Sign-Regressor Adaptive Filtering Algorithms Using Averaged Iterates and Observations
Functional relationship to describe temporal statistics of soil moisture averaged over different...
-
Upload
independent -
Category
Documents
-
view
1 -
download
0
Transcript of Functional relationship to describe temporal statistics of soil moisture averaged over different...
Advances in Water Resources 28 (2005) 553–566
www.elsevier.com/locate/advwatres
Functional relationship to describe temporal statistics ofsoil moisture averaged over different depths
Michael J. Puma a,*, Michael A. Celia a, Ignacio Rodriguez-Iturbe a, Andrew J. Guswa b
a Department of Civil and Environmental Engineering, Princeton University, Princeton, NJ 08544, USAb Picker Engineering Program, Smith College, Northampton, MA 01063, USA
Received 4 March 2004; received in revised form 28 July 2004; accepted 5 August 2004
Available online 21 January 2005
Abstract
Detailed simulation studies, highly resolved in space and time, show that a physical relationship exists among instantaneous soil-
moisture values integrated over different soil depths. This dynamic relationship evolves in time as a function of the hydrologic inputs
and soil and vegetation characteristics. When depth-averaged soil moisture is sampled at a low temporal frequency, the structure of
the relationship breaks down and becomes undetectable. Statistical measures can overcome the limitation of sampling frequency,
and predictions of mean and variance for soil moisture can be defined over any soil averaging depth d. For a water-limited ecosys-
tem, a detailed simulation model is used to compute the mean and variance of soil moisture for different averaging depths over a
number of growing seasons. We present a framework that predicts the mean of soil moisture as a function of averaging depth given
soil moisture over a shallow d and the average daily rainfall reaching the soil.
� 2004 Elsevier Ltd. All rights reserved.
Keywords: Soil moisture; Remote sensing; Richards equation; Root-water uptake; Plant model; Modeling
1. Introduction
Soil moisture is the critical variable that dynamically
links plants to the overall water balance, thereby influ-
encing feedbacks to the atmosphere [33]. Soil moisture
is controlled by complex interactions involving soil,
plants, and climate. Plants connect the soil to the atmo-
sphere through their active roots, which provide path-
ways for water transport from the root zone to the
atmosphere [9]. Therefore, knowledge of soil moisturewithin the root zone, the region where active roots reside
in the soil, is essential for estimation of fundamental
hydrological and atmospheric processes. Accurate esti-
mation of these processes is important for large-scale cli-
mate models as well as for ecohydrological models.
0309-1708/$ - see front matter � 2004 Elsevier Ltd. All rights reserved.
doi:10.1016/j.advwatres.2004.08.015
* Corresponding author. Tel.: +1 609 258 7819; fax: +1 609 258
2799.
E-mail address: [email protected] (M.J. Puma).
Soil moisture can be estimated by in situ measure-
ments, by remote sensing, or by hydrological modeling.For large-scale applications, in situ methods cannot be
used because an in situ measurement network does not
exist over large land surface areas and the technique is
expensive [16]. Microwave remote sensors have been
successful because they are sensitive to soil moisture
through the effects of moisture on the dielectric constant
and, consequently, the soil�s emissivity [30,23]. However,
remote sensing has uncertainty that depends on the sen-sor type (active or passive), vegetation cover, landscape
roughness, and soil type. Yet the primary shortcoming
of remote sensing is that soil moisture is inferred only
for the top few centimeters of the soil column e.g.,
[8,22]. Consequently, remote sensing must be used in
conjunction with some other information to estimate
soil-moisture values over the entire root zone.
Hydrological models often have uncertainpredictions that result from model assumptions and
554 M.J. Puma et al. / Advances in Water Resources 28 (2005) 553–566
parameterization [16]. Kostov and Jackson [19] sug-
gested that combining remotely sensed data with
hydrological models is the most promising approach
for soil-moisture estimation. Accordingly, recent re-
search has focused on this combination e.g., [28,6,16,
27,14] with a general strategy of guiding hydrologicalmodels with periodic remote sensing of soil moisture.
The instantaneous soil-moisture data inferred from
remote sensing observations are assimilated into hydro-
logical models with the hope that model biases will be
corrected. Fundamental to this strategy is the relation-
ship between the measured instantaneous soil moisture
and instantaneous soil moisture at deeper locations.
This paper first investigates the relationship betweeninstantaneous values of soil moisture over different aver-
aging depths, d, using a detailed simulation model with a
focus on temporal resolution of measurements. This
instantaneous soil-moisture analysis illustrates the diffi-
culties in relating instantaneous soil-moisture values
averaged over the top 5 cm to the instantaneous soil-
moisture values averaged over the top 30 cm. We then
consider an alternative approach to relating instanta-neous soil-moisture values over different averaging
depths using statistical measures of soil moisture. In par-
ticular, we investigate statistical soil-moisture measures
as functions of depth, d, focusing on the relationships
among the soil-moisture means over different d as well
as the variances over different d. Based on this approach,
a methodology is presented that enables prediction of
the mean of soil moisture as a function of averagingdepth, d, as a function of soil and climate parameters.
2. Simulation description
A one-dimensional model, based on the Richards
equation, in combination with a model of water uptake
by plants and stochastically generated rainfall, is used tosimulate soil-moisture dynamics in a water-limited eco-
system, which we take to be a savannah. Assumptions
have been made regarding the model�s resolution and
complexity with some processes simplified. For example,
plant growth and nutrient uptake are not modeled,
because we assume that these processes are not impor-
tant to estimate soil moisture for a mature plant in a
savanna. We take the output from our highly resolved(in space and time) model to represent actual field
conditions.
Data for the model correspond to measurements ta-
ken at Nylsvley, South Africa [34]. Since this model only
considers the vertical spatial dimension of the soil, an
inherent assumption is that only vertical soil-moisture
dynamics are important, and, therefore, lateral soil-
moisture dynamics do not have to be resolved. The rootzone of the vegetation (typically 30–100 cm) is resolved
into 1 cm layers. We use highly resolved time discretiza-
tion, with a maximum time step of approximately 3 min,
and finer resolution around storm events. The model
represents water uptake by the plant using the so-called
�Type I� model [15,13], in which water uptake is con-
trolled by differences in fluid potential between the soil
and the plant [1,38,24,12,13,15,17]. The model may beinterpreted as representative of either a single plant or
a homogeneous stand of vegetation.
2.1. Infiltration and redistribution
We model vertical infiltration and redistribution,
including evapotranspiration, using the one-dimensional
Richards� equation with suitable sink terms to accountfor evapotranspiration. The governing equation has
the following form,
oð/SÞot
� o
ozKoWoz
� �þ oK
oz¼ �e0 � u0 ð1Þ
where S is the relative soil-moisture content or satura-
tion (L3 water/L3 voids), / is the porosity (L3 voids/L3
soil), K is the unsaturated hydraulic conductivity (L/
T), W is the fluid pressure head (L), e 0 is the rate of evap-
oration (L3 evaporated water/L3 soil/T), u 0 is the rate of
plant uptake (L3 plant-extracted water/L3 soil/T), z isthe vertical dimension designated to be positive down-
ward (L), and t is time (T) e.g., [3,4,37]. In unsaturated
soil, the fluid pressure head, W, is negative and is often
referred to as the soil matrix potential. Suction is defined
for unsaturated soil as the absolute value of W.
2.2. Rainfall model and boundary conditions
Rainfall input is treated as an external random forc-
ing that is independent of soil moisture. The rainfall
input is modeled with the storm occurrence, depth,
and duration represented as random variables. Storm
occurrence is modeled as a Poisson process with rate kwith a duration and intensity associated with each
occurrence. The duration is obtained from a beta distri-
bution. The total storm rainfall depth is taken from anexponential distribution with mean rainfall depth a.
The aboveground portion of plants intercepts a sig-
nificant amount of rainfall, especially in arid and semi-
arid ecosystems where rainfall duration is short and
evaporation demand is high e.g., [21,7]. Following the
simplified approach of Laio et al. [21], interception is
modeled by setting a fixed threshold rainfall depth, D.If a simulated storm produces a total storm rainfalldepth less than D, then no rain reaches the soil due to
interception. For storm depths greater than D, the depthof rainfall reaching the soil surface is simply the total
storm rainfall depth minus D. The rate of arrival of
storms with rainfall reaching the soil surface, k 0, be-
comes [32,21]
M.J. Puma et al. / Advances in Water Resources 28 (2005) 553–566 555
k0 ¼ ke�D=a ð2ÞSince both the evaporation and transpiration functions
do not depend on whether it is raining, k is important
in this model only as it relates to k 0.
Two types of boundary conditions, fixed pressure
head and flux, are imposed at the top of the soil column
in this model. The flux is set to zero during periods whenno precipitation reaches the soil. When a storm with rain
reaching the soil surface occurs, the top boundary condi-
tion is changed to a non-zero flux condition. This flux
boundary condition is equal to the rainfall rate at the soil
surface, as long as the surface remains unsaturated. If the
top boundary becomes fully saturated, then the bound-
ary condition switches to a pressure head condition and
remains so as long as full-saturation or ponding condi-tions exist at the surface [12]. The bottom boundary con-
dition for the soil is a fixed pressure head condition. The
location of the bottom boundary of the soil is set to a
depth where the bottom boundary condition has minimal
impact on soil-moisture dynamics in the root zone [12].
2.3. Discretization and solution
The governing equation, Eq. (1), is highly non-linear,
because it must be combined with its constitutive rela-
tionships for the soil�s hydraulic properties, evaporation,and transpiration. Consequently, Eq. (1) must be solved
using numerical methods. The soil column is discretized
into layers, or grid cells, with Dzi denoting the thickness
of layer i. Following the procedure of Celia et al. [3], a
backward Euler approximation in time with a modifiedPicard iteration scheme is applied to the governing equa-
tion, using uniform grid cell sizes. In the simulations,
evaporation and transpiration are calculated explicitly
in time, and unsaturated conductivities are determined
using upstream weighting.
2.4. Hydraulic properties
An empirically derived constitutive equation for the
soil–water retention curves approximates the relation-
ship between W and S. The Brooks and Corey formula-
tion [31] is used here,
WðSÞ ¼ We
S � Sh
1� Sh
� ��b
ð3Þ
where We is the air-entry pressure head (L), Sh is the
hygroscopic saturation, and b is a parameter that con-
trols the curve�s shape. We use a hydraulic conductivity
function that is also based on the formulation by Brooks
and Corey [31],
KðSÞ ¼ Ksat
S � Sh
1� Sh
� �2bþ3
ð4Þ
where Ksat is the saturated hydraulic conductivity (L/T).
2.5. Evaporation and transpiration
The sink terms in Eq. (1) are represented empirically
and account for water loss from the soil column due to
evaporation and transpiration. These two processes are
governed by the atmospheric demand, the amount ofwater in the soil, and the characteristics of the vegeta-
tion. We model a completely vegetated surface. There-
fore, transpiration is dominant relative to evaporation,
so we assume that evaporation accounts for 10% of total
evapotranspiration. Consequently, we model evapora-
tion in a simplistic manner relative to evaporation mod-
els that take into account heat and moisture transport,
but this simplification is justified given the dominanceof transpiration.
2.5.1. Evaporation
In our model, evaporation is constrained to occur
over depth Ze with the greatest evaporative losses from
soil closest to the surface. The local evaporation rate is a
function of depth, saturation, and time, and is repre-
sented by the following functional form [12,21],
ei ¼ Dzi � e0i ¼ Dzi � ewi �EðSi; tÞPme
i¼1
Dzi � ewi
ð5Þ
where ei is the extraction rate in cell i, ewi is a depth-
dependent weighting function for evaporation evaluated
at the centroid of cell i, Si is saturation in cell i, E(Si, t) isan empirical evaporation function, and me is the number
of soil layers over which the evaporation is non-zero.
The evaporative weight, ew(z), constrains the evapora-
tion function so that shallow depths have the greatest
evaporation potential. It is approximated with a beta
distribution [12] as
ewðzÞ ¼ CðAþ BÞCðAÞCðBÞ z
A�1ð1� zÞB�1 ð6Þ
where C(Æ) is the gamma function and A [-] and B [-] are
positive parameters. E(Si, t) is given by [21]
EðSi; tÞ ¼0 Si 6 Sh
Si�ShSw�Sh
� EmaxðtÞ Sh < Si < Sw
EmaxðtÞ Si P Sw
8><>:
9>=>; ð7Þ
where Emax (L3 evaporated water/L2 soil/T) is the max-
imum instantaneous evaporation rate for an average
growing season day, Sw is the wilting saturation, and
Sh is the hygroscopic saturation. Emax is varied toapproximate atmospheric demand with a sinusoidal
function over 12 h representing daytime. The maximum
value of Emax(t) is reached at noon, and it is zero for the
12 night hours. The wilting saturation, Sw, is the satura-
tion at which the plant can no longer extract soil water,
because the suction required to do so is so high that the
plant tissue is damaged e.g., [21,10]. While Sw as a
556 M.J. Puma et al. / Advances in Water Resources 28 (2005) 553–566
threshold value for evaporation is arbitrary, dynamics
of evaporation are not crucial for the simulations in this
paper because the soil is covered by vegetation, and
transpiration is the dominant process.
2.5.2. Transpiration
The effect of vegetation is incorporated into the
model through the plant water uptake rate, u 0, which
accounts for plant transpiration T(S). Plant water up-
take is represented by a simplified mathematical expres-
sion based on experimental and field observations. We
use a semi-mechanistic (Type I) model where water up-
take is assumed proportional to the difference between
the soil and plant water potentials with no change instorage within the plant. This formulation assumes that
an Ohm�s law analogy for water uptake may be applied
to the transport of water from the soil to the atmosphere
through the plant [1,24,15,38,13]. Water uptake is lim-
ited by the resistance of water flow through the soil to
the root surface and from the root surface to the root
xylem:
ui ¼ Dzi � u0i ¼ Dzi �Wi �Wp
rs;iðtÞ þ rr;ið8Þ
where Wi is the local fluid pressure head (L), Wp is the
fluid potential in the plant (L), rs,i is the local soil resis-
tance (T/L), and rr,i is the local root resistance (T/L).
Fluid potential is assumed to be constant throughout
the plant. The soil resistance can be expressed as [12]
rs;i ¼Cs
rwi �RW0 � KðSiðtÞÞð9Þ
where Cs is a dimensionless constant that accounts for
root diameter, geometry and arrangement, rwi is the rel-
ative root density as a function of depth, and RW0 is the
average root-length density in the root zone (L roots/L3
soil). Similarly, the root resistance is [12]
rr;i ¼Cr
rwi �RW0
ð10Þ
where Cr is a constant parameter of the plant (T/L). As
for ewi, the relative root density, rwi, is approximated by
as beta distribution as
rwðzÞ ¼ CðAþ BÞCðAÞCðBÞ z
A�1ð1� zÞB�1 ð11Þ
The extraction rate ui is constrained such that it cannot
be negative, and, consequently, hydraulic lift is not rep-
resented explicitly in this model e.g., [25]. Transpiration
is the sum of the water uptake from all layers i in the
root zone and is expressed as
T actðtÞ ¼Xzi6Zr
ui ¼Xzi6Zr
Wi �Wp
rs;iðtÞ þ rr;ið12Þ
where Tact is the actual transpiration and Zr is the max-
imum depth of active roots. Since Eq. (8) contains three
unknowns, Wp, Cr, and Cs, additional constraints are
necessary to solve the system of equations.
2.5.3. Solution of transpiration expression
Maximum total transpiration, Tmax, is dependent
upon atmospheric conditions, while actual transpira-tion, Tact, is controlled by soil moisture in the root zone,
soil properties, and water uptake characteristics of the
plant with an upper limit of Tmax. We approximate Tmax
in the same manner that Emax is approximated, with a
diurnal sinusoidal variation. The plant is modeled as a
single point described by Wp, which must be constrained
based on plant physiology. First, the lower limit for Wp
is the plant fluid potential at wilting, Ww, which preventsthe plant from extracting water from a layer if Wi is
more negative than Ww. The other constraint restricts
the total transpiration, so that it cannot exceed Tmax.
Consequently, Wp is either Ww or the fluid potential in
the plant for which the water uptake from the root zone
equals Tmax, whichever is larger (i.e., less negative) [12].
The parameter, Cr, a coefficient in the root resistance
term, can be estimated experimentally, and some valuescan be found in the literature. Yet these estimations are
species specific, are known for a small number of plants
under conditions that do not necessarily exist in the
field, and are sometimes different for different root resis-
tance formulations. In addition, no reliable data exist
for the parameter Cs. Therefore, we seek two constraints
that will enable solution for these unknown parameters
Cr and Cs.The first constraint considers the plant�s physiologi-
cally controlled ability to extract water from the soil.
We define a plant compensation factor, c, as a specific
characteristic of the plant�s ability to uptake water. It is
derived using the Ohm�s law formulation for transpira-
tion by considering the minimum fraction of the root
zone, 1/c, that needs to be saturated in order for the plant
to meet Tmax. If a hypothetical situation is consideredwhere soil layers are either at W = 0 (full saturation) or
below Ww, then water uptake occurs only from the satu-
rated layers. In order to find the minimum fraction of the
root zone that must be saturated, the fluid potential in the
plant should be at the minimum value, Ww, which pro-
duces the greatest difference in potential between the soil
and the plant (and hence the greatest force the plant can
produce for water uptake). Consequently, 1/c becomesthe minimum fraction of roots that must be saturated
for the plant to withdraw enough water to meet the tran-
spiration demand if extraction from elsewhere in the soil
column is zero [12]. This scenario is expressed as
Tmax ¼Xzi6Zr
ui ¼Zr
c� ð�WwÞ
Cs
KsatRW0þ Cr
RW0
ð13Þ
When the soil is at or near full saturation, the soil resis-
tance is assumed to be small relative to the root resis-
0Weight
Dep
th
Z r
Z e
ew(z)
rw(z)
124 8
Fig. 1. Plot of evaporation and root-length-density weights as
functions of depth for Burkea africana in Nylsvley soil.
M.J. Puma et al. / Advances in Water Resources 28 (2005) 553–566 557
tance. Therefore, the plant compensation factor, c, iscontrolled by the root resistance term, rr = Cr/RW0.
The plant compensation factor, c, is physically
meaningful in that it is a measure of the plant�s ability
to extract water at high rates from wet sections of the
soil to compensate for dry sections [12]. A plant witha value of c greater than one can increase water uptake
from wet portions of the soil column when soil mois-
ture is non-uniformly distributed in the root zone. If
c equals one, then the plant cannot compensate, and
it would need the entire root zone to be saturated to
meet atmospheric transpiration demand. Since the
model only includes a static root system, root response
to available soil moisture through growth or changes inroot resistance, rr, is not considered. Therefore, c can
also be considered an empirical way to account for root
growth or for rr changes in response to available soil
moisture.
The other constraint specifies that if the soil column is
uniformly saturated at the saturation where stomata clo-
sure commences, S* in the model of Laio et al. [21], then
the total plant uptake is equal to the transpirationdemand when the plant fluid potential is at its lowest
value, Ww:
Tmax ¼ Zr �ðW� �WwÞCs
KðS�ÞRW0þ Cr
RW0
ð14Þ
where W* is the fluid pressure head that corresponds to
S*. While the condition in Eq. (14) facilitates compari-
son with the bucket model of Laio et al. [21] and Guswa
et al. [12], it is a constraint not based on the physical sys-
tem and made only for convenience. Specifying c, com-
bining Eqs. (13) and (14) with the conditions imposed on
Wp, and using the constitutive relationships in Eqs. (3)–(11) enable solution of Eq. (1).
2.6. Description of soil, plant, and climate in simulations
As an initial investigation, the analysis of this paper
will be applied to the savanna in Nylsvley, South Africa,
which is a water-limited ecosystem that has been exten-
sively studied [34] and has been the focus of previousmodeling efforts e.g., [20,12]. The Nylsvley site has a cli-
mate typical of a savanna, characterized by a hot and
rainy growing season and a warm and dry dormant sea-
son [34]. The simulations in this paper will focus on the
soil-moisture dynamics during the hot and rainy grow-
ing season.
This analysis will consider only one vegetation type,
Burkea africana, which is a dominant broad-leafed woo-dy species in this savanna [34]. The roots of B. africana
generally extend to a depth of 100 cm and are distrib-
uted with the root density having a minimum (equal to
zero) at the soil surface and at all depths greater than
100 cm, and a maximum at a depth of 50 cm [18,34].
To match this root-distribution data, in Eq. (11) we
set A and B both equal to 2 with RW0 equal to
0.02 cm/cm3 and Zr equal to 100 cm [20]. The depth of
the bottom boundary is set to 200 cm for the simula-
tions. For the evaporative weights in Eq. (6), A is 0.9
and B is 5 [12]. This distribution assures that water loss
from the soil due to evaporation is highest from the soillayers closest to the surface. Fig. 1 shows the two
weighting functions, rw(z) and ew(z).
The water usage of B. africana is reported in Scholes
and Walker [34], which enables estimation of Tmax, Ww,
W*. Laio et al. [20] adjusted the maximum instantaneous
transpiration rates found in Scholes and Walker [34] to
evapotranspiration values that are representative for
average conditions for a 24-h period over a growing sea-son, hETmaxi1 day, where hÆi1 day denotes averaging over a
day. In Laio et al. [20], hETmaxi1 day for B. africana is
0.475 cm/day. Because our model has a temporal resolu-
tion finer than a day, evapotranspiration is partitioned
into two separate functions. The maximum depth from
which evaporation can occur, Ze, is estimated to be
20 cm and the average daily evaporation for a mean
growing season day under well-watered conditions,hEmaxi1 day, is approximated as 0.046 cm/day for a com-
pletely vegetated surface. This hEmaxi1 day value is based
on the assumption that if the soil is at full saturation for
an entire day, then evaporation accounts for 10% of
evapotranspiration. The maximum transpiration,
hTmaxi1 day, is then the difference between hETmaxi1 dayand hEmaxi1 day. However, because both maxima (evapo-
ration and transpiration) are dependent upon atmo-spheric conditions, the instantaneous maximum values,
Emax and Tmax, should change over a day. Fig. 2 shows
the diurnal maximum instantaneous evapotranspiration
function over a day for B. africana. The sinusoidal func-
tion is subject to the constraint that if the instantaneous
evaporation and transpiration are meeting atmospheric
demand for an entire day, then 0.475 cm is the maxi-
mum amount of water loss over a day from the soildue to evaporation and transpiration.
The plant fluid potential at wilting, Ww, for B. afri-
cana is �31,600 cm, and W* is �730 cm [34]; from these
0
0.4
0.8
1.2
1.6
0 0.2 0.4 0.6 0.8 1Time [days]
ET
max
[cm
/day
]
ET max = constant
ET max(t )
Fig. 2. Plot of maximum instantaneous evapotranspiration, ETmax,
over a day for Burkea africana. ETmax is assumed diurnal and is
approximated with a sinusoidal function for 12 h each day.
Table 1
Soil parameters values for Nylsvley, South Africa [34,20]
Parameter Symbol Value
Saturated conductivity Ksat 109.8 cm/day
Porosity / 0.42
Retention curve parameter b 2.25
Air-entry pressure head We �3.0 cm
Hygroscopic saturation Sh 0.02
0
10
20
30
0 0.1 0.2
S
Dep
th [
cm]
0.35S30S
558 M.J. Puma et al. / Advances in Water Resources 28 (2005) 553–566
it follows that Sw is approximately 0.06, and S* is 0.105
[20]. The fixed interception threshold, D, is estimated to
be 0.2 cm for B. africana [20]. The Nylsvley soil is rela-
tively homogeneous and generally has a high sand con-
tent (85–90%). Following Laio et al. [20], weighted
averages of the properties of the A and B soil horizons
reported in Scholes and Walker [34] are calculated and
applied uniformly. Table 1 presents these soil parametervalues [34,20,5].
The rainfall in the Nylsvley savanna occurs primarily
as convective storms with high intensity and short dura-
tion [34]. Therefore, a marked Poisson process is used to
simulate the Nylsvley rainfall [20]. For the instantaneous
soil-moisture analysis in Section 3, average climate
parameters typical of Nylsvley are used. Following Laio
et al. [20], storm arrival rate, k, is 0.167 day�1, and thetotal depth of rainfall produced by a storm is generated
from an exponential distribution with a mean of 1.5 cm.
The duration is from a beta distribution with shape
parameter A equal to 2 and B equal to 4.67, assuming
a mean storm duration of 0.0625 days, a minimum of
0.0125 days, and a maximum of 0.25 days.
40
Fig. 3. Sketch of saturation as a function of depth to show
instantaneous saturation values for two averaging depths, 5 cm and
30 cm.
3. Instantaneous soil-moisture analysis
Let Sd denote a spatial average of instantaneous soil
moisture with the average taken between the soil surface
and a depth d. A special value of this average corre-
sponds to observations of soil moisture, denoted by
Sobs, which are measurements taken over depth dobs.
An example of Sobs derives from remote sensing mea-
surements, which provide estimates of soil moisture for
the upper few centimeters of soil at an instant in time.The corresponding averaging depth, dobs, depends on
sensor type, vegetation cover, landscape roughness,
and soil type. For our purposes, we take dobs equal to
5 cm as a characteristic depth for remote-sensed data,
although the depth is often less than this value. Fig. 3
shows a soil-moisture profile demonstrating vertical
averaging of instantaneous soil moisture for S30 and S5.
Recent research has focused on use of the instanta-neous measurement Sobs in conjunction with a model
or other information to improve prediction of Sd , where
d is deeper than dobs e.g., [28,6,16,27]. A recent field
study by Wilson et al. [39] investigated the relationship
between S6 and S30 using in situ measurement tech-
niques, but those authors were unable to find a clear
relationship between the two variables.
The relationship among instantaneous soil-moisturevalues integrated through different soil depths is dy-
namic, and it evolves in time as a function of hydrologic
inputs, atmospheric conditions, soil properties, and
plant characteristics. Because of its dynamic nature,
the time interval between soil-moisture measurements
is an important quantity. To demonstrate the impor-
tance of this time interval, we simulate a 200-day grow-
ing season for B. africana in the Nylsvley savanna,assuming an average Nylsvley climate realization with
k equal to 0.167 storms/day and a equal to 1.5 cm. Addi-
tionally, the plant compensation factor, c, is set equal to2, which means the plant has a moderate ability to ex-
tract water at high rates from wet sections of the root
zone to compensate for dry sections.
Fig. 4a shows a scatter plot of instantaneous soil
moisture for S30 versus S5. We plot S30 versus S5 (both
0
0.2
0.4
0.6
0 0.2 0.4 0.6
30S
5S
(a)
0
0.2
0.4
0.6
0 0.2 0.4 0.6
30S
5S
(b)
0
0.2
0.4
0.6
0 0.2 0.4 0.6
30S
5S
(c)
Fig. 4. Scatter plots of instantaneous saturation for 30 cm versus 5 cm averaging depth for Burkea africana (c = 2) from 200-day growing season
simulation with (a) dt equal to 0.1 day, (b) dt equal to 1 day, and (c) dt equal to U{0.1,5}.
M.J. Puma et al. / Advances in Water Resources 28 (2005) 553–566 559
from model output at the same instant in time) to gen-
erate each data point with a temporal interval between
output data, dt, of 0.1 day. This interval between output
data, dt, is interpreted to represent the time between soil-
moisture measurements, Sobs. As shown in Fig. 4a, bothS30 and S5 immediately increase when a storm with rain-
fall reaching the soil occurs. The amount of this increase
depends on the storm�s intensity, duration, and the ini-
tial soil conditions. The S5 value declines much faster
than S30 following a large storm event, which is a conse-
quence of the infiltrating front advancing downward in
the soil column coupled with drying in the upper layers.
Drying curves for individual storms are evident in Fig.4a with points forming a path from right to left in the
figure. This relationship is a function of the soil, plant,
and climate characteristics of the system.
In Fig. 4b, dt is increased to 1 day. We observe that
the relationship between S30 and S5 is still detectable.
However, because the instantaneous soil-moisture data
are available less frequently, detail is lost from the dry-
ing curves. Consequently, it becomes difficult to deter-mine to which drying curve some of the data points
belong, especially immediately after large storm events.
If dt is randomly chosen from a uniform distribution be-
tween 0.1 and 5 days, then the structure of the relation-
ship breaks down and becomes undetectable, as seen in
Fig. 4c. Clearly, a relationship is evident only when dt isconsistently small enough to capture the infiltration
dynamics of the system.
Wilson et al. [39] could not find a structured relation-
ship between instantaneous soil-moisture values with d
equal to 6 cm and 30 cm. One probable reason for the
unstructured relationship is that the dt between mea-
surements was not small enough. The dt necessary to de-tect the relationship between different instantaneous
soil-moisture averaging depths will be a function of
the soil, plant, and climate characteristics. A second rea-
son for the unstructured relationship is that the scatter
plots of the S6 versus S30 field measurements were for
different horizontal spatial locations in each of the
watersheds studied. Undoubtedly, soil profile, vegeta-
tion, and infiltrating water is variable over a watershed.Consequently, their finding of no clear relationship be-
tween pairs of S6 and S30 measurements is consistent
with the variability among the different site locations.
Notwithstanding the challenge of measurement fre-
quency, it is worth exploring the relationship between
S30 and S5 in Fig. 4a. In Fig. 5, the ratio of S5 to S30
is plotted over a 200-day growing season for a dt equalto 0.1 day. The ratio increases during and immediatelyfollowing a storm event, where the increase depends
on the storm�s intensity, duration, and the initial soil
conditions. From Fig. 5, a storm with a high intensity
and long duration will lead to S5 : S30 ratios greater than
3:1. For less intense and shorter storms with dry initial
conditions, the ratio of S5 to S30 is between 1:1 and
2:1. If initial soil-moisture conditions and the storm�sintensity and duration are known, then, in theory, one
0
1
2
3
4
0 50 100 150 200Time [days]
S /S 5
30
Fig. 5. Trace of the ratio of instantaneous saturation averaged over
5 cm to instantaneous saturation averaged over 30 cm for a 200-day
growing season with dt equal to 1 day for Burkea africana in Nylsvley
soil.
560 M.J. Puma et al. / Advances in Water Resources 28 (2005) 553–566
should be able to predict the relationship between S30
and S5. In addition, if we have frequent remote sensing
measurements, then data assimilation techniques, such
as the Kalman filter technique, would be able to estimate
accurately soil-moisture profiles from the remote sensingdata as demonstrated by Galantowicz et al. [11].
In summary, two major challenges prevent the use of
Sobs measurements to estimate Sd , where d is deeper than
dobs. First, dt must be sufficiently small relative to the
infiltration dynamics, as demonstrated in Fig. 4a–c. A
primary constraint for remote sensing or any other data
collection scheme is the measurement frequency. Cur-
rent remote sensing measurement intervals are greaterthan daily [30], while field measurements are often
obtained at irregular intervals when large horizontal
spatial areas are studied [39]. Second, regarding horizon-
tal spatial scale, instantaneous soil-moisture values can
only be related if hydrologic inputs, soil properties,
and plant characteristic are similar for locations being
considered. In light of these challenges in relating instan-
taneous soil-moisture measurements, we propose a sta-tistical approach to address the problem of the data�stemporal resolution.
0
20
40
60
80
0.15 0.2 0.25 0.3λ [1/day]
Extreme Savanna Parameters
Simulation Parameters
Tot
al R
ainf
all f
or 2
00-D
ayG
row
ing
Seas
on [
cm/2
00da
ys]
Fig. 6. Climate sample space for model simulations of Burkea africana
in Nylsvley soil used in the statistical soil-moisture analysis.
4. Statistical soil-moisture analysis
This alternative analysis investigates the relationship
between the temporal mean of soil moisture and averag-ing depth, d, as well as between the variance of soil mois-
ture and d. The overall goal is to predict these
relationships from the statistics of soil-moisture mea-
surements over averaging depth dobs and from known
parameters that characterize the soil, plant, and climate
system. Although issues such as scaling of soil proper-
ties, vegetation, and precipitation in horizontal space
need to be considered, we propose that if these statisticalrelationships can be predicted, then this information
could assist in the validation of soil-moisture estima-
tions produced by large-scale models. That is, if the
statistics of the large-scale model�s soil-moisture estima-
tions do not match the statistics of the estimations ob-
tained through our methodology for a given location
in horizontal space, then the validity of the large-scale
model�s estimations are questionable.As in Section 3, we will use our model to simulate
soil-moisture dynamics during the growing season at
Nylsvley. Herein we consider two basic statistical mea-
sures: the mean of soil moisture, lSd, and the variance
of soil moisture, r2
Sdover 1000 consecutive growing sea-
son days. We calculate the mean and variance of soil
moisture from instantaneous soil-moisture output
during 1000-day simulations for various climate realiza-tions. Since the dormant season is dry, the soil-moisture
conditions at the beginning of a growing season are sig-
nificantly changed after the first rainfall event. That is,
the initial soil-moisture conditions for each growing
season do not last long and do not change the soil-
moisture statistics of the growing season. Consequently,
lSdand r2
Sdare obtained by simulating a continuous
1000-day growing season with our model.Because instantaneous values of soil moisture are
dependent on hydrologic inputs, atmospheric condi-
tions, soil properties, and plant characteristics, the sta-
tistical characteristics of soil moisture throughout the
soil profile must depend on these same variables. For
this initial investigation, we vary rainfall frequency
and intensity, while soil and plant properties remain
constant. The range of values for k and total growingseason rainfall (assuming a 200-day growing season)
commonly found in a water-limited ecosystem are
shown in Fig. 6. We sample the parameter space system-
atically to define 25 different simulations. These 25 pairs
of parameters are shown in Fig. 6. Specification of k and
total growing season rainfall allows calculation of a. Theprobability distribution of storm duration is not chan-
ged, and, therefore, the analysis focuses only on the role
M.J. Puma et al. / Advances in Water Resources 28 (2005) 553–566 561
of k and a in a statistical analysis of soil moisture versus
averaging depth.
From model output of instantaneous soil moisture
with a dt of 0.1 day over 1000 growing season days,
we calculate lSdand r2
Sdfor averaging depths ranging
from dobs to the total root-zone depth, Zr, taken in5 cm increments. In all of the simulations, dobs is consid-
ered equal to 5 cm, while Zr is 100 cm. Fig. 7a shows a
plot of lSdversus d for four of the simulations that are
representative of the relationships found among the 25
simulations, where the markers are the simulation data.
Fig. 7b shows the same simulations for r2
Sdversus d.
Piecewise linear interpolation in Fig. 7a and b produces
smooth and well-behaved curves that reveal the exis-tence of relationships among lSd
for a range of d and
r2
Sdfor a range of d. If these relationships can be related
to soil, plant, and climate parameters, then soil-moisture
mean and variance over any d might be estimated from
easily measured environmental parameters.
A functional form describing the relationships in Fig.
7a and b is sought. In particular, we wish to relate sta-
tistics for the depth of observation, dobs, to statisticsfor other depths of interest. We begin by focusing on
the behavior of the means. We assume that we know
0
0
0.05
0.1
0.15
0.2
0.25
0.3
5d [cm]
αλ' = 0.27 cm/dayαλ' = 0.21 cm/dayαλ' = 0.16 cm/dayαλ' = 0.10 cm/day
0.005
0.01
0.015
0.02
0.025αλ' = 0.27 cm/dayαλ' = 0.21 cm/dayαλ' = 0.16 cm/dayαλ' = 0.10 cm/day
20 35 50 65 80 95
5d [cm]
20 35 50 65 80 95
(a)
(b)
µ sd
σ sd2
Fig. 7. Plots of (a) the mean of saturation over 1000 growing season
days as a function of averaging depth and (b) the variance of
saturation over 1000 growing season days as a function of averaging
depth for Burkea africana in Nylsvley soil for four climate scenarios
from model simulations.
dobs, and that we can use a large number of remote sens-
ing observations (for example, every 2 days for several
years) to estimate lSobs. The curves in Fig. 7a can be fit
with the following expression [35]:
lSd¼
lSobsðdj
obs þ vÞ1=j
ðdj þ vÞ1=jð15Þ
where lSobsis the mean of soil moisture over observation
averaging depth, dobs, and v and j are parameters that
affect the curvature and asymptote of the function.
For this analysis, we assume that lSobsis known, and,
consequently, model output is used to compute lSobs.
Eq. (15) provides excellent fits to the data in Fig. 7a
through implementation of the Levenberg–Marquardt
algorithm [29], which solves for the parameters v and
j. The values of the parameters v and j from the four
climate realizations in Fig. 7a are presented in Table 2.
Fig. 8 shows the data and fits for the mean saturation
versus averaging depth from these climates, where thesolid lines represent the fits using Eq. (15). The parame-
ters v and j were also calculated for the other climate
realizations. The goal is to relate the unknown parame-
ters v and j to measurable properties of the ecosystem
under consideration.
Since soil and plant properties remain constant in
the simulations presented, the parameters v and j,which determine the shape of the curve relating the
Table 2
Parameter values of Eq. (15) for the four climate realizations in Fig. 7a
Average daily rainfall reaching
the soil, ak0 [cm/day]
Parameter v Parameter j
0.27 2890 2.06
0.21 652 1.67
0.16 181 1.46
0.10 32.7 0.91
The mean of saturation versus averaging depth curves, which are
produced using these parameters in Eq. (15), are shown in Fig. 8.
0
0.05
0.1
0.15
0.2
0.25
0.3
5d [cm]
αλ' = 0.27 cm/dayαλ' = 0.21 cm/dayαλ' = 0.16 cm/dayαλ' = 0.10 cm/day
20 35 50 65 80 95
µ sd
Fig. 8. Plot of the fits of Eq. (15) (solid lines) to the simulation data
(markers) of Fig. 7a, which shows the mean of saturation versus
averaging depth.
562 M.J. Puma et al. / Advances in Water Resources 28 (2005) 553–566
soil-moisture mean to averaging depth, are controlled by
rainfall frequency and intensity. We would like to use
these climate parameters to predict the parameters vand j. We expect that v and j will depend on the aver-
age daily rainfall reaching the soil for a growing season,
which is given by a Æ k 0. Alternatively, these parameterscould depend on the index of dryness, DI, from Milly
[26], which is defined as DI = ETmax/(ak 0). Since ETmax
does not change because the vegetation is not varied in
this study, we focus on relating the parameters v and jto a Æ k 0. Fig. 9a presents the relationship for v versus
a Æ k 0, while Fig. 9b shows j versus a Æ k 0. The relation-
ship in Fig. 9a can be fit by the equation
v ¼ 1:85� 105 � ða � k0Þ3:90; a � k0 < 0:17 cm=day
1:40� 107 � ða � k0Þ6:59; a � k0 P 0:17 cm=day
(
ð16Þwhile for Fig. 9b the equation is
j ¼ 1:20 lnða � k0Þ þ 3:57 ð17ÞEqs. (16) and (17) were obtained empirically through
regression. Knowledge of a Æ k 0, used in conjunction with
Eqs. (16) and (17), allows for estimation of the parame-
1
10
100
1000
10000
0 0.1 0.2 0.3αλ'
αλ' = 0.17
(a)
-0.5
0.0
0.5
1.0
1.5
2.0
2.5
0 0.1 0.2 0.3
αλ'
κχ
(b)
Fig. 9. Plots of (a) the parameter v versus ak 0 and (b) the parameter jversus ak 0 for the 25 climate scenarios for Burkea africana in Nylsvley
soil. The parameters v and j are obtained from fitting Eq. (15) to the
mean of saturation versus averaging depth curves.
ters v and j. By incorporating the estimated v and j into
Eq. (15), we can predict the lSdversus d relationship for
a specific vegetation and soil. That is, if we have accu-
rate knowledge of the average daily rainfall reaching
the soil, a Æ k 0, for B. africana in Nylsvley soil and obtain
lSobsfrom remote sensing observations, then we can pre-
dict the mean soil-moisture value for any averaging
depth d.
We test this prediction procedure by considering the
four climate realizations plotted in Fig. 7 and attempt-
ing to predict the lSdversus d curves. We know a Æ k 0
and lSobsfor each of these realizations, while Eqs. (16)
and (17) estimate the parameters v and j. Then Eq.
(15) predicts the lSdversus d curves shown in Fig. 10.
We find close approximations for the lSdversus d rela-
tionship for three of the four climate realizations. The
exception is the climate with a Æ k 0 equal to 0.21 cm/
day, where the predicted curve provides a somewhat
poorer fit to the data. The reason is that the parameters
v and j for that climate realization are not well esti-
mated by Eqs. (16) and (17).
Eqs. (16) and (17) are not based on any physical char-acteristics of the system but were chosen to balance
complexity and prediction accuracy. Future research
could focus on increasing the number of simulations
and improving the fits for Eqs. (16) and (17) to improve
prediction ability. Another concern is accurate knowl-
edge of a Æ k 0 for a given ecosystem. If the actual a Æ k 0
does not match the predicted a Æ k 0, problems could arise
when seeking a relationship between parameters of Eq.(15) and the parameters of the climate realizations.
Alternatively, we attempted to predict the parameters
v and j from either the mean rainfall depth, a, or the
mean storm arrival rate, k. However, the relationship
between a Æ k 0 and the parameters was much stronger.
It is also possible to predict the parameters v and j
650
0.05
0.1
0.15
0.2
0.25
0.3
5d [cm]
αλ' = 0.27 cm/dayαλ' = 0.21 cm/dayαλ' = 0.16 cm/dayαλ' = 0.10 cm/day
20 35 50 80 95
µ sd
Fig. 10. The curves for mean of saturation versus averaging depth are
predicted from ak 0 using the fits in Fig. 9a and b to estimate the
parameters of Eq. (15). The Eq. (15) predictions (solid lines) are
compared to the simulation data (markers).
M.J. Puma et al. / Advances in Water Resources 28 (2005) 553–566 563
from, as an example, both a Æ k 0 and k, but we focused
on using one variable for simplicity in this initial inves-
tigation. Despite the challenges, the results in Fig. 10
demonstrate that this methodology has the potential to
estimate effectively the lSdversus d relationship using
readily available data on soil, plant, and climatecharacteristics in a model of appropriate resolution
and detail.
The curves for variance of instantaneous soil mois-
ture versus averaging depth in Fig. 7b require a fitting
equation other than Eq. (15), because this equation fails
to provide close fits to the data. Therefore, further work
must be done to obtain a function that accurately fits the
variance versus averaging depth curves. An alternativeapproach to looking for a particular functional form
for the curves in Fig. 7a and b is to look at lSdversus
a Æ k 0 for different d and r2
Sdversus a Æ k 0 for different d.
Fig. 11a shows that lSdincreases linearly as a Æ k 0 in-
creases for two averaging depths, 30 cm and 100 cm.
In Fig. 11b, r2
Sdincreases as a Æ k 0 increases, but the
spread of the simulation data increases as a Æ k 0 in-
0
0.05
0.1
0.15
0.2
0.25
0.3
0
αλ' [cm/day]
d = 30 cmd = 100 cm
0
0.004
0.008
0.012
0.016
0.02
d = 30 cmd = 100 cm
0.1 0.2 0.3
0
αλ' [cm/day]0.1 0.2 0.3
µ sd
σ sd2
(a)
(b)
Fig. 11. Plots of (a) the mean of saturation for different averaging
depths as a function of average daily rainfall reaching the soil, ak 0, and
(b) the variance of saturation for different averaging depths as a
function of ak 0 for Burkea africana in Nylsvley soil.
creases. Fig. 11 highlights the opportunity for explora-
tion of the statistical properties of soil moisture in the
root zone through a variety of approaches.
5. Discussion
The statistical analysis presents a methodology that
allows prediction of the mean of soil moisture, lSd,
for any averaging depth in the root zone. We consider
a specific vegetation and soil, while simulating a range
of climate realizations. We would like to extend the
current analysis to predict the relationships between sta-
tistical measures of soil moisture and averaging depthfor a range of vegetation and soil subjected to various
climates. Extension of this analysis highlights the impor-
tance of understanding the effect of the different pro-
cesses considered in soil, plant, and climate models on
lSd. In fact, it is important to recognize that the model
used for a given soil, plant, and climate system depends
on the characteristics of the system and the model out-
put of interest [36]. Therefore, while this model is appro-priate for simulating a completely vegetated area of B.
africana in Nylsvley soil, it might not be appropriate
for another system.
As the simulations of B. africana in Nylsvley soil
demonstrate, the average rainfall reaching the soil,
a Æ k 0, has a significant impact on lSdfor all d in the root
zone. The average amount of water in the root zone in-
creases for higher values of a Æ k 0. Consequently, all lSdmust increase. The shape of each curve depends on the
average infiltration and redistribution characteristics of
the system, which are controlled by the rainfall, soil,
and vegetation. We would like to identify parameters
that are dominant in controlling the shape of these
curves. Measurable parameters that have a significant
impact on lSdinclude Ksat, ETmax, a, k 0, average rainfall
intensity, and root distribution. The dimensionlessgroups of parameters discussed by Guswa et al. [12]
could be useful measures in determining what controls
lSd. For example, the spatial infiltration index is defined
as [12]
I I;x ¼Zi
Zr
¼ a=½/ðSfc � ShÞ�Zr
ð18Þ
where Zi is a measure of the depth of infiltration of an
average storm and Sfc is the saturation at field capacity(0.29 for Nysvley soil) [12]. This index is a measure of
the infiltration depth relative to the depth of the root
zone. II,x values for the simulations range from 0.03 to
0.15, which means that, on average, water is infiltrating
to very shallow depths relative to Zr. This finding is con-
sistent with the results of Fig. 7a, which show that lSddecrease rapidly as d increases.
564 M.J. Puma et al. / Advances in Water Resources 28 (2005) 553–566
When considering different soil, plant, and climate
systems, the proposed statistical approach may vary
depending upon the properties of the system. For exam-
ple, if an ecosystem has high interannual climatic
variability, then using a Æ k 0 to predict lSdmight not be
accurate. It is necessary to identify the parameters thatprovide meaningful information on lSd
for each system.
Then, in terms of extending the proposed analysis to
other systems, simulations can be run for a variety of
soil and vegetation types to obtain statistical soil-mois-
ture data for different averaging depths. A function
can be fitted to the data from the simulations, and we
can then look for relationships between the parameters
of the fitting equation and measurable soil, plant, andclimate parameters (or dimensionless groups of these
parameters).
Next, we consider the uses of these statistical soil-
moisture measures for different averaging depths. To ad-
dress this issue, we must again consider the suggestion
that the most promising approach to estimate soil mois-
ture for large-scale applications is through combination
of remotely sensed data and hydrological models [19].The instantaneous soil-moisture analysis (Fig. 4a–c)
highlights the challenges that prevent the use of shallow
instantaneous soil-moisture values, inferred from re-
mote sensing observations, in the estimation of instanta-
neous soil moisture for deeper averaging depths. The
methodology that we propose, which uses the statistics
of these shallow instantaneous soil-moisture values,
can be used as a diagnostic tool to evaluate the estima-tions of soil moisture made by large-scale land surface
models. Because computation time typically constrains
the complexity and resolution of large-scale models, a
large-scale model might not be sufficiently resolved in
the vertical spatial dimension or might not include all
processes necessary to estimate accurately soil-moisture
dynamics. In our methodology, the appropriate com-
plexity and resolution of the model depends on the soil,plant, and climate characteristics of the system under
consideration. The statistics of soil-moisture estimations
for each horizontal grid cell from a coarse and simplified
large-scale model could be validated through compari-
son with the soil-moisture statistics estimated using
our methodology, where the estimations will be from
statistics of remotely sensed shallow-layer soil moisture
and a model of proper complexity and resolution forthe system under consideration. Through our statistical
methodology, we avoid the difficulties associated with
using instantaneous soil-moisture values and demon-
strate a technique that could assess the validity of the
soil-moisture estimations from large-scale models. We
do not guide model estimations through assimilation
of soil-moisture values from remote sensing. However,
the proposed validation scheme could elucidate sourcesof inaccuracy in large-scale models. In addition to large-
scale model validation, statistics of soil moisture ob-
tained through the proposed analysis are important for
applications in ecohydrology. For example, if we want
to estimate plant stress in a water-limited ecosystem
due to climate change, soil-moisture statistics would be
useful. On the other hand, soil-moisture statistics and,hence, this methodology would not be very useful for
applications such as weather forecasting or real-time
flood predictions, because they do not provide event-
based information on soil moisture.
Finally, it is important to discuss two issues regarding
soil-moisture values obtained from remote sensing.
First, we consider the case where dobs is extremely shal-
low, less than 5 cm, corresponding to cases where asharp vertical gradient in the top few centimeters of soil
occurs [2]. The methodology presented can be adjusted
easily by using a dobs less that 5 cm with a smaller inter-
val between values of d in the simulations. The function
in Eq. (15) might change, but the overall approach will
remain the same. In this manner, we overcome the prob-
lematic situation of very shallow dobs. Second, the hori-
zontal scale of soil-moisture values inferred from remotesensing measurements is typically much larger than the
horizontal scale of the model used in this initial investi-
gation. Therefore, we must make certain that the hori-
zontal spatial scale of the soil-moisture values from
remote sensing measurements is consistent with the
horizontal scale of the model in our methodology.
Future work must examine the validity of our method-
ology when the model, used to simulate different soil,plant, and climate systems, has a large horizontal spatial
scale.
6. Conclusions
It is not generally feasible to obtain remote sensing
measurements at the highly resolved temporal scalesnecessary to see a structured relationship among Sobs
and Sd for different values of d. Because of this limi-
tation, we have presented and tested a statistical
methodology. The results of this initial investigation
demonstrate that a simulation model, combined with
known climate parameters and soil moisture from re-
mote sensing measurements, allows accurate prediction
of the mean of root-zone soil moisture for any averagingdepth. Initial attempts to predict the r2
Sdversus d rela-
tionship have been less successful than those reported
herein for the mean, yet there is evidence for optimism
that further work will enable accurate prediction of this
relationship as well. While much research needs to be
done to validate these soil-moisture predictions, the
methodology presented provides a promising pathway
to obtain statistical information about soil moistureover the entire root zone.
M.J. Puma et al. / Advances in Water Resources 28 (2005) 553–566 565
Acknowledgments
The authors gratefully acknowledge the support of
the William Clay Ford, Jr �79 and Lisa Vanderzee Ford
�82 Princeton Graduate Fellowship, and I.R-I. gratefully
acknowledges the support of the National ScienceFoundation through grants DEB-0083566 (Biocomplex-
ity) and EAR-0120914 (National Center for Earth Sur-
face Dynamics).
References
[1] Campbell GS. Simulation of water uptake by plant roots.
Modeling plant and soil systems—agronomy monograph, no.
31. Madison, WI, USA: ASA-CSSA-SSSA; 1991.
[2] Capehart WJ, Carlson TN. Decoupling of surface and near-
surface soil water content: a remote sensing perspective. Water
Resour Res 1997;33(6):1383–95.
[3] Celia MA, Bouloutas ET, Zarba RL. A general mass-conservative
numerical solution for the unsaturated flow equation. Water
Resour Res 1990;26(7):1483–96.
[4] Celia MA, Binning P. A mass conservative numerical solution for
two-phase flow in porous media with application to unsaturated
flow. Water Resour Res 1992;28(10):2819–28.
[5] Clapp RB, Hornberger GN. Empirical equations for some
soil hydraulic properties. Water Resour Res 1978;14(8):
601–4.
[6] Crow WT, Wood EF. The assimilation of remotely sensed
soil brightness temperature imagery into a land surface
model using ensemble Kalman filtering: a case study based on
ESTAR measurements during SGP97. Adv Water Res 2003;26:
137–49.
[7] Dingman SL. Physical hydrology. 2nd ed. Upper Saddle River,
NJ, USA: Prentice-Hall; 2002.
[8] Entekhabi D, Nakamura H, Njoku EG. Solving the inverse
problem for soil moisture and temperature profiles by sequential
assimilation of multifrequency remotely sensed observations.
IEEE Trans Geosci Remote Sens 1994;32:438–47.
[9] Feddes RA, Hoff H, Bruen M, Dawson T, de Rosnay P, Dirmeyer
P, et al. Modeling root water uptake in hydrological and climate
models. Bull Am Meteor Soc 2001;82(12):2797–809.
[10] Fitter A, Hay R. Environmental physiology of plants. 3rd ed. San
Diego: Academic Press; 2002.
[11] Galantomicz JF, Entekhabi D, Njoku EG. Test of sequential data
assimilation for retrieving profile soil moisture and temperature
from observed L-band radiobrightness. IEEE Trans Geosci
Remote Sens 1999;37(4):1860–70.
[12] Guswa AJ, Celia MA, Rodriguez-Iturbe I. Models of soil
moisture dynamics in ecohydrology: a comparative study.
Water Resour Res 2002;38(9):1166. doi:10.1029/2001Wrooo826.
[13] Guswa AJ, Celia MA, Rodriguez-Iturbe I. Effect of vertical
resolution on predictions of transpiration in water-limited eco-
systems. Adv Water Res 2004;27:467–80.
[14] Hoeben R, Troch PA. Assimilation of active microwave obser-
vation data for soil moisture profile estimation. Water Resour Res
2000;36(10):2805–19.
[15] Hopmans JW, Bristow KL. Current capabilities and future needs
of root water and nutrient uptake modeling. In: Sparks DL,
editor. Advances in agronomy, vol. 77. San Diego: Academic
Press; 2002. p. 103–81.
[16] Houser PR, Shuttleworth J, Famiglietti JS, Gupta HV, Syed KH,
Goodrich DC. Integration of soil moisture remote sensing and
hydrologic modeling using data assimilation. Water Resour Res
1998;34(12):3405–20.
[17] Jones HG. Plants and microclimate: a quantitative approach to
environmental plant physiology. 2nd ed. Cambridge, UK: Cam-
bridge University Press; 1992.
[18] Knoop WT, Walker BH. Interactions of woody and herbaceous
vegetation in two savanna communities at Nylsvley. J Ecol 1984;
73:235–53.
[19] Kostov KG, Jackson TJ. Estimating profile soil moisture from
surface layer measurements—a review. In: Proceedings of the
international society of optical engineering. Orlando, FL,
USA: SPIE; 1993. p. 125–36.
[20] Laio F, Porporato A, Fernandez-Illescas C, Rodriguez-Iturbe I.
Plants in water-controlled ecosystems: active role in hydrologic
processes and response to water stress IV. Discussion of real cases.
Adv Water Res 2001;24(7):745–62.
[21] Laio F, Porporato A, Ridolfi L, Rodriguez-Iturbe I. Plants in
water-controlled ecosystems: active role in hydrological processes
and response to water stress II. Probablistic soil moisture
dynamics. Adv Water Res 2001;24(7):707–23.
[22] Li J, Islam S. Estimation of root zone soil moisture and surface
fluxes partitioning using near surface soil moisture measurements.
J Hydrol 2002;259:1–14.
[23] Lin DS. Microwave remote sensing of surface soil moisture and its
application to hydrologic modeling. PhD dissertation. Princeton,
NJ: Princeton University; 1994.
[24] Lhomme J-P. Formulation of root water uptake in a multilayer
soil-plant model: does van den Honert�s equation hold. Hydrol
Earth Syst Sci 1998;2(1):31–40.
[25] Mendel M, Hergarten S, Neugebauer HJ. On a better under-
standing of hydraulic lift: a numerical study. Water Resour Res
2002;38(10):1183. doi:10.1029/2001WR000911.
[26] Milly PCD. A minimalist probabilistic description of root zone
soil water. Water Resour Res 2001;37(3):457–64.
[27] Mohanty BP, Skaggs TH, Famiglietti JS. Analysis and mapping
of field-scale soil moisture variability using high-resolution,
ground-based data during the Southern Great Plains 1997
(SGP97) hydrology experiment. Water Resour Res 2000;36(4):
1023–31.
[28] Montaldo N, Albertson JD. Multi-scale assimilation of surface
soil moisture data for robust root zone moisture predictions. Adv
Water Res 2003;26:33–44.
[29] More JJ. The Levenberg–Marquardt algorithm: implementation
and theory. In: Watson GA, editor. Numerical analysis. Lecture
notes in mathematics 630. Berlin: Springer Verlag; 1977. p.
105–16.
[30] Njoku EG, Jackson TJ, Lakshmi V, Chan TK, Nghiem SV. Soil
moisture retrieval from AMSR-E. IEEE Trans Geosci Remote
Sens 2003;41(2):215–29.
[31] Rawls WJ, Ahuja LR, Brakensiek DL, Shirmohammadi A.
Infiltration and soil water movement. In: Maidment DR, editor.
Handbook of hydrology. New York: McGraw-Hill; 1993. p.
5.17–39.
[32] Rodriguez-Iturbe I, Porporato A, Ridolfi L, Isham V, Cox D.
Probabilistic modelling of water balance at a point: the role of
climate, soil, and vegetation. Proc R Soc London Ser A
1999;455:3789–805.
[33] Rodriguez-Iturbe I, Porporato A, Laio F, Ridolfi L. Plants in
water-controlled ecosystems: active role in hydrologic processes
and response to water stress: I. Scope and general outline. Adv
Water Res 2001;24(7):745–62.
[34] Scholes RJ, Walker BH. An African savanna. Cambridge,
UK: Cambridge University Press; 1993.
[35] Spanier J, Oldham KB. An atlas of functions. Washing-
ton: Hemisphere Pub Corp; 1987.
566 M.J. Puma et al. / Advances in Water Resources 28 (2005) 553–566
[36] Thornley JHM, Johnson IR. Plant and crop modelling: a
mathematical approach to plant and crop physiology. Caldwell,
NJ, USA: The Blackburn Press; 2000.
[37] Touma J, Vaulin M. Experimental and numerical analysis of two-
phase infiltration in partially saturated soil. Trans Porous Media
1986;1:22–55.
[38] Verburg K, Ross PJ, Bristow KL. SWIMv2.1 user manual,
Divisional Report 130. CSIRO Division of Soils, 1996.
[39] Wilson DJ, Western AW, Grayson RB, Berg AA, Lear MS,
Rodell M, et al. Spatial distribution of soil moisture over 6 and
30 cm depth, Mahurangi river catchment, New Zealand. J Hydrol
2003;276:254–74.