mean, m(u,t) = E{M(u,t)}, can then be further

decomposed either as
m(u , t ) = p(u ) + q(t ) ,
[13]
or as
6 Sta
s cal Modeling Temporally

Stable SWC Fields
Statistical models of TS SWC may find several applications.

Having the statistical model of the field will allow a substantially
more accurate estimate of the error in the mean SWC. Ignoring
the TS pattern results in the inclusion of the deterministic variability in these patterns into the standard error of the average
SWC. The standard error becomes then excessively high, and
this compromises evaluation of remote sensing products (Cosh et
al., 2004) and assimilation of soil water content data (Pan et al.,
2012). Statistical models of TS SWC can be used in geostatistical
simulations or stochastic imaging, providing multiple equiprobable realizations of the field (Pachepsky and Acock, 1998). These
realizations can be used to test sampling strategies and to evaluate
the scale effects on sampling results. Yet another reason to develop
statistical models of the TS SWC is to define parameters of these
models and to attempt to relate them to parameters of potential
temporal stability controls such as soil texture, land use, etc. To
improve the readability of the mathematical expressions in this
section and to acknowledge the spatial and temporal dependencies of the variables, (u, t) is used instead of the subscripts i and j.
m(u , t ) = p(u ) q(t ) ,
where p(u) and q(t) are functions of space and time, respectively.
A similar decomposition can be obtained for M(u,t) and R(u,t).
According to Kyriakidis and Journel (1999), the adoption of one
of these decompositions is a modeling decision, rather than a
data-based hypothesis, since measurements of M(u,t), p(u), and
q(t) are generally not available. They suggest basing this decision
on secondary information, using deterministic physically-based
relationships between the trend and the observations. An additive
model of particular interest with respect to TS analysis, reported
by Kyriakidis and Journel (1999), is
m(u , t ) = m + v(u ) + w(t ) ,
Spatiotemporal models have been extensively explored in geostatistical literature. Kyriakidis and Journel (1999) reviewed
geostatistical space-time models for environmental data, including SWC. They recalled that features such as time order-past,
present, and future- and isotropy are only meaningful in the time
and space domain, respectively, while distance units in space and
time are unrelated. Therefore, time cannot be considered as just
an extra dimension and models do generally not account for the
full space-time dependence. One way they suggested to represent
environmental data was by using a nonstationary spatiotemporal
random function model, Z(u, t) , which is decomposed into either
a non-stationary stochastic mean, M(u, t) , or a deterministic mean,
m(u, t) , and a zero-mean stationary residual random function component R(u, t) , which both depend on the location, u, within a
two-dimensional space A, and the moment, t, during a period of
time T:
[15]
where m is the stationary space-time mean and v(u) and w(t) are
smooth functions of u and t for which the following conditions
apply over the Ns spatial locations and the Nt observation times,
Ns
v (u ) = 0
Spa otemporal Models
[14]
and
=1
Nt
w ( t ) = 0 ,
[16]
=1
where and are indices. Further details on the estimation of M(u,

t), v(u), and w(t) can be found in Kyriakidis and Journel (1999).
Spa otemporal Model of SWC and MRD

In this section, a spatiotemporal model for TS SWC is presented,
with the aim of providing a framework to further analyze spatiotemporal SWC data sets, and identify the main controls on TS
SWC. Consider SWC as a random function, (u, t) , at location
u and time t, according to Eq. [12], with a deterministic mean
further decomposed as shown in Eq. [15]:
(u , t ) = m + v(u ) + w(t ) + R(u , t )
[17]
Here, m is the average of M(u, t) across all the observation locations,

u = 1,, Ns in A, and over all the observation times, u = 1,,
Nt during period T. v(u) is the deviation of the SWC at location
u, relative to the average across the Ns observation points at any
moment of observation and w(t) is the deviation of the average
SWC across the Ns observation points on time t from the average
SWC over all Nt observation times. Both v(u) and w(t) may have
positive and negative values and must have average values of zero
(Eq. [16]).
Assuming that the residual component, R(u,t), is zero, and combining the deterministic form of Eq. [17] with Eq. [1], the RD defined
by Vachaud et al. (1985) can be written as:
RD(u , t ) =
(u , t ) w(t ) m
v(u )
=
,
w(t ) + m
w(t ) + m
[18]
with (u, t) the SWC at location u and observation time t. Then,

according to Eq. [2], the MRD over the whole observation period,
T, is defined as
MRD(u ) =
1
T
v(u )
T RD(u , t )dt = mT T
1
dt
w(t )
1+
m
[19]
It appears that the MRD as a function of spatial coordinates

depends not only on the spatial distribution of the bias v(u) but
also on the particular soil water regime. If |w(t)| << m, i.e., average
SWC has not varied much during the observation period, then
1
SDRD(u ) =
T
12
2
2
RD(
u
,
t
)
d
t
MRD(
u
)
[23]
Deriva on of Mean and Variance of MRD(u)

By integrating Eq. [22] across the spatial domain we obtain the
mean value, MRD, for MRD(u):
v
u
(
)
1
1
1
MRD = MRD(u )du =

d
t
du
w (t )
A A
A A mT T
+
1
R (u , t )
1
1
+
dt du
A A mT T 1 + w(t )
[24]
m
1
1
dt v(u )du
=
w (t ) A
mTA T
1+
m
1
1
+
dt
R ( u , t ) dt du
w (t ) A T
mTA T
1+
m
Both terms at the right-hand side are equal to zero. Thus the mean
value, MRD, across the domain is also zero. The variance of
MRD(u), SDMRD2 is then equal to
SDMRD2 =
1
MRD(u ) 2 du .
A A
[25]
Substituting MRD(u) by Eq. [22] we obtain:
MRD(u )
v(u )
m
[20]
However, if the variations of the SWC in time were substantial

(and this has to be expected if weather conditions substantially
vary during the observation period), then MRD(u) can be both
larger than and smaller than v(u)/m depending on the shape of w(t)
over the observation period. If values of |w(t)| are predominantly
larger than 0, than MRD(u) > v(u)/m, and vice versa.
If the integral of the residual component, R(u,t), is included then
Eq. [18] and Eq. [19] become:
RD(u , t ) =
MRD(u ) =
=
v(u ) + R(u , t )
w (t ) + m
1
T
[21]
T RD(u , t )dt
v(u )
1
1
dt +
mT T 1 + w(t )
mT
m
R (u , t )
dt .
w(t )
1+
m
[22]
SDMRD2 =
1
1
1
v(u )
dt +
A
T
(
)
w
t
A mT
mT
1+
or also
2
v ( u ) + R ( u , t ) dt 2
1
1
2
SDMRD =
dt
du [27]
T
A
w (t )
A
mT
1+
Th is shows that the variance of MRD(u) is determined by a temporal and a spatial component.
If the residual term, R(u,t) is small with respect to v(u), and |w(t)|
<< m we obtain:
v (u ) 2
1
du
SDMRD =
A A m
2
The SDRD(u) can be obtained as:
2
R (u , t )
T w(t ) dt du [26]
1+
[28]
This means that if the average SWC remains rather constant during
the observation period, the variance is mainly determined by the
spatial variation of SWC. If the SWC variation is large during
the observation period, the term w(t) influences the variance of
MRD(u). In case that only the residual term is small we find:
v (u ) 2
1
1
du
SDMRD2 =
dt
w (t ) A mT
A T
1+
[29]
The model Eq. [17] is additive as the spatial and temporal effects
are separated. Multiplicative models, of the type shown in Eq. [14]
can be developed in which the deterministic component in SWC
can be obtained by multiplication of the same spatial pattern by
the time dependent component. The EOF modeling of the TS
SWC presents one example of multiplicative models (Perry and
Niemann, 2007; Korres et al., 2010).
Es ma ng Determinis c Spa al and Temporal

Components
The estimation is straightforward:
m=
1
Ns Nt
v(u ) =
w(t ) =
1
Nt
Ns Nt
(u , t )
[30]
=1=1
Nt
(u , t ) m
[31]
1 Ns
(u , t ) m
N s =1
[32]
=1
Quantitative evaluation of statistical models of TS SWC can be

done in two ways. Part of the available data can be used to parameterize the model, i.e., to fi nd v(u) and the SDRD(u), while the
remaining part is used to test the model. On the other hand, the
statistical behavior of predicted and measured SWC can be compared. It has to be recalled that the model in Eq. [17] does not
predict the spatial variability of SWC as a function of its average,
for which typical concave relationships have been found (Penna
et al., 2009).
The proposed model can explain the implications of rs in the relationship between the MRD distribution and spatial distributions
of SWC. If the correlation is strong then the spatial distribution
of SWC has a strong deterministic component, which in case of
the above discussed model, is approximated by the function v(u) in
Eq. [17]. If MRD follows Eq. [22] and MRD can be approximated
with the Gaussian model, then v(u) is also approximately Gaussian
and SWC will have the spatial distributions close to normal at any
observation time. If the Spearman rank correlations are low then
the random values of SWC and RD can be considered as independent random variables in space. If RD is normally distributed in
space at each measurement time, then the central limit theorem

(Pollard, 2002) predicts that MRD as the sum of a large number of
normally distributed random values with zero means will approximate a Gaussian distribution having a zero mean. The converse
statement, i.e., concluding that SWC are distributed normally
because MRD are distributed normally, follows from the Cramrs
decomposition theorem (Pollard, 2002). This theorem states that
if the RD distributions are independent and MRD has normal
distribution then RD have to be also normal.
6 Conclusions and Future
Research Avenues
A review of published work shows that TS SWC has been observed

under a wide range of conditions, from the field-plot to the basin
scale, for measurement periods of a few days to several years, under
a wide range of terrain, soil and vegetation types, and using different SWC measurement methods and spatial designs. The spatial
variability of MRD generally increased with increasing extent
of the study area, although some deviations of that trend were
observed due to the extremely heterogeneous nature of the data
set. While finding time-stable locations for monitoring continues
to be the important application of TS SWC, the question whether
the most representative points for estimating the spatial average
SWC are maintained in space as scale increases or under which
conditions this might happen, remains unanswered. In addition,
research is now needed to improve the rapid identification of such
sites given limited data.
Seven potential key factors in controlling TS SWC that were
identified can be classified in three groups, according to measurement strategy, terrain, and climate. Generally there exists much
contradicting evidence and a lack of information on potentially
controlling factors such as the spatial variability of soil hydraulic
properties and the homogeneity or heterogeneity of the boundary
forcings (rainfall and transpiration) during the measured periods.
However, published results suggest the occurrence of combined
effects of controlling factors rather than single factors dominating TS SWC. Modeling appears to be a logical step to proceed to
research the interplay of the TS SWC controls. In addition, additive or multiplicative statistical models for TS SWC, such as the
one proposed here, can be used to untangle controls on the spatial
and temporal components of TS SWC. Also, a further exploration
of the conditions under which the Gaussian model for the distribution of MRD is valid, is expected to contribute to this.
An interesting application of TS SWC arises when controls by
soil properties such as texture and structure, can be identified or
isolated. Under such circumstances, information on TS SWC, provided by sensor networks or remote sensing, could be used to infer
the spatial variations of those soil properties.
Overall, the results of this review call for a focused research effort
directed towards identifying the interactions and effects of measurement design, topography, soil, vegetation and climate on TS
SWC. More detailed studies are needed to identify the effect of
local and non-local controls on TS SWC. Soil moisture networks
providing spatially and temporally highly resolved soil moisture
data should be combined with detailed on-site characterization.
This may contribute to identifying the major controls and to study
the temporal dynamics of the rank stability curves in more detail.
A matter related to this is the development of better methods for
defining the spatial properties of soils and vegetation as they affect
the TS SWC.
Acknowledgments
This research was made possible thanks to funding provided by the Spanish Ministry
of Science and Innovation and FEDER through grant AGL2009-12936-C03-03 and
by the Junta de Andaluca though grant AGR-4782. Also mobility grants PR20100191 and of EX2009-0433 the Spanish Ministry of Education are acknowledged.
References
Biswas, A., and B.C. Si. 2011a. Depth persistence of the spa al pa ern of soil
water storage in a Hummocky landscape. Soil Sci. Soc. Am. J. 75:10991109.
doi:10.2136/sssaj2010.0399
Biswas, A., and B.C. Si. 2011b. Scales and loca ons of me stability of soil water
storage in a hummocky landscape. J. Hydrol. 408:100112. doi:10.1016/j.
jhydrol.2011.07.027
Biswas, A., and B.C. Si. 2011c. Iden fying scale specific controls of soil water
storage in a hummocky landscape using wavelet coherency. Geoderma
165:5059. doi:10.1016/j.geoderma.2011.07.002
Bosch, D., V. Lakshmi, T. Jackson, M. Choi, and J. Jacobs. 2006. Large scale measurements of soil moisture for valida on of remotely sensed data: Georgia
soil moisture experiment of 2003. J. Hydrol. 323:120137. doi:10.1016/j.
jhydrol.2005.08.024
Brocca, L., F. Melone, T. Moramarco, and R. Morbidelli. 2009. Soil moisture
temporal stability over experimental areas in Central Italy. Geoderma
148:364374. doi:10.1016/j.geoderma.2008.11.004
Brocca, L., F. Melone, T. Moramarco, and R. Morbidelli. 2010. Spa al-temporal
variability of soil moisture and its es ma on across scales. Water Resour.
Res. 46:W02516. doi:10.1029/2009WR008016
Canton, Y., A. Solebenet, and F. Domingo. 2004. Temporal and spa al pa erns
of soil moisture in semiarid badlands of SE Spain. J. Hydrol. 285:199214.
doi:10.1016/j.jhydrol.2003.08.018
Cassel, D.K., O. Wendroth, and D.R. Nielsen. 2000. Assessing spa al variability
in an agricultural experiment sta on field: Opportuni es arising from spaal dependence. Agron. J. 92:706714. doi:10.2134/agronj2000.924706x
Chen, Y. 2006. Le er to the editor on rank stability or temporal stability. Soil
Sci. Soc. Am. J. 70:306. doi:10.2136/sssaj2005.0290l
Choi, M., and J.M. Jacobs. 2007. Soil moisture variability of root zone profiles
within SMEX02 remote sensing footprints. Adv. Water Resour. 30:883896.
doi:10.1016/j.advwatres.2006.07.007
Choi, M., and J.M. Jacobs. 2011. Spa al soil moisture scaling structure during soil moisture experiment 2005. Hydrol. Processes 25(6):926932.
doi:10.1002/hyp.7877
Comegna, V., and A. Basile. 1994. Temporal stability of spa al pa erns of
soil water storage in a cul vated Vesuvian soil. Geoderma 62:299310.
doi:10.1016/0016-7061(94)90042-6
Coppola, A., A. Comegna, G. Dragone , N. Lamaddalena, A. Kader, and V.
Comegna. 2011. Average moisture satura on eects on temporal stability
of soil water spa al distribu on at field scale. Soil Tillage Res. 114:155164.
doi:10.1016/j.s ll.2011.04.009
Cosh, M.H., T.J. Jackson, R. Bindlish, and J.H. Prueger. 2004. Watershed scale
temporal and spa al stability of soil moisture and its role in valida ng
satellite es mates. Remote Sens. Environ. 92:427435. doi:10.1016/j.
rse.2004.02.016
Cosh, M.H., T.J. Jackson, S. Moran, and R. Bindlish. 2008. Temporal persistence
and stability of surface soil moisture in a semi-arid watershed. Remote
Sens. Environ. 112:304313. doi:10.1016/j.rse.2007.07.001
Cosh, M., T. Jackson, P. Starks, and G. Heathman. 2006. Temporal stability of
surface soil moisture in the Li le Washita River watershed and its applicaons in satellite soil moisture product valida on. J. Hydrol. 323:168177.
doi:10.1016/j.jhydrol.2005.08.020
da Silva, A.P., A. Nadler, and B. Kay. 2001. Factors contribu ng to temporal stability in spa al pa erns of water content in the llage zone. Soil Tillage Res.
58:207218. doi:10.1016/S0167-1987(00)00169-0
De Lannoy, G.J.M., P.R. Houser, N.E.C. Verhoest, V.R.N. Pauwels, and T.J. Gish.
2007. Upscaling of point soil moisture measurements to field averages at
the OPE3 test site. J. Hydrol. 343:111. doi:10.1016/j.jhydrol.2007.06.004
de Rosnay, P., C. Gruhier, F. Timouk, F. Baup, E. Mougin, P. Hiernaux, L. Kergoat, and V. LeDantec. 2009. Mul -scale soil moisture measurements at
the Gourma meso-scale site in Mali. J. Hydrol. 375:241252. doi:10.1016/j.
jhydrol.2009.01.015
Dumedah, G., and P. Coulibaly. 2011. Evalua on of sta s cal methods for infilling missing values in high-resolu on soil moisture data. J. Hydrol. 400:95
102. doi:10.1016/j.jhydrol.2011.01.028
Famiglie , J.S., D. Ryu, A.A. Berg, M. Rodell, and T.J. Jackson. 2008. Field observa ons of soil moisture variability across scales. Water Resour. Res. 44:116.
Gmez-Plaza, A., J. Alvarez-Rogel, J. Albaladejo, and V.M. Cas llo. 2000. Spa al
pa erns and temporal stability of soil moisture across a range of scales
in a semi-arid environment. Hydrol. Processes 14:12611277. doi:10.1002/
(SICI)1099-1085(200005)14:7<1261::AID-HYP40>3.0.CO;2-D
Gmez-Plaza, A., M. Mar nez-Mena, J. Albaladejo, and V. Cas llo. 2001. Factors regula ng spa al distribu on of soil water content in small semiarid
catchments. J. Hydrol. 253:211226. doi:10.1016/S0022-1694(01)00483-8
Grant, L., M. Seyfried, and J. McNamara. 2004. Spa al varia on and temporal
stability of soil water in a snow-dominated, mountain catchment. Hydrol.
Processes 18:34933511. doi:10.1002/hyp.5798
Grayson, R.B., and A.W. Western. 1998. Towards areal es ma on of soil water
content from point measurements: Time and space stability of mean response. J. Hydrol. 207:6882. doi:10.1016/S0022-1694(98)00096-1
Grayson, R.B., A.W. Western, F.H.S. Chiew, and G. Blschl. 1997. Preferred
states in spa al soil moisture pa erns: Local and nonlocal controls. Water
Resour. Res. 33:28972908. doi:10.1029/97WR02174
Guber, A.K., T.J. Gish, Y.A. Pachepsky, M.T. van Genuchten, C.S.T. Daughtry, T.J.
Nicholson, and R.E. Cady. 2008. Temporal stability in soil water content
pa erns across agricultural fields. Catena 73:125133. doi:10.1016/j.catena.2007.09.010
Heathman, G.C., M. Larose, M.H. Cosh, and R. Bindlish. 2009. Surface and profile soil moisture spa o-temporal analysis during an excessive rainfall period in the Southern Great Plains, USA. Catena 78:159169. doi:10.1016/j.
catena.2009.04.002
Herbst, M., N. Prolingheuer, A. Graf, J.A. Huisman, L. Weihermller, and J.
Vanderborght. 2009. Characteriza on and understanding of bare soil
respira on spa al variability at plot scale. Vadose Zone J. 8:762771.
doi:10.2136/vzj2008.0068
Hu, W., M.A. Shao, F.P. Han, K. Reichardt, and J. Tan. 2010a. Watershed
scale temporal stability of soil water content. Geoderma 158:181198.
doi:10.1016/j.geoderma.2010.04.030
Hu, W., M.A. Shao, and K. Reichardt. 2010b. Using a new criterion to iden fy
sites for mean soil water storage evalua on. Soil Sci. Soc. Am. J. 74:762
773. doi:10.2136/sssaj2009.0235
Hu, W., M. Shao, Q. Wang, and K. Reichardt. 2009. Time stability of soil water storage measured by neutron probe and the eects of calibra on
procedures in a small watershed. Catena 79:7282. doi:10.1016/j.catena.2009.05.012
Ibrahim, H.M., and D.R. Huggins. 2011. Spa o-temporal pa erns of soil water storage under dryland agriculture at the watershed scale. J. Hydrol.
404:186197. doi:10.1016/j.jhydrol.2011.04.029
Jacobs, J.M., B.P. Mohanty, E. Hsu, and D. Miller. 2004. SMEX02: Field scale variability, me stability and similarity of soil moisture. Remote Sens. Environ.
92:436446. doi:10.1016/j.rse.2004.02.017
Jawson, S.D., and J.D. Niemann. 2007. Spa al pa erns from EOF analysis of
soil moisture at a large scale and their dependence on soil, land-use, and
topographic proper es. Adv. Water Resour. 30:366381. doi:10.1016/j.advwatres.2006.05.006
Joshi, C., B.P. Mohanty, J.M. Jacobs, and A.V.M. Ines. 2011. Spa otemporal analyses of soil moisture from point to footprint scale in two dierent hydroclima c regions. Water Resour. Res. 47:W01508. doi:10.1029/2009WR009002
Kachanoski, R.G., and E. de Jong. 1988. Scale dependence and the temporal
persistence of spa al pa erns of soil water storage. Water Resour. Res.
24:8591. doi:10.1029/WR024i001p00085
Kaleita, A.L., L.F. Tian, and M.C. Hirschi. 2004. Iden fica on of Op mal Sampling Loca ons and Grid Size for Soil Moisture Mapping. ASAE Paper No.
041146. h p://asae.frymul .com/azdez.asp?JID=5&AID=16171&CID=can2
004&T=2 (accessed 13 Sept. 2012).
Kamgar, A., J.W. Hopmans, W.W. Wallender, and O. Wendroth. 1993. Plot size
and sample number for neutron probe measurements in small field trials.
Soil Sci. 156:213224. doi:10.1097/00010694-199310000-00001
Kirda, C., and K. Reichardt. 1992. Comparison of neutron moisture gauges with
nonnuclear methods to measure field soil water status. Sci. Agric. (Piracicaba, Braz.)
49:111121. doi:10.1590/S0103-90161992000400015
Korres, W., C.N. Koyama, P. Fiener, and K. Schneider. 2010. Analysis of surface
soil moisture pa erns in agricultural landscapes using Empirical Orthogo-
nal Func ons. Hydrol. Earth Syst. Sci. 14:751764. doi:10.5194/hess-14751-2010

Kyriakidis, P.C., and A.G. Journel. 1999. Geosta s cal space- me models: A review. Math. Geol. 31:651684. doi:10.1023/A:1007528426688
Lin, H. 2006. Temporal stability of soil moisture spa al pa ern and subsurface
preferen al flow pathways in the shale hills catchment. Vadose Zone J.
5:317340. doi:10.2136/vzj2005.0058
Manfreda, S., and I. Rodrguez-Iturbe. 2006. On the spa al and temporal sampling of soil moisture fields. Water Resour. Res. 42:110.
doi:10.1029/2005WR004548
Mar nez, G., K. Vanderlinden, J.V. Giraldez, A.J. Espejo, and J.L. Muriel. 2010.
Field-scale soil moisture pa ern mapping using electromagne c induc on.
Vadose Zone J. 9:871881. doi:10.2136/vzj2009.0160
Mar nez-Fernndez, J., and A. Ceballos. 2003. Temporal stability of soil moisture in a large-field experiment in Spain. Soil Sci. Soc. Am. J. 67:16471656.
doi:10.2136/sssaj2003.1647
Mar nez-Fernndez, J., and A. Ceballos. 2005. Mean soil moisture es ma on
using temporal stability analysis. J. Hydrol. 312:2838. doi:10.1016/j.jhydrol.2005.02.007
Minet, J., E. Laloy, S. Lambot, and M. Vanclooster. 2011. Eect of high-resolu on spa al soil moisture variability on simulated runo response using a distributed hydrologic model. Hydrol. Earth Syst. Sci. 15:13231338.
doi:10.5194/hess-15-1323-2011
Mohanty, B.P., and T.H. Skaggs. 2001. Spa o-temporal evolu on and mestable characteris cs of soil moisture within remote sensing footprints
with varying soil, slope, and vegeta on. Adv. Water Resour. 24:10511067.
doi:10.1016/S0309-1708(01)00034-3
Pachepsky, Ya., and B. Acock. 1998. Stochas c imaging of soil parameters to assess variability and uncertainty of crop yield es mates. Geoderma 85:213
229. doi:10.1016/S0016-7061(98)00021-4
Pachepsky, Ya., A. Guber, and D. Jacques. 2005. Temporal persistence in ver cal distribu ons of soil moisture contents. Soil Sci. Soc. Am. J. 69:347352.
doi:10.2136/sssaj2005.0347
Pan, F., Ya. Pachepsky, D. Jacques, A. Guber, and R. Hill. 2012. Data assimilaon with soil water content sensors and pedotransfer func ons in soil
water flow modeling. Soil Sci. Soc. Am. J. 76::829844. doi:10.2136/sssaj2011.0090.
Penna, D., M. Borga, D. Norbiato, and G. Dallafontana. 2009. Hillslope scale
soil moisture variability in a steep alpine terrain. J. Hydrol. 364:311327.
doi:10.1016/j.jhydrol.2008.11.009
Perry, M., and J. Niemann. 2007. Analysis and es ma on of soil moisture at
the catchment scale using EOFs. J. Hydrol. 334:388404. doi:10.1016/j.jhydrol.2006.10.014
Petrone, R.M., J.S. Price, S.K. Carey, and J.M. Waddington. 2004. Sta s cal
characteriza on of the spa al variability of soil moisture in a cutover peatland. Hydrol. Processes 18:4152. doi:10.1002/hyp.1309
Pollard, D. 2002. A users guide to measure theore c probability. Cambridge
Univ. Press, Cambridge.
Reichardt, K., O. Bacchi, M. Villagra, A.L. Tura , and Z.O. Pedrosa. 1993. Hydraulic variability in space and me in a dark red latosol of the tropics. Geoderma 60:159168. doi:10.1016/0016-7061(93)90024-F
Reichardt, K., O. Portezan, O. Bacchi, J. Oliveira, D. Dourado-Neto, J. Pilo o, and
M. Calvache. 1997. Neutron probe calibra on correc on by temporal stability parameters of soil water content probability distribu on. Sci. Agric.
(Piracicaba, Braz.) 54:1721. doi:10.1590/S0103-90161997000300003
Rocha, G.C., P.L. Libardi, L. Alves de Carvalho, and A.C. Rodrigues Cruz. 2005.
Temporal stability of the spa al distribu on of water storage in a soil under citrus cul va on. Rev. Bras. Cienc. Solo 29:4150. doi:10.1590/S010006832005000100005
Rolston, D.E., J.W. Biggar, and H.I. Nigh ngale. 1991. Temporal persistence of
spa al soil-water pa erns under trickle irriga on. Irrig. Sci. 12:181186.
doi:10.1007/BF00190521
Ruiz-Sinoga, J.D., J.F. Mar nez-Murillo, M.A. Gabarrn-Galeote, and R. GarcaMarn. 2011. The eects of soil moisture variability on the vegeta on pattern in Mediterranean abandoned fields (Southern Spain). Catena 85:111.
doi:10.1016/j.catena.2010.11.004
Schneider, K., J. Huisman, L. Breuer, Y. Zhao, and H. Frede. 2008. Temporal
stability of soil moisture in various semi-arid steppe ecosystems and its
applica on in remote sensing. J. Hydrol. 359:1629. doi:10.1016/j.jhydrol.2008.06.016
Starks, P., G. Heathman, T. Jackson, and M. Cosh. 2006. Temporal stability of soil moisture profile. J. Hydrol. 324:400411. doi:10.1016/j.jhydrol.2005.09.024
Starr, G. 2005. Assessing temporal stability and spa al variability of soil water
pa erns with implica ons for precision water management. Agric. Water
Manage. 72:223243. doi:10.1016/j.agwat.2004.09.020
Tallon, L., and B.C. Si. 2003. Representa ve soil water benchmarking for environmental monitoring. J. Environ. Inf. Arch. 1:581590.
Teuling, A.J., and P.A. Troch. 2005. Improved understanding of soil moisture
variability dynamics. Geophys. Res. Le . 32:710.
Teuling, A.J., R. Uijlenhoet, F. Hupet, E.E. van Loon, and P.A. Troch. 2006. Es ma ng spa al mean root-zone soil moisture from point-scale observa ons.
Hydrol. Earth Syst. Sci. 10:755767. doi:10.5194/hess-10-755-2006
Thierfelder, T.K., R.B. Grayson, D. van Rosen, and A.W. Western. 2003. Inferring the loca on of catchment characteris cs soil moisture monitoring
sites. Covariance structure in the temporal domain. J. Hydrol. 280:1332.
doi:10.1016/S0022-1694(03)00077-5
Tomer, M.D., and J.L. Anderson. 1995. Varia on of soil water storage across
a sand plain hill slope. Soil Sci. Soc. Am. J. 59:10911100. doi:10.2136/
sssaj1995.03615995005900040021x
Vachaud, G., A. Passerat De Silans, P. Balabanis, and M. Vauclin. 1985. Temporal
stability of spa ally measured soil water probability density func on. Soil Sci.
Soc. Am. J. 49:822828. doi:10.2136/sssaj1985.03615995004900040006x
Venkatesh, B., N. Lakshman, B.K. Purandara, and V.B. Reddy. 2011. Analysis
of observed soil moisture pa erns under dierent land covers in Western
Ghats, India. J. Hydrol. 397:281294. doi:10.1016/j.jhydrol.2010.12.006
Yoo, C., and S. Kim. 2004. EOF analysis of surface soil moisture field variability.
Adv. Water Resour. 27:831842. doi:10.1016/j.advwatres.2004.04.003
Zehe, E., T. Grae, M. Morgner, A. Bauer, and A. Bronstert. 2010. Plot and field
scale soil moisture dynamics and subsurface wetness control on runo
genera on in a headwater in the Ore Mountains. Hydrol. Earth Syst. Sci.
14:873889. doi:10.5194/hess-14-873-2010
Zhu, Q., and H. Lin. 2011. Influences of soil, terrain, and crop growth on soil
moisture varia on from transect to farm scales. Geoderma 163:4554.
doi:10.1016/j.geoderma.2011.03.015

Temporal Stability of Soil PDF

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Temporal Stability of Soil PDF

Uploaded by

Copyright:

Available Formats

Review and Analysis

Soil Science Society of America

All rights reserved. No part of this periodical may

Temporal Stability of Soil

Two groups of methods have been proposed in literature to

Mean Rela ve Dierences

The MRD for the location i becomes then

where Nt is the number of observation times. The standard

The value of SDRDi serves as one of the measures of the TS

j throughout time. To simplify notations, the subscript i will

Pa ern Analysis with EOF

Temporal Persistence: Regression and

A close linear regression relationship between ij1 and ij2 was

to analyze the scale-dependent persistence of SWC across depth

Spearman Rank Correla on

6 i=1 s [ R(i , j1 ) R(i , j2 )]

with Ns the number of spatial observation locations, i = 1,2,,

Representa ve Loca ons

RMSE i = MRDi 2 + SDRD2i

The location with the minimum value of MABE was suggested to

Gaussian Approxima on of the Distribu on

could be fitted to the experimental dependencies of MRD on i

and the sample standard deviation,

respectively. Here, Ns is the number of locations and sample mean

6 Experimental Data on TS SWC

Data Set and Descrip ve Sta s cs

average separation distance between sampling locations, in case

Mean relative difference and standard deviation of relative difference

MRD: spatial mean of MRD, SDMRD: standard deviation of MRD,

Avg. across Loc. 1 to 3

Partly absent measurements (marked as ND)

Figure 2a shows that non-zero MRD are

The accuracy of fitting Eq. [8] to

Dispersion in the overall linear relationship between sample and fitted

of positive extreme MRD values did not

Spearman Rank and

Pearson correlation coefficients were used (Lin, 2006; Cosh et

Factors Aec ng TS SWC

considered representative for situations with high, intermediate,

In the complete data set, the Theta-probe was used in 45 cases,

Fig. 7. Distribution of the sample standard deviation of the mean

(1992). They compared rs derived from SWC measurements using

Depth and Layer Thickness

measurements generally show a larger SDMRD than values

Fig. 9. Relationship between the sample standard deviation of the

an increasing extent of the study area, additional factors of spatial

data from Choi and Jacobs (2011), measured at 14 moments in

Walnut Creek Watershed

obtained from the other data sets, as a consequence of

content (Venkatesh et al., 2011). Existing feedbacks between TS

during the growing season, especially at 0.1- and 0.4-m depths,

Climate and Seasonality

Topographic effects on time stability appear to be scale dependent.

The intertwined influences of scale, topography, soil properties,

Topography was described as a non-local control of soil water

Rela ve Role of Controls

demonstrates that rainfall and topography explain surface soil

Z(u , t ) = m(u , t ) + R(u , t ) ,

The deterministic mean, m(u,t) = E{M(u,t)}, can then be further

s cal Modeling Temporally

Statistical models of TS SWC may find several applications.

m(u , t ) = p(u ) q(t ) ,