WATERRESOURCES

RESEARCH,
VOL.28,NO.7, PAGES
1955-1974,
JULY1992

A CriticalReviewof DataonField-Scale
Dispersion
in Aquifers
LYNN W. GELHAR

Department
ofCivilEngineering,
Massachusetts
Institute
of Technology,
Cambridge
CLAIRE WELTY

Department
of CivilandArchitectural
Engineering,
DrexelUniversity,
Philadelphia,
Pennsylvania
KENNETH R. REHFELDT

IllinoisStateWaterSurvey,Champaign

A critical review of dispersivityobservations

from 59 differentfield siteswas developedby
compilingextensivetabulations
of information
onaquifertype,hydraulicproperties,
flowconfiguration, type of monitoringnetwork,tracer,methodof datainterpretation,overallscaleof observation
and longitudinal,horizontaltransverseand verticaltransverse
dispersivities
from originalsources.
This informationwas then usedto classifythe dispersivitydatainto three reliabilityclasses.Overall,
the data indicatea trend of systematicincreaseof the longitudinaldispersivitywith observationscale
but the trend is much less clear when the reliability of the data is considered.The longitudinal

dispersivities
ranged
from10-2 to 104 m forscales
ranging
from10-1 to 105m, butthelargest
scale
for high reliability data was only 250 m. When the data are classifiedaccordingto porousversus
fractured media there doesnot appearto be any significantdifferencebetweentheseaquifertypes. At
a given scale, the longitudinaldispersivityvaluesare foundto rangeover 2-3 ordersof magnitudeand
the higherreliability data tend to fall in the lower portionof thisrange.It is not appropriateto represent
the longitudinaldispersivitydata by a singleuniversalline. The variationsin dispersivityreflect the
influence of differing degrees of aquifer heterogeneityat different sites. The data on transverse
dispersivitiesare more limitedbut clearlyindicatethat verticaltransversedispersivities
are typically
an order of magnitudesmaller than horizontaltransversedispersivities.Reanalysesof data from
severalof the field sitesshowthat improvedinterpretationsmostoftenlead to smallerdispersivities.
Overall, it is concludedthatlongitudinaldispersivities
in thelowerpart of the indicatedrangearemore
likely to be realistic for field applications.

meaningfulif their variable quality was recognized, we

assembledthe dispersivitydata alongwith related informaThephenomenon
of dispersive
mixingof solutesin aquifers tion from the originalsourcesand evaluatedthe reliability or
hasbeenthe subjectof considerable
researchinterestoverthe qualityof thesedata [Getharet al., 1985].The graphical
past10 years. Characterizingthe dispersivityat a particular resultsof that work have been widely used by both practifieldsiteis essentialto any effortin predictingthe subsurface tioners and theoreticians,often without appropriate considmovement
and spreadingof a contaminant
plumeat that eration of the reliability of the data. For example, recent
basedon fractal concepts[Philip,
location.
Boththeoreticalandexperimental
investigations
have theoreticaldevelopments
foundthatfield-scale
dispersivities
are severalordersof.mag- 1986; Wheatcraftand Tyler, 1988; Neuman, 1990] have
nitude
greaterthanlab-scalevaluesfor the samematerial;it is relied on information similar to that in the work by Gelhar et
generally
agreedthat this differenceis a reflectionof the al. [1985]but those studiesdisregardedthe issueof the
INTRODUCTION

influence
of naturalheterogeneities
whichproduceirregular
flow patternsat the field scale. Consequently,laboratory
measurements
of dispersivity
cannotbe usedto predictfield
valuesof dispersivity.Insteadfield-scaletracertestsare sometimesconducted
to estimatedispersivity
at a particular
site.
Earlyeffortsto documentthe scaledependence
of dispersivity[Lallemand-Barresand Peaudecerf, 1978;Anderson,
1979;Pickensand Grisak, 1981;Beims,1983;Neretnieks,
1985]werebasedon fieldvaluesof dispersivity
reportedin
the literature and the test scales associated with those

reliabilityof the data.We feel that it is importantto update

the dispersivity
information
includingresultsfrom recent
comprehensive
fieldexperiments
andat thesametimefocus
on the interpretations
of the reliabilityof the data. With
thesegoalsin mind,thisworkdevelops
thefollowing:(1) an
outlineof the theoreticaldescriptionof dispersivemixingin

porousmedia;(2) a tabularsummary
of existingdataon
valuesof field-scaledispersivityandrelatedsiteinformation

reported
in theliterature;
(3)anevaluation
of thereliability
or qualityof thesevalues
basedonclearlydelineated
crite-

andinterpretation
of the appliedand
values.Thesestudieswere usefulin that they indeeddocu- ria;and(4) discussion
of the data.
mentedfield evidence of the scale effect, but they were theoreticalimplications

lacking
in thattheydidnotassess
thereliabilityof thedata
THEORETICAL
CONCEPTS
OF FIELD-SCALE

presented.Because we felt that the data would be more

DISPERSIVEMIXING

Copyright
1992by theAmerican
Geophysical
Union.
Papernumber92WR00607.
0043.1397/92/92
WR-00607505.00

The massbalanceequationgoverningthe transportof an

idealchemically
nonreactive
conservative
soluteby a homo1955

1956

GELHAR ET AL.: FIELD=SCALEDISPERSIONIN AQUIFERS

geneous fluid (constant density and viscosity) that flows

through a rigid saturated porous medium is commonly expressed as [e.g., Bear, 1972; de Marsily, 1986]

--+vi=
at

Oxi

Oxi

Dij

i,j=

1, 2, 3

(1)

!981;DieMin,1980].A thirdgroup[e.g.,GelharandAxness,
1983;Dagan, 1982;Neuman et al., 1987]has examinedmore
general three-dimensional heterogeneity with stochastic
methodsand concludedthe classicalequation with constant

field-scaledispersivitiesis applicableto describetransport

over largedistances.These stochasticapproachesincorp0.

rate the effects of practically unknowable small-scale variawhere c is the solute concentration, v i is the seepage tions in flow by meansof macrodispersivitieswhich are used
velocitycomponentin the xi direction,and Dij are the in a deterministictransport model describingthe large-scale
componentsof the dispersion coefficient tensor. The fight- variations in flow by means of the convection terms. None.
a field-scaledispersivity
hand side of (1) representsthe net dispersivetransportwhich theless,underwhat circumstances
is presumed to be Fickian, i.e., the dispersive mass flux is can be used to describefield-scalesolute transport is stillan
proportional to the concentration gradient. Some investiga- open question. Until the issue is resolved, the field-scale
tors [e.g., Robertson and Barraclough, 1973;Bredehoeft and dispersivityconceptcan be regardedas a working hypothe.
Pinder, 1973] alternatively define the dispersion coefficient sis which has a sound theoretical basis and finds wide
tensor
including
theporosity
n asDj =nD ij. Whenit was application.

clearthatDj wasusedina study,

weconverted
tothemore
common form used in (1). The mean flow direction is taken

to be Xl, with vl = v, v2 = v3 = 0. Assumingthat xl, x2,

and x 3 are principaldirections,the dispersivityis simplythe
ratio of the appropriate component of the dispersivecoefficient tensor divided by the magnitude of the seepagevelocity, v. To distinguish the field-scale dispersivities from
laboratory values, the field-scale values are designatedby
the uppercase letter A [see Ge!har and Axness, !983] and, to
allow for anisotropy of transverse dispersion, a third dispersivity coefficient is used as follows:

Dll = ALV

D22 = Arv

D33 = Avv

(2)

where A t is the longitudinalmacrodispersivity(field scale),

and A T is the horizontal transversemacrodispersivity,and
A v is the vertical transversemacrodispersivity.
The classical equation (1) with macrodispersivities (2) is
standardly used for applied modeling of field-scale solute
transport. The macrodispersivities are considered to be a
property of some region of the aquifer. Although the macrodispersivitymay be a function of space, in most applications it is assumedconstant over a region of the aquifer that
encompassesthe entire plume both horizontally and vertically. Real solute plumes are observed to be threedimensional [LeBlanc, 1982; Perlmutter and Lieber, 1970;

FIELD DATA ON DISPERSIVITY

Summary of Observations

A literature review was conducted to collect reported

values of dispersivity from published analyses of field-scale
tracer tests and contaminant transport modeling efforts. The
literature sources and pertinent data characterizing each
reviewed study are summarized in Table 1 which includes
information

on 59 different

field

sites.

The

information

compiled from each study includes site location, description

of aquifer material, average aquifer saturated thickness,
hydraulic conductivity or transmissivity, effective porosity,
mean pore velocity, flow configuration, dimensionality of
monitoringnetwork, tracer type and input conditions, length
scale of the test or problem, reported values of longitudinal
and horizontal and vertical transverse dispersivities, and
classificationof the reliability of the reported data. Blank
entriesindicate that the information was not provided in the
cited documents.

This

table

summarizes

information

for

purposesof comparisononly. More detail regarding a particular study may be found in the original sources.
Aquifer characteristics. As indicated by the second
through sixth columnsfrom the left, the study sites represent
a wide variety of aquifer conditions and settings. SummaMacFarlane et al., 1983] and often of limited vertical extent.
rized in these columns is information on aquifer material,
Although the classical equation is three-dimensional, the saturatedthickness,hydraulic conductivity or transmissivtwo-dimensional form is most commonly applied. Reasons ity, and velocity. The aquifer thicknessfor each site is the
for the use of the two-dimensional form of the equation arithmeticaverageof the range,at that site. Hydraulic
include lack of three-dimensional
data and in the case of
conductivity and transmissivity values show the range renumerical models, restrictions on the size of data arrays in portedat the site. Reportedvaluesfor effectiveporosityvary
the model. Seldom is the two-dimensional form justified on from 0.5% (for fracturedmedia)to 60% (for porousmedia).
the basis of site conditions or plume observations.
Whena valuewas reportedas "porosity," we interpreted
A number of theoreticalstudieshave proposedmethodsof thisasthe effectiveporosity(interconnected
porespace),the
describing field-scale dispersive mixing. All of the theories valueusedin analysisof the advection-dispersion
equation.
view field-scale dispersionas being produced by some kind Where porositywas reported as "total porosity," we have
of small-scaleheterogeneity or variability of the aquifer. At indicatedthisin the table. The velocitycolumnindicatesthe
present there is considerable debate concerning how to meanpore or seepagevelocity at a site. In somecasesthe
parameterize the variability and model field-scale solute values were calculated from information provided on avertransport. Assuming a perfectly layered aquifer, one group age specificdischarge,q, and effective porosity, n, as v =
[Molz et aI., 1983, 1986]suggestsmeasuringthe variability in q/n. Velocities ranged from 0.0003 to 200 m/d.
detail and modelingthe transport in each layer with localMethods of determiningdispersivity. The seventh
scaledispersivities,thuseliminatingthe needfor a field-scale throughtenthcolumns
from the left summarizethe method
dispersivity. Again assuminga layered aquifer, a second usedto determine
thedispersivity
for eachsite.The seventh,
group suggests the use of a scale-dependent or time- eighth, and ninth columnsfrom the left describeexperimendependentfield-scaledispersivity[e.g., Pickensand Grisak, tal conditions'
flow configuration,
monitoring,
tracerand

GELHAR
ETAL.:
FIELD-SCALE
DISPERSION
INAQUIFERS

1957

input;
thetenthcolumn
fromtheleftsummarizes
methods
of andtwo-welltracertests,analysis
of thedatamustaccount
datainterpretation.
Dispersivity
values
werecalculated
or for thiseffectto determine
dispersivity
properly
for.such

inferred
fromoneoftwotypes
ofsubsurface
solute
transportcases.Nonuniformvelocityeffectshavealsobeenobserved
events:
large-scale,
uncontrolled
contamination
(naturallyin ambientflow tracer tests.
occurring
or human-induced)
events,or controlled
tracer
tests.

Thetypesoftracers
usedtodetermine
dispersivity
ateach
sitearesummarized
inthe"tracerandinput"column
along

Uncontrolled
events
arecharacterized
bya source
input withtheinputconditions.
A varietyof chemical
andmicrohistory
thatis unknown,
transport
of contaminants
bythe biological
tracers
havebeenemployed
forcontrolled
tracer

ambient
flowof groundwater,
andsoluteplumes
thatoften tests.Discussions
of thesuitabilityof differentchemicaland
extend
overregionalscales(hundreds
of meters).
We de- microbial
species
for tracertestsarepresented
by Daviset
scribe
naturallyoccurringeventsas "environmental"
trac- al. [1980,1985]andBetson
et al. [1985].A primaryconsiders,implyingchemicalconstituents
associated
withuncon- erationin designinga controlledtracer test is whetherthe

trollednaturalchangesoccurringin groundwater
beforethe speciesis conservativeor nonconservative.A conservative
startof a study.Examples
of naturallyoccurring
events traceris one that moveswith the samevelocityas the
include
tritiumin groundwater
fromrecharge
containinggroundwater
and doesnot undergoradioactivedecay,adatmospheric
bombtritium,seawaterintrusion,andmineral sorption,degradation,
chemicalreaction(or in the caseof
dissolution.
These eventsare indicatedin the "tracer and microorganisms,
death).If anyof theseeffectsarepresent,
input"column
by thenotation
"environmental"
alongwith theymustbeaccounted
forin evaluation
of thedispersivity.
thetypeof chemicalspecies
reported.Examples
of human- Anotherfactorimportantin the choiceof a tracer is that it is

induced
contamination
eventsincludeleaksandspillsto not presentin naturallyoccurringgroundwater,or that it is
groundwater
fromlandfills,
storage
tanks,surface
impound-injectedat concentrations
muchhigherthan naturalbackments,and infiltrationbasins.Thesetypesof eventsare ground levels.

indicated
by the notation "contamination"in the tracerand
The "monitoring" column indicates whether two- or
inputcolumn.Values of dispersivityfor uncontrolled
events three-dimensional
monitoring
was employedat a site. By
arecommonlydeterminedby fittinga one-, two-, or three- two-dimensional
monitoring
we meandepth-averaged
(ver-

dimensional
solutetransportmodelto historicaldata;i.e., ticallymixed).Three-dimensional
monitoring
impliespoint
values
of dispersivity
arealtereduntilmodeloutputmatches samples with depth. This information is noted because
historical solute concentration measurements.

vertical mixing in an observation well influences the concen-

The main features distinguishing

controlledtracer tests trationof tracerin a watersample.Severalstudies[Meyeret
fromuncontrolledones is that in the former, both the al., 1981;Pickensand Grisak, 1981]have shownthat when a
quantityand duration of solute input are known. This is traceris not injectedoverthe full aquiferdepth,vertically

indicated
by "step" (continuous
inputof mass)or "pulse" mixed samplesunderestimatethe tracer concentration and
("instantaneous"
or sluginput)in the "tracerandinput" as a result the longitudinaldispersivityis overestimated.
column.Controlledtracer tests may be conductedunder Thisoccursbecausethetraceroccupies
onlya portionof the
ambientgroundwaterflow conditions(also referredto as verticalthickness.When a samplefrom the entire thickness
naturalgradient tests), or under conditionswhere the flow

is taken, the true tracer concentration is diluted in the well

configuration
is inducedby pumping
orrecharge.
Thetypeof withtracer-freewater.If an attemptis madeto interpretthe
testis reportedin the "flow configuration"
column.Induced diluted("measured")concentration,
the dispersivitywill be
flow configurationsinclude radial, two-well, and forced overestimated. At many sites there was no indication
uniform
flow.In radialflowtracertests,a pulseor stepinput whetherpointor fully mixedsamplingwasperformed.From
oftraceris injectedat a rechargewell andthetimedistribu- examination of the cases where three-dimensional measuretionof traceris recordedat an observation
well (diverging ments of solute concentrationswere made, it is clear that
radialflow test), or the traceris injectedat an observation verticalmixingof the traceras it travelsthroughthe aquifer
well and the time distribution is recorded at a distant

is often very small [Sudickyet al., 1983; LeBlanc, 1982;

pumpingwell (convergingradial flow test). In a two-well Freyberg, 1986; Garabedian et al., 1988, 1991].
test,both a rechargewell and pumpingwell are operating; Field dispersivities and scale. The "scale of test" coltraceris injected at the rechargewell and tracer break- umn represents the distance traveled from the source for
through
isobserved
at thepumping
well.Recirculation
ofthe ambientconditions,or the distancebetween injection and
water (containing tracer) from the pumping well to the observationwells for the caseof an inducedflow configurarecharge
well is often employed."Forceduniformflow" tion. The values of dispersivity reported at the indicated

refersto the flow regimeat the Bonnaudsite in France, scale are given in the secondcolumn from the right. Data
wherea uniformflow fieldwasgenerated
betweentwo lines from the 59 sites yielded 106 valuesof longitudinaldisperof equallyspacedwells, one line recharging
and one line sivity, sinceoften multipleinvestigations
or multipleexperpumping,with both screenedto the full depthof the aquifer.

iments by one investigator were performed at one site. A

A discussion
of the advantages
anddisadvantages
of differ- plot of the longitudinaldispersivityvalues as a function of
enttypesof tracer testsis presentedby Weltyand Gelhar scaleis presentedin Figure t. The arithmeticaveragewas
[1989].
plotted in caseswhere a range was reported either for the
A number of methods have been used to evaluate the data

scale or dispersivity in Table 1. In some cases, values of

fromcontrolled
tracertests,asindicated
by columnheaded dispersivityfor individuallayers were reportedas well as an
"method
of datainterpretation."
Theseincludefittingof average "aquifer" value. In these casesthe latter value was
one-,or two- or three-dimensional
solutetransportanalytical plottedfor the given scale.The symbolson Figure 1 indicate
solutions,
andthe methodof spatialmoments.
It shouldbe whetherthe dispersivityvalue is tbr fractured media (open
notedthat sincethe velocityis nonuniform
for bothradial symbols, 18 values) or porous media (solid symbols, 88

1958

GELHARET AL.: FIELD-SCALEDISPERSION

IN AQUIFERS

TABLE 1. Summary0f
Hydraulic

Reference

and Site Name

Adams and Gelhar [1991],

Columbus,Mississippi
Ahlstrom et al. [1977],
Hanford, Washington
Bentley and Walter [1983],

Aquifer Material

very heterogeneoussand
and gravel
glaciofluviatile sandsand
gravels
fractured

Average
Aquifer

Conductivity
(m/s)or

Effective

Thickness,

Transmissivity

Porosity,

Velocity,

(m2/s)

m/d

10-5 to 10-3 m/s

35

0.03-0.5

64

5.7 x 10-4 to 3.0

ambient

18

0.3

WIPP

Bierschenk [1959] and Cole

[1972], Hanford,
Washington
Bredehoeft and Pinder
[ 1973], Brunswick,
Georgia
Claasen and Cordes [!975],
Amargosa, Nevada

glaciofiuviatile sandsand
gravels

64

1.7 x 10-1 m2/s

10

limestone

50

6.5 x !0 -7 to 8.6
x 10-7 m2/s

35

15

5 x 10-2to 11
x 10-2 m2/s

Daniels [198!, 1982],

alluvium
tuff

500

1.7 x 10-5 m/s

Nevada

Test Site

Dieulin [1981], Le Cellier

(Lozere, France)
DieMin [1980], Torcy,

fractured dolomite
limestone
derived

and

from

ambient

x 10-2 m/s
5.5

dolomite

Flow Configuration

fractured granite

20

alluvial deposits

3 x 10-4 to 9
X 10-4 m/s
3 x 10-4 m/s

6-60

2-8

26
31

two-wellrecirculating
ambinet

radial converging
0.14-3.4

two-wellrecirculating

0.04

radial converging

radial converging

0.5

ambient

France

Egboka et al. [1983],

glaciofluvial sand

10-5 to 10-7 m/s

38

0.01-0.04

ambient

53

4.7 x 10-6 m/s

23

1.2

radial diverging

7.2 x 10-5 m/s

33

0.09

ambient

9.6

radial diverging

7-27

Borden

Fenske [1973], Tatum Salt

limestone

Dome, Mississippi
Freyberg [1986], Borden

glaciofluvial sand

(total)

Fried and Ungemach

[ 1971], Rhine aquifer
Fried [ 1975], Rhine aquifer
(salt mines) southern
Alsace, France
Fried [1975], Lyons,
France (sanitary landfill)

sand, gravel, and

12

cobbles

alluvial; mixture of sand,

125

10'-3 m/s

ambient

gravel, and pebbles

with clay lenses
alluvial, with sand and
gravel and slightly

20

5.0

ambient

0.43

ambient

stratified clay lenses

Garabedian et al. [1988]
Cape Cod,
Massachusetts

medium

to coarse sand

7O

1.3 x 10-3 m/s

39

with some gravel

overlying silty sand

Gelhar [1982], Hanford,

Washington

and till
brecciated basalt
interflow zone

Goblet [1982], site B,

fractured granite

50

10-5 to 10-7 m/s

basaltic lava and

76

1.4 x 10-1 to 1.4

x 101 m2/s

two-well without
recirculation
84

radial converging

France

Grove [1977], NRTS, Idaho

sediments

Grove and Beetera [1971],

Eddy County (near
Carlsbad), New Mexico

fractured dolomite

Gupta et al. [1975], Sutter

sandstone, shale, sand,

Basin, California
Halevy and Nir [1962] and
Lenda and Zuber [1970],
Nahal Oren, Israel
Harpaz [1965], southern
coastal plain, Israel

and alluvial
dolomite

12

ambient

10
12

3.5

two-well recirculating

ambient

sediments

sandstone with silt and

100

3.4

90

4.0

14

radial converging

radial diverging

clay layers
ambient

Helweg and Labadie

[1977], Bonsall subbasin,
California

Hoehn [1983], lower Glatt

Valley, Switzerland

layered gravel and silty

sand

25

9.2 x 10-4 to 6.6

x 10-3 m/s

3.4
1.8
1.2
8.6
4.1
1.7

ambient

GELHAR
ETAL.'FIELD-SCALE
DISPERSION
INAQUIFERS

1959

FieldObservations
Classification

of Reliability
of

Monitoring

Tracer and Input*

three-dimensionalBr- (pulse)

Methodof DataInterpretation Scale of Test, m

spatialmoments

two-dimensional
3H (contamination)
two-dimensional
numerical

200

20,000

Dispersivity

AL/AT/Av

A L/A r/A v,t m

(I, II, I!I)$

7.5
30.5/18.3

IiI

model

two-dimensional PFB, SCN (step)

one-dimensional
quasi-uniform
flow solution [Grove and

two-dimensional fluorescein(pulse)

one-dimensional
uniformflow

23

5.2

III

Beetera, 1971]

solution

two-dimensional C1- (contamination) two-dimensional

numerical

3,500
4,000

6
460

IIi
III

2,000

170/5211

IlI

model

two-dimensional
3H (pulse)

one-dimensional
quasi-uniform

122

15

III

91

10-30

III

0.5

flow solution [Grove and

Beetera, 1971]

two-dimensional
3H (contamination)
radialflowtypecurve[Sauty,
1980]

two-dimensional

CI-, I- (pulse)

radial flow type curve [Sauty,

II

1980]

two-dimensional
(resistivity)

C1- (pulse)

one-dimensionaluniform flow
solution

three-dimensional
3H (environmental) one-dimensional
uniformflow

15

III

600

30-60

III

91

11.6

III

solution

3H (pulse)

one-dimensional
uniformflow
solution

three-dimensional Br-, C1- (pulse)

C1- (pulse)

spatialmoments

90

III

one-dimensional
radial flow
numerical

0.43/0.039

model

three-dimensional C1- (contamination) two-dimensionalnumerical

8O0

15/1

III

12/4

III

model

two-dimensional EC (contamination) two-dimensional

numerical

600-1000

model

three-dimensional Br- (pulse)

two-dimensional

1311
(pulse)

spatialmoments

one-dimensional
nonuniform

250

17.1

0.96/0.018/
0.0015

0.60

flow solution along

streamlines[Gelhar, 1982]
two-dimensional

RhWt,SrC1(pulse) one-dimensional
uniformflow

two-dimensional

CI- (contamination)two-dimensional
numerical

III

17

solutionincludingborehole
flushingeffects
20,000

91/91

III

38.1

III

80-200/

III

model

two-dimensional

3H (step)

one-dimensional
quasi-uniform

55

flow solution [Grove and

Beetera, !971]

C1- (environmental)
three-dimensional
numerical

50,000

model

two-dimensional

two-dimensional

two-dimensional

250

8-20
6

II

6Co
(pulse)

one-dimensional
uniform
flow

C1- (step)

one-dimensional
radialflow

28

0.1-1.0

II

TDS
two-dimensional
numerical
(contamination) model

14,000

30.5/9.1

III

uranine (pulse)

solution

solution

one-dimensional
uniformflow
solution for layers

4.4
4.4
4.4
10.4

10.4
10.4

0.!
0.01
0.2
0.3
0.04
0.7

III

lI!

1960

GELHARET AL.: FIELD-SCALEDISPERSION

IN AQUIFERS

TABLE 1,
Hydraulic

Reference
Hoehn

and Site Name

and Santschi

[1987], lower Glatt

Valley, Switzerland

Aquifer Material
layered gravel and silty

Average
Aquifer

Conductivity
(m/s)or

Thickness,

Transmissivity

Porosity,

Velocity,

(m2/s)

rn/d

27.5

8.1 x 10-5 to 6.6

Effective

x 10-3 m/s

sand

Flow Configuration

1.5
3.2

ambient

5.6
3.9

ambient

3.2

Huyakorn et al. [1986],

Mobile, Alabama
Iris [!980], Campuget
(Gard), France

layered medium sand

Ivanovitch

fractured

and Smith

alluvial deposits

3.6 x 10-3 m2/s

2.2 x 10-3 m/s
(fast pulse)
3.6 x 10-4 m/s
(slow pulse)
9.55 x 10-5 m/s

chalk

[1978], Dorset, England

chalk

Kies [1981], New Mexico

State University, Las

0.35

21.6

fluvial sands

two-well without
recirculation
0.05

radial diverging

0.5

57.6

radial converging

2.3

9.6

radial converging
ambient

42

(total)

Cruces

Klotz et al. [1980],

Dormach, Germany
Konikow [1976], Rocky
Mountain

fluvioglacialgravels

20

14

radial converging

30

ambient

2.4 x 10-4 to 4.2

x 10-3 m/s

20

ambient

3.1 x 10-5 to
1.5 x 10-4

24

alluvium

Arsenal

Konikow and Bredehoeft

[1974], Arkansas River
valley (at La Junta,

alluvium,

inhomogeneousclay,
silt, sand and gravel

Colorado)

Kreft et aI. [1974], Poland

2.5

sand

29

radial converging

m/s; 1.2

Kreft et al. [1974], Zn-Pb

deposits, Poland
Kreft et aI. [1974], sulfur
deposits, Poland
Lau et al. [1957],

University of California,
Berkeley
Lee et al. [1980], Perch
Lake, Ontario, (lake
bed)
Leland and Hillel [1981],
Amherst, Massachusetts
Mercado [1966], Yavne
region, Israel

limestone

X 10-4 m2/s
2.5 x 10-4to
4.7 x 10-4 m/s
2.5 x 10-4to
4.7 x 10-4 m/s
1.1 x 10-4 m/s

12.3

limestone

1.1 x 10-4 m/s

12.3

9 x 10-4 m/s

30

fractured

dolomite

57

fractured

dolomite

48

sand and gravel with

clay lenses

2.4

3.2 x 10-5 m/s

sand

fine sand and glacial till

0.75

2.4 to3 x 10-5

4O

7.5
100
60.1
22.7
10
10.8
8.6

radial converging
radial converging

radial converging
radial converging
radial diverging

0.14

ambient

0.3-0.6

ambient

m/s

sand and sandstone with

8O

some silt and clay

Meyer et al. [1981];

Koeberg Nuclear Power
Station, South Africa

sand

Molinari and Peaudecerf

[1977] and Sauty [1977],
Bonnaud, France

sand

Moltyaner and Killey

[1988a, b l, Twin Lake
aquifer (Chalk River)
Naymik and Barcelona
[ 1981], Meredosia,
Illinois (Morgan County)

1.5

2.4

2.1 x 10-8 to
2.4 x 10-8
m2/s

23.3

radial

diverging/converging

20

8.3 x 10-4 to
1.I x 10-3
m2/s

40.8

fluvial sand

0.84-3.4

0.12

ambient

2.7
1.0
2.4

forced uniform

1.0
2.0
2.0
!.2

ambient

(total)
unconsolidated

gravel

sand and

27

2.2 x 10-2 to
4.3 x 10-2
m2/s

ambient

GELHAR
ETAL.:FIELD-SCALE
DISPERSION
INAQUIFERS

1961

(continued)
Classification

of Reliability
of

Monitoring
two-dimensional

Tracer
andInput* Methodof DataInterpretation
uranine (pulse)

temporal moments

Scale of Test, m
4.4
10.4

two-dimensional

3H (environmental)

temporal moments

two-dimensional

Br- (pulse)

two-dimensional numerical

100
110
500

38.3

Dispersivity
A L/A r/A v,? m

A /Ar/A v
(I, II, III):
II
II
III
III
III
I

1.1
1.2

6.7
10.0
58.0
4.0

model

three-dimensional heat (pulse)

two-dimensional radial

40

3/1.5

II

numerical model

82Br(pulse)

one-dimensional uniform flow

3.1

III

1.0

Iii

25

1.6/0.76

II!

10

5, 1.9

solution

82Br(pulse)

one-dimensional uniform flow

solution

two-dimensional NO/- (pulse)

two-dimensional uniform flow

solution

two-dimensional

82Br,uranine(pulse)

one-dimensional uniform flow

II

solution

C1- (contamination)

two-dimensionalnumerical

!3,000

30.5

II!

18,000

30.5/9.1

III

model

two-dimensional

dissolved

sol/ds

(contamination)

two-dimensional1311
(pulse)

1311
(pulse)
]31I(pulse)

58Co(pulse)
58Co(pulse)
C1- (step)

three-dimensional

C1- (pulse)

two-dimensionalnumerical
model

one-dimensional
solution

uniform flow

one-dimensional uniform
solution
one-dimensional uniform
solution
one-dimensional uniform
solution
one-dimensional uniform
solution
one-dimensional radial
numerical model

5-6

0.18

flow

22

flow

21.3

2.1

flow

27

2.7-27

flow

41.5

20.8

19

2-3

44-110

0.012

one-dimensional uniform flow

solution

three-dimensional

C1- (pulse)

three-dimensional

6Co,C1- (step)

three-dimensional

131i(pulse)

two-dimensional uniform flow

solution
one-dimensional radial flow
solution

one-dimensional uniform flow

solution for layers

-<115 (observation
wells)
2-8

0.05-0.07

III

0.5-1.5 (injection
phase)
0.01, 0.03,
0.01, 0.05

II!

for layers;
0.42 for

depth
average

two-dimensional uniform flow

two-dimensional

3H
1311
1311
1311

131I(pulse)
three-dimensional 131i(pulse)
two-dimensional

solution

two-dimensionaluniform flow

33.2
32.5

0.79
1.27
0.72
2.23
1.94/0.11
2.73/0.11

40

0.06-0. ! 61" '/

13
13
13
26

0.0006---0.002

solution

NH 3 (contamination) two-dimensional numerical

model

I
I
I
I
I
I
I1

16.4

2.13-3.35/
0.61-0.915

III

1962

GELHARET AL.: FIELD-SCALEDISPERSION

IN AQUIFERS

TABLE I.
Hydraulic

Reference

and Site Name

Aquifer Material

Average
Aquifer

Conductivity
(m/s)or

Thickness,

Transmissivity

(m2/s)

Effective

Porosity,

Velocity,
m/d

Flow Configurati0

New Zealand Ministry of

Works and Development
[1977] Heretaunga
aquifer, New Zealand:
Roys Hill site

gravel with cobbles

100

0.29 m2/s

22

!50-200

ambient

Flaxmere

alluvium (gravels)

120

0.37 m2/s

22

20-25

ambient

site 2

Hastings City rubbish

dump
Oakes and Edworthy
[1977], Clipstone, United
Kingdom

Papadopulos and Larson

[1978], Mobile, Alabama

0.14,0.35m2/s

alluvium (gravels)

44

sandstone

medium to fine sand

21

2.4 x 10-6 to
1.4 x 10-4 m/s

5 x 10-4m/s

32-48

20

ambient

5.6, 4.0

radial diverging

9.6

2.4, 3.6

radial converging

25

0.05

radial diverging

38

0.15

two-well recirculating

38

0.!5

radial

35

0.43

regional

!1-21

regional

0.09

ambient

(horizontal)
and

interspersed with clay

5.1 x 10-m/s

and silt

(vertical)
Pickens and Grisak [1981],

sand

8.5

sand

8.5

Chalk River

Pinder [1973], Long Island

glacial outwash

43

Rabinowitz

fractured

61

and Gross

limestone

[1972], Roswell Basin,

New

2 x 10-5 to
2 X 10-4 m/s
2 x 10-5 to
2 x 10-4 m/s
7.5 x 10-4 m/s
1.1 x 10-2 to
2.9 x 10-1 m2/s

diverging/convergi

Mexico

Rajaram and Gelhar

[ !991], Borden
Roberts et al. [198 !], Palo
Alto bay lands

Robertson [1974] and

Robertson

and

7.2 x 10-5 m/s

glaciofluvialsand

33
(total)

sand, gravel, and silt

1.25 x 10-3 m2/s

(lower aquifer)'
5.0 X 10-4 m2/s
(upper aquifer)

25

15.5
12.0
3.5

25.6
7.9
! .5-8

radial diverging

basaltic lava and

sediments

76

1.4 x 10-l to
1.4 x 101 m2/s

10

alluvial

27

2.1 x 10-4 to
I x 10-2 m2/s

40

two-well recirculating

40

regional
regional

regional

BarracIough [1973],
NRTS, Idaho
Robson [1974, 1978],
Barstow, California

Robson [1978], Barstow,

alluvial

sediments

sediments

30.5

5 x 10-4 m/s

40

!2

6.5 x 10-3 to
1.5 x 10-2 m/s

14
2.1-18
1.8-5.9
1!-24

California

Rousselot et al. [1977],

Byles-Saint Vulbas near
Lyon, France

clay, sand, and gravel

Saucy [1977], Corbas,

sand and gravel

18

24

125, 100
15.5, 78

12

France

radial converging

11.5,
46.7, 16

radial converging

6.9

sand

Saucy et al. [1978],

Bonnaud, France
Segol and Pinder [1976],
Cutler area, Biscayne
Bay aquifer, Florida

fractured limestone and

calcareous sandstone

Sudicky et al. [1983],

glaciofluvial sand

30.5

8.3 x 10-4to
1.1 x 10-3 m2/s
0.45 x 10-2 m/s
(horizontal) and
0.09 X 10-4 m/s

radial diverging

25

20

ambient

(vertical)
7-27

Borden

Sykeset al. [1982, 1983],

sand

Alabama

38

0.070.25

35

ambient

ambient

m/s

Borden

Sykes et a/.[1983], Mobile,

4.8 x 10-5 to
7.6 x 10-5 m/s
5.8 to 7.2 x 10-5

sand, silt, and clay

21

5 x 10-4 m/s

(horizontal)
and

2.5 x 10 -m/s

(vertical)

25

0.05

radial diverging

1963

GELHAR
ETAL.' FIELD-SCALE
DISPERSION
IN AQUIFERS

(continued)
Classification

of Reliability
of

Monitoring

TracerandInput*

Method of Data Interpretation

Scale of Test, m

three-dimensional
1311,
RhWt,82Br, three-dimensional
uniform
flow
C!-, E. Coli (pulse)

1.4-11.5/
0.1-3.3/
0.04-0.10
0.3-1.5/
./0.06
41/10/0.07

54-59

solution

three-dimensional
RhWt,82Br
(pulse) three-dimensional
uniform
flow

25

three-dimensional
C1- (contamination)
three-dimensional
uniform
flow

290

Dispersivity
A z,/Ar/A v,? m

solution

A :/Ar/A v
(I, II, III):

II

II
III

solution

two-dimensional
82Br(pulse)

0.16, 0.38

radialflownumerical
model

two-dimensional CI-, I- (pulse)

two-dimensional heat (step)

two-dimensional numerical

three-dimensional
51Cr(step)

one-dimensionalquasi-uniform

three-dimensional
1311
(step)

one-dimensional radial flow

three-dimensional
Cr+6
(contamination)

two-dimensional numerical

0.31
6
3
57.3

0.6
0.6
1.5

model

0.5

III

0.03

III

flow solution
solution

1,000

21.3/4.2

III

20-23

III

model

two-dimensional 'H (environmental) one-dimensionaluniform flow

32,000

solution

three-dimensionalBr-, C1- (pulse)

spatialmoments

two-dimensional

one-dimensionaluniform flow

C1- (step)

solution

two-dimensional
C!- (contamination)
two-dimensional
numerical

0.50/O.05/

0.0022
11
20
40
16
43

20,000

5
2
8

III
III
III

4
1!

III
III
III

91o/137oll

model

two-dimensional

C1- (step)

one-dimensional
quasi-uniform

two-dimensional

TDS

two-dimensional
numerical

three-dimensional

TDS

(contamination)
(contamination)
two-dimensional I- (pulse)

two-dimensional I- (pulse)

two-dimensional
heat(step)

6.4

flow solution

lO,OOO

model

two-dimensional
numerical

model(verticalsection)
one-dimensional
uniformflow
solution for layers

one-dimensional
uniformflow
solutionfor layers

one-dimensional
radial
flow
solution

three-dimensional
C1-(environmental)
two-dimensional
numerical
model
three-dimensional C1- (pulse)

three-dimensional
uniformflow

three-dimensional C1- (pulse)

two-dimensional
numerical

three-dimensional heat (step)

three-dimensional
numerical

solution
model

model

3,2oo
9.3
5.3
10.7
7.1
25
50
150
13

15.2

III

61/18

I!I

61/. 40.2

III

6.9

0.3, 0.7
0.46, 1.1
0.37

1!, 1.25
25, 6.25
12.5
1.0

II
III
III
II
III
III
I!
II

490

6.7/. ./0.67

III

11
0.75
700

0.08/0.03
0.01/0.005

II
II
IIi

57.3

7.6/.'

40.31

0.76/- - ./0.15

II

1964

GELHAR ET AL.: FIELD-SCALEDISPERSIONIN AQUIFERS

TABLE 1.
Hydraulic

Reference

and Site Name

Vaccaro and Bolke [1983],

Spokane aquifer,
Washington and Idaho
Valocchi et al. [1981], Palo
Alto bay lands

Walter [1983], WlPP

Aquifer Material

Average
Aquifer

Conductivity
(m/s)or

Thickness,

Transmissivity

Porosity,

Velocity,

(m2/s)

m/d

glaciofiuvial sand and

gravel

152

sand, gravel, and silt

9 x 10-5 m2/s to
6.5 m2/s

1.25 x 10-3 m2/s

Effective

7-40

25

0.003-2.8

27

Flow Configuration
ambient

radial diverging

(lower aquifer);

fractured

5.0 x 10-4 m2/s

(upper aquifer)
8.0 x 10-5 m2/s

dolomite

0.7 and

4.7, 2.4

radial converging

1.3

two-well
recirculating

9.1

ambient

11 (along
separate

paths)

crystalline, fractured
schist and gneiss

76

Werner et al. [1983],

Hydrothermal Test Site,
Aefligen, Switzerland

gravel

20

Wiebenga et al. [1967] and

Lenda and Zuber [1970],
Burdekin Delta,

sand and gravel

Webster et al. [1970],

Savannah River Plant,

3.6x 10-7 m/s

21.4

South Carolina

6.1

6 x 10-3 m/s

17

5.5 x 10-3 m/s

32

5.75 x 10-3
m2/s

38

29

radialconverging

Australia

Wilson [1971] and Robson

[1974], Tucson, Arizona

unconsolidatedgravel,
sand, and silt

two-well without
recirculation
radial diverging

Wood [1981], Aquia

Formation, southern
Maryland
Wood and Ehrlich [1978]
and Bassett et al. [1980],
Lubbock, Texas

sand

1,000

2.9 x 10-4 to
8.7 x 10-4 m2/s

sandandgravel

17

3.2 x 10-3 to

4.4 x 10-3 m2/s

35

0.0003-

ambient

0.0007

78

radialconverging

*TDS denotestotal dissolvedsolids;EC, electricalconductivity;PFB, pentafluorobenzoate;

MTFMB, metatrifluoromethylbenzoate;
MFB, metafluorobenzoate;
Para-FB, paraftuorobenzoate;
RhWT, rhodamine-WTdye; and SCN, thiocyanate.

?AL denotes
longitudinal
dispersivity;
AT, horizontal
transverse
dispersivity;
andA v, vertical
transverse
dispersivity.
Reported
values
for A,Ar, andAv areseparated
by slashes.
Absence
of slashes
means
thatvalueswerereported
forAL only.A comma
ora dash
separating
entriesmeansthatmultiplevaluesor a rangeof values,respectively,
werereported
for a particular
dispersivity
component.
$For descriptionof classificationcriteria, see text.

E. E. AdamsandL. W. Gelhar,Fieldstudyof dispersion

in a heterogeneous
aquifer:Spatialmomentsanalysis(submitted
to Water
Resources Research, 1991).

IlPorosity-corrected
dispersivity
value.

values).The type of eventevaluatedis indicatedby a circle scatterover a similar range, althoughat a smallerscale
(tracer test, 83 values), triangle (contaminationevent, 15 fracturedmediaseemto showhighervalues.At eachscale
rangein dispersivvalues), or square(environmentaltracer, eight values).The thereis at leasta two-order-of-magnitude
total numbers of values of dispersivity for each type of ity. Because
we noteda numberof problems
withdataand
medium and test are shown in Table 2. Any reported values their interpretationas we gatheredthem for Table 1, we
of horizontal transversedispersivityor vertical transverse wouldregardanyconclusions
aboutFigure1 withskepti-

dispersivityare also listed in the dispersivitycolumnof

cismuntilfurtherqualifyingstatements
canbe madeabout

Table 1. For the cases examined, 24 values of horizontal

thedatapoints.Typicalproblems
that we foundwiththe
transversedispersivityandnine valuesof verticaltransverse studiesreportedin Table 1 includethe following:
data
dispersivity
werereported.In nearlyall cases,thehorizontal analysisnot matchedto flow configuration;
massinput
historyunknown;nonconservative
effectsof tracernot
the longitudinalvalues,and the verticalvaluessmallerby accounted
for;dimensionality
ofthemonitoring
notmatched

values were found to be 1-2 orders of magnitude less than

another order of magnitude.

to thedimensionality
of the analysis;
andassumption
of
distinctgeologiclayersin analysiswhentheir actualpres-

Evaluation of Dispersivity Data

ence was not documented.Based on these problems,we

toratethedataashigh(I), medium
(II), orlow(III)
From Figure 1, it appearsthat longitudinaldispersivity decided
increases with scale. Field observations of dispersivity

reliability
according
tothecriteria
setforthbelow.
Table
3

rangedfrom0.01m to approximately
5500m at scales
of 0.75 lists the criteriausedto designateeither high-or lowm to 100km. The longitudinaldispersivityfor the two types

of aquifermaterial(porousversusfracturedmedia)tendsto

reliability
data.No specific
criteriaweredefined
forthe
intermediate
classification;
it encompasses
thedispersivity

GELHAR
ETAL.:FIELD-SCALE
DISPERSION
INAQUIFERS

1965

(continued)
Classification

of Reliability
of

Monitoring

Tracer and Input*

Methodof DataInterpretation Scaleof Test, m

C1- (contamination)two-dimensional
numerical

43,400

model

C1- (step)

two-dimensional
numerical

16

Dispersivity
A L/A r/A v,' m
9!.4/27.4

AL/A T/A V
(I, II, III)$
III

1.0/0.1

model

two-dimensional

MTFMB, PFB,
MFB, para-FB
(pulse)

two-dimensional 85Sr

85Br(pulse)
three-dimensional heat (step)

one-dimensional
uniformflow

one-dimensional
quasi-uniform

10-15

III

538

134

III

700
37
105
200
18.3

130-234
131
208
234
0.26

!II
III
I!!
!1I
II

79.2

15.2

III

flow solution

one-dimensional numerical
model

13I, 3H (pulse)

30

solution

one-dimensional uniform flow

solution

three-dimensional

C1- (step)

one-dimensional
quasi-uniform
flow solution

two-dimensional

C1- (step)

one-dimensionalradial flow

4.6

0.55

III

solution

Na +
(environmental)

two-dimensional

I- (pulse)

one-dimensional
uniformflow

105

5,600-40,000

III

solution

one-dimensionalradial flow

1.52

0.015

II

solution

valuesthat do not fall into the high or low groups. These

classifications
do not placestrictnumericalconfidence
limits

onreporteddispersivities,but ratherare intendedto provide

an order-of-magnitudeestimate of the confidencewe place
on a given value. In general, we considerhigh-reliability
dispersivityvalues to be accurate within a factor of 2.
Low-reliability
valuesareconsidered
to be nomoreaccurate
thanwithin1 or 2 ordersof magnitude.Intermediate
reliability falls somewhere between the extremes. We wish to make

a distinction
betweenthejudgmentof the reliabilityof the
reporteddispersivityandthe worthof a study.Often,the
purpose
of a studywas for something
otherthanthe determination
of dispersivity.
Ourclassification
of dispersivity
is
notintendedas a judgmenton the qualityof a studyas a

1.

The tracer test was either

ambient

flow with known

input, divergingradial flow, or a two-well pulsetest (without

recirculation).These three test configurationsproduce breakthroughcurveswhichare sensitiveto the dispersioncoefficient
and appear to work well in field applications[Wel.ty and
Gelhar, 1989]. The radial converging flow test is generally
consideredless satisfactorythan the divergingtest because
breakthroughcurvesat the pumpingwell for the converging
testfrequentlyexhibittailing,which complicatesthe interpretation of these tests. Some researchers attribute this behavior

to two or morediscretegeologiclayersandtry to reproducethe

observed breakthrough curve by superposition of breakthroughcurvesin each layer, where the propertiesof each

layer may differ [e.g., lvanovitchand Smith, 1978; Sauty,

whole,
butratherto provideuswithsomecriteriawithwhich 1977].The problemwith this interpretationis that there are

heterogeneities
ona smallscalethatcannot
toscreenthe largenumberof datavaluesobtained.By then typicallynumerous
examining
the morereliabledata,conclusions
whichevolve be attributedsolelyto identifiablelayers. One possibleexplafromthe datawill be more soundlybasedandalternative nation of the tailing in radial convergenttests is sometimes
termed "boreholeflushing,"where the tail of the breakthrough
interpretations
maybecome
apparent.
High-reliability
dispersivity
data. Fora reported
disper- curveisattributedto the slowflushingof the inputslugof tracer
flow.
sivityvalueto be classified
as highreliability,eachof the outof theinjectionboreholeby the ambientgroundwater
Goblet[12] measuredthe slow flushingof tracer out of the
following
criteriamusthavebeenmet.

1966

GELHARET AL.: FIELD-SCALEDISPERSIONIN AQUIFERS

104

103
102
lO1
lO0

1o'1

tracer

tests

contam.
events

-2

envir.

tracers

10

-3

10-1

100

101

102

103

104

105

106

scale (rn)

Fig. !. Longitudinal
dispersivity
versusscaleofobservation
identified
bytypeof observation
andtypeof aquifer.The
data are from 59 field sitescharacterizedby widely differinggeologicmaterials.

5. The analysisof the concentrationdata was appropriinput. His solutionreproducedthe tailing observedat the ate. Sincethe interpretationof the tracer data is necessarily
pumping
well. In caseswhereboreholeflushing
wasobserved linkedto the type of tracer test to whichthe interpretation
and accountedfor, dispersivities
obtainedfrom a radialcon- methodis applied,these two featuresof the field studies
of
vergentflow test werenot excludedfromthe high-reliabilitywere evaluatedtogether.The three general categories
data interpretationcan be groupedas follows:(1) breakcategory.
2. The tracer inputmustbe well defined.Both the input throughcurveanalysis,usuallyappliedto uniformambient
concentrationand the temporal distribution of the input flowtestsandradialflowtests[e.g., Sauty, 1980];(2) method
concentrations must be known (measured). If not, the input of spatialmoments,appliedto uniform ambientflowtests
is another unknown in the solution of the advection[Freyberg,1986];and (3) numericalmethods,applied
to
events [e.g., Pinde.
r, 1973; Konikowand
dispersion
equation,andwe are lessconfident
in theresult- contamination

boreholeandmodeledthe effectas an exponentiallydecreasing

ing value of dispersivity.

3.

The tracer must be conservative. A reactive or non-

Bredehoeft, 1974].

A commondifficultywith the interpretation

of concentra-

curvematchingto determine
conservative
tracercomplicates
the governingequations
and tion datausingbreakthrough
is the assumption
that the dispersivity
is conresulted in additionalparametersthat must be estimated. dispersivity
Consequently,
we are lessconfidentin the resultingdisper- stant. The field data assembledin this review suggestthat
is not valid, at leastfor small-scale
tests
sivity. Tracers such as CI-, I-, Br-, and tritium were this assumption
(tensof meters).At largerscales(hundreds
of meters)an
considered to be conservative.
constant
valueof dispersivity
is predicted
by
4. The dimensionalityof the tracer concentrationmea- asymptotic
surementswas appropriate.A tracer introducedinto an sometheories.However, at most sites the displacement
aquifer will spread in three spatial dimensions.High- distanceafter which the dispersivityis constantis not
reliability dispersivitieswere judged to be those where
three-dimensional monitoring was used in all cases except
TABLE2. Numbers
ofDispersivities
for Different
Types
of
where the aquifertracer had been injectedand measured
over the full depth of the aquifer; in this case twodimensionalmonitoringwas acceptable.In all other cases,
where the dimension of the measurement was either not

reportedor where two-dimensional

measurements
were
used where three-dimensionalmeasurementsshould have

beenused,thedispersivity
valueswerejudgedto beof lower
reliability.

Tests and Media

Tracer Type

Media
Type ArtificialContamination
Environmental
Tctal
Porous
Fractured
Total

68
15
83

14
1
15

6
2
8

88
!8
106

GELHAR
ETAL.:FIELD-SCALE
DISPERSION
INAQUIFERS
TABLE3. CriteriaUsedto Classify
theReliability
of the
ReportedDispersivityValues

Classification

Criteria
,,

Highreliability

Tracertest waseitherambientflow, radial

divergingflow, or two-wellinstantaneous
pulsetest (withoutrecirculation).

to the productionwell. As a result, the mixing processis

partiallyreversibleandthe dispersivitywouldbe underestimated relative to the value for unidirectional flow. Heller

[1972]has cardedout experimentswhich demonstratethe

reversibilityeffect on a laboratoryscale.
3. The tracer input was not clearly defined. When a

Tracer input was well defined.

contamination event or environmental tracer is modeled, the

Tracer was conservative.

tracer input (both quantityand temporal distribution)is not

well definedand becomesanother unknown in solving the
advection-dispersion
equation.
4. The tracer breakthroughcurve was assumedto be the
superpositionof breakthrough curves in separate layers
when there was little or no evidence of such layers at the
field site. These studiesgenerallyassumethat the porous
mediumis perfectlystratified,which, especiallyat the field
scale,may notbe a validassumption.At a smallscale(a few

Spatialdimensionalityof the tracer

concentration measurements was
appropriate.

Analysis of the tracer concentrationdata was

appropriate.
Low reliability

1967

Two-wellrecirculating
testwith stepinput
was used.

Single-well injection-withdrawaltest with

tracer monitoringat the singlewell was

used.

Tracer input was not clearly defined.

Tracer breakthrough curve was assumedto be
the superpositionof breakthroughcurvesin
separate layers.
Measurementof tracer concentration
in space
was inadequate.
Equationusedto obtaindispersivitywasnot
appropriate for the data collected.

meters) where the existence of continuous layers may be a

reasonableassumption,the dispersivity of each layer does
not representthe field-scaleparameter. The field-scaledispersivity is a result of the spreading due to the different
velocities in each layer.
5. The measurement of tracer concentration in space
was inadequate.Under ambientflow conditionsthe tracer is
usually distributed in three-dimensionalspace, but if the
measurements

are two-dimensional

then the actual

tracer

cloudcannotbe analyzedlackingthe appropriatedata. If the

known. Data for which no a priori assumptionswere made tracer is introduced over the entire saturated thickness, then
regardingthe dispersivity were consideredto be highly two-dimensional measurementswould be adequate.
reliable.
6. The equation used to obtain dispersivity was not
A second major problem with many of the analyses appropriate for the data collected. Various assumptions
reviewed was that a one- or two-dimensional
solution to the
regardingflow and solutecharacteristicsare made in obtainequation.To apply
advection-dispersion
equationwas usedwhen the spreading ing a solutionto the advection-dispersion
of the plume under considerationwas three-dimensional
in a particularsolutionto the data from a field experiment,the
nature.High-reliability dispersivitieswere those for which assumptionsin that solution must be consistentwith the
the dimensionalityof the solute plume, the solute measure- experimentalconditions.One common example is the case
of applyinga one-dimensional
{,uniformvelocity) flow soluments,and the data analyseswere consistent.
Low-reliabilitydispersivitydata. A reporteddispersivity tion to a radial flow test in which the converging (or
was classified as being of low reliability if one of the diverging)flow field aroundthe pumpingor injectionwell is
clearly nonuniform.
followingcriteria was met.
Results of classification. From the classificationpro1. The two-well recirculatingtest with a step input was
used.The problemwith thisconfiguration
is that, exceptfor cess, 14 dispersivity values were judged to be of high
very early time where concentrationsare low, the break- reliability. The sites where these values were determined

throughcurve is not stronglyinfluencedby dispersion,

but
ratheris determinedby the differenttraveltimesalongthe
flowpathsestablished
by injectionandpumping
wells[Wetty
andGeIhar, 1989].As a result,the two-welltest with a step
inputis generallyinsensitiveto dispersion.
For thisreason

include Borden, Ontario, Canada; Otis Air Force Base, Cape

Cod, Massachusetts;Hanford, Washington; Mobile, Alabama; University of California, Berkeley; Yavne region,
Israel; Bonnaud,France (six tests); and Palo Alto bay lands.

reliability.

datajudgedto be of intermediatevalue. Figure2 depictsthe

There were 61 valuesjudgedto be of low reliability for one

alltestsof thistypewereconsidered
to produce
dataof low or more of the reasons discussed above; 31 sites provided

dispersivity
datareplottedwith symbolsreflect2. The single-well
injection-withdrawal
test was used longitudinal
the largestsymbolsindicate
withtracer monitoringat the pumpingwell. A difficulty ingthe reliabilityclassification;
encounteredin the small-scale, single-well, injection- datajudgedto be of highestreliability.
The generalcompilationof all dispersivitydatain Figure1
withdrawal
test(wherewateris pumped
intoandoutof one
well)is that if observationsare madeat the productionwell, indicatesthat dispersivitymight increaseindefinitelywith

thedispersion
process
observed
is different
fromoneof scale, but after critically evaluating the data in terms of
unidirectional
flow.Theproblem
stems
fromthefactthat reliabilityas shownin Figure2, it is evidentthat this trend
macrodispersion
nearthe injectionwellis dueto velocity cannot be extrapolatedwith confidenceto all scales. The

dispersivityvalue is 4 m (Mobile,
differences
associated
with layeredheterogeneity
of the largesthigh-reliability
Alabama)
and
the
largest
scaleof high-reliabilityvaluesis
hydraulic
conductivity.
In thesingle-well
testwithobserva-

tions
madeattheproduction
well,theeffect
observed
isthat
ofreversing
thevelocity
ofthewater.If thetracertravels
at
different
velocities
in layersasit radiates
outward,
it will
also
travelwiththesamevelocity
pattern
asit isdrawnback

250m (CapeCod,Massachusetts).
It is notclearfromthese
datawhetherdispersivityincreasesindefinitelywith scaleor
whether the relationshipbecomesconstantfor very large
scales,as wouldbe predictedby sometheories.This points

1968

GELHARET AL.: FIELD=SCALE

DISFERSION
IN AQUIFERS

104 :- ........

, 'i,,,,,,i

"['['"'1

[ : ["'",','[[['j'
,, [ ["[":',[,[ ,,, ,[

' ''"'"l

103

- 102

.%

' 101

a 100
....

RELIABILITY
c::

o10

-1

10'2

10'1

100

101

low

intermediate

high

102

103

104

105

106

Scale (m)

Fig. 2. Longitudinal
dispersivity
versusscalewith dataclassified
by reliability.

to a need for reliable data at scales larger than 250 m.

Whether conducting controlled tracer tests at these very

large scalesis feasible is open to question.

When the reliability of the data is considered,the apparent
difference between fractured and porous media at small
scales(Figure 1) is regardedto be less significantbecause
none of the fractured media data are of high reliability.
Reanalyses of Selected Dispersivity Data
In cases where the concentration data collected were of

a value of 2.4 m without the assumptionof layers. At 50 m,

Sautycalculateddispersivityvaluesof 25 m and6.25 m forthe

two layers;we calculatean overallvalueof 4.6 m. At a scale
0f
150m, Sautycalculateda dispersivityvalue of 12.5m without
the assumption
of layers;our calculationof 10.5m is in close
agreement.
Our calculations
indicatethatdispersivity
increases
with scale,accounting
for nonuniformflow effectsandwithout
the arbitraryassumptionof geologiclayers.
SavannahRiver Plant, Georgia. Websteret al. [1970]
evaluateddata from a two-well recirculatingtest usingthe

methodology
of Groveand Beetern[1971].This analysis

high reliability but the method of analysiscould be imassumes

uniformflowalongstreamtubesandsumsindividproved,we reevaluatedthe datato determinea dispersivity
ual breakthrough
curvesalongthe streamtubesto obtain
a
valuewhichwe judgedto be of higherreliability.The details
composite
breakthrough
curve.
A
dispersivity
value
of
134
rn
of theseanalysesare reportedby Welty and Gelhar [1989].
at a scaleof 538 m was obtainedusingthis method.We
The results are summarized here.
Corbas, France. The data from this converging radial
flow tracer test are reportedby Sauty [1977].Thesedataare

reevaluated
thedatausing
themethodology
ofGelhar[1982]
which accountsfor nonuniformflow effects. We obtained
a

of particularinterestbecausetestswereconductedat three dispersivityvalueof 47 m from our analysis.We havemore

different scales in the same aquifer matehal; tracer was

confidence in this value because the analysis more accu-

injectedat 25, 50, and 150 m from a pumpingwell. Sauty rately representsthe actual flow configuration.
Tucson,
Arizona. Thedatareported
by Wilson[1971]
for
[1977]evaluatedthese data usinguniformflow solutionsto
a
two-well
test
were
also
evaluated
by
Robson
[1974]
using
a
the one-dimensional advection-dispersionequation. At the
two smaller-scale tests, he assumed a two-layer scheme,

Grove
andBeetem-type
analysis.
Wilson
reported
avalue
of

11m and 1.25m for thetwo hypothesized

layers;we calculated

attheColumbus
site(E.E. Adams
andL. W. Ge!har,
Fidd

dispersivity
of !5.2mata scaleof79.2m.Using
althoughthis assumptionwas not supportedby geologic longitudinal
a nonuniform
flow solutionbasedon that of Gelhat[1982],
evidence. For this reason the data at the smaller scales were
a valueof longitudinal
dispersivity
of 1.2m,an
rated to be of lower reliability than the data at !50 m. We we calculated
smallerthanthatof Robson.Again,we
reevaluatedthese data using a solution that accountsfor orderof magnitude
in this valuebecausethe analysis
nonuniform,convergentradial flow effectsand that makes have more confidence
no assumptions
aboutgeologiclayers [Welty and Gelhat, moreaccuratelyreflectsthe actualflow situation.
Columbus,
Mississippi.
Thenaturalgradient
tracer
test
1989].The valuesof dispersivity
reportedby Sautyat 25m are

GELHAR
ETAL.'FIELD-SCALE
DISPERSION
1NAQUIFERS

1969

1o4

103
o

EI

02

'

01

>
''

0 o

oO
,
o

RELIABILITY

o 0.1

--1

intermediate

high
reanalysis

10'2
10-3

10-1

100

101

102

103

104

105

106

Scale (m)

Fig.3. Longitudinal
dispersivity
versus
scale
ofobservation
withadjustments
resulting
fromreanalyses.
Arrows
indicatereported
valuesat tailsandcorresponding
valuesfromreanalyses
at heads.Dashedline connects
two
dispersivity
valuesdetermined
at theHanfordsite.

studyof dispersion
in a heterogeneous
aquifer:Spatial structedas the averageof breakthroughcurves in three
arbitrarilydefinedlayers.We suspectthat this
moments
analysis,submitted
to WaterResources
Research, somewhat
kind
of
localized
observationwill producea significantly
1991;hereinafter
AdamsandGelhar,submitted
manuscript,
thanwouldresultfroma spatialmoments
1991)is uniquein that the large-scale
ambientflow field lowerdispersivity
whichconsiders
theoverallspreading
of theplume.
exhibits
strongnonuniformity
andtheaquiferis veryheter- analysis
of thepossible
increase
in the dispersivity
ogeneous.
A superficial
spatialmoments
interpretation,
ig- The magnitude
becausethe samplingnetwork did not
noting
theflownonuniformity,
indicated
a longitudinal
dis- cannotbe assessed
encompass
theplumeat theTwinLakesite.
persivty
of around70 m, whereas
a morerefinedanalysis completely
thatexplicitly
includes
theinfluence
of flownonuniformity Anotherexampleis thatof the firstBordensitenatural
experiment
[Sudick.w
et al., 1983]whichwasanayieldsa dispersivity
of around7 m (AdamsandGelbar, gradient
lyzed
using
an
analytical
solution
with spatiallyconstant
submitted
manuscript,
1991).This refinedestimate
is reIn thenear-source
region
where
dispersivities
gardedto be of intermediate
reliabilitybecauseof the dispersivities.
increasing
withdisplacement,
thisapproach
will
uncertainty
regarding
themass
balance
attheColumbus
site. areactually
tend
to
underestimate
the
magnitude
of
the
dispersivity.
Fromthe abovereanalyses,
all valuesof dispersivity
the firstBordenexperiment
calculated
were smallerthan the originalvalues.We have Gelharet al. [1985]reanalyzed

of spatial
moments
andfoundthatthe
higher
confidence
in thesevaluesbecause
theyareassoci- usingthemethod
dispersivity
at 11mwas2-4times
thatfound
by
atedwithsolutions
to theadvection-dispersion
equation
with longitudinal
Sudicky
et
al.
[1983].
The
resulting
increase
in
the
dispersivmorerealistic
assumptions.
In all cases
wewouldratethe

inFigure
3connected
totheoriginal
pointby
newvaluesto be of intermediate
reliabilityinsteadof low ityisillustrated
a verticalline. Because
of the incomplete
plumesampling

reliability.
Thereevaluated
dataareshown
assolidsymbols
bifurcation
inthistest(onlythe"slowzone"was
onFigure
3 connected
to theiroriginal
values
by verticalandplume
analyzed),
thispointisstillregarded
to beof intermediate
arrows.
Based
ontheabovereanalyses,
we suspect
thatit ismost

reliability.

Dispersivities
atsmall
displacements
willalsobeundereslikelythatimproved
analyses
wouldreduce
manyof the timated
if basedonbreakthrough
curvesmeasured
in 1ocaIlower-reliability
dispersivities
in Figure2. However,
there
inindividual
layers.
Such
effects
arelikely,for
area few casesfor whichmoreappropriate
observationsizedsamplers
in thePerchLake[Leeet al., 1980]
andLower
and/or
interpretations
would
most
likelyleadtolarger
dis- example,
Glatt
Valley
[Hoehn,
1983]
interpretations.
Later
interpretapersivities.
Forexample,
theTwinLakenatural
gradienttionoftheLowerGlattValleydatausingtemporal
moments
tracer
test[Moltyaner
andKilley,
1988a,
b]wasinterpreted
[Hoehn
and
Santschi,
19871
shows
vues
an
orderof
byusing
breakthrough
curves
at individual
boreholes
con-

1970

GELH^R ET AL.' FIELD-SCALEDISPERSIONIN AQUIFERS

magnitude larger; these are connected with the original

values by vertical lines in Figure 3.
As a further illustration of the uncertainty in the longitudinal dispersivityvaluesin Figure 2, considerthe datafor the

104

103

RELIABILITY

Hanford site. The tracer test [Bierschenk, 1959; Cole, 1972]

low

interpretedfrom breakthroughcurves at two differentwells

at roughly the same distance (around 4000 m) from the lO2
injection point produced values differing by 2 orders of
magnitude (see dashed line in Figure 3). This difference
illustrates the difficulty in interpreting point breakthrough 5
101
curvesin heterogeneousaquifers,even at this large displace-

intermediate

O high

.t

ment. The numerical simulationsof the contaminationplume

[Ahlstrom et al., 1977] extendingto 20,000 m used a disper-

oo

oo

sivity of 30.5 m (100 feet) as identifiedby the bold arrow in

Figure3. Evidentlythisroundnumber(100feet) waspopular
in several different simulations of contaminant plumes.
In none of the cases of simulations of contamination events

c: 0'1

C) o

Ol
N
O

is thereany explicitinformationon howthe dispersivityvalues

were selected or in what sense the values may be optimal.

Consequently
it is not possibleto quantifythe uncertainty
in
dispersivity
valuesbasedon contamination
eventsimulations.
However,experiencesuggests
that, becauseof the possible
tendencyto selectlargedispersivities
whichavoidthe numer-

10-2

10'3

10'1

100

101

ical difficultiesassociatedwith largegrid Pecletnumbers,some

of the dispersivityvaluesbasedon contaminant

plumesare
likely to be biasedtowardhighervalues.Suchoverestimates
would occur mainly at larger scales.
The resultsof these reanalysesprovide an explicit indica-

102

103

104

105

Scale (rn)

Fig. 4. Horizontaltransverse
dispersivity
as a functionof observation

scale.

tion of the uncertaintyin the dispersivityvaluesin Figure2

and suggest
that for largedisplacements
the low-reliability
dispersivities
are likely to decreasewhereasfor smalldis- als. All of the vertical dispersivitiesare less than 1 m and
high-reliability
valuesare only a few millimeters,thisbeing
placements
someincreasescan be expected.
the sameorder of magnitudeas the local transversedisperTransverse Dispersivities

sivity for sandy materials.

The ratioof longitudinal

dispersivity
to the horizontal
and
vertical
transverse
dispersivities
is
shown
in
Figure
6.
This
Althoughthe dataontransverse
dispersivity
are muchmore
isusedbecause
it iscommon
practice
to
limited,they reveal somefeatureswhichare importantin formofpresentation
applications.
The dataon horizontalandverticaltransverse select constant values for the ratio of longitudinal to transFor onething,thisplot illustrates
the
dispersivities
aresummarized
in Figures4 and5, whichshow versedispersivities.
of using,in numerical
simulations,
a horizontal
theseparameters
asa function
ofscaleofobservation.
Thedata popularity
dispersivity
which is about one third of the
are portrayedin termsof reliabilityclassification
with the transverse
longitudinal
dispersivity
(thehorizontal
dashed
lineinFigure
largestsymbols
identifying
thehigh-reliability
points.
for
In the case of the horizontal dispersivity, there appearsto 6). Theredoesnot appearto be any realjustification

usingthisratio.We arenotawareof anysimulation

work
which
systematically
demonstrates
the
appropriateness
of
appearance
results
fromlow-reliability
datawhichfindstheir

be sometrend of increasingdispersivitywith scalebut this

originlargelyin contaminant
eventsimulations
usingtwo- thisvaluefor the horizontaltransversedispersivity.Thetwo
pointsshowan orderof magnitude
higher
dimensional
depth-averaged
descriptions.
In thesecontami- high-reliability
to horizontal
transverse
dispersivities.
nation situationsthe sourcesare often ill-defined; if the ratioof longitudinal
linesin Figure6 are usedto identify
actual sourcearea is larger than that representedin the The verticaldashed
monitored
sitesforwhichallthreeprinmodel there will be greater transversespreadingwhich three-dimensionally
cipalcomponents
ofthedispersivity
tensor
havebeen
estiwouldincorrectlybe attributedto transverse
dispersion.
In the caseof vertical transversedispersion(Figure 5), the
data are even more limited and certainly do not imply any

mated.In all of thesecases,the verticaltransversedisper-

sivity
is1-2orders
ofmagnitude
smaller
thanthehorizontal

dispersivity.
Thisbehavior
furtheremphasizes
significant
trendwithoverallscale.Notethatthereareonly transverse
of verticalmixingwhichis frequently
two pointsof highreliability,thosecorresponding
to the the smalldegree
in naturally
stratified
sediments.
Thissmall
Borden[Freyberg,
1986]andCapeCod[Garabedian
et al., encountered

of vertical
mixing
is clearly
animportant
consider1988, 1991]sites.The estimateof the verticaltransverse degree

inmany
applications,
such
asthedesign
ofobservation
dispersivity
for the Bordensiteis fromthe recentthree- ation
to monitorcontamination
plumes
andthedevelop'
dimensional
analysisof Rajaramand Gelhar[1991].The networks

in orderto
verticaltransversedispersivityis seento be muchsmaller mentof remediationschemes.Consequently,
model
many
field
situations
realistically,
it
will
be
necessary
than the horizontaltransversedispersivity,apparentlyre-

transport
models
which
adequately
flectingthe roughlyhorizontal
stratification
of hydraulictousethree-dimensional
conductivity
encountered
inpermeable
sedimentary
materi- representthe smallbut finite verticalmixing.

GELHAR
ETAL.:FIELD-SCALE
DISPERSION
INAQLilFERS

1971

10o

c lO-2

I:

I in.
termediate
I

10-3
101

102

103

104

105

Scale (m)
Fig. 5. Vertical transversedispersivityas a functionof observationscale.

high-reliability
pointsat scalesgreaterthan300 m and the
high-reliability
pointsaresystematically
in thelowerportion
of
the
scattering
of
data.
The
lack
of
high-reliability
data at
aquifersdemonstrates
severaloverallfeatureswhich are
scales
greater
than
300
m
reflects
the
fact
that
the
data
evidentfrom the graphical and tabular informationdevelbeyond
that
scale
are
almost
exclusively
from
contamination
opedhere.Takenin aggregate,
withoutregardfor reliability,
INTERPRETATIONS

This review of field observations of dispersive mixing in

simulations or environmental tracer studies for which the

the data indicate a clear trend of systematic increase of

longitudinal
dispersivity
with scale.In termsof aquifertype soluteinputis typicallyill-defined.Becauseof the very long
(porous
versusfracturedmedia)the dataat smallerscale periodof timerequiredto carryout controlledinputtracer
at theselargerscales,suchexperiments
have
mayseemto behigherfor fracturedmediabut,in viewof the experiments
lowerreliabilityof thefracturedmediadata,thisdifference
is not been undertaken.
Althoughthe data shownin Figure 2 suggestthat some
of minimal significance.
overall
trendof increasingdispersivitywith scaleis plausiWhenthe data on longitudinaldispersivityare classified
to concludethat a single
according
to reliability,the patternregarding
scaledepen- ble, it doesnot seemreasonable
identified
dence
of dispersivity
islessclear(seeFigure2). Thereareno universalline[Neuman,1990]canbe meaningfully

. 103

o AL/A
H
A AL/A
v

102

>,

10

--

'

ol

II
II

II
II

II

,6

:.

o 0

oo

'

100 -

t ,I tttlltl

.- 0_1
..a
o 1 I 0'1

10
0

I f illitel t i I,ttll

101 10
2

I I IIIIlil

!03

I I II,I!

10
4

I._!

!ltll

105

Scale (m)

Fig.6. Ratio
oflongitudinal
tohorizontal
and
vertical
transverse
dispersivities;
largest
symlxls
are
high
reliability
and
smallest
symbols
are
low
reliability.
Vertical
dashed
lines
connecting
two
points
indicate
sites
where
allthree
principal
components
ofthe
dispersivity
tensor
have
been
measured.
Horizontal
dashed
line
indicates
aratio
ofA/A
r
= 1/3,which
hasbeen
widely
used
innumerical
simulations.

1972

GELHARET AL.: FIELD-SCALEDISPERSIONIN AQUIFERS

by applying standard linear regressionto all of the data.

Rather we would expect a family of curvesreflectingdifferent dispersivitiesin aquiferswith differentdegreesof heterogeneity. At a given scale, the longitudinal dispersivity
typically rangesover 2-3 ordersof magnitude.This degreeof
variation can be explained in terms of the establishedstochastictheory [e.g., Gelhat and Axness, 1983;Dagan, 1984]
which shows that the longitudinal dispersivity is proportional to the product of the variance and the correlationscale
of the natural logarithmof hydraulic conductivity.A compilation of data on these parameters [Gelhat, 1986] showsthat
they vary over a range that can easily explain the range of
variation in Figure 2. The theoretical results for the developing dispersionprocess [Gelhat et al., 1979;Dagan, 1984;
Gelhat, 1987; Naff et al., 1988] show that the longitudinal
dispersivity initially increases linearly with displacement
distance and gradually approaches a constant asymptotic
value [see Gelhat, 1987, Figure 9]. One could visualize the
behavior of Figure 2 as being the result of superimposing
several such theoretical curves with different parameters
characterizing aquifer heterogeneity.
The results of reanalyses for several of the individual sites
serve to illustrate explicitly the uncertainty involved in the
estimates of longitudinal dispersivity. The reanalyses indicate that, for the most part, improved analysis will lead to
decreasesin the longitudinal dispersivity except possiblyfor
very small displacementswhere limited localized sampling
can produce underestimatesof the bulk spreadingand mixing. In caseswhere the dispersivity estimateswere basedon
numerical simulations of contamination events, the degreeof
uncertaintyis likely large and ill-determined,but bias in some
of the estimatestoward the high side seemsmost likely.
From an applicationsperspective, the information assembled here should serve as a strong cautionary note about
routinely adopting dispersivitiesfrom Figure 2 or a linear
regressionrepresentationthrough the data. We feel that the
preponderanceof evidence favors the use of dispersivity
values in the lower half of the range at any given scale. If
values in the upper part of the range are adopted, excessively large dilution may be predicted and the environmental

Acknowledgments.
Theworkwas'supported
in partbythe

ElectricPowerResearchInstitute(EPRI), project2485-5,which

wasa jointeffortof theMassachusetts

Instituteof Technology
(MIT) andthe Tennessee
Valley Authority(TVA). Thisportion
of
the work was doneat MIT undercontractTV-61664AwithTYA.

Theworkwasalsosupported
bytheNationalScience
Foundation,
grant CES-8814615.

REFERENCES

Ahlstrom, S. W., H. P. Foote, R. C. Arnett, C. P. Cole, andR. J.

Serne, Multicomponent mass transport model: Theory and nu.

mericalimplementation
(discrete-particle-random-walk-version),
Rep. BNWL-2127, Battelle Pac. Northwest Lab., Richland,
Wash., 1977.

Anderson, M.P.,
Using models to simulate the movement0f
contaminantsthrough groundwater flow systems, CRC Crit. Rev.
Environ. Control, 9, 97-156, 1979.

Bassett, R. L., et al., Preliminary data from a series of artificial

rechargeexperimentsat Stanton,Texas, U.S. Geol. Surv.Open

File Rep., 81-0149, 1980.

Bear, J., Dynamicsof Fluids in Porous Media, Elsevier Scientific,

New York, 1972. (Reprinted by Dover, New York, 1988.)
Beims, U., Planung, Durchfihrung und Auswertung yon
Gtepumpversuchen, Z. Angew. Geol., 29(10), 482-490, 1983.
Bentley, H. W., and G. R. Walter, Two-well recirculatingtracer
tests at H-2: Waste Isolation Pilot Plant (WlPP), southwestNew
Mexico, draft paper, Hydro Geochem., Inc., Tucson, Ariz., 1983.
Betson, R. P., J. M. Boggs, S.C. Young, W. R. Waldrop, andL. W.
Gelhar, Macrodispersion experiment (MADE): Design of a field
experiment to investigate transport processes in a saturated
groundwater zone, Rep. EPRI EA-4082, Elec. Power Res. Inst.,
Palo Alto, Calif., June 1985.
Bierschenk, W. H., Aquifer characteristics and ground-watermove.
ment at Hanford, Rep. HW-60601, Hanford At. Products Oper.,
Richland, Wash., 1959.
Bredehoeft, J. D., and G. F. Pinder, Mass transport in flowing
groundwater, Water Resour. Res., 9(!), 144-210, 1973.
Claasen, H. C., and E. H. Cordes, Two-well recirculating tracertest
in fractured carbonate rock, Nevada, Hydrol. $ci. Bull.., 20(3),
367-382, 1975.

Cole, J. A., Some interpretations of dispersion measurements

in
aquifers,GroundwaterPollution in Europe, edited by J. A. Cole,
pp. 86-95, Water ResearchAssociation,Reading,England,1972.
Dagan, G., Stochasticmodelingof groundwaterflow by unconditional and conditionalprobabilities,2, The solutetransport,Water
Resour. Res., 18(4), 835-848, 1982.

Dagan,G., Solutetransportin heterogeneous

porousformations,
J.
Fluid Mech., 145, 151-177, 1984.

Daniels,W. R. (Ed.), Laboratoryfield studiesrelatedto theradionuclide migration project, Progress Rep. LA-8670-PR, Los
consequencesmisrepresented.In the case of transverse

dispersivities,it is particularly important to recognizethe

very low vertical transverse dispersivitiesthat have been
observed at several sites. As a result, many contamination

Alamos Sci. Lab., Los Alamos, N.M.,

1981.

Daniels,W. R. (Ed.), Laboratoryfield studiesrelatedto theradionuclidemigrationproject(draft),ProgressRep. LA-9192-PR,Los

Alamos Sci. Lab., Los Alamos, N.M.,

1982.

plumes will exhibit very limited vertical mixing with high Davis, S. N., G. M. Thompson,H. W. Bentley, and G. Stiles,
Groundwater tracers--A short review, Ground Water, 18(1),
concentrations at a given horizon. The recognition of such
14-23, 1980.
features is of obvious importance in designingmonitoring Davis,S. N., D. J. Campbell,H. W. Bentley,andT. J. Flynn,An
schemesand implementingaquifer remediation.Horizontal
introduction
to groundwater
tracers,Rep. EPA/600/2-85/022,
Environ.Prot.Agency,Washington,
D.C., 1985.(Available
asNTIS
transversedispersivitiesare typically an order of magnitude
PB86-10059Ifrom Natl. Tech. Inf. Serv., Springfield,Va.)
smaller than the longitudinaldispersivitywhereas vertical
de Marsily,G., Quantitative
Hydrogeology,
Academic,SanDiego,
transversedispersivities
are anotherorderof magnitudelower.
Calif., 1986.
From a research perspective, the data reviewed here Dieulin,A., Propagation
de pollutiondansun aquiferealluvial:
L'effetdeparcours,
doctoral
dissertation,
Univ.PierreetMarie
suggesta need for someskepticismregarding"universal"
Curie-Paris
VI and l'Ecole Natl. Super.des Minesde Paris,
modelswhich representthe scattereddata of varying reliFontainebleau, France, 1980.
.ability by a singlestraightline. The presumptionof sucha Dieulin,A., Lixiviationin situd'ungisement
d'uranium
enmilieu
universalmodel ignoresthe fact that different:.tquiferswill
granitique,
Draft Rep. LHM/RD/81/63,
EcoleNatl. Super.
des
Mines de Paris, Fontainebleau, France, 1981.
have differentdegreesof heterogeneityat a given scale.The
B.C. E., J. A. Cherry,
R. N. Farvolden,
andE. O.Ffind,
data suggestthat there is a scaledependence
of longitudinal Egboka,
Migration
of
contaminants
in
groundwater
at
a
landfill:A case
dispersivity
butreliabledatamustbedeveloped
at largerscales

study,3, Tritiumasanindicator
of dispersion
andrecharge,
J.

in order to establishthe nature of the dependence.Clearly,

Hydrol., 63, 51-80, 1983.
there is a need for very large scale, long-term, carefully Fenske,
P. R., Hydrology
andradionuclide
transport,
monitoring

plannedexperiments
extending
to severalkilometers.

wellHT-2m,TatumDome,Mississippi,
Proj.Rep.25,Tech.
Rep.

GELlIAR
ETAL.:FIELD-SCALE
DISPERSION
IN AQUIFERS
NVD-1253-6,Cent. for Water Resour.Res., DesertRes. Inst.,
Univ. of Nev. Syst., Reno, 1973.

1973

Ivanovitch, M., and D. B. Smith, Determination of aquifer param-

Freyberg,
D. L., A naturalgradient
experiment
onsolute
transport

etersby a two-wellpulsedmethodusingradioactive
tracers,J.
ttydrol., 36(1/2), 35-45, 1978.

in a sand aquifer, 2, Spatial momentsand the advectionand

Kies,B., Solutetransport
in unsaturated
fieldsoilandin groundwater,Ph.D. dissertation,
Dep. of Agron.,N.M., StateUniv.,
dispersion
of nonreactive
tracers,WaterResour.Res., 22(13),
2031-2046,1986.

Las Cruces. 1981.

Klotz, D., K. P. Seller,H. Moser,and F. Neumaier,Dispersivity

Fried,J. J., GroundwaterPollution,Elsevier,New York, 1975.
fromlaboratory
andfieldexperiments,
J.
Fried,J. J., andP. Ungemach,Determination
in situdu coefficient andvelocityrelationship
Hydrol., 45(3/4), 169-184, 1980.
de dispersion
1ongitudinale
d'un milieu poreuxnaturel,C. R.
Acad. Sci., $er. 2,272, 1327-1329, 1971.
Konikow,L. F., Modelingsolutetransportin groundwater, in
Garabedian,
S. P., L. W. Gelhar, and M. A. Celia, Large-scale Environtnental
Sensingand Assessment:
Proceedingsof the
Conference,
article20-3,Institutefor Electricaland
dispersive
transportin aquifers:Fieldexperiments
andreactive International
ElectronicEngineers,Piscataway,N.J., 1976.
transporttheory, Rep. 315, RalphM. ParsonsLab. for Water
Resour. and Hydrodyn., Mass. Inst. of Technol., Cambridge, Konikow,L. F., andJ. D. Bredehoeft,Modelingflowand chemical
1988.
quality changesin an irrigatedstream-aquifersystem, Water
Resour. Res., I0(3), 546-562, 1974.
Garabedian,
S. P., D. R. LeBIanc, L. W. Gelhat, andM. A. Celia,
Large-scale
naturalgradienttracertestin sandandgravel,Cape Kreft, A., A. Lenda, B. Turek, A. Zuber, and K. Czauderna,
Determinationof effectiveporositiesby the two-well pulse
Cod,Massachusetts,
2, Analysisof tracermomentsfor a nonremethod,lsot. Tech.GroundwaterHydrol., Proc. Syrup.,2, 295active tracer, Water Resour. Res., 27(5), 911-924, 1991.
Gelhar,L. W., Analysisof two-well tracer testswith a pulseinput,

3 !2, 1974.

A., andP. Peaudecerf,Recherchedesrelations

Rep.RHO-BW-CR-I31P, RockwellIntl., Richland,
Wash.,1982. Lallemande-Barres,
entre la valeurde la dispersivit6macroscopique
d'un milieu
Gelhar,L. W., Stochasticsubsurfacehydrologyfrom theory to
aquifre,sesautrescaractristiques
et lesconditions
de mesure,
applications,
WaterResour.Res.,22, 135S-145S,
1986.

Gelhar,L. W., Stochasticanalysisof solutetransportin saturated Bull. Bur. Rech. Geol. Min., Sect. 3, Ser. 2, 4, 1978.
and unsaturatedporous media, NATO ASI Ser., $er. E, 128, Lau, L. K., W. J. Kaufman,andD. K. Todd, Studiesof dispersion
in a radialflow system,Canal SeepageResearch:Dispersion
657-700, 1987.
in Flow ThroughPorousMedia. ProgressRep. 3,
Gelhar,L. W., and C. L. Axness,Three dimensional
stochastic Phenomena
I.E.R. Ser. 93, Issue3, Sanit.Eng. Res. Lab., Dep. of Eng. and
analysisof macrodispersion
in aquifers,Water Resour.Res.,
Schoolof PublicHealth, Univ. of Calif., Berkeley, 1957.
19(1), 161-180, 1983.
Gelhar,L. W., A. L. Gutjahr,andR. L. Naff, Stochastic
analysis
of LeBlanc,D. R., Sewageplumein a sandandgravelaquifer,Cape

macrodispersion
in a stratified
aquifer,WaterResour.Res.,15(6),

Cod,Massachusetts,
U.S.Geol.Surv.OpenFileRep.,82-274,35

1387-1397, 1979.

pp., 1982.

itique(siteB), Rep.LHM/RD/82/11,
Cent.d'Inf. Geol.,Ecole
Natl.Super.desMinesde Paris,Fontainebleau,
France,1982.

Chapman
Conference
onSpatial
Variability
in Hydrologic
Mod-

fromNatl. Tech. Inf. Serv., Springfield,Va.)

Migration
of contaminants
at a landfill:A casestudy,1, Ground-

Gelhar,L. W., A. Mantaglou,C. Welty, and K. R. Rehfeldt,A Lee, D. R., J. A. Cherry,andJ. F. Pickens,Groundwatertransport
of a salt tracerthrougha sandylakebed,Limnol. Oceanogr.,
reviewof field-scalephysicalsolutetransportprocesses
in satu25(1), 46-61, 1980.
ratedand unsaturatedporousmedia,EPRI Rep. EA-4190,Elec.
Leland,D. F., and D. Hillel, Scaleeffectson measurement
of
PowerRes. Inst., Palo Alto, Calif., Aug. 1985.
dispersivity
in a shallow,
unconfined
aquifer,paperpresented
at
Goblet,
P., Interpretation
d'experiences
de tracageenmilieugran-

eling,AGU, FortCollins,Colo.,July21-23, 1981.

Grove,D. B., The useof Galerkinfinite-element
methods
to solve Lenda,A., andA. Zuber,Tracerdispersionin groundwaterexperiments,in !sot. Hydrol.Proc.Symp.1970,619-641,1970.
masstransportequations,
Rep. USGS/WRD/WRI-78/011,
U.S.
D. S., J. A. Cherry,R. W. Gilham,andE. A. Sudicky,
Geol.Surv.,Denver,Colo.,1977.(AvailableasNTIS PB277-532 MacFatlane,

J. Hydrol.,63, 1-29, 1983.

Grove,D. B., andW. A. Beetem,Porosityanddispersion
constant waterflowandplumedelineation,
andmixingtestsat Yavne20 well field,
calculations
for a fracturedcarbonate
aquiferusingthe two-well Mercado,A., Recharge
Underground
WaterStorage
StudyTech.Rep.12, Publ.611,
tracermethod,WaterResour.Res., 7(1), 128-134,!971.

Gupta,
S. K., K. K. Tanji,andJ. N. Luthin,A three-dimensionalTahal-Water Plann. for Isr., Tel Aviv, 1966.
andD. Stephenson,
finiteelementgroundwatermodel,Rep. UCAL-WRC-C-152,Meyer,B. R., C. A. R. Bain,A. S. M. DeJesus,
Calif.WaterResour.Cent.,Univ. of Calif.,Davis,1975.(Avail-

Radiotracerevaluationof groundwaterdispersionin a multi-

layered
aquifer,
J. Hydrol.,50(1/3),259-271,1981.
Molinari,J., andP. Peaudeceff,
Essaisconjoints
en laboratoire
et
simplefi6e
dela pr6visiondes
Halevy,
E., andA. Nir, Determination
of aquifer
parameters
with surle terrainenrue d'uneapproach
desubstances
miscibles
danslesaquifres
r6els,
theaidof radioactive
tracers,J. Geophys.
Res.,67(5),2403-2409, propagations
ableasNTIS PB 248-925fromNatl. Tech. inf. Serv.,Springfield,
Va.)

paper
presented
at Symposium
onHydrodynamic
Diffusion
and
1962.
inPorous
Media,
Int.Assoc.
forHydraul.
Res.,Pavis,
Harpaz,
Y., Fieldexperiments
in recharge
andmixingthrough Dispersion
wells,Underground
WaterStorage
StudyTech.Rep.17,Publ. Italy, 1977.
Mo!tyaner,
G. L., andR. W. D. Kil!ey,TwinLaketracertests:
483,Tahal-WaterPlann.for Isr., Tel Aviv, 1965.
dispersion,
WaterResour.
Res.,24(10),16!3-1627,
Heller,
J.P.,Observations
ofmixing
anddiffusion
inporous
media, Longitudinal
1988a.
Proc.Syrup.
Fundam.
Transp.
Phenom.
Porous
Media,2nd, Moltyaner,
G. L., andR. W. D. Kil!ey,TwinLaketracertests:
1-26, 1972.
dispersion,
WaterResour.Res.,24(10),1628-1637,
Helweg,
O.J.,andJ.W.Labadie,
Linked
models
formanaging
river Transverse
1988b.

basin
saltbalance,
WaterResour.
Res.,13(2),329-336,
!977.
F. J., O. Given,andJ. G. Melville,An examination
of
Hoehn,
E., Geological
interpretation
of local-scale
tracerobserva-Molz,
scale-dependent
dispersion
coefficients,
Ground
Water,21,715tionsin a river-ground
waterinfiltration
system,
draftreport, 725,
1983.
Swiss
Fed.Inst.Reactor
Res.(EIR),Wtirenlingen,
Switzerland,
Molz,F. J., O. Given,
J. G. Melville,R. D. Crocker,
andK. T.
1983.
Matteson,
Performance,
analysis,
andsimulation
of a two-well
Hoehn,
E., andP. H. Santschi,
Interpretation
of tracer
displace- tracertestat the Mobilesite,WaterResour.Res.,22(7), 1031-

mentduring
infiltration
of riverwaterto groundwater,
Water

Resour.Res., 23(4), 633-640, 1987.

1037, 1986.

R. L., T.-C.J. Yeh,andM. W. Kemb!owski,

A noteonthe
Huyakorn,
P.S.,P.F. Anderson,
F.J.Motz,O.Gtiven,
andJ.G. Naff,
recentnaturalgradienttracertest at the Bordensite, Water
Melville,
Simulations
oftwo-well
tracer
tests
instratified
aquifers Resour.Res,, 24(12),2099-2103,1988.
at theChalkRiverandthe Mobilesites,WaterResour.Res.,

22(7), 1016-1030, 1986.

Iris,P.,Contribution
l'rtudedela valorisation
energetique
des

Naymik,
T. G.,andM.J.Barcelona,
Characterization
ofa contaminantplume
inground
water,Meredosia,
Illinois,
Ground
Water,

!9(5), 517-526, 1.98!.

aquiffres
peuprofonds,
thsededocteur-ingenieur,
Ecole
des Neretnieks,
!., Transport
infractured
rocks,paper
presented
atthe

Minesde Paris,Fontainebleau,
France,1980.

1974

GELHAR
ETAL.: FIELD-SCALE
DISPERSION
IN AQUIFERS

17thInternationalCongresson the Hydrologyof Rock of Low

Permeability,Intl. Assoc. of Hydrogeol., Tucson,Ariz., Jan.

Sauty,
J.P., Ananalysis
of hydrodispersive
transfer
inaquifers,

7-12, 1985.

Sauty,J.P., A. C. Gringarten,
andP. A. Landel,Theeffects
of

Water Resour. Res., 16(1), 145-158, 1980.

Neuman, S. P., Universal scalingof hydraulicconductivitiesin

thermal
dispersion
oninjection
of hotwaterin aquifers,
paper
geologicmedia, Water Resour.Res., 26(8), 1749-1758,1990.
presented
at Invitational
Well-Testing
Symposium,
Lawrence
Neuman, S. P., C. L. Winter, and C. M. Newman, Stochastic
Berkeley Lab., Berkeley, Calif., 1978.
theory of field-scaledispersionin anisotropicporousmedia, Segol,G., andG. F. Pinder,Transientsimulation
of saltwater
Water Resour. Res., 23(3), 453-466, 1987.

intrusion
in southeastern
Florida,WaterResour.
Res.,12(1),

65-70, 1976.
New ZealandMinistryof Worksand Development,Water andSoil
Division,Movementof contaminants
into andthroughthe Here- Sudicky,E. A., J. A. Cherry,and E. O. Frind, Migration
of
taungaPla.insaquifer, report, Wellington, 1977.
contaminants
in groundwater
at a landfill:A casestudy,4,A
Oakes,D. B., andD. J. Edworthy,Fieldmeasurement
of dispersion natural-gradient
dispersiontest, J. Hydrol., 63, 81-108, 1983.
coefficientsin the United Kingdom,in GroundWater Quality, Sykes,J. F., S. B. Pahwa,R. B. Lantz, andD. S. Ward,Numerical
Measurement,Prediction, and Protection, pp. 327-340, Water
simulation
of flowandcontaminant
migration
at anextensively
ResearchCentre, Reading, England, 1977.
monitoredlandfill,WaterResour.Res., 18(6),1687-1704,1982.
Papadopulos,S.S., and S. P. Larson, Aquifer storageof heated Sykes,J. F., S. B. Pahwa,D. S. Ward, and D. S. Lantz,The

water: II, Numerical simulationof field results, Ground Water,

16(4), 242-248, 1978.

Perimutter,N.M., and M. Lieber, Dispersalof platingwastesand

sewage contaminants in the groundwater and surface water:

South Farmingdale-Massapequid
area, Nassau County, New
York, U.S. Geol. Surv. Water SupplyPap., 1879-G, 1970.
Philip, J. R., Issuesin flow and transportin heterogeneous
porous
media, Transp. Porous Media, 1,319-338, 1986.

Pickens, J. F., and G. E. Grisak, Scale dependentdispersionin a

stratifiedgranularaquifer, Water Resour. Res., 17(4), 1191-12!!,
1981.

validation
of SWENT,a geosphere
transport
model,in Scientific
Computing,editedby R. Staplemanet al., pp. 351-361,IMAESI

North-Holland, Amsterdam, 1983.

Vaccaro,J. J., and E. L. Bolke,Evaluationof waterquality

characteristics
of part of the Spokaneaquifer,Washington
and
Idaho, using a solute transport digital model, U.S. Geol. Surv.
Open File Rep., 82-769, 1983.

Valocchi,A. J., P. V. Roberts,G. A. Parks, and R. L. Street,

Simulation
of the transportof ion-exchanging
solutesusinglabo.
ratory-determined
chemicalparametervalues, GroundWater,
I9(6), 600-607, 1981.

Pinder, G. F., A Galerkin-finite element simulationof groundwater Walter, G. B., Convergent flow tracer test at H-6: Waste isolation
contaminationon Long Island, Water Resour. Res., 9(6), 1657pilot plant (WlPP), southeastNew Mexico (draft), Hydro
1669, 1973.
Geochem, Inc., Tucson, Ariz., 1983.
Rabinowitz, D. D., and G. W. Gross, Environmental tritium as a
Webster, D. S., J. F. Procter, and J. W. Marine, Two-well tracer
hydrometeorologictool in the Roswell Basin, New Mexico, Tech.
testin fracturedcrystallinerock, U.S. Geol. Surv., WaterSupply
Completion Rep. OWRR:A-O37-NMEX, N.M.
Water Resour.
Pap., 1544-1, 1970.
Res. Inst., Las Cruces, 1972.
Welty, C., andL. W. Gelhar,Evaluationof longitudinaldispersivity
Rajaram, H., and L. W. Gelhar, Three-dimensionalspatialmoments
from tracer test data, Rep. 320, Ralph M. ParsonsLab. for Water
analysis of the Borden tracer test, Water Resour. Res., 27(6),
Resour. and Hydrodyn., Mass. Inst. of Technol., Cambridge,
1989.
1239-!251, 1991.
Roberts, P. V., M. Reinhard, G. D. Hopkins, and R. S. Summers, Werner, A., et al., Nutzung von Grundwasserfur Warmepumpen,
Advection-dispersion-sorption models for simulating the transVersickerrungstestAefligen, Versuch 2, 1982/83, Water andEnport of organic contaminants,paper presentedat International
ergy Manage. Agency of the State of Bern, Switzerland, 1983.
Conference on Ground Water Quality Research, Rice Univ.,
Wheatcraft, S. W., and S. W. Tyler, An explanation of scaleHouston, Tex., 1981.
dependentdispersivity in heterogeneousaquifers usingconcepts
Robertson, J. B., Digital modeling of radioactive and chemical
of fractal geometry, Water Resour. Res., 24(4), 566-578, 1988.
waste transport in the Snake River Plain aquifer of the National Wiebenga,W. A., et al., Radioisotopesas groundwatertracers,J.
Reactor Testing Station, Idaho, U.S. Geol. Surv. Open File Rep.,
Geophys. Res., 72(16), 4081-4091, 1967.
ID0-22054, 1974.
Wilson, L. G., Investigations on the subsurfacedisposal of waste
Robertson, J. B., and J. T. Barraclough, Radioactive and chemical
effluentsat inland sites, Res. Develop. Progress Rep. 650, U.S.
waste transport in groundwater of National Reactor Testing
Dep. of Interior, Washington, D.C., 1971.
Station: 20-year case history and digital model, Underground Wood, W., A geochemical method of determining dispersivityin
Waste Manage. Artif. Recharge Prepr. Pap. Int. Symp. 2nd, 1,
regionalgroundwater systems,J. Hydrol., 54(!/3), 209-224, 1981.
291-322, 1973.
Wood, W. W., and G. G. Ehrlich, Use of baker's yeast to trace
Robson, S. G., Feasibility of digital water quality modeling illusmicrobial movement in ground water, Ground Water, I6(6),
398-403, 1978.
trated by application at Barstow, California, U.S. Geol. Surv.
Water Resourc. Invest., 46-73, 1974.
Robson, S. G., Application of digital profile modelingtechniquesto
L. W. Gelhar,RalphM. ParsonsLaboratory,DepartmentofCivil
ground-water solute transport at Barstow, California, U.S. Geol. Engineering,MassachusettsInstitute of Technology, Cambridge,
MA 02139.
Surv. Water Supply Pap., 2050, 1978.
Rousselot, D., J.P. Sauty, and B. Gaillard, Etude hydrog6ologique
K. R. Rehfeldt,
IllinoisStateWat'rSurvey,
2204Griffith
Drive,
de la zone industrielle de Blyes-Saint-Vulbas, rapport pr61iminaire Champaign, IL 61820.
no. 5: Caracteristiques hydrodynamiques du systme aquifre,
C. Welty, Departmentof Civil and ArchitecturalEngineering,
Rep. Jal 77/33, Bur. de Rech. Geol. et Min., Orleans, France,
Drexel University, Philadelphia, PA 19104.
1977.

S.auty, J.P., Contribution b. l'identification des parameters de

dispersiondansles aquifres par interpr6tationdesexp6riencesde
tracage, dissertation, Univ. Sci. et Med. et Inst. Natl. Polytech.
de Grenoble, Grenoble, France, 1977.

(Received April 8, 1991;

revised March 4, 1992;

accepted March !2, 1992.)