Download as pdf or txt
Download as pdf or txt
You are on page 1of 11

VOL. 5, NO.

2 WATER RESOURCES RESEARCH APRIL 1969

TheEffecto[ BendsonDispersion
in Streams

HUGO B. FISCHER

University of California, Berkeley, California 9•7œ0

Abstract. Bends in streams induce secondary currents that alter the rates of both transverse
mixing and longitudinal dispersion.Within a bend the transversemixing coefficientdependson
the square of the mean velocity, cube of the depth, and inversely on the shear velocity and
square of the radius of curvature, as verified by a laboratory experiment.The longitudinal
dispersioncoefficientdependson the channelgeometry, velocity distribution, rate of transverse
mixing, and a dimensionlessparameter that includes the mean velocity and length of an
average bend. A numerical program for predicting dispersion coefficientsin streams differs
from previous theoriesby including the effect of the alternating direction of curvature in a
series of bends; the difference is important for wide streams but not for narrow ones. Experi-
mental findings in two streams, the Green-Duwamish in Washington (width 20 meters), and
the Missourinear Omaha, Nebraska (width 180 meters), verify the numericalprogram.

INTRODUCTION Fischer [1968b] suggested


that in a meander-
Longitudinaldispersion is the actionby which ing stream one should measure a number of
a flowing stream spreadsout and dilutes an typical channel geometriesand velocity distri-
initially concentratedtracer material. If the butions at various cross sections within the
stream is infinitely long and of uniform geom- bends,determinea dispersioncoefficientfor each
etry, Taylor [1953, 1954] has shownthat for one individually, and average the results. For
asymptoticallylong time the dispersionprocess each calculation this procedure effectively as-
may be describedby a one-dimensional diffusion sumesthat the stream consistsof one infinitely
equation long, uniform bend in one direction,sinceonly
one geometry is used at a time. The method is
06/Orq- a(O6/Ox) = D(O"6/Ox • (1) only approximatebut has yielded accuratepre-
where • is the cross-sectional mean concentra- dicted dispersioncoefficientsin 7 small streams
tion, x is distancealong the stream, t is time, (maximum width 70 meters).
•2 is the cross-sectionalmean flow velocity, and This paper presentsan extensionof the previ-
D is termed the longitudinal dispersioncoeffi- ous analysis to the case of an infinitely long
cient. Taylor gave an analysis that predicted series of uniform bends in alternate directions.
the dispersioncoefficientfor either laminar or It will be shownthat in narrow streams,suchas
turbulent flow in a round pipe. Aris [1956] ex- thosepreviouslytested,it makeslittle difference
tended Taylor's work to a general cross-sec- whether the bends alternate in direction or are
tional diffusion coet•cient. Taylor's conceptwas continuousand uniform in one direction only.
applied to an infinitely wide, two-dimensional In wider streams, however (a strict definition
streamwith power-lawvelocitydistributionby of 'wider' is given later), the predicteddisper-
Thomas [1958], and to the same flow with sion coefficient for a series of bends that alter-
logarithmic velocity distribution by Elder nate in direction may be much lower than that
[1959]. Fischer [1967a] showedhow the con- for the same series of bends all in the same di-
cept couldbe further extendedand appliedto rection. A computer program is describedthat
the type of crosssectionoftenfoundin natural has predictedthe dispersioncoefficientin a 200-
streams. However, all of the above work was meter-wide, meandering reach of the Missouri
limited to infinitely long streams of uniform River.
geometry,whereasin all natural streamsthe
CONCEPT OF CONVECTIVE DISPERSION
cross-sectionM
geometryvaries along the course
of the stream. The convectivediffusionequation,as pre-
496
Dispersion in Streams • 497
sentedby numerouswriters, has the physical stream velocities cause transverse mixing be-
meaningthat marked fluid particlesmove along tween stream tubes,which tendsto equalizethe
the mean flow stream lines, while at the same net downstream motion of the ensemble of
time exchangingbetweenstreamlinesvia molec- tracer particles. To determine the rate of longi-
ular or turbulent diffusion.This conceptcan be tudinal dispersionit is necessaryto know both
expanded in scale to a convective-dispersion the geometry and velocity distribution at each
model by visualizinga flow made up of a num- crosssection and the rate of transversemixing.
ber of separatestreamtubesboundedby stream In this paper, we first considerthe effect of
surfaces.Within a given stream tube all the bends on transversemixing; then we consider
fluid particles are carried downstreamat the the effect of a seriesof bends on longitudinal
stream tube velocity, which may vary along dispersion.
the stream.At the sametime, marked fluid par-
EFFECT OF BENDS ON TRANSVERSE NIIXING
ticlesarecontinuously
exchanged
between
ad-
jacent stream tubes, at a rate proportional to Theoretical analysis. No theory exists at
the concentration gradient at the stream sur- present to describethe rate of mixing of a
face. Thus a particular marked fluid particle tracer in the directiontransverseto a straight,
may migrate from stream tube to stream tube uniform flow. In the vertical direction,for an
but moves downstreamat the velocity of the infinitely wide flow, the Yon Karman loga-
stream tube in which it happensto be. rithmic velocitydistributionleadssemiempiri-
The use of this conceptin a meanderingflow cally to an averagevertical mixingcoefficient
is illustrated by Figure 1. The flow has been
divided into five stream tubes of equal discharge e• = 0.067 dU* (2)
but varying velocity and area. The dividing
stream surfacesare assumedto be vertical, as in
where e• is the vertical mixing coefficient,d is
the local depth of flow, and U• is the bed shear
the writer's previous work, but the widths,
velocity. One might expect, becauseof the ab-
depths, and velocitiesof the stream tubes vary
senceof restricting boundaries,that the trans-
in the streamwisedirectionaccordingto position
verse mixing coefficient would be somewhat
along the curve. Around a curve to the left, the
dividing stream surfacesare displacedtowards greater; experiments in uniform flow [Orlob,
the right (outside) bank, where the velocities 1959; Sayre and Chang, 1968; Fischer, 1967b]
and depths are greater; around a curve to the have given as an acceptableformula
right the oppositeis true.
ez = 0.23 dU* (3)
Longitudinaldispersionis the sum of two ef-
fects: (a) the average velocities of the stream where e• is the transverse mixing coefficient.In
tubes in the center of the flow are higher than a curving reach of the Missouri River, Yotsu-
these at the side, so that tracer particles in the kura el al. [1968] reported the value
center stream tubes are carried downstream
faster; and (b) the transverse concentration ez = 0.6 dU* (4)
gradients induced by the differential down- A probable explanationfor the higher rate of

SECTION A-A

Fig. 1. Schematicdivision of meanderingflow into stream tubes of equal discharge.Not to


_scale.
498 HUGO B. FISCHER

transversemixingin a curvingflow is the exist- where ur is the radial velocity, k is Von Kar-
ence of spiral secondarycurrents,termed by man's constant,u• is the depth-meanvelocity in
Prandtl [1952] 'secondarycurrentsof the first the longitudinaldirection,d is the local depth of
kind.' These currentsare inducedby the vari- flow, R is the radius of curvature, f• is the
ation in centrifugalforce whenevera flow is re- Darcy-Weisbachfraction factor, • is a dimen-
quiredto rounda bend.The net flowthrougha sionlessvertical coordinate (y/d), and F• and
dividing stream surfacewill be zero by the F, are dimensionless functionsplotted by Roz-
definition of the stream surface,but there will ovskii (page 42), that have zero mean and
in generalbe a flowthroughthe surfacetowards range from valuesof approximately1.3 and 0.4
the outsideof the bend in the upper portion and at the water surface to --1.9 and --1.8 at the
toward the inside of the bend in the lower por- bottom, respectively.
tion. These velocity deviations will produce a Rozovskii's radial velocity distribution may
transversedispersion,just as deviationsfrom be used to compute a transversedispersionco-
the mean velocity in the longitudinaldirection efficient,just as Elder usedthe logarithmic dis-
producea longitudinaldispersion. tribution to compute a longitudinal dispersion
The patternof secondary velocities
inducedin coefficient.Following Elder, we obtain
the bend of a natural stream may be too com-
plex to permit detailedanalysis;nevertheless,
Rozovskii [1957] has given formulasfor simpli- I 2d
4fo

fied conditionsthat have agreed well with ex-
periment. Rozovskii assumesthat within the "1dy F(V)
dv (6)
central portion of a wide stream rounding a
long bend an equilibriumcan be established whereF(?) is the quantity in bracketsin equa-
tween the driving centrifugal forces, the pres- tion 5. Using the vertical mixing coefficient
sure forces, and the turbulent shear stresses. given by the logarithmicvelocity distribution
This leads analytically to an expressionfor the
radial velocity componentof flow over a rough e• -- k(1 -- •7)•d U* (7)
surface as gives the result

= 1 d[F•(•)_
14• 1 (5) e•= -kSR•_U
,I (8)
where I/kdU • is the result of the triple integra-
tion shownin equation6. Values of I for v•rious
choices of Von Karman constant and friction
factor are shownin Figure 2.
Agreementwith field experiments. Table 1
showsthe conditionsobservedby Yotsukura et
al. in the Missouri River and the transverse
• 0.08 mixing coefficientpredictedby equation8. The
table shows two sets of hydraulic conditions'
•0.• firstly, the cross-sectional
mean valuesof depth
m 0.04 and velocity and the average radius of curva-
ture, from which a low value of the mixing co-
efficientis obtained,and secondly,the observed
• 0.02 combinationof depth, velocity, and radius of
curvature from which the maximum value of
the mixing coefficientcould be predicted. The
o.o I 1 t maximum predictablecoefficientis greater than
-0.2 -0.25 -0.5 -.0.•5 what was actually observed,suggestingat least
the possibilitythat secondarycurrents are re-
Fig. 2. Result of triple integrationof Rozovskii'a sponsiblefor the increasein the mixing coeffi-
radial velocity profile for use in Equation 8. cient,
Dispersion in Streams 499
TABLE 1. Hydraulic Conditions and Predicted Water enteredthe flume through a woodenstill-
Transverse Mixing Coefficients in the Missouri ing box attached to the upstream end of the
River, November 17, 1967
curve, built slightly wider than the flume itself,
Observed and approximately i meter deep. The exit was
Values controlledby a free overfall over the top of a
Cross- That Yield
perforatedplate, so that someof the flow went
sectional Maximum
Mean Transverse
throughthe plate and someover the top. The
Value Mixing experimentswere conductedstarting approxi-
mately 90ø aroundthe curve from the inlet box,
Depth-meanvelocity,u, with the furthest downstreammeasuringstation
(meters/see) 1.75 2.5 approximately 180ø around the curve. The
Depth, d (meters) 2.7 4.9
flume and dye injection apparatusare shownin
Radius of curvature, R
(meters) 3400 1500
Figure 3.
Shear velocity The spreadof the dye plume was measured
(meters/see) 0.074 0.074 by takingsamplesfrom the flow at a numberof
Von Karman constant crosssectionsdownstreamfrom the injector. At
(estimated) 0.4 0.35
each crosssectionsampleswere taken at inter-
Transverse mixing co-
efficient predicted by vals of approximately5 cm acrossthe flume in
equation 8 (cm2/sec) 22.2 2740 two traverses,onenear the surfaceand onenear
the bottom. The average concentration was
Observed transverse
given by the average of surface and bottom
mixing coefficient
(cm2/sec) 1200 samples.Both surface and bottom traverses
gaveapproximatelyGaussiantransverseconcen-
tration distributions, the surface distribution
Laboratory experiments. Because of the being displacedsomewhattowards the outside
number of assumptions
involvedin the deriva- of the curve and the bottom distribution to-
tion of equation 8 and the inconclusivenature wards the inside. Variances of the concentration
of the field experiments,it seemeddesirableto distributions were calculated from the averaged
test the analysisunder laboratory conditions.A curves,and the transversemixing coefficientwas
circular flume was available in the Berkeley obtained from the rate of increase of the vari-
laboratory, which, although not ideally suited ance, according to the method of Fischer
for the presentstudy,allowedexperiments to be [1967b].
conductedover a limited range of flow condi- The depth of flow and slopewere obtainedby
tions. measuring surface and bottom elevationsat a
The experimentsconsistedof injectinga con- number of points,usinga point gage.Discharge
stant stream of Rhodamine WT dye into the was measuredby diverting the flow into a vol-
flow at mid-depth and mid-width. The injector umetric tank and measuring the rate of rise.
was a 5/16-inch glasstube, bent 90ø and extend- Velocity was not measured,but was calculated
ing 11 inches downstreamfrom the bend, the by dividing dischargeby cross-sectional area.
end being necked down to form a fine nozzle. Table 2 showsthe hydraulic parametersand
During the runs at higher flow rates the maxi- experimentalresultsof each of the five experi-
mum injection rate was approximately0.1 ments. A range of depth, velocity, and slope
ml/sec,the exit velocitybeingmatchedvisually was obtained for which the mixing coefficient
with the local velocity of flow. The flume con- predictedby the analysiswould vary by a fac-
sistedof a curvethroughapproximately330ø, tor of nearly 30; within this range the observed
with smooth vertical steel sides and a wooden coefficientalways matched the prediction within
bottom roughened by the addition of sand a factor of 2, and at lower flow rates within
grains glued to the wood. The flume was 76.3 25%. The error can be explainedby the short-
cm wide, with a center-line radius of curvature ness of the flume; particularly at the higher
of 206 cm and a fixed bottomslopeof approxi- flow rates, sufiqcientlength may not have been
mately 0.002 (this varied somewhatalong the available to establishthe equilibrium assumed
curve,due to warpingof the woodenbottom). in the analysis. Table 2 also shows the ratio
500 •UGO B. F•SCX4r•n

given by the result for a straight, uniform flow


.:!:.:'(equation
3), thensecondary
flowisunlikely
to
i•" be important,andthe uniformflowresultmay
::. be usedwith confidence.
On the other hand, if
:•.: for a reasonable
choiceof parametersequation
:
.:
8 yieldsa higher•oefficient
than equation3,
: secondaryflow must be considered, and at the
presenttime (196.8) the only reliablemeansof
findingthe true value of the mixingcoefficient
appearsto be by experiment.
EFFECT O,F BENDS ON LONGITUDINAL DISPERSION

In the case of a uniform stream or a stream


bending uniformly in one direction with a uni-
form cross-sectionaldistributionof velocity, the
writer [Fischer, 1967a] has explainedhow Tay-
lor's conceptmay be used to calculate a cross-
sectional concentrationdistribution, and hence
to predict a dispersioncoefficient.In brief, the
concentrationprofile is obtained from the re-
lation

Fig. 3a. The circular flume and dye injection


apparatus.
06foZ
1
of the observedmixing coefficientto dU*; the
'fo'
Ifo'•U'(Y,Z)
dY
1az - (o)

value of this ratio rangesup to 10' times what


onewould expectin a straight channel.
Summary. Equation 8 gives an analytical
result for the value of the transversemixing'co-
efficientin a curvingflow. However,the analysis
assumesan equilibrium betweentransversecen- . i.:'.."•ii
." ' • ......... ::::".'
:':.:
': :."•:.•i?'' • .• ........
............

trifugal forces and vertical mixing, which is


only likely to exist some distance downstream
•:...:....
•"•:•:•,:::::•::•:
from the beginnin• of a bend, in the center sec- .:•,•-::':.".::::•.::
.&•.•:•.:•.:•::::;
:..•?:•:•:•::•::•:•:•::.•::::•:•::•?•::::::::::::::::::::::::::::::::::
....................
'.•- ..:'"'
'.:•5.......s....
..... ......:• •:•........-..**•,.
• ...• . ,,.... j• ß- ::•.,.

tion of a flow of large width-to-depthratio. The


result has been verified by laboratory experi-
ments for conditions that approximated the
requirementsof the theory.
For flow in natural streams, equation 8 is
probably not a practical method of estimating
the rate of transversemixing, partly becauseof
the assumptionsin its derivation, and partly
becauseof the sensitivity of the result to small
changes in the input parameters, which are
often not accurately known. Nevertheless,the
result is of interest in showingwhether secon-
dary currents of the first kind are likely to be
important in the transversemixing process.If
for a given stream the maximum coefficientthat
can be obtainedby equation8 is lessthan that Fig. 3b. Detail of the dye injection.
Dispersion
in Strea•s 501

• o 00 o
,4 _,4 ,.:; o•

o • • o o
502 HUGO B. FISCHER
z

riod of time. During this periodthe flow con-


tinues downstream and may passout of the
bendto the left and entera bendto the right.
The concentrationprofile that will be estab-
lishedby an indefinitelylongbendto the right
,," \\ is the mirror imageof that of a bendto the left
(dashedline, Figure 4). If the flow consists of
an infinitesequenceof bendsalternatelyto left
and to right, the concentration
profilewill at-
N_
,.,.,
•.•-'•,..,.,
i, _20 tempt to assumealternatelythe pattern shown
by the solid and by the dashedline in the
0 -:50
z
i i I
figure. If the flow time within the bend is in-
0 50 I00 150 sufficient for cross-sectionaldiffusion to estab-
DISTANCE FROM LEFT BANK, IN METERS
lisheitherpattern,the profilewill be a compro-
Fig. 4. Profiles of concentration deviation in mise, skewedfirst to one side and then to the
the Missouri River: infinitely long bend other,but neverreachingthe extremein either
to the left; ....... infinitely long bend to the
right; ....... end of set of bends,first to left and direction.The longitudinal
dispersion
coefficient
then to right. canstill be calculated
by equation10,but since
the extremes of concentrationdeviation have
where u' is the deviationof the longitudinal beenreduced,
the longitudinal
dispersion
coeffi-
velocity and c' the deviation of concentration cient will be lessthan for a continuousbend in
one direction.
from their cross-sectional mean vaues,and as
beforex, y, and z are the longitudinal,vertical, Whethera bend is sufficiently long for the
and transversecoordinates. c'(O) is determined steady-state concentrationprofile(equation9)
by the requirement that the cross-sectional to be establisheddependson the ratio of the
meanof c' be zero.The longitudinaldispersion cross-sectional diffusion time to the time re-
coefficient
is obtainedby quired for flow to round the bend. This ratio is
expressedby the formula

= I f.4
u•c•
dA(10)where•, is the 3'desired
= (12/e•)(a/L) (11)
ratio, 1 is a characteristic
where A is the total cross-sectionalarea.
The effectof meanderswill be illustratedby transverselength (previouslydefinedby the
two streams,the Missouri River betweenBlair writer, 1967a, as the distancefrom the thread
and Omaha,Nebraska,as reportedby ¾otsu- of maximum velocityto themostdistantbank),
kura et al., and the Green-Duwamishnear and L is the length of the bend. Table 3 shows
Renton Junction,Washington,as reportedby the value of the parametersand the ratio for
Fischer[1968a].Figure4 shows(solidline) the
concentrationprofile calculatedby equation9 TABLE 3. Hydraulic Parameters for the Missouri and
Green-Duwamish Rivers
using the velocity distributionmeasuredin the
MissouriRiver 1000 feet upstreamfrom the Missouri River Green-Duwamish
Blair Bridge.Sincethe fiver is makinga curve near Blair, River near
Nebraska, RentonJunction,
to the left (lookingdownstream), most of the November17, Washington,
discharge is concentratednearthe right bank; 1967 August 17, 1965
this gives a calculatedconcentration
profile Characteristic transverse
with positive relative concentrationson the length, • (meters) 140 14
right and negative on the left. If the curve to Transverse mixing co-
efficient • (cm2/sec) 1200 120
the left wereto continueindefinitely,
the con- Mean velocity of flow,
centrationprofile shownin the figurewould be
--

u (meters/sec) 1.75 .28


established,
and the dispersioncoefficientcould Average length of curve
(meters) 4400 380
be obtainedby equation10. However,the con- Ratio of cross-sectional
centrationprofileshownmustbe establishedby diffusion time to flow
time around curve, •, 65 12
cross-sectional
diffusion,which requiresape-
Dispersionin Streams 503
the two illustrative streams;the high value of If • is a longitudinaldistancecoordinatein a
7 for the MissouriRiver suggests that the con- frame of referencemoving downstreamat the
centrationprofile will not have time to reach mean flow velocity, and if, as in the writer's
the extremesshownin Figure 4; thus the true previouswork, one neglectsvariations of con-
value of the longitudinaldispersioncoefficient centrationand velocity in the vertical direction,
will be lower than would be given by the previ- the convectivediffusionequation in the moving
ous theory. In the Green-Duwamishstream, on frame of reference has the form
the other hand, the low value of 7 suggests that
the previoustheory may have given an accurate Oc/Ot-- --u'(Oc/Otl)+ (O/Oz)(e.Oc/Oz) (12)
result. This equation may be written in finite-differ-
To determine the effect of bends more ex- enceform to apply to eachstreamtube as
actly,a computeranalysiswasmadeasfollows:.
a streamwashypothesized to consistof an in- /•(c,
A,) ' A Oc
finite sequence
of identicalcurvesin alternate At = --ui • 0'•
directions.Each curve, of length L, was as-
sumedto consistof two parts; for the first %
L, the cross-sectionalgeometryand velocity +[e.
dO•zzl.
--[e.d-•z]_
(13)
distributionare exactly that measuredwithin where A indicates the incremental change of a
the curve of the real stream (i.e., the Missouri
quantity, the subscripti means the subscripted
or Green-Duwamish).In the remainingx/3 L, quantity for the ith stream tube, and the sub-
the geometryand velocitydistributionchange scripts + and -- on the brackets mean the
linearlyfromthe measured valuesto theirmir- quantity within the brackets evaluated first at
ror image,so that at the end of the curveall the uppermostvalue of z for the stream tube,
quantitiesoriginallymeasuredfrom the left and then at the lowermost value of z. (For the
bank are now measured from the right bank. first stream tube the -- bracket would be taken
The schemethen proceedswith a new curve equ•l to zero, as would the q- bracket for the
section;for the first 2/00
L the flow modelsa final stream tube, to satisfy the conditionof no
curve with the samepropertiesas the previous transverse transport through the boundaries).
one,but in the oppositedirection,and in the At this point in the analysis,Taylor's solution
final x/3L the geometryis againreversed.This for uniform flow is to divide concentration into
proceduremay be repeatedas often as neces- a cross-sectional mean value • and a deviation
sary until quasi-steady-state conditionsare from the mean c', to assumethat Oc'/O• and
reached,i.e., until the concentrationpattern at Oc/O•are zero, that OWO•is a constant,and to
the end of one set of curves is the same as at integrate equation 13 to obtain a concentration
the end of the previousset of curves. profile (i.e., equation9). For curvingflow one
The two measured cross sections used to may still assumethat 0•/0• is a constantand
verify the program(velocitydistributionsmea- seek a solutionfor which the profile of concen-
sured 1000 feet upstream from the Blair Bridge, tration deviation c'(z) is periodic, i.e., varies
in the Missouri River and at the Renton Junc- within the bends but is repeated with each set
tion Bridge in the Green-Duwamish)were di- of bends.Defining
vided into 20 stream tubes of equal discharge
similar to thoseshownin Figure 1. Input to the (1
computerprogram is the dischargeper stream and letting ? be the variation of • from its
tube, the averagedepth of each tube, and the cross-sectionMmean, equation 13 may be
location of its end point relative to one bank. written as
From this information the computer generates
the curve pattern describedabove. The com-
puter then executesan iterative computation
basedon Taylor's originalhypothesisof a quasi-
of
steady-stateequilibrium existing in a slice of
fluid moving downstream at the mean flow --(ezd
•)_)]
fit (15)
velocity, describedas follows; Here •[, is •e change of normafized concert-
504 HUGO B. FISCHER

tration •, in the ith stream tube in time incre- TABLE 5. Predicted Dispersion Coefficient for
the Missouri River for Various Choices of Ratio of
ment At as one movesalongthe stream tube at
Curve Length to Cross-OverLength
the cross-sectional
meanvelocity•. Equation 15
can be iterated over as many time steps and
Run I Run 2 Run 3
curvesas necessary,using for each iteration the
value of A]J/A• obtainedin the previousitera- Number of stepstaken
tion, until a periodically repetitive solution for using observed cross
? is obtained. Then the dispersioncoefficient section 56 48 24
Number of stepstaken
for the stream is calculatedby for crossover to mir-
ror image of observed
cross section 14 24 48
D--• I •. i•u• (16)Predicted dispersion
coefficient,m•'/sec 850 860 926
The result of this equationis also periodic,but
its averagevalue taken over all positionson the Note' For all runs the cross-sectionalparam-
curve gives the averagerate of dispersion. eters were those given in Tables 3 and 4 for the
The results of applying the computer pro- Missouri River near Blair, Nebraska, November 17,
1967.
gram to the Missouri and the Green-Duwamish
are given in Table 4. The lengths of the dis-
tance and time steps are chosen to provide another 48 steps are taken during which the
numerical stability, accordingto the usual cri- stream parametersmatch the mirror image of
terion for stability of solutionof diffusionequa- the observedcrosssection; and finally another
tions. Initially the values of L' in all stream 24 steps return all parameters to their initial
tubes are set equal to zero, and the values of values. The set of 144 steps was repeated 9
A•, e, and d for each stream tube are taken times; on the 9th round values of f' for each
from the observedcrosssection.For instance, stream tube at each point in the curve sequence
in the Missouri the first 48 stepsare taken with were nearly identical with the corresponding
all stream tube parameters matching the ob- values in the 8th round, so that the sequence
served crosssection; in the next 24 steps the was stopped. The concentration profile gener-
cross-overto the mirror image occurs; then ated by the program at the end of the 9th
round is shownin Figure 4 (dotted line).
The decision to divide each curve into two
TABLE 4. Predicted and Observed Longitudinal Dispersion
Coefficients for the Missouri and Green-Duwamish Rivers parts, one of length• L and oneof length•3
L, having been arbitrary, two additional com-
Missouri River Green-Duwamish puter runs were made to show the effect of
near Blair, River near
Nebraska, Renton Junction,
changingthe division.In one run eachcurve of
November 17, Washington, length L was assumedto consistof a sectionof
1967 August 17, 1965 length 4/5 L having the cross-sectional geom-
etry and velocity distributionmeasuredin the
Parameters for use with
curve of the real stream, whereasin the remain-
computer program de-
scribed herein: ing 1/5 L t•he distribution changedlinearly to
Length of time step, its mirror image. In the secondrun the division
sec 35.2 4.6
was constant geometry for only • L, and the
Length of distance
step, meters 61 1.28 linear reversal was over • L. For each of these
Number of steps per runs all other parameters were exactly those
curve 72 150
Predicted dispersion co-
used for the study of the Missouri River de-
efficient by computer scribed in the previous paragraph. The results,
program, m•-/sec 860 7.6
given in Table 5, show that the predicteddis-
Observed dispersion co-
efficient, m•-/sec 1500 6.5-8.4 persion coefficientis insensitiveto the internal
Dispersion coefficient division of the curve, so long as the total length
predicted by previous of the curve plus reversalremainsconstant.
method, [Fischer,
1967a] (m•-/sec) 7OOO 7.8 It has been suggestedthat the alternating
characterof the bendsmight be important in
Dispersion in Streams 505
the Missouri, but not in the Green-Duwamish;
direction of curve. A computer program has
Table 4 shows this to be the case. In the Green-
been describedthat has predicted the disper-
Duwamish the new program and the previous sion coefficient in two streams, one approxi-
method give nearly identical results, which are mately 20 meters wide, and the other 180
verified by experiment. In the Missouri the meters wide. I am not aware of other experi-
previousmethod gives a higher result than ex- ments in which sufficient data were taken to
periment, but the new program is in reasonable apply the program; however, it is concluded
agreement.The error probably occursbecause that at the present time (1968) the program
the Missouri does not, of course,consistof a described herein represents the most accurate
uniform sequenceof exactly mirror image bends available method for predicting dispersion co-
of equal length; an exact analysiswould have eftqcients in natural streams.
to consider the detailed geometry of every
Acknowledgments. This analysis was moti-
bend, which is not feasible.It is therefore en- vated by the findings of an experimental study
couraging that the new program is verified conducted in the Missouri River in November,
within reasonablelimits by both setsof experi- 1967, in which I participated as a Research Hy-
ments and appears capable of predicting dis- draulic Engineer, U.S. Geological Survey. I wish
persion coefficients both in uniform and in to thank Nobuhiro Yotsukura,who organizedand
directed the study, Paul C. Benedict, whose sup-
meanderingstreams. port and enthusiasmmade my participation pos-
sible, and all those who participated in the study.
CONCLUSIONS I am particularly indebted to Harry M. Nichan-
dros, who carried out the laboratory experiments
Stream meanders influence longitudinal dis- as an undergraduate researchproject in the Hy-
persion in two ways: first, by concentratingthe draulic Laboratory of the University of California,
zone of high velocity towards the outsideof the Berkeley. The description of the laboratory ex-
curve they greatly increasethe rate of disper- periments is a condensationof his report. I am
also indebted to H. A. Einstein, whose helpful
sion, and second,by inducing secondaryspiral suggestionsincluded the use of the circular flume,
currents they increase the transverse mixing, and to E. A. Prych, who read and criticized the
which tends to reducethe dispersion.An anal- manuscript. The laboratory and computer work
ysis of the effect of secondarycurrentson trans- required by this study was supported by the
Water Resources Center of the University of
verse mixing has been attempted herein, with California under project WRC-165.
limited success.Equation 8, which predicts a
transversemixing coeftqcientfor a curving flow, REFERENCES

has been found reliable under laboratory condi- Aris, R., On the dispersion of a solute in a fluid
tions. For a natural stream, its result may be flowing through a tube, Proc. Roy. $oc. (Lon-
compared with the result of equation 3, for don) A, 235, 67-77, 1956.
straight streams,to tell whether or not secon- Elder, J. W., The dispersionof marked fluid in
turbulent shear flow, J. Fluid Mech., 5, 544-560,
dary currents are likely to increasethe rate of 1959.
transversemixing; if they are, the value of the Fischer, H. B., The mechanics of dispersion in
transversemixing coefficientshouldbe found by natural streams, J. Hydraul., Div., Am. $oc.
experiment. Civil Eng., 93, (HY6), 187-216, 1967a.
It has been shown that a succession of curves Fischer, H. B., Transverse mixing in a sand-bed
channel, U. $. Geol. $urv. Pro•ess.Paper 575-D,
in alternating directions,as is usual in meander- D267-D272, 1967b.
ing streamsmay producea lower rate of longi- Fischer, H. B., Methods for predicting dispersion
tudinal dispersionthan the same succession of coefficients in natural streams with applications
curves in the same direction. Whether the dif- to lower reaches of the Green and Duwamish
Rivers, Washington, U. $. Geol. $urv. Pro,es.
ference is important dependson the ratio of
Paper 582-A, 1968a.
cross-sectionalmixing time to time of flow Fischer, H. B., Dispersion predictions in natural
around an individual curve. In a narrow stream streams,J. Sanit. Eng. Div., Am. $oc. Civil
a reasonably accurate assumptionis that the Engrs., 94, 927-943, 1968b.
flow consistsof a uniform, infinitely long curve Orlob, G. T., Eddy diffusionin open channel flow,
Contribution q•19, Water Resources Center,
in one direction, as was implicit in previous
University of California, 1959.
theories; however,in wider streamsit is neces- Prandtl, L., Essential of Fluid Dynamics, Hafner,
sary to considerthe effect of the alternating New York, p. 149, 1952.
506 HUGO B. FISCHER

Rozovskii, I. L., Flow o• Water in Bends o• Open lent flow through a pipe, Proc. Roy. $oc. (Lon-
Channels,Academy of Sciencesof the Ukrainian don) (A), 223, 446-468, 1954.
SSR, 1957 (translation No. OTS60-51133,Office Thomas, I.E., Dispersion in open-channel flow,
of TechnicalServices,U.S. Dept. of Commerce, Ph.D. thesis, Northwestern University, Evans-
Washington, D.C.). ton, Illinois, August 1958.
Sayre, W. W., and F. M. Chang, A laboratory in- ¾otsukura, N., H. B. Fischer, and W. W. Sayre,
vestigation of open-channeldispersionprocesses Measurement of mixing characteristics of the
for dissolved,suspended,and floating dispers- Missouri River between Sioux City, Iowa, and
ants, U. $. Geol. $urv. Pro•ess. Paper 433-E, Plattsmouth, Nebraska, manuscript prepared
1968. for intended publication as a U.S. Geological
Taylor, G.I., Dispersion of soluble matter in Survey Circular, 1968.
solvent flowing slowly through a tube, Proc.
Roy. Soc. (London), A., 219, 186-203, 1953. (Manuscript received August 12, 1968;
Taylor, G.I., The dispersioaof matter in turbu- revised November 12, 1968.)

You might also like