Advances in Water Resources 23 (2000) 799810

www.elsevier.com/locate/advwatres

On the physical geometry concept at the basis of space/time

geostatistical hydrology
G. Christakos a,*, D.T. Hristopulos a, P. Bogaert b
a
Environmental Modeling Program, Department of Environmental Science and Engineering, Center for the Advanced Study of the Environment,
University of North Carolina at Chapel Hill, 111 Rosenau Hall, CB#7400, Chapel Hill, NC 27599-7400, USA
b
Facult
e des Sciences Agronomiques, Unite de Biom
etrie et Analyse des Donn
ees, Universit
e Catholique de Louvain, Louvain-la-Neuve, Belgium
Received 4 November 1999; received in revised form 17 March 2000; accepted 17 March 2000

Abstract
The objective of this paper is to show that the structure of the spatiotemporal continuum has important implications in practical
stochastic hydrology (e.g., geostatistical analysis of hydrologic sites) and is not merely an abstract mathematical concept. We
propose that the concept of physical geometry as a spatiotemporal continuum with properties that are empirically dened is im-
portant in hydrologic analyses, and that the elements of the spatiotemporal geometry (e.g., coordinate system and space/time metric)
should be selected based on the physical properties of the hydrologic processes. We investigate the concept of space/time distance
(metric) in various physical spaces, and its implications for hydrologic modeling. More specically, we demonstrate that physical
geometry plays a crucial role in the determination of appropriate spatiotemporal covariance models, and it can aect the results of
geostatistical operations involved in spatiotemporal hydrologic mapping. 2000 Elsevier Science Ltd. All rights reserved.

Keywords: Spatiotemporal; Random eld; Hydrology; Mapping; Geostatistics

1. Introduction metric structures. We will show that these features can

have important consequences in geostatistical analysis
Spatiotemporal random eld (S/TRF) modeling of and mapping of hydrologic processes.
hydrologic phenomena has led to considerable advances
over the last few decades, e.g., [3,30,34,35]. The fol-
lowing question can now be posed: Which are the fun-
2. Spatiotemporal continuum and its physical geometry
damental concepts responsible for the success of S/TRF
modeling? From our perspective, there are three fun-
The majority of applied scientists today view space/
damental concepts [8]: (a) the spatiotemporal continuum
time as a continuous spatial arrangement combined with
concept (i.e., a set of points associated with a continuous
a temporal order of events. In other words, space rep-
spatial arrangement of events combined with their
resents the order of coexistence and time represents the
temporal order), (b) the eld concept (which associates
order of successive existence. In the natural sciences,
mathematical entities scalar, vector, or tensor with
space/time is viewed as the union of space and time,
space/time points), and (c) the complementarity concept
dened in terms of their Cartesian product. Spatiotem-
(according to which uncertainty manifests itself as an
poral continuity implies an integration of space with
ensemble of possible eld realizations that are in
time and is a fundamental property of the mathematical
agreement with what is known about the hydrologic
formalism of natural phenomena [6]. The continuum
phenomenon of interest). In this work, we will discuss
idea implies that continuously varying spatiotemporal
certain features of the spatiotemporal continuum con-
coordinates are used to represent the evolution of a
cept (a), including suitable coordinate systems and
system's properties. The operational importance of the
spatiotemporal continuum concept is its book-keeping
*
Corresponding author. Tel.: +1-919-966-1767; fax: +1-919-966- eciency that permits ordering hydrologic measure-
7911. ments and establishing relations among them by means
E-mail address: george_christakos@unc.edu (G. Christakos). of physical theories and mathematical expressions. This
0309-1708/00/$ - see front matter 2000 Elsevier Science Ltd. All rights reserved.
PII: S 0 3 0 9 - 1 7 0 8 ( 0 0 ) 0 0 0 2 0 - 8
800 G. Christakos et al. / Advances in Water Resources 23 (2000) 799810
description of space/time suces for data analysis and
mapping of macroscopic processes in hydrologic geo-
statistical applications.
The systematic study of a spatiotemporal continuum
requires the introduction of two important entities: (a) a
suitable coordinate system with a measure of space/time
distance (metric), and (b) models and techniques that
establish linkages between spatiotemporally distributed
hydrologic data. These entities require the development
of a physical geometry model, i.e., a spatiotemporal
continuum that has a structure with empirically dened
properties. In geostatistical studies of hydrologic
phenomena, one may consider dierent coordinate sys-
tems that allow representations of spatiotemporal ge-
ometry based on the underlying symmetry of the
hydrologic processes involved, the topography, etc. In
addition to coordinate systems, an important issue is the
measurement of distances (metrics) in space, or more
general, in space/time. The denition of an appropriate
metric depends on both the local properties of space and
time (e.g., the curvature of space/time) as well as on the
physical constraints imposed by the specic hydrologic
process (e.g., many-scale obstacles on fractal structures).
Mathematical models that establish linkages between
spatiotemporally distributed data include covariance
functions of various forms (ordinary and generalized
covariances, structure functions, etc.). These covariance
functions need to satisfy certain permissibility criteria
[6,8]. The permissibility conditions depend crucially on
the space/time metric, as we further discuss in Section 5.
The denition of a space/time metric is important in
formulating parametric models for these covariance
functions, which are then used in hydrologic estimation
and simulation studies. A metric may be dened ex-
plicitly or implicitly. Explicit expressions for the space/
time metric are generally obtained on the basis of
physical considerations, invariance principles, etc. If
such expressions are not available, it is still possible to
obtain the covariance functions for specic hydrologic
variables from numerical simulations or experimental
observations (variables that occur in fractal spaces are
an example of the latter).
3. Spatiotemporal coordinate systems
It is important to identify points on a continuum by
means of an unambiguous address. However, it is often
taken for granted because it seems so obvious. The in-
troduction of a coordinate system is essential in deter-
mining the `addresses' of dierent points in a
spatiotemporal continuum. Generally speaking, a co-
ordinate system is a systematic way of referring to
places, times, things and events. The choice of the co-
ordinate system depends on the pertinent information
about the system (natural laws, topographical features,
etc.), as well as on the mathematical convenience re-
sulting from a particular choice of the coordinate system
(e.g., a spherical spatial coordinate system may simplify
calculations in the case of an isotropic problem). Below,
we discuss two types of coordinate systems, Euclidean
and non-Euclidean, which are of interest in geostatistical
applications.
3.1. Euclidean coordinate systems
In the classical Euclidean space, a point p in the
spatiotemporal continuum is identied by means of the
spatial coordinates s s1 ; . . . ; sn in Rn (i.e., s 2
S  Rn ), and the time coordinate t along the time axis
T  R1 , so that
p s; t: 1
For example, the `address' of a point in an aquifer over
time is characterized by n 1 numbers (n 2 or 3) that
depend on the coordinate system. For many applica-
tions it is sucient to investigate the temporal evolution
after an initial time, set equal to zero, so that T  0; 1.
Depending on the choice of the spatial coordinates
s s1 ; . . . ; sn , Eq. (1) suggests more than one way to
dene a point in a spatiotemporal domain as described
in the following. In the commonly used Euclidean
rectangular (Cartesian) coordinate system, the si
s1 ; . . . ; sn i and ti are the orthogonal projections of a
point Pi on the spatial axes and temporal axis, respec-
tively, so that the following mapping is dened:
Pi ! si ; ti s; ti pi : 2
In an alternative notation, a point is denoted by Pij ,
where its spatial coordinates are si 2 S and its time co-
ordinate is tj 2 T , i.e., a point Pij in Cartesian space/time
is dened as
si ; tj pij ! Pij : 3
In a non-Cartesian environment, the Euclidean curvi-
linear spatial coordinates are dened by means of a
spatial transformation of the form
s1 ; . . . ; sn ; ti ;
T : si Ti  4
where the  s1 ; . . . ; sn denote the rectangular spatial co-
ordinates (note that the time coordinate tj is not aected
by the transformation). In the polar coordinate system:
n 2 and s s1 ; s2 r; h with r > 0. In cylindrical
coordinates: n 3 and s s1 ; s2 ; s3 r; h; s3 . In
spherical coordinates: n 3 and s s1 ; s2 ; s3
q; u; h. Physical data must be associated with a space/
time coordinate system that is appropriate for the ob-
served process. Geographic coordinates are used in
some water resources management systems which in-
volve the latitude / and the longitude h of a point P on
the surface of the earth (both expressed in radians). The
latitude is dened as the angle between P and the
 G. Christakos et al. / Advances in Water Resources 23 (2000) 799810
equator along the meridian (meridians are lines dened
by the intersection of the Earth's surface and any given
plane passing through the North and South poles). The
longitude is dened as the angle between the meridian
through P and the central meridian (through Green-
wich, UK) in the plane of the equator.
In practice, one may need to establish a transfor-
mation of the original coordinate system into one that
provides the most realistic representation of the hyd-
rologic phenomenon under consideration. While the
use of a specic coordinate system is determined from
the physical processes involved, the mathematical
convenience aorded by the specic system will also
play a role. For example, in the case of a hydrologic
process that has cylindrical symmetry (e.g., ow in a
well), the cylindrical coordinate system captures the
underlying symmetry and is thus more convenient for
mathematical analysis than a rectangular coordinate
system. In this case the latter is inecient, but it is not
ruled out.
3.2. Non-Euclidean coordinate systems
A rectangular Euclidean coordinate system is not
appropriate for physical processes that occur in curved
spaces. Non-Euclidean coordinate systems are not
constrained to rectangular coordinates. For a curved
two-dimensional surface, a Gaussian coordinate system
may be appropriate. In the Gaussian coordinate sys-
tem, the rectangular grid of the Euclidean space is re-
placed by an arbitrary dense grid of ordered curves
(Fig. 1) generated as follows: Fixing the value of one
coordinate, s1 or s2 , produces a curve on the surface in
terms of the free coordinate. In this way, two families
of one-parameter, non-intersecting curves are generated
on the surface. Only one curve of each family passes
through each point. The s1 -curves intersect the s2 -
curves, but not necessarily at right angles. Neither the
s1 - nor the s2 -curves are uniformly spaced. This type of
grid permits locating points, but not a direct mea-
surement of the distance between them. If a global
Fig. 1. A Gaussian coordinate system.
801
coordinate system does not suce to entirely cover a
given surface, local coordinate systems should be used
instead.
For natural processes that take place on the Earth's
surface, a Cartesian coordinate system with origin at the
center is not convenient. In addition, a rectangular grid
is not appropriate for processes aected by the earth's
curvature (see, global hydrological modeling, climatic
processes, etc.; e.g., [19,21,22]). Instead, the two-di-
mensional continuum of the earth's surface is described
by a non-Euclidean geometry of the Gaussian type.
Gaussian geometry oers an internal visualization of the
earth's surface (to visualize a surface internally is
equivalent to living on such a surface; to visualize a
surface externally is to view it from a higher dimensional
space that includes it). In this case, things are simplied
considerably by using curvilinear coordinate systems. In
this case, straight lines are replaced by arcs, for these are
the shortest distances between points (geodesics). A
triangle consists of three intersecting arcs, and the sum
of its angles is greater than 180. Every surface has a set
of properties, called intrinsic (or internal), that remain
invariant under transformations preserving the arc
length (e.g., [24]). The above example points out an
important consideration in the choice of a coordinate
system for a natural process: it is more ecient to use
internal, as opposed to external, properties of the
physical space.
Riemann generalized Gauss' analysis by introducing
the concept of a continuous manifold as a continuum of
elements, such that a single element is dened by n
continuous variable magnitudes. This denition in-
cludes the analytical conception of space in which each
point is dened by n coordinates. Since two Gaussian
coordinates, (s1 ; s2 ), are required to locate a point on a
surface in three-dimensional space, the surface is a two-
dimensional space or manifold (note that in Cartesian
coordinates a relation of the form f s1 ; s2 ; s3 0 is
required to describe such a manifold). Riemann ex-
tended Gauss' two-dimensional (n 2) surface to n-di-
mensional manifolds (n > 2) in Riemannian coordinate
systems. Thus, the Riemannian coordinate system con-
sists of a network of si -curves (i 1; . . . ; n). A detailed
mathematical presentation of the Riemannian theory of
space may be found in [5]. Other types of non-Euclidean
coordinate systems are discussed in [7], including sys-
tems of coordinates with particular physical properties
such as the geodesic, the Glebsch, and the toroidal
systems. For geostatistical applications, it is important
to realize that the Riemannian coordinates specify the
position by consistently assigning to each point on a
manifold a unique n-tuple, but they do not automati-
cally provide a measure of the distance between points.
If explicit relations or measures are required, the con-
cept of spatiotemporal metric structure should be
introduced.
802 G. Christakos et al. / Advances in Water Resources 23 (2000) 799810

4. The spatiotemporal metric structure
Central among the quantitative features of a physical
geometry is its metric structure, that is, a set of math-
ematical expressions that dene spatiotemporal dis-
tances. These expressions cannot always be dened
unambiguously. The expression for the metric in any
continuum depends on two entirely dierent factors: (a)
a relative factor the particular coordinate system; and
(b) an absolute factor the nature of the continuum
itself. The nature of the continuum is imposed by
physical constraints, such as the geometry of the space
in which a given process occurs (i.e., whether it is a
plane, a sphere, or an ellipsoid). Other constraints are
imposed by the physical laws governing the natural
processes. If a natural process takes place inside a
three-dimensional medium with complicated internal
structure, the appropriate metric for correlations is sig-
nicantly inuenced by the structure of the medium. We
further investigate this issue in relation with fractal
spaces in Section 6. Below, we discuss the separate and
the composite metric structures which are often used in
spatiotemporal geometry.
4.1. Separate metric structures
These metrics may be more convenient for geosta-
tistical applications, because they reat the concept of
distance in space and time separately. The separate
metric structure includes an innitesimally small spatial
distance jdsj P 0 and an independent time lag dt, so that
dp : jdsj; dt:
In Eq. (5), the structures of space and time are intro-
duced independently. For a xed point in space `dis-
tance' means `time elapsed', while for a xed time it
denotes the spatial `distance between two locations'. The
distance jdsj can have dierent meanings depending on
the particular topographic space used. In Euclidean
space the jdsj is dened as the length of the line segment
between the spatial locations s1 and s2 s1 ds, i.e., the
Euclidean distance in a rectangular coordinate system is
dened as
s
X n
jdsj ds2i : 6
i1
moving from point P1 to point P2 , if the particle is
constrained by the physics of the situation to move
along the sides of the grid. This distance measure is thus
more appropriate for processes that actually occur on a
discrete grid or network of some sort (this is not
necessarily true for continuous processes simulated on
numerical grids, since in this case the grid is only a
convenient modeling device and does not change the
space/time metric). We consider the impact of the metric
(7) on the permissibility of covariance functions in
Section 5, and we investigate the dierence between the
metrics of Eqs. (6) and (7) from a mapping perspective
in Section 7. Yet another distance metric jdsj is dened
by
jdsj maxjdsi j; i 1; . . . ; n: 8
The distance jdsj between two geographical locations on
the surface of the earth (considered as a sphere with
radius r) is dened by
q
jdsj r d/2 cos2 / dh2 ; 9
where d/ and dh are the latitude dierence and longi-
tude dierence, respectively (both expressed in radians).
Note that the spatiotemporal metric and the coordinate
system in which the metric is evaluated are independent.
An exception is the rectangular coordinate system, the
denition of which involves the Euclidean metric. The
following example illustrates how the metrics considered
above can lead to very dierent geometric properties of
space. In the geostatistical analysis of spatial isotropy in
5 R2 one needs to dene the set H of points at a distance
r jdsj from a reference point O. In Fig. 2 it is shown
that in the case of the metric (6) the set H is a circle of
radius r, while
p in the case of the metric (7) H is a square
with sides 2r. Note that the Hs may represent isoco-

Non-Euclidean distance measures may be more appro-

priate for particular applications. For example, the
distance between points P1 and P2 with spatial coordi-
nates s1 and s2 s1 ds, respectively, can be dened by
X
n
jdsj jdsi j: 7 Fig. 2. The set H of points at a distance r jdsj from O: (a) when r is
i1 the Euclidean distance of Eq. (6) with n 2; and (b) when r is the
absolute distance of Eq. (7) with n 2. The set H denes an isoco-
The distance measure of Eq. (7) may represent, e.g., the variance contour. Such isocovariance contours may be associated with
length of the shortest path traveled by a uid particle the spatial distribution of a hydraulic head eld in an aquifer, etc.
 G. Christakos et al. / Advances in Water Resources 23 (2000) 799810
variance contours associated with the spatial distribu-
tion of a hydraulic head eld in an aquifer, etc.
A general form for distance metrics Euclidean or
non-Euclidean can be summarized in terms of the
Riemannian distance dened as
v knowledge, the Euclidean and non-Euclidean ge-
uX
u n
jdsj t gij dsi dsj ; 10
i;j1
where gij are coecients that, in general, depend on the
spatial location. The tensor g gij is called the metric
tensor. Although from the dierential geometry view-
point the metric tensor gives innitesimal length ele-
ments, the mathematical form of Eq. (10) may be used
to dene nite distances as well (see, e.g., Eq. (22)). A
metric tensor satises certain physical and mathematical
conditions [5]. A metric of the form (10) is Euclidean if a
coordinate transformation exists such that Eq. (10) is
expressed in Cartesian form. The Euclidean metric in a
rectangular coordinate system is a special case of
Eq. (10) for gii 1 and gij 0 (i 6 j). In a polar coor-
dinate system, the metric is obtained from Eq. (10) for
n 2, g11 1, g22 s21 and gij 0 (i 6 j). Eq. (10) for
n 3, g11 g33 1, g22 s21 and gij 0 (i 6 j) provides
the metric in a cylindrical coordinate system. In a
spherical coordinate system, the metric is obtained from
2
Eq. (10) for n 3, g11 1, g22 s21 , g33 s1 sin s2 and
gij 0 (i 6 j). The metric structures of Gaussian and
Riemannian coordinate systems are also represented by
means of Eq. (10). For n 2, Eq. (10) gives the local
distance on a curved surface (e.g., a hill); the metric co-
ecients gij are functions of the spatial coordinates si
(i 1; 2), and g12 g21 . Thus, the curvature of a Gauss-
ian (or Riemannian) surface is reected in the metric.
4.2. Composite metric structure
A composite metric structure requires a higher level
of physical understanding of space/time, which may in-
volve theoretical and empirical facts about the investi-
gated hydrologic process. The metric is determined by
the geometry of space/time and also by the physical
processes and the space/time structures that they gen-
erate. This is expressed by the following denition: In
the composite metrics the structure of space/time is in-
terconnected by means of an analytical expression, i.e.,
dp : jdpj gds1 ; . . . ; dsn ; dt;
where g is a function determined by the available
physical knowledge (topography, physical laws, etc.;
[7]). Consider, e.g., a point P in the space/time contin-
uum R2  T with coordinates p s1 ; s2 ; t. A hydrologic
process that varies within this continuum is denoted by
X p X s1 ; s2 ; t. If! the separate metric structure is
used, the distance jOP j is dened in terms of two inde-
pendent space and time distances forming the pair
803
jsj; t, where jsj has one of the spatial forms discussed
above. If, however, the composite metric structure is
used, the function g should be determined by means of
the dynamic structure of the hydrologic process
X s1 ; s2 ; t. Concerning the representation of physical
ometries display important dierences. Euclidean ge-
ometry determines the metric, which constrains the
physics. In this case, a single coordinate system implying
a specic metric structure covers the entire spatiotem-
poral continuum. Non-Euclidean geometries clearly
distinguish between the spatiotemporal metric and the
coordinate system, thus allowing for choices that are
more appropriate for certain physical problems.
In several problems the separate metric structure (5)
is adequate. In other cases, however, the more involved
composite structure (11) is necessary. In the latter case,
considering the several existing spatiotemporal ge-
ometries that are mathematically distinct but a priori
and generically equivalent, the spatiotemporal metric
structure (i.e., function g) that best describes physical
reality must be determined. Mathematics describes the
possible geometric spaces, and empirical knowledge
determines which best represents the physical space.
Axiomatic geometry is not sucient for physical appli-
cations in space/time, and it is required to establish a
relationship between the geometric concepts and the
empirical investigation of space/time as a whole. The
term `empirical' includes all available physical knowl-
edge bases (observational data, covariance functions,
physical laws, etc.). A special case of Eq. (11) is the
space/time generalization of the distance (10) that leads
to the spatiotemporal Riemannian metric
v
uX X
u n n
jdpj t gij dsi dsj 2dt g0i dsi g00 dt2 ; 12
i;j1 i1
where the metric coecients gij (i; j 1; . . . ; n) are
functions of the spatial location and time.
We can learn about the nature of the spatiotemporal
continuum by studying the characteristics of the physi-
cal system it describes. Hydrologic processes are subject
to constraints imposed in the form of physical laws.
Assume that the distribution of a hydrologic eld X p is
expressed by the law
X p Lm; BC; IC; p; 13
11
where m m1 ; . . . ; mk are known coecients, BC and IC
are given boundary and initial conditions, p are space/
time coordinates, and L
is a known mathematical
functional. The law (13) can play an important role in
the determination of a physically consistent spatiotem-
poral metric form. Often Eq. (13) leads to an explicit
expression for the metric
jdpj jp0 pj gv0 ; v; m; BC; IC; 14
804 G. Christakos et al. / Advances in Water Resources 23 (2000) 799810
where the g
has a functional form that depends on the
L
-operator. Eq. (14) restricts the number of possible
metric models. It determines the metric jdpj of the space/
time geometry from the hydrologic eld values at p and
p0 , the coecients m, the BC, and the IC. Assuming that
Eq. (14) is valid, one cannot specify both the spatio-
temporal metric and the hydrologic eld values inde-
pendently, since they are connected via Eq. (14). In some
other cases, the metric form is obtained indirectly from
the eld equations. This happens if the solution of the
physical law is such that
X p Xgs1 ; . . . ; sn ; t: 15
Solution (15) puts restrictions on the geometrical fea-
tures of space/time and suggests a metric of the form
jpj gs1 ; . . . ; sn ; t, where jpj denes the space/time
distance from the origin. It is possible that the physical
law could lead to a solution (14) that oers information
about the coecients gij of the metric (11). These
possibilities are demonstrated with the help of the fol-
lowing examples. Consider the hydraulic head eld
hs1 ; s2 , the spatial distribution of which is governed by
the Laplace equation
r2 hs1 ; s2 0:
In the case of radial ow, Eq. (16) admits a solution of
p
 1 ; s2 h
the form hs1 ; s2 hgs  s2 s2 . Hence,
1 2
the spatial metric suggested by Eq. (16) is the Euclidean
p
jsj s21 s22 . In the case of two-phase ow in a porous
domain the governing equations for phases a ( water
and oil) are [9]
dfa
/ea ; K a fa 0;
dla
where ea is the direction vector of the a-owpath tra-
jectory, fa the magnitude of the pressure gradient in the
direction ea , Ka denotes the intrinsic permeabilities of the
phases, and / is a function of ea and Ka . The solution of
Eq. (17) is of the form fa fa jsj, where the corre-
sponding metric jsj la is the distance along the a-
owpath. Next, let us assume that the geophysical eld
X s1 ; s2 ; t is governed by the ux-conservative equation
oX =ot m
rX 0;
where m m1 ; m2 is an empirical velocity to be deter-
mined from the data. By means of a coordinate trans-
formation from the rectangular Euclidean system (si ) to
the system of coordinates dened by si si mi t, the
solution of Eq. (18) has the form
X s1 ; s2 ; t Xs1 m1 t; s2 m2 t; 19
i.e., it depends on the space/time vector p s mt.
Therefore, in the rectangular coordinate system a geo-
physical eld governed by Eq. (18) may have a metric of
the Riemannian form (12), where n 2, g00 m21 m22 ,
g11 g22 1, g10 g01 2m1 , g20 g02 2m2 , and
g12 g21 g01 g10 g02 g20 0. This expression
demonstrates how the physical law determines the geo-
T
metric metric through the empirical vector m m1 ; m2 .
The adoption of the spatiotemporal metric above could
be usefully exploited in Eulerian/Lagrangian schemes of
hydrodynamics. Below, we discuss how the covariance
function can be instructive in determining the appro-
priate geometry in a spatiotemporal continuum.
Geostatistical analysis usually includes a covariance
model tted to the data or derived from a physical
model. The covariance can be helpful in determining the
space/time geometry. In particular, the form of the
metric k is sought such that
cx h1 ; . . . ; hn ; s cx k: 20
The metric may be viewed as a transformation
k T h1 ; . . . ; hn ; s of the original coordinate system,
where T has a Riemannian structure and the forms of
the coecients gij are sought on the basis of physical
and mathematical facts. In particular, let k be of the
form
v
uX X
u n n
kh; s t gij hi hj 2s g0i hi g00 s2 : 21
i;j1 i1
16
While the nite space/time distance (21) has the same
form as the innitesimal Riemannian distance (12), the
gij s do not necessarily coincide with the metric coef-
cients of (12). In Eq. (21) the gij denote functions of
the spatial and lag distances rather than the local co-
ordinates, that is gij gij hi ; hj , g0i g0i s; hi , i; j
1; . . . ; n, and g00 g00 s. Clearly, the determination of
17 the gij may require additional assumptions based on
theoretical and experimental facts. If the gij are space
and time independent, Eqs. (20) and (21) give the fol-
lowing set of equations:
Pn
ocx =ohi j1 gij hj g0i s
Pn
ocx =ohj i1 gij hi g0j s
and 22
Pn
ocx =ohi j1 gij hj g0i s
Pn :
ocx =os i1 g0i hi g00 s
18
For illustration, consider a covariance function in
R1  T that satises the following physical model:
ocx =oh h
2 ; 23
ocx =os m s
where h Ds1 s01 s1 > 0, s Dt t0 t > 0, m a=b,
a and b are empirical covariance coecients. Note that
determining the covariance from physical equations,
whenever possible, avoids common problems of empir-
ical covariance estimation, and eliminates the circular
problem of standard geostatistics (i.e., estimating the
covariance from the same dataset that is also used to
obtain the kriging estimates). We seek a metric form
 G. Christakos et al. / Advances in Water Resources 23 (2000) 799810 805

k such that cx h; s cx k. In view of Eqs. (22) and (23),
the metric coecients are such that
g11 h g01 s h a separable space/time covariance, i.e., cx h; s
2 :
g01 h g00 s m s
Therefore, a geometric space/time metric that satises
the last relationship and, thus, is consistent with the
physical equation (23) is of the form (21) with n 1,
g00 m2 , g01 0 and g11 1; i.e.,
p p
kh; s h2 m2 s2 h2 a2 s2 =b2 : 25
Eq. (25) shows how the covariance coecients deter-
mine the spatiotemporal metric. A function which is a
permissible covariance model and has a metric of the
form (25) is cx h; s c0 exp h2 m2 s2 . In light of the
above analysis, the choice of a space/time geometry
must be compatible with the `natural' geometry as
revealed by the physical equations.
5. Spatiotemporal geometry and permissibility criteria
The choice of the spatiotemporal geometry has sig-
nicant consequences in geostatistical analysis. One
such consequence is related to the permissibility of a
covariance model cx h; s in Rn  T : The permissibility
criteria that determine if a function can be used as a
covariance, semivariogram, generalized covariance, etc.
model depend on the assumed metric structure. In-
deed, a covariance that is permissible for one spatio-
temporal geometry may not be permissible for another
geometry. According to Bochner's theorem (e.g., [8]) a
necessary and sucient condition for a spatiotemporal
function cx h; s to be permissible is that its spectral
density
Z Z
c~x k; x dh dseik
hxs cx h; s 26
means of a numerical calculation of the Fourier trans-
form that gives the spectral density. Since Eq. (27) is
24 cx hc0x s, we focus on the spatial component cx h. The
covariance cx h is related to the spectral density c~x k as
follows:
Z
c~x k dh exp ik
hcx h: 29
R2
Hence, cx h is a permissible covariance if the c~x k is
non-negative. In the case of the spatial component
2
cx h c0 exp jhj of the covariance (27), the spectral
density (29) is negative in parts of the frequency domain
(see Fig. 3). We have calculated the Fourier transform
using a GaussLegendre quadrature method [31] with 80
abscissas in each direction. This involves a total of
25,600 function evaluations using double-precision
arithmetic. The Fourier transform exhibits negative
valleys near the corners of the spatial frequency domain.
We used dierent numbers of abscissas to verify that the
negative areas are true features of the Fourier transform
and not artifacts of the numerical integration due to
oscillations of the integrand. We have also veried that
the FT is accurate using the MATLAB double integra-
tion function `dblquad' with the adaptiverecursive
Newton Cotes algorithm that allows relative and abso-
lute error control (we set both to 1
E 5). Thus, the

be a real-valued, integrable and non-negative function

of the spatial frequency k and the temporal frequency x.
An important issue is whether the type of the coordinate
system or the distance metric considered modify the
permissibility of a function.
As we saw above, in relation with Eq. (20), the co-
variance model is a function of the spatiotemporal
metric, which may have a variety of forms (Euclidean or
non-Euclidean). In the following, we will show that the
spatiotemporal metric aects the permissibility of the
covariance model. For example, in R2  T the Gaussian
function
cx h; s c0 expjhj2 m2 s2 ; 27
where the spatial distance is dened as
jhj jh1 j jh2 j 28
Fig. 3. The Fourier transform c~x k of Eq. (29) using the metric of
is not a permissible covariance model (a mathematical Eq. (28). Note the islands of negative values at the four corners of
proof can be found in [7]). This result is veried by the frequency domain.
806 G. Christakos et al. / Advances in Water Resources 23 (2000) 799810
Gaussian function (27) is not a permissible covariance
for the distance metric (28), even though it is permissible
for the Euclidean metric. Next, we consider the expo-
nential function in R2  T
cx h; s c0 expjhj ms; 30
where the spatial distance is dened as in Eq. (28). The
spectral density c~x k; x of the function (30) is non-
negative for all k 2 R2 , x 2 R1 . Hence, the exponential
covariance is permissible for the metric of Eq. (28).
In conclusion, the permissibility of a covariance
model cx h; s with respect to the Euclidean metric does
not guarantee its permissibility for a non-Euclidean
metric. Hence, the permissibility of each model cx h; s
must be tested with respect to the corresponding non-
Euclidean metric.
6. Fractal geometry
Many physical processes that take place in non-uni-
form spaces with many-scale structural features (e.g.,
within porous media) are better represented by fractal
rather than Euclidean geometry. In fractal spaces it is
not always possible to formulate explicit metric expres-
sions, such as Eq. (12), since the physical laws may not
be available in the form of dierential equations. Geo-
metric patterns in fractal space/time are self-similar (or
statistically self-similar in the case of random fractals)
over a range of scales (e.g., [11,26]). Self-similarity im-
plies that fractional (fractal) exponents characterize the
scale dependence of geometric properties.
A common example is the percolation fractal (e.g.,
[11,37]) generated by the random occupation of sites or
bonds on a discrete lattice. In the site percolation model,
each site is occupied with probability p and empty with
probability 1 p. Similarly, in the bond percolation
model, conducting and non-conducting bonds are ran-
domly assigned with probabilities p and 1 p. The
medium is permeable if p > pc , where pc is a critical
threshold that depends on the connectivity and dimen-
sionality of the underlying space (for a table of pc values
on dierent lattices see [20]). The percolation model has
applications in many environmental and health pro-
cesses that occur at various scales. These applications
include single and multiphase ow in porous media
[1,2,12,15], the geometry [27,32] and the permeability of
hard and fractured rocks [23,25,28,39,40]). Percolation
models are also used to model the spread of forest res
and epidemics [18,33], tumor networks [14], and anti-
genantibody reactions in biological systems [38].
Length and distance measures on a percolation clus-
ter, denoted by lr, scale as power laws with the Eu-
clidean (linear) size of the cluster. Power-law functions
are called fractal if the scaling exponents are non-in-
teger. The fractal functions are homogeneous (e.g., [4]),
i.e., they satisfy
lbr bd0 lr; 31
where r is the appropriate Euclidean distance, d0 the
fractal exponent for the specic property, and b a scaling
factor. In practice, scaling relations like Eq. (31) only
hold within a range of scales bounded by lower and
upper cutos. For a small change db, the scaling factor
becomes b0 b db and Eq. (31) leads to
d
lb0 r b db 0 lr: 32
Expanding both sides around b 1, we obtain
dlr lr
d0 : 33
dr r
By integrating Eq. (33), we obtain
 d0
lr r
: 34
lrco rco
where rco is the lower cuto for the fractal behavior. For
example, the length of the minimum path on a perco-
lation fractal scales as lmin r / rdmin , where r denotes the
Euclidean distance between the points. The fractal di-
mension dmin of the percolation fractal on a hypercubic
lattice satises 1 6 dmin 6 2, where dmin 1:1; 1:3 for
d 2; 3 [17,36]. Thus, if the minimum path length be-
tween two points at Euclidean distance r is on average
2 miles, the length of the minimum path between two
points separated by 2r is, on average, more than 4 miles.
In Fig. 4, we show the minimum path length between
two points separated by a Euclidean distance r in Eu-
clidean space (curve 1) and in a fractal space with
d0 1:15 (curve 2). The path length in the Euclidean
space is a linear function of the distance between the two
points, for all types of paths (e.g., circular arcs, or linear
segments). The fractal path length increases nonlinearly,
because the fractal space is non-uniform and obstacles
to motion occur at all scales.
Space/time covariance functions in fractal spaces
have dynamic scaling forms (e.g., [10]) that can be quite
dierent than Euclidean covariance functions. The self-
similarity of fractal processes implies that covariance
functions decay as power laws. This means that the tail
of the covariance function carries more weight than the
tail of short-ranged models (e.g., exponential, Gaussian,
spherical, etc.) An example of a fractal process that
generates power-law correlations is invasion percolation
(e.g., [13]) in which a defending uid (e.g., oil) is dis-
placed from a porous medium saturated by an invading
uid (e.g., water).
Below, we investigate an example of a composite
space/time covariance model for fractal spaces. Within
the fractal range, we consider a covariance function of
the form
 G. Christakos et al. / Advances in Water Resources 23 (2000) 799810
Fig. 4. Minimum path length between two points separated by a
Euclidean distance r in Euclidean space (curve 1) and in a space with
fractal length dimension d0 1:15 (curve 2).
cx h; s / ra s=rb z ;
where r jhj, s0  s  sm and r0  r  rm dening the
space/time fractal ranges. General permissibility con-
ditions for unbounded fractals [20], i.e., s0 r0 0 and
rm sm 1 impose certain constraints on the expo-
nents a, b and z. The permissibility conditions can be
relaxed by using nite cutos. In addition, cutos ensure
that cx h; s tends to a nite variance at zero lag and
drops o faster than a power-law for lags that exceed the
cutos. As we discussed above, Bochner's theorem re-
quires that the spectral density c~x k; x be a monotoni-
cally decreasing function of bounded variation. The
spectral density of the function (35) is given by
Z Z
c~x k; x dh eik
h rabz ds eixs sz : 36
It follows that the permissibility conditions are
1 < z < 0 and n 1=2 < a bz < 0. If b > 0, the
last inequality implies that a < 0.
As is shown in Appendix A, a covariance function
that has the fractal behavior of Eq. (35) and a nite
variance r2 is given by
cx h; s; uc ; wc r f^z s=r ; uc f^a r; wc ;
2 b
where r2 is the variance. The covariance function is
plotted in Fig. 5 for r 1, z a 1=2, b 1:1, and
cutos uc 25; wc 25. The axes used in the picture
are r and s=rb . The function f^z s=rb ; uc has an unusual
dependence on the space and time lags through s=rb .
807
Fig. 5. Plot of the fractal correlation function of Eq. (37) for
z a 1=2 and b 1:1. The correlation function is dened as the
covariance normalized by the variance.
35
For large s, the ratio s=rb is close to zero if r is su-
ciently large, and the value of the function f^z s=rb ; uc is
close to one. With regard to f^z s=rb ; uc two pairs of
space/time points are equidistant if s1 =r1b s2 =r2b .
Hence, the equation for equidistant space/time contours
is s=rb c. This dependence is physically quite dierent
than the one implied by, e.g., a Gaussian space/time
covariance function. In the latter, equidistant lags satisfy
the equation r2 =n2r s2 =n2s c. The dierence is shown
in Fig. 6, in which we plot the equidistant space/time
contours for f^z s=rb ; uc (solid lines) and for
exp r2 =n2r s2 =n2s (dots) as a function of the space
and time lags. The contour labels represent the values of
c0 s=rb (solid lines), and r2 =n2r s2 =n2s (dots), obtained
using c0 62:95, nr 10 and ns 5.
7. Metric structure and spatiotemporal mapping
Space/time estimation and simulation depend on the
metric structure assumed, since the covariance (or
37
semivariogram, etc.) are used as inputs in most mapping
techniques (e.g., kriging estimation of precipitation
distribution, turning bands simulation of hydraulic
conductivity). Hence, the same dataset can lead to dif-
ferent space/time maps if estimation is performed using
dierent metric structures.
808 G. Christakos et al. / Advances in Water Resources 23 (2000) 799810

Fig. 6. Equidistant contours for fractal space/time dependence (solid

contours) and for Gaussian dependence (dotted contours).

For example, consider a two-dimensional eld

X s1 ; s2 with a constant mean and an exponential
covariance

cx h exp1:5jhj; 38

where h h1 ; h2 . The X s1 ; s2 may denote, e.g., the

concentration of a groundwater pollutant. Since the
subsurface is a medium with complicated internal
structure, it is likely that a non-Euclidean metric is a
more appropriate measure of distance. The metric
should in principle be derived based on a physical model
of the subsurface medium and the dynamics of trans-
port. For the sake of illustration, assume that the ap-
propriate metric for this eld is the non-Euclidean form
jh1 j jh2 j. Spatial estimates using this metric were gen- Fig. 7. Maps obtained using: (a) the appropriate non-Euclidean
erated on the basis of a hard dataset vhard using a geo- metric; (b) the inappropriate Euclidean metric.
statistical kriging technique [29]. This led to the contour
map of Fig. 7(a). Practitioners of geostatistics or spatial
statistics often favor a theory-free approach which fo- quate choice of metric, while the correct one (Fig. 7(a))
cuses solely on the dataset available and ignores physical accounts for the underlying physical geometry.
models. The standard commercial software for geo-
statistics
p restricts the user to the Euclidean metric
h21 h22 for covariance estimation and kriging. If this 8. Discussion and conclusion
metric were used, the same dataset vhard as above results
in the contour map in Fig. 7(b). As expected, the two In this paper, we investigated the important role of
maps show considerable dierences. The Euclidean- space/time coordinate systems and distance metrics in
based map (Fig. 7(b)) assumes a convenient but inade- the geostatistical analysis of hydrologic systems. In
 G. Christakos et al. / Advances in Water Resources 23 (2000) 799810 809

particular, we presented several examples for metrics
and covariance functions in Euclidean and non-Euclid-
ean spaces, including fractal spaces. We showed that
these metrics can lead to very dierent geometric prop-
erties of space. We also showed that covariances which
are permissible for one type of metric are not de facto
permissible for a dierent metric. We investigated a
composite fractal covariance model with a new de-
nition of space/time metric. A characteristic of this co-
variance function is that the correlations decay
asymptotically much slower than short-range models
with the usual Euclidean metric. Finally, we showed that
under dierent assumptions about the type of the met-
ric, the same dataset can lead to very dierent maps of
the hydrologic processes under consideration. In such
cases, the physical models governing the hydrologic
processes could play a pivotal role in determining the
appropriate space/time metric. These considerations
also imply that users of commercial geostatistical soft-
ware should be aware of the limitations of these pack-
ages. One of these limitations, discussed in this work, is
that the Euclidean metric is chosen by default regardless
of the physical situation.
Acknowledgements
We would like to thank the four anonymous ref-
erees for their helpful comments. We are also grateful
to Dr. Cass T. Miller for reading an earlier version of
the paper and making valuable observations. This
work has been supported by grants from the Army
Research Oce (Grant nos. DAAG55-98-1-0289 and
DAAH04-96-1-0100), the Department of Energy (Grant
no. DE-FC09-93SR18262), and the National Institute
of Environmental Health Sciences (Grant no. P42
ES05948-02).
This function has power-law behavior for x  xc  1=yc ,
and at the same it is nite at the origin. The power-law
behavior can be shown as follows: For xyc  1, Eq. (A2)
can be replaced with Eq. (A1) without considerable er-
ror, due to the fact that the exponential term in the in-
tegral is essentially zero. Then, it is straightforward to
show by a change of variables x ! kx in Eq. (A1) that
fm kx; yc km fm x; yc , which characterizes a power law
with exponent m. We express fm x; yc as fm x; yc
xm cm; xyc =Cm, where c denotes the incomplete
gamma function [16]. The integrand of Eq. (A2) has an
integrable singularity at y 0. By means of the trans-
formation y 1=y 0 , we avoid the singularity and obtain
Z 1
1 0
fm x; yc dy 0 y 0m1 ex=y : A3
Cm 1=yc
Eq. (A3) is more convenient than (A2) for numerical
calculations. In view of (A3), the value at the origin is
1
fm 0; yc mCm ycm . In Fig. A1, we plot the nor-
malized function
f^m x; yc fm x; yc =fm 0; yc A4
for yc 1 and the exponential function exp x=n with
n 6:5. Note that the power-law decays asymptotically
much slower than the exponential. In Fig. A2, we plot
f^m x; yc for dierent values of 1=yc . Increasing yc (that is,

Appendix A

Our aim is to construct a fractal covariance that has a

nite variance and behaves like Eq. (35) within a fractal
range. A power-law function of a general argument x
(where x stands for the spatial lag, the time lag, or some
combination thereof) with a negative exponent m < 0
can be expressed as
Z 1
1
xm dy eyx y m1 ; A1
Cm 0
where C is the gamma function. Because m < 0, the
function xm is singular at r 0. The singularity is tamed
by imposing an upper cuto yc on the integral, thus
leading to the function
Z yc Fig. A1. Plot of the fractal correlation function f^m x; yc with m 1=2
1
fm x; yc dy eyx y m1 : A2 and yc 1 (solid) and the exponential correlation function expx=n
Cm 0 with n 6:5 (dots) vs. the lag x.
810 G. Christakos et al. / Advances in Water Resources 23 (2000) 799810
Fig. A2. Plot of the function f^m x; yc with m 1=2, vs. the lag x for
three dierent values of the cuto 1=yc .
decreasing 1=yc ) leads to a steeper slope of f^m x; yc at the
origin. Based on the above, the covariance function
given by Eq. (37) has the fractal behavior of Eq. (35) and
a nite variance r2 .
References
[1] Acuna JA, Yortsos YC. Application of fractal geometry to the
study of networks of fractures and their pressure gradients. Water
Resour Res 1995;31(3):52740.
[2] Berkowitz B, Balberg I. Percolation theory and its application to
groundwater hydrology. Water Resour Res 1993;29(4):77594.
[3] Bell TL. A space-time stochastic model of rainfall for satellite
remote-sensing studies. J Geophys Res 1987;92(D8):963143.
[4] Bunde A, Havlin S, Fractals and disordered systems. New York:
Springer, 1996.
[5] Chavel I. Riemannian geometry: a modern introduction. New
York: Cambridge University Press, 1995.
[6] Christakos G. Random eld models in earth sciences. San Diego,
CA: Academic Press, 1992.
[7] Christakos G. Modern spatiotemporal geostatistics. New York:
Oxford University Press, 2000.
[8] Christakos G, Hristopulos DT. Spatiotemporal environmental
health modelling: a tractatus stochasticus. Boston, MA: Kluwer
Academic Publishers, 1998.
[9] Christakos G, Hristopulos DT, Li X. Multiphase ow in
heterogeneous porous media from a stochastic dierential ge-
ometry viewpoint. Water Resour Res 1998;34(1):93102.
[10] Family F. Dynamic scaling and phase transitions in interface
growth. Physica A 1990;168:56180.
[11] Feder J. Fractals. New York: Plenum Press, 1988.
[12] Ferer M, Sams WN, Geisbrecht RA, Smith DH. Crossover from
fractal to compact ow from simulations of two-phase ow with
nite viscosity ratio in two-dimensional porous media. Phys Rev E
1993;47(4):271322.
[13] Furuberg L, Feder J, Aharony A, Jossang T. Dynamics of
invasion percolation. Phys Rev Lett 1988;61(18):211720.
[14] Gazit Y, Berk DA, Leunig M, Baxter LT, Jain RK. Scale
invariant behavior and vascular network formation in normal and
tumor tissue. Phys Rev Lett 1995;75(12):242831.
[15] Guyon E, Mitescu CD, Hulin JP, Roux S. Fractals and
percolation in porous media and ows. Physica D 1989;38:1728.
[16] Gradshteyn IS, Ryzhik IM. Table of integrals, series and
products. New York: Academic Press, 1994.
[17] Hermann HJ, Stanley E. J Phys A 1988;21:L829833.
[18] Hermann HJ. Growth: An introduction. In: Stanley H, Ostrowsky
N, editors. Growth and form. Boston, MA: Kluwer Academic
Publishers, 1986. p. 320.
[19] Houghton JT. The physics of atmospheres. Cambridge: Cam-
bridge University Press, 1991.
[20] Isichenko MB. Percolation, statistical topography, and transport
in porous media. Rev Mod Phys 1992;64(4):9611043.
[21] Jones JAA. Global hydrology. Essex, UK: Longman, 1997.
[22] Kalma JD, Sivapalan M, Scale issues in hydrological modelling.
New York: Wiley, 1995.
[23] Katz AJ, Thompson AH. Quantitative prediction of rock per-
meability in porous rock. Phys Rev B 1986;34(11):817981.
[24] Lipschutz M. Dierential geometry. New York: McGraw-Hill,
1969.
[25] Makse HA, Davies GW, Havlin S, Ivanov PCh, King PR, Stanley
HE. Long-range correlations in permeability uctuations in
porous rock. Phys Rev E 1996;54(4):312934.
[26] Mandelbrot BB. The fractal geometry of nature. New York:
Freeman, 1982.
[27] Margolin G, Berkowitz B, Scher H. Structure, ow and general-
ized conductivity scaling in fracture networks. Water Resour Res
1998;34(9):210321.
[28] Mourzenko VV, Thovert JF, Adler PM. Percolation and conduc-
tivity of self-ane fractures. Phys Rev E 1999;59(4):426584.
[29] Olea RA. Geostatistics for engineers and earth scientists. Boston,
MA: Kluwer Academic Publishers, 1999.
[30] Polyak I, North GR, Valdes JB. Multivariate spacetime analysis
of PRE-STORM precipitation. J Appl Meteor 1994;33:107987.
[31] Press WH, Flannery BP, Teukolsky SA, Vettering WT. Numerical
recipes. Cambridge: Cambridge University Press, 1986.
[32] Radlinski AP, Radlinska EZ, Agamalian M, Wignall CD, Linder
P, Randl OG. Fractal geometry of rocks. Phys Rev Lett
1999;82(15):307881.
[33] Rhodes CJ, Anderson RM. Power laws governing epidemics in
isolated populations. Nature 1996;381:6002.
[34] Rodriguez-Iturbe I, Mejia JM. The design of rainfall networks in
time and space. Water Resour Res 1974;10(4):71328.
[35] Rouhani S, Hall TJ. Space-time kriging of groundwater data. In:
Armstrong M, editor. Geostatistics. Boston, MA: Kluwer Aca-
demic Publishers, 1989;2:63951.
[36] Stanley E. An introduction to self-similariry and fractal behavior.
In: Stanley H, Ostrowsky N, editors. Growth and form. Boston,
MA: Kluwer Academic Publishers, 1986. p. 2153.
[37] Stauer D, Aharony A. Introduction to percolation theory.
London, UK: Taylor & Francis, 1992.
[38] Stauer D. Celluar automata. In: Bunde H, Havlin S, editors.
Fractals and disordered sytems. New York: Springer, 1996,
p. 35159.
[39] Thompson AH, Katz AJ, Krohn CE. The microgeometry and
transport properties of sedimentary rock. Adv Phys 1987;36:
62594.
[40] Wong P, Koplik J, Tomanic JP. Conductivity and permeability of
rocks. Phys Rev B 1984;30(11):660614.