Advanced Environmental Hydraulics - SNU

Ch 1.
Concepts and Definitions
Chapter 1 Concepts and Definitions
Contents
1.1 The Role of Hydrology and Hydraulic Engineering in Environmental Management
1.2 Environmental Hydraulics
1.3 Strategies and Approaches for Problems Solving
1.4 Basic Definitions and Concepts
1-1
Ch 1. Concepts and Definitions
1.1 The Role of Hydrology and Hydraulic Engineering in Environmental Management
• Water quality = water quantity
+ hydrodynamics of transport and mixing
+ chemistry and biology of natural water systems
atmosphere - air motion
hydrosphere - water motion
Fluid motions transport and disperse disposal of residuals
- massive discharges of wastewater into rivers and oceans
Without fluid motion, there is no transport and dispersion
[Ex] estuarine modification or barriers → radically change the circulation patterns decreasing
flushing of pollutants
1.1.1 Overall Framework for Environmental Management
•Environmental Fluid Mechanics
- study of fluid motions in the lower atmosphere, in the ground, and in rivers, lakes, and seas
that relate to problems connected to human activities within the environment.
•Environmental Hydraulics
- study of water motions in the ground, and in rivers, lakes, and seas that relate to problems
connected to human activities within the environment.
1-2
•Transport - Hydraulic engineer
Transformation - Chemist
Accumulation - Biologist
⇒ processes that takes places between the point where a pollutant is discharged into the
water environment and some other sites (downstream site in rivers) where the ambient water
quality is observed.
• Role of hydraulic engineer in environmental management
1) design of hydraulic structure (outfalls, diffusers)
-> jets and plumes
Wastewater Treatment
Plant
Water Surface
Wastefield
Sea Bed Port
Riser
Feed
e r Pi
pe
Ports Diffuser Pipe
Riser
"Detail of Diffuser"
Diffuser Pipe
1-3
2) analyze water quality processes -> diffusion and dispersion
Conceptual View of Pollutant Inputs

Rainfall
Agricultural Urban Area

Area
Agricultural Combined Sewer Separate

Runoff Runoff Sewer
Wastewater
Storage Treatment
Plant
Non-Point
Source
Overflow Effluent
Receiving Stream
Flow
Effluent
Wastewater
Treatment Industry Accident
Plant Spills
Point Conservative Miscible

Non-point Reactive Immiscible
Instantaneous Dissolved Organic
Continuous Suspended Inorganic
1-4
• Environmental Control System
- Optimization of
(a) control of pollutants at the source (pre-treatment, clean technology)
(b) wastewater treatment (primary, secondary treatment)
(c) disposal in the environment (post-treatment, wastewater outfalls)
• Role of Hydraulic Engineer (Environmental Hydraulics)
- make interface between man's activities and the natural environment
- draw water supply from natural water bodies
- develop technology how wastewater is returned → design of outfalls
1-5
Flow Diagram of Water Quality Engineering
Desirable Water Use: Water Quality

Water Supply Standard
Fisheries
Recreational
Ecological Balance
Pollutant Inputs: Aquatic

Point Ecosystem
Non-Point
Transport
Dispersion
Reaction
Environmental
Engineering
Control
Water Quality
Concentration
Desired vs. Actual
Actual > Desired
1-6
1.1.2 Using the Water Environment for Waste Assimilation
• Types of pollutant (from the least dangerous to the most hazardous)
(1) Natural inorganic salts and sediments
- not toxic unless in excessive doses
(2) Waste heat or heated water discharges
- cooling water for electric generating plants
- decrease water's assimilative capacity for oxygen
(3) Organic wastes
- domestic sewage → biochemical oxygen demand (BOD)
- carbon, nitrogen, phosphorous: nutrients → eutrophication
• BOD
- amount of dissolved oxygen for bacteria to oxidize the organic wastes in the water
bacteria
Organic waste  O2  CO2  H 2O
• COD
- amount of dissolved oxygen to oxidize the organic wastes using chemicals
COD > BOD

1-7
Yangz River
Nakdong River
1-8
Cooling system for nuclear power plant
LNG plant (Manzanillo, Mexico)
1-9
• Drinking water quality
class BOD
1 < 1ppm
2 < 3ppm
3 < 6ppm
4 < 8ppm
• ppm = parts per million = mg/ℓ
1ppm = 1g of BOD/1,000,000g of water
BOD-DO coupled system
1-10
(4) Trace metals
- industrial wastewater – electro-planting, battery manufacturing, mining, smelting, refining
- lead, mercury, cadmium, selenium
- toxic in high concentration (accumulation)
(5) Synthetic organic chemicals
- slow to degrade in environment
- bioaccumulate in the aquatic food chain
- industrial chemicals: phenol, benzenes, PCB(Poly-Chlorinated Biphenyls),
- agricultural chemicals: pesticides, herbicides, DDT
* biological process (multiplying the concentration by a factor of 105 in successive food
chain steps)
⇔ physical process of mixing (= high dilution reduces the concentration)
(6) Radioactive materials
- resulting from production of nuclear energy, nuclear weapons, and production of radioactive
materials for industrial use
- plutonium - 239/240, strontium-90, cesium-137
- long-term storage of radio-active wastes w/o leakage
[Cf] radioisotopes for tracer materials: I-131 (half-life - 8.3 days)
1-11
(7) Chemical and biological warfare agents
- exceedingly toxic; cannot be dispersed in the environment
[Re] "Dilution is the solution to pollution."
- suitable only for heat and natural organic materials
[Re] Toxic substances: trace metals, synthetic chemicals, radioactive materials
- results in acute effects of mortality and long-term chronic effect
- tendency to sorb to particulates in the water body
- tendency to be toxic at relatively low concentrations of g/l or ng/l
- tendency to be concentrated by aquatic organisms and transferred up the food chain
1-12
1.1.3 Mass Balance Concepts in Residual Management
• Conservation of mass
- Flux of substance source must balance the fluxes for subsequent transport and diffusion
with adjustments for chemical and biological conversions and sinks, such as deposition on the
river bed or sea floor.
• flux of solute mass = mass of a solute crossing a unit area per unit time in a given direction
• For steady state, total influx = total efflux
• Types of pollutant source
1) point source
- discharge from a structure which is specially designed for the outflow of wastewater or
accidental spill
[Ex] - industrial and municipal sewerage system
- accidental spill of chemicals or oil from a ship
- release of heated water from power plant
2) nonpoint source
- widely distributed points where pollutants are introduced into the hydrologic cycle
[Ex] runoff of salts, soil erosion, acid rainfall, street drainage
1-13
[Cf] classification by input period
continuous input - municipal sewerage system, heated water from power plant
instantaneous input - accidental spills
Effluent discharge channel (Tancheon STP)
1-14
1.1.4 Impacts of Some Traditional Activities of Hydraulic Engineers
• Adverse effects of traditional hydraulic works on water quality
(1) man-made reservoirs → summertime thermal stratification → oxygen depletion in the
lower layer
(2) diversion water for consumptive uses → reduce river flow (inflow) and its ability to
provide natural flushing
(3) canals → transport huge amount of dissolved salts, sediment, nutrients and parasites
(4) agricultural drainage system → accelerate the leaching of nutrients and salts from land to
natural hydrologic systems
(5) breakwater for harbors → interfere with natural nearshore circulation which could
otherwise carry away pollutants
(6) estuarine modification or barriers → radically change the circulation patterns decreasing
flushing of pollutants
Panama Canal
1-15
Breakwater for harbors
SaemanGeum
1-16
SaemanGeum Barrier
1-17
1.2 Environmental Hydraulics
1.2.1 Hydrologic transport processes
• Hydrologic transport processes
- physical processes of flow of natural water bodies which cause pollutants or natural
substances to be transported and mixed, or exchanged with other media
man-made unit process (chemical plant)
(1) Advection: transport by an imposed current system of receiving (ambient) water bodies
(2) Convection: vertical transport induced by hydrostatic instability (buoyancy)
[Ex] - flow over a heated plate
- flow below a chilled water surface in a lake (winter time)
(3) Molecular diffusion
- scattering of particles by random molecular motion
- Brownian motion
- described by Fick's law
- molecular diffusivity
[Re] diffusion (physics): the process in which there is movement of a substance from an
area of high concentration of that substance to an area of lower concentration

1-18
(4) Turbulent diffusion
- random scattering of particles by turbulent motion
- analogous to molecular diffusion
- molecular diffusion ≪turbulent diffusion
[Ex] mixing in coffee cup: in rest vs. stirring
1-19
(5) Shear (shear flow)
- advection of fluid at different velocities (direction and magnitude) at different positions
[Ex] - velocity distribution over stream bed
- complex flow in estuary or coastal areas
c’(z)
x
v(z) u(z)
(6) Dispersion
- scattering of particles or cloud of contaminants by the combined effects of shear and
diffusion
- shear advection + vertical and/or transverse diffusion
- molecular diffusion ≪ turbulent diffusion < dispersion
[Re] In optics, dispersion is the phenomenon in which the phase velocity of a wave depends
on its frequency, or alternatively when the group velocity depends on the frequency.
1-20
1-21
1-22
(7) Mixing
- diffusion or dispersion
- turbulent diffusion in buoyant jets and plumes
- any process which causes one parcel of water to be mixed with or diluted by another
(8) Evaporation
- transport of water vapor from a water or soil surface to the atmosphere
(9) Radiation
- flux of radiant energy at a water surface
(10) Particle settling
- sinking (or rising) of particles having densities different from the ambient fluid
[Ex] sand grains, dead plankton → downward transport of nutrients in lakes and ocean
(11) Particle entrainment
- picking up of particles (sand, organic detritus) from the bed by turbulent flow past the bed
• flow analysis - mean velocity is important
• pollutant analysis - fluctuation and irregularities in hydrologic systems are equally
important
1-23
DNS
RANS
LES
1-24
1.2.2 Buoyant Jets and Plumes
• submerged (momentum) jet
- increase the dilution of effluent discharge with the surrounding waters
• submerged buoyant jet
- when discharge fluid is lighter or heavier than surrounding waters
- heated water discharge (momentum jet) vs. wastewater discharge (buoyant jet)
• plume
- initial momentum is not important
1-25
●Analysis of buoyant jets and plumes
a) jet parameters : initial momentum flux, mass flux, buoyancy flux
b) ambient conditions : ambient density stratification, ambient velocity
c) geometric factors : jet shape, angle, orientation
1-26
1.2.3 Density-Stratified Flows in a Natural Environment and Geophysical Fluid
Mechanics
• Density stratification
→ lake, reservoirs - due to temperature variation
→ estuary - salinity profiles
- internal structure causes effect on both mean flow fields and turbulent mixing and
dispersion
1.2.4 Sedimentation and Erosion
• particle setting and entrainment, stream morphology
• erosion, transport and deposition → 『Advance River Engineering』
1.2.5 Interdisciplinary Modeling
• fate of pollutants - governed by combination of physical, chemical and biological processes
• different time scales between the dominance of the various processes
- hydrodynamic mixing in the jets - min
- dispersion, biochemical reaction - hours, days
- biological or ecological effects - weeks, months
1-27
1.3 Strategies and Approaches for Problems Solving
1.3.1 Strategies
• Strategy:
a. Identify problems
b. For large complicated problems, break a problem (of mixing) into submodels
(component parts)
c. Use two or more approaches (mixed approaches/hybrid modeling)
- interweaving of all of the approaches
- better than single approach, computer model, hydraulic model, and field studies
[Re] Approaches:
- computer modeling
- hydraulic modeling
- field experimentation
(1) Problem identification
a) For acute toxic effects
→ predict maximum instantaneous point concentration of a pollutant

1-28
b) For long-term ecological effects
→ predict changes in monthly averages over broad areas
At different scales of length and time, different processes will be important.
→ Table 1.1 Effluent flow from a sewer outfall (ocean)
length scale time scale

phase phenomenon
(m) (sec)
1 initial jet mixing < 102 < 103
2 establishment of sewage field 10 ∼ 103 102 ∼ 103
3 natural lateral diffusion / dispersion 102 ∼ 104 103 ∼ 105
4 advection by currents 103 ∼ 105 103 ∼ 106
5 large scale flushing (by tidal motion) 104 ∼ 106 106 ∼ 108
(2) Definition of submodels
a) Omnibus model
- cover all steps
b) Component models
- break problems into components for different length and time scales
- simplifying (idealized) representations can be made
- concentrate on the dominant processes and important features of the environment

1-29
[Ex] Thermal discharge problem
∙ near-field mixing:
- initial jet and plume mixing occurs; momentum and buoyancy of the jet are important
(active mixing) → use hydraulic modeling
∙ far-field mixing:
- heat loss, natural lateral dispersion and advection by currents are dominant (passive mixing)
→ use computer model
- results from near-field model are used as input of far-field model
[Re] Model
- idealized representations
- concentrate on the dominant processes and important features of the water domain
- omit many secondary details or interactions
1-30
• Computer model:
- numerical solution of mathematical equations
→ cannot be better that the validity of the underlying approximations
- can include meteorological factors (wind, surface cooling)
- can avoid scaling errors
1-31
• Physical model:
- reproduce complex 3-D flows (density-stratified flows)
- large scale phenomena (large scale vortices, internal waves and hydraulic jumps, multilayer
shear flows, gravitational spreading) are represented
- scaling errors → viscous effects are too strong in reduced laboratory models
[Re] scaling based on Froude laws
- Reynolds numbers are much reduced from the prototype
- alter turbulence and resistance characteristics
- distorted models are used for big estuary (river) model
• Field Studies
-can be used to verify or adjust numerical models
a. Eulerian-type measurements
- fixed-location recording meter → time series data at a fixed point
b. Lagrangian experiments
- follow drogues to track flow trajectories and dispersion → time series data for a given
parcel of water
1-32
1-33
1.3.2 Approaches
• Order-of-magnitude analysis
- quick approximate solution (quick-and-dirty)
- scaling
- powers of ten
- show the correct dependence on the most important parameters
- based on dimensional analysis
[Ex] complete vertical mixing
Find longitudinal distance required for complete vertical mixing of surface discharge
pollutants
d2
mixing time T   (time scale)

d = depth (cm)
 = vertical eddy diffusivity (m2 / s)
set   0.07u*d
f
u * = shear velocity = u
8
f = Darcy – Weisbach friction factor
d2 8
then T  
0.07du f
1-34
8
 15 ;   0.35
f
d
T  75
u
substitute X  uT
x
  75  102
d
x = longitudinal distance required for complete vertical mixing
1-35
1.4 Basic Definitions and Concepts
1.4.1 Concentration
• Concentration
- units of mass of tracer or contaminant per unit volume
M
C  lim
V 0 V
where M = tracer mass in elemental volume V
• Time average of C
C  C ( x, y, z, t )
→ include turbulent fluctuations
1 t 0 T
T t0
C ( x, y, z, t0 )  C ( x, y, z, t )dt
where T = averaging time interval
- sec, min for turbulence fluctuation
C  fn( x, y, z, t0 ,T )
→ slowly varying function; reflects only change of flow rate and ambient water conditions
- hours, day for unsteady flow
1-36
Instantaneous input
Thalweg line
L  V t
Section 1 Section 2 Section 3
advection diffusion/
dispersion
C
1-37
• Spatial average of C
1
V 
Cv ( x0 , y0 , z0 , t )  V C ( x, y , z , t )dV
1
3
- wipes out turbulent fluctuations occurring on scales smaller than V
• Flux average of C  C f
◦ flux = mass per unit area per unit time
Flux of contaminant mass through AA
= C f • (flux of water through AA)
 A C u dA  C f  A u dA  C f Q
1
Q
 C f (t )  A C u dA
• Total mass M
T T
M   C f (t ) Qdt    A C u dAdt
0 0
1-38
1.4.2 Dilution
• Dilution: rate at which tracer is diluted, S
total volume of sample( vol. of mixture)

S
volume of effluent contained in the sample
S=1 → undiluted effluent
p = volume fraction of effluent in a sample
= 1/S = relative concentration
• mixture of effluent with ambient water of background concentration Ca
items effluent ambient water mixture
vol. vole vola vole  vola
vole Ce  vola Ca
conc. of
Ce Ca vole  vola
contaminant
(harmonic mean)
vole Ce  vola Ca (vole  vola )Ca  vole (Ce  Ca )

C  
vole  vola vole  vola
vole
 Ca  (Ce  Ca )
vole  vola
 Ca  P(Ce  Ca )
1
 Ca  (Ce  Ca )
S
→ increment of concentration above background is reduced by the dilution factor S or p from
the point of discharge to the point of measurement of effluent
1-39
C  Ca
 p 
Ce  Ca
Ce  Ca
S 
C  Ca
Mass balance at effluent outfall
● After effluent is fully mixed across the section,
mass rate of substance upstream + mass rate added by outfall
= mass rate of substance downstream from outfall
Qu Cu  QeCe  QC
w ⇒ Qe Ce = impact waste load [M/T]
• dilution of a composite sample
total vol. 1  vol i

S    N
i 1
1
 vol
total effluent vol. P
i
i 1 Si
1-40
1.4.4 Density
• weight density of water plays an important role in mixing in water body
weight density = weight per unit volume =  g = weight/vol.
where  = mass density = M/vol.
g = gravitational acceleration
• Variation of  is less than 3% in estuary and ocean
⇒ unimportant for fluid acceleration (fluid dynamics)
⇒ however, weight (buoyancy) difference ( = g  ) is important for buoyancy of
discharges of stability of density-stratified flows (mixing mechanics)
• buoyancy per unit mass

g  g '  modified gravitational acceleration

•  t    units for water density
t
  1 ( g / cm3 )  1000   t (kg / m3 )
1000
 t    units
 f n (temperature, salinity)  See App.1

1-41
1.4.5 Density Stratification
∙ density profile a ( z )
z = vertical coordinates
d a
 0
dz
∙ linear density stratification
d a
g  const.
dz
1.4.6 Dynamically Active versus Passive Substances
• Dynamically active substance:
- cause significant density changes to affect the flow dynamics
- massive heated water discharge
- need to recalculate flow fields at each time step → coupled model
• Dynamically passive substance:
- does not cause density changes
- wastewater discharge
- flow fields are separately calculated and used as given input to mixing analysis → scalar
transport model
[Cf] conservative substance: conserve mass
non-conservative substance: reactive, BOD
1-42
1.4.7 Velocity Distribution in Turbulent Shear Flow
• turbulent shear flow in a long pipe or channel
- driving force: pressure gradient and gravity
- resisting force: shear stresses at the wall
• velocity contour map (Fig.5.11)
- u  u ( y, z )
• Vertical velocity distribution
- approximated by a logarithmic function
3 u* 2.30 * z
a) Pipe : u  u   u log10 (1.27)
2  R
u* 2.30 z
b) Wide channel : u  u   u* log10 (1.28)
  d
where z = distance from the wall
R = pipe radius
d = channel depth
= Von Karman constant ≈ 0.4
• mean velocity in the cross section related to mean wall shear stress
1
o  f u 2 → [Re1]
8
1-43
where  o = mean wall shear stress
f = Darcy-Weisbach friction factor
f - estimated from Moody diagram for circular pipes
- use Moody diagram Rh  D / 4 with for open channels
1
A
u  A u ( y , z ) dA
o f
 u
 8
define
o
 u*  shear (friction) velocity → [Re2]

• mean shear stress ← balance of force
 o   gRh S
 u*  gRh S → [Re3]
u 8

u* f
1-44
[Re1]
L v2  L
hL  f , hL  0
D 2g  Rh
L L v2
o  f
D D 2g

4
1
o  f  v2
8
[Re2]
o
Shear velocity  u   gRS for steady unsteady flow
*
D v v 1 v
 gD ( So    ) for
x g x g x
unsteady flow
- dimensions of velocity
- varies with the boundary friction o
[Re3]
hL
 o   Rh
l
S0 = channel slope for uniform flow
1-45
Homework #1
Due: 1 Week from Today
Upstream flow with a background level of chlorides, a conservative substance, of
30mg/ℓ is supplemented by an industrial discharge of 0.3m³/s carrying 1,500mg/ℓ chlorides
and a downstream tributary of 0.15m³/s with background chlorides concentration of 30mg/ℓ.
Assume downstream tributary chlorides concentration does not vary with flow.
To maintain a desired chlorides concentration of 250mg/ℓ at the water intake,
determine: (a) the required industrial reduction in chloride concentration (b) the required
increase in tributary flow.
1-46
Ch 2. Fickian Diffusion
Chapter 2 Fickian Diffusion
Contents
2.1 Fick’s Law of Diffusion
2.2 The Random Walk and Molecular Diffusion
2.3 Some Mathematics of the Diffusion Equation
2.4 Solutions of the Diffusion Equation
2-1
▪ diffusion
fluid at rest → diffusion
moving fluid → diffusion + advection
∙ molecular diffusion - only important in microscopic scale; not much important in
environmental problems
∙ turbulent diffusion - large scale; analogous to molecular diffusion
2.1 Fick's Law of Diffusion time rate of heat per unit area in a given
direction is proportional to the temperature
2.1.1 Diffusion Equation
gradient in direction
Fick (1855) adopt Fourier's law of heat flow (1822) to diffusion
Fourier Fick
transport heat mass
gradient temp. conc.
1) Fick's law
→ flux of solute mass, that is, the mass of a solute crossing a unit area per unit time in a
given direction, is proportional to the gradient of solute concentration in that direction.
∂C
q∝
∂x
∂C
q = −D (2.1)
∂x
2-2
q = solute mass flux
C = mass concentration of dispersing solute
D = coefficient of proportionality
→ diffusion coefficient (m²/s), molecular diffusivity
- Minus sign indicates transport is from high to low concentrations
[Re] Fick’s law in 3D


q =− D∇C (2.1a)
   
q =iqx + jq y + kqz → vector
∂  ∂  ∂
=
∇ i + j +k
∂x ∂y ∂z
 ∂  ∂  ∂ 
∇C
= i + j + k C
 ∂x ∂y ∂z 
 ∂C  ∂C  ∂C
=i + j +k → vector
∂x ∂y ∂z
∙ gradient of scalar, C → vector, q
2-3
2) Conservation of Mass
Consider mass conservation for 1-D transport process
∆M=C ∆V
∂C
i) time rate of change of mass in the volume= (∆x ⋅1)
∂t
ii) net change of mass in the volume = {( flux)in − ( flux)out }× unit area
 ∂q 
=q −  q + ∆x 
 ∂x 
∂q
=
− ∆x
∂x
Now, combine (i) and (ii)
∂C ∂q
∴ ∆x =− ∆x
∂t ∂x
∂C ∂q
= − (2.2)
∂t ∂x
2-4
3) Diffusion Equation
Combine Eq. (2.1) and (2.2)
∂C ∂  ∂C 
=
−  −D  q
∂t ∂x  ∂x 
∂C ∂ 2C
=D 2 ⇒ Diffusion Equation (Heat Equation)
∂t ∂x
▪Diffusion Equation = Fick's law of diffusion + Conservation of mass
Differentiate Eq. (2.2) w.r.t. x
∂  ∂C  ∂2q
 = − 2
∂x  ∂t  ∂x
∂  ∂C  ∂  q  1 ∂q
LHS =  = −  = −
∂t  ∂x  ∂t  D  D ∂t
1 ∂q ∂2q
∴ − =− 2
D ∂t ∂x
[Re] Conservation of mass in 3D
∂C   ∂q ∂q ∂q 
= −∇ ⋅ q = −  x + y + z 
∂t  ∂x ∂y ∂z 
∂C 
+ ∇⋅ q = 0 (i)
∂t

Then consider q by various transport mechanisms
2-5

- molecular diffusion (Fickian diffusion) → q =− D∇C
 
- advection by ambient current → q = Cu
 
∴ q = Cu − D∇C (ii)
Substitute (ii) into (i)
∂C 
∂t
(
+ ∇⋅ Cu − D∇C = 0 )
∂C 
∂t
( )
+ ∇ ⋅ Cu = D∇ 2C
(iii)
  
( )
∇ ⋅ Cu = ( )
( ∇C )⋅u + C ∇ ⋅u
Continuity
 ∂u ∂u y ∂u
∇⋅
=u x
+ + = z
0
∂x ∂y ∂z
 
∴ ∇ ⋅ (Cu ) = ∇C ⋅ u
 ∂C  ∂C  ∂C     
= 
 ∂x
i+
∂y
j+ (
k ⋅ u x i + u y j + u z k
∂z 
)
∂C ∂C ∂C
= ux + uy + uz
∂x ∂y ∂z
     
( i ⋅ i = j ⋅ j = k ⋅ k = i i cos 0° = 1
     
i ⋅ j = j⋅ k = k ⋅ i = 0)
2-6
Thus, (iii) becomes
∂C 
+ ∇C ⋅ u = D∇ 2C
∂t
write out fully in Cartisian coordinates
∂C ∂C ∂C ∂C  ∂ 2C ∂ 2C ∂ 2C 
+ ux + uy + uz = D  2 + 2 + 2 
∂t ∂x ∂y ∂z  ∂x ∂y ∂z 
→ 3D advection-diffusion equation
For molecular diffusion only
∂C
= D∇2C
∂t
∂C  ∂ 2C ∂ 2C ∂ 2C 
= D 2 + 2 + 2 
∂t  ∂x ∂y ∂z 
→ linear, 2nd order PDE
[Re] Vector notation of conservation of mass
Consider a fixed volume V with surface area S


total mass in the volume = ∫ V
C ( x, t )dV
 
mass flux = q ( x, t )
2-7
Consider conservation of mass
∂    
∂t ∫V ∫S ( x, t ) ⋅ndS =
C ( x , t ) dV + q 0 (a)

n = unit vector normal to surface element dS
Green's theorem
  
∫S
q⋅ndS= ∫ V
∇ ⋅ qdV
(b)
Substitute (b) into (a)
 ∂C 
∫V  ∂t + ∇⋅ q  dV = 0

∂C 
+ ∇⋅q = 0
∂t
2-8
2.1.2 Diffusion Process
1. diffusion process
= process by which matter is transported from one part of a system to another as a result
of random molecular motions
(i) Watch individual molecules of ink
→ Motion of each molecule is a random one.
→ Each molecule of ink behaves independently of the others.
→ Each molecule of ink is constantly undergoing collision with other.
→ As a result of collisions, if moves sometimes towards a region of higher, sometimes of
lower concentrations, having no preferred direction of motion.
→ The motion of a single molecule is described in terms of random walk model
→ It is possible to calculate the mean-square distance travelled in given interval of time. It is
not possible to say in what direction a given molecule will move in that time.
(ii) On the average some fraction of the molecules in the lower element of volume will
2-9
cross the interface from below, and the same fraction of molecule in the upper element will
cross the interface from above in a given time.
(iii) Thus, simply because there are more ink molecules in the lower element than in the
upper one, there is a net transfer from the lower to the upper side of the section as a result of
random molecular motions.
(iv) Transfer of ink molecules from the region of higher to that of lower concentration is
observed.
2. Molecular Diffusion
(i) Fick’s 1st Law:
→ Rate of mass transport of material or flux through the liquid, by molecular diffusion is
proportional to the concentration gradient of the material in the liquid.
∂C
Diffusive mass flux, q = − D (1)
∂x
(negative sign arises because diffusion occurs in the direct opposite to that of increasing
concentration)
2-10
(ii) Fick’s 2nd Law:
∂C ∂ 2C
Conservation of mass + Fick’s 1st Law → =
D 2
∂t ∂x
●Assumption for Fick’s Law
→ Fick’s 1st law is consistent only for an isotropic medium, whose structure and diffusion
properties in the neighborhood of any point are the same relative to all directions.
In molecular diffusion: D=
x D=
y D=
z D
In turbulent diffusion: ε x ,ε y ,ε z
In shear flow dispersion: Kx, K y , Kz
[Cf] anisotropic medium
→ diffusion properties depend on the direction in which they are measured
3. 3-D differential equation of diffusion
2-11
(i) Rate at which diffusing substance enters the element through the face ABCD in the x
direction
 ∂q 
=
Influx 4dydz  qx − x dx 
 ∂x 
In which qx = rate of transfer through unit area of the corresponding plane through P
(ii) Rate of loss of diffusing substance through the face A’B’C’D’
 ∂q 
=
Outflux 4dydz  qx + x dx 
 ∂x 
(iii) Contribution to the rate of increase of diffusing substance in the element from these two
faces
 ∂q   ∂q  ∂q
Netflux =
4dydz  qx − x dx  − 4dydz  qx + x dx  =
−8dxdydz x
 ∂x   ∂x  ∂x
(iv) Similarly from the other faces we obtain
∂q y
−8dxdydz and
∂y
∂qz
−8dxdydz
∂z
(v) Time rate at which the amount of diffusing substance in the element increases
2-12
∂ ∂
= ( mass ) ( volume × conc.)
∂t ∂t
∂c
= 8dxdydz
∂t
(vi) Combine (iii), (iv), and (v)
∂c  ∂q ∂q ∂q 
8dxdydz =
−8dxdydz  x + y + z 
∂t  ∂x ∂y ∂z 
∂c ∂qx ∂q y ∂qz
+ + + =
0
∂t ∂x ∂y ∂z (2)
(vii) Substitute Fick’s law into Eq.(2)
∂C ∂ ∂C  ∂  ∂C  ∂  ∂C  ∂  ∂C  ∂  ∂C  ∂  ∂C 
=
−  −D  −  −D  −  −D = D + + D 
∂z  ∂x  ∂x  ∂y  ∂y  ∂z  ∂z 
D
∂t ∂x  ∂x  ∂y  ∂y  ∂z 
Remember D is isotropic for molecular diffusion.
For homogeneous medium; D ≠ f n ( x, y , z )
∂C  ∂ 2C ∂ 2C ∂ 2C 
= D 2 + 2 + 2 
∂t  ∂x ∂y ∂z 
For 1-dimensional system
∂C ∂ 2C
= D 2 → Fick’s 2nd law of diffusion
∂t ∂x
2-13
[Re] Vector Operations
Vectors = magnitude + direction
~ velocity, force
Scalar = magnitude
~ pressure, density, temperature, concentration
Vector F
F = Fx ex + Fy ey + Fz ez
ex , ey , ez = unit vectors
Fx , Fy , Fz = projections of the magnitude of F on the x, y, z axes
(1) Magnitude of F
F = F = ( Fx2 + Fy2 + Fz2 )

1/2
(2) Dot product = Scalar product
S = F ⋅G = F G cos φ
(3) Vector product = Cross product
V= F × G → vector
magnitude of V= V= F G sin φ
2-14
direction of V = perpendicular to the plane of F and G
→ right hand rule
(4) Derivatives of vectors
∂ F ∂Fx ∂F ∂F
= ex + y e y + z ez
∂s ∂s ∂s ∂s
(5) Gradient of F (scalar) → vector
∂F ∂F ∂F
grad F =∇ F = ex + ey + ez → vector
∂x ∂y ∂y
[∇ ]= pronounced as ‘del’ or ‘nabla’

∂
∇ ≡ ei
∂xi
[Re] grad(scalar) → vector grad(F+G)=grad F+grad G
grad(vector) → tensor grad CF=c grad F
(6) Divergence of F (vector) → scalar
 ∂ ∂ ∂ 
divF = ∇⋅F =  ex + e y + ez  ⋅ F
 ∂x ∂y ∂z 
 ∂ ∂ ∂ 
 ∂x ∂y ∂z 
(
=  ex + ey + ez  ⋅ Fx ex + Fy ey + Fz ez )
∂ ∂F ∂F
= ex Fx ex cos 0 + y ex ey cos90° + z ex ez cos90°
∂x ∂x ∂x
2-15
∂F ∂
+ ey Fy ey cos 0 + ez Fz ez cos 0
∂y ∂z
∂Fx ∂Fy ∂Fz

= + + → scalar
∂x ∂y ∂z
i j k
∂ ∂ ∂
(7) Curl V = ∇ × V =
∂x ∂x ∂x
ν 1 v2 v3
(8) div ( grad F ) = ∇ ⋅ ∇F = ∇ 2 F ≡ Laplacian of F
∂ 2 Fx ∂ Fy ∂ 2 Fz
2
= 2 + + 2
∂x ∂y 2 ∂z
 ∂F ∂F ∂F 
[Pf] div ( grad F
= ) div  ex + ey + ez 
 ∂x ∂y ∂z 
∂  ∂F  ∂  ∂F  ∂  ∂F 
=  + +  
∂x  ∂x  ∂y  ∂y  ∂z  ∂z 
∂ 2 Fx ∂ Fy ∂ 2 Fz
2
= 2 + + 2
∂x ∂y 2 ∂z
 ∂ ∂ ∂  ∂ ∂ ∂ 
∇ ⋅ ∇ =  i + j + k  i + j + k 
 ∂x ∂y ∂z  ∂x ∂y ∂z 
∂2 ∂2 ∂2
=2 + 2 + 2 =
∇2
∂x ∂y ∂z
2-16
2.1.3 Analytical Solution of Diffusion Equation
● Problem 1: Consider diffusion of an initial slug of mass M introduced instantaneously at
time zero at the x origin
[Cf] Continuous input → initial concentration specified as a function of time
i) Governing equation:
∂C ∂ 2C
=D 2 (2.3)
∂t ∂x
ii) Initial & Boundary conditions:
-Spreading of an initial slug of mass M introduced instantaneously at time zero at the x origin
Direct delta function=spike
1
function =
C (= t 0)= M δ ( x)
x 0, = ∆x
C ( x = ±∞, t ) = 0
Mass/area
iii) Solution by dimensional analysis

- dimensional analysis
- separation of variables
C ( x, t ) = f ( M , x, t , D ) - Laplace transformation
M  x 
C= f  (2.4)
4π Dt  4 Dt 
x
Set η= (2.5)
4 Dt
2-17
Substitute Eq. (2.4) into Eq. (2.5) and then into Eq. (2.3)
Cp
Eq. (2.4):
M  x  M
=C =f  = f (η ) C p f (η )
4π Dt  4 Dt  4π Dt
x ∂η η ∂η 1
η= → =
− , =
4 Dt ∂t 2t ∂x 4 Dt
∂f ∂η
'
∂C M ∂f  M 1   1
= + = f Cp + Cp  −  f
∂t 4π Dt ∂t  4π D t  ∂η ∂t  2t 
∂f  η   1
= Cp  −  + Cp  −  f
∂η  2t   2t  (a)
∂C ∂f ∂f ∂η ∂f 1
= C= Cp = Cp
∂x ∂x ∂η ∂x ∂η
p
4 Dt
∂ 2C ∂2 f 1
= C (b)
∂x 2 ∂η 2 4 Dt
p
Substitute (a) and (b) into Eq. (2.3)
∂f  η   1 ∂2 f 1
Cp  −  + Cp  −  f =
DC p 2
∂η  2t   2t  ∂η 4 Dt
∂f ∂2 f
2η +2f + 2 = 0
∂η ∂η
∂ ∂2 f
( 2η f ) + 2 = 0
∂η ∂η
2-18
Integrate once w.r.t. η
df
2η f + =
0 (2.6)
dη
Apply separation of variables to Eq. (2.6)
df
= −2η dη
f
Integrate both sides
ln f =−η 2 + C
−η +C
=f e= C0 e −η
2 2
(2.7)
Total mass, M
∞
∫ −∞
Cdx = M (2.8)
Substituting Eq.(2.4) and Eq.(2.7) into Eq.(2.8) yields C0 = 1
Then, (2.4) becomes
M  x2 
=
C ( x, t ) exp  −  (2.9)
4π Dt  4 Dt 
C p - peak exponential decay
2-19
2-20
2-21
2-22
[Re] Analytical solution by separation of variables
∂C ∂ 2C
=D 2 (2.10)
∂t ∂x
C ( x,= = M δ ( x)
t 0) (2.11a)
C ( x = ±∞, t ) = 0 (2.11b)
ε
M = lim ∫ f ( x)dx (2.11c)
ε →0 −ε
Separation of variables
C ( x, t ) = F ( x)G (t ) (2.12)
Substitute Eq.(2.12) into Eq.(2.10)
∂G ∂2 F
F ( x) = DG ( t ) 2
∂t ∂x
FG ' = DGF ''
1 G ' F ''
= = k
DG F
where k = const. ≠ f n ( x or t )
i) k > 0
k = ω2
1 G ' F ''
= = ω2
DG F
2-23
F ''− ω 2 F =
0 (2.13a)
G '− Dω 2 G =
0 (2.13b)
=
Solution of (2.13a) is F C1e wx
+ C2 e − wx (a)
Substituting (2.11b) into (a) yields C1 = 0
Then
F = C2e − wx
Dωt
Solution of (2.13b) is G = C3e
Substituting B.C. (2.11b) gives C3 = 0
This means that C = F∙G = 0 at all points, which is not true.
Therefore, k ≤ 0
ii ) k = 0
F '' = 0 → F = ax + b → a = 0 ∴F = b
G ' =0 → G =k
∴ C = FG = bk = const. → not true
Therefore, k < 0
2-24
iii ) k < 0
k = − p2
1 G ' F ''
= = − p2
DG F
F ''+ p 2 F =
0 (2.13c)
G '+ Dp 2 G =
0 (2.13d)
λx
Assume solution of Eq. (2.13c) as F = e
Substitute this into Eq. (2.13c) and derive characteristic equation (2.14)
λ 2 + p2 =
0
∴ λ=
± pi
∴ F= C1e pxi + C2 e − pxi
= C1 (cos px + isin px) + C2 (cos px − i sin px)
= A cos px + B sin px (2.14)
λt
Assume solution of Eq. (2.13d) as and G = e
Substitute this into Eq. (2.13d) and derive characteristic equation
λ + Dp 2 =
0
∴ λ =− Dp 2
C1e − Dp t
∴G =
2
(2.15)
2-25
Substitute Eq. (2.14) and (2.15) into Eq. (2.12)
C ( x, t ) F (=
= x ) G ( t ) ( A cos px + B sin px)e − Dp t
2
(2.16)
Use Fourier integral for non-periodic function.
Assume A, B = f n ( p )
C ( x, t ; p ) ={ A ( p ) cos px + B ( p ) sin px}( − Dp 2 t ) (2.17)
∞
C ( x, t ) = ∫ C ( x, t ; p ) d p
0
∫ { A( p) cos px + B ( p ) sin px} exp ( − Dp t ) dp

∞
= 2
(2.18)
0
Since Eq. (2.10) is linear and homogeneous, integral of Eq. (2.18) exists
I.C.: Eq. (2.11a) and Eq. (2.11c)
∫ { A( p) cos px + B ( p ) sin px} dp =

∞
C ( x, t =
0) = f ( x)
0
where f(x)= Fourier integral
≡
π
1
∫
0
∞
{ ∞
cos px ∫ f (v)cos ( pv ) dv + sin px ∫
0
∞
−∞ }
f ( v ) sin ( pv)dv dp
1 ∞
A( p ) = ∫ f ( v ) cos ( pv ) dv
π −∞
1 ∞
B ( p) = ∫ f ( v ) sin ( pv ) dv
π −∞
2-26
Use Trigonometric rule
∫ {∫ }
1 ∞ ∞ ∞
C ( x,0) f (v) cos px sin pvdv + ∫ f (v)sin px sin pvdv dp
π 0 −∞ −∞
∫ {∫ }
1 ∞ ∞
f ( v ) cos( px − pv)dv dp (2.19)
π 0 −∞
Substitute Eq. (2.19) into Eq. (2.18)
π ∫ {∫ }
1 ∞ ∞
=
C ( x, t ) f (v) cos( px − pv) exp(− Dp 2 t )dv dp
0 −∞
Switch order of integral
C(=
x, t )
π∫
1
0
∞
f (v ) {
∞
−∞
) cos( px − pv)dv} dp
∫ exp(− Dpt
2

(2.20)
(e)
∞
Let (e) = ∫ 0
exp(− Dp 2 t ) cos( px − pv)dp
Use residue theorem to get integral of (e)
∞ πy
∫ e − s cos 2bsds = e−b
2 2
(2.21)
0 2
x−v
=
Set s p=
Dt , b
2 Dt
Then 2bs =
( x − v) p, ds =Dtdp
2-27
(e) becomes
∞ π
 ( x − v) 2 
∫ exp(− Dp t ) cos( px − pv)= exp −
2
dp  (2.22)
0
2 Dt  4 Dt 
Substitute Eq. (2.22) into Eq. (2.20)
1 ∞  ( x − v) 2 
=C ( x, t ) ∫ f (v)exp − dv
4π Dt −∞
 4 Dt 
1 ε  ( x − v) 2 
lim ∫ f (v)exp −  dv
4π Dt ε →0 − ε
 4 Dt 
M  x2 
= exp  −  (2.23)
4π Dt  4 Dt 
2-28
2.2 The Random Walk and Molecular Diffusion
2.2.1 The Random Walk
Think motion of a tracer molecule consists of a series of random steps
→ whether the step is forward or backward is entirely random
Use central limit theorem → in the limit of many steps probability of the particle being
m∆x between and (m + 1)∆x is the normal distribution
mean: µ =0
t (∆x) 2
variance: σ =
2
∆t
1  x2 
normal distribution:=
p( x, t )dx exp  − 2 
σ 2π  2σ 
t ( ∆x ) 2
σ
= = 2 Dt
2
Designate ∆t
1  x2 
Then p( x, t )dx == exp  −  dx (2.15)
4π Dt  4 Dt 
Now think whole group of particles
C ( x, t ) = ∫∫ p ( x, t )dxdn
M  x2 
=
C ( x, t ) exp  − 
4π Dt  4 Dt 
→ Random walk process leads to the same result that a slug of tracer diffuses according to
the diffusion equation, Eq. (2.4).
2-29
2.2.2 The Gradient-Flux Relationship
Think random motion of large number of molecules at the same time.
→ probability of a molecule passing through the surface is proportional to the average
number of molecule near the surface
→ differences in mean concentration are, on the average, always reduced, never increased.
Consider flux of material across the bounding surface
ql
ql = kM l - flux of material from left to right
qr = kM r - flux of material from right to left
where k = transfer probability [1/t]
M l = mass of the tracer in the left-hand box
M r = mass of the tracer in the right-hand box
2-30
q = net flux = net rate at which tracer mass is exchange per unit time and per unit area
=q k (M l − M r ) (a)
M=
l Cl Vol= Cl ∆x
Define
Ml
Cl = (b)
∆x
Mr
Cr =
∆x (c)
M l = average masses in the left-hand box
M r = average masses in the right-hand box
Combine (b) and (c)
Ml − Mr =
∆x(Cl − Cr )
C −C
( ∆x )  − r l 
=
2
 ∆x 
2  ∂C 
≈ ( ∆x )  − if ∆x is small
 ∂x 
(d)
Substitute (d) into (a)
∂C
q =−k (∆x) 2
∂x
∂C
= −D ⇒ Fick’s law
∂x
D= k (∆x) 2 ⇒ Diffusion coefficient (constant)

2-31
2.3 Some Mathematics of the Diffusion Equation
2.3.1 Concentration Distribution
i) G.E.
∂C ∂ 2C
=D 2
∂t ∂x
ii) Initial condition for instantaneous point source
C ( x, 0) = M δ ( x) (2.18)
M = initial slugs of mass introduced at time zero at the origin
1
δ = Dirac delta function (= )
∆x
→ representing a unit mass of tracer concentrated into an infinitely small space with an
infinitely large conc.
→ spike distribution
[Ex] bucket of concentrated dye dumped into a large river
iii) Solution
M  x2 
C ( x, t )
= exp  −  (2.14)
4π Dt  4 Dt 
→ Gaussian distribution (Normal if M = 1)
[Re] mass M
For plane source, M = mass
2-32
For line source, M = mass / length
For point source, M = mass / area
2.3.2 Moments of Concentration Distribution
1. Moments
∞
Zeroth moment = M 0 = ∫ C ( x, t ) dx
−∞
∞
1st moment = M 1 = ∫ C ( x, t ) xdx
−∞
∞
2nd moment = M 2 = ∫ C ( x, t ) x 2 dx
−∞
∞
pth moment = M p = ∫ C ( x, t ) x p dx
−∞
x
C(x, t)
i) Mass M = M 0
ii) Mean µ = M1 / M 0
∞
∫ ( x − µ ) 2 C ( x, t ) dx M
iii) Variance σ
= 2 −∞
= 2
− µ2
M0 M0
2-33
M3 M
− 3µ 2 + 2 µ 3
M0 M0
iv) Skewness St =
(σ )
2 3/ 2
- measure of skew
- For normal dist., St = 0
For a normal distribution, M = 1
 ( x − µ )2 
1
N (µ , σ =
2
) ; f ( x) exp − , − ∞ < x < ∞
σ 2π  2σ 2 
E ( x) = µ ; Var ( x) = σ 2
µ 0,=
For concentration distribution, substitute= σ 2Dt
1  x2 
C ( x, t )
= exp  − 
4π Dt  4 Dt 
Then, M 0 = 1
µ = 0 → location of centroid of concentration distribution
σ 2 = 2Dt → measure of the spread of the distribution
2. Diffusion coefficient
Measure of spread of dispersing tracer
σ = 2Dt ⇒ standard deviation (see Table 2.1) (a)
2-34
4σ = 4 2Dt ⇒ estimate of the width of a dispersing cloud
⇒ include 95% of the total mass
[Cf] 6σ = 6 2Dt ⇒ include 99.5% of the total mass
95 %
Calculation of diffusion coefficient
→ change of moment method
Start from (a)
1
Dt = σ 2
2
1 dσ 2
D= (2.22)
2 dt
i) For normal distribution: it is obvious
ii) Eq. (2.22) can be also true for any distribution, provided that it is dispersing in
accord with the Fickian diffusion equation.
2-35
[Pf] Start with Fickian diffusion equation
∂C ∂ 2C
=D 2 (a)
∂t ∂x
2
Multiply each side by x
∂C ∂ 2C
x2 = Dx 2 2
∂t ∂x
Integrate from to −∞ and +∞ w.r.t x
∞ ∂C 2 ∞
2 ∂ C
2
∫−∞ ∂t x dx = ∫−∞ Dx ∂x 2 dx
∫ uv=' uv − ∫ u ' v
Apply integration by parts into right hand side
∂ ∞ 2   2 ∂C  ∞ ∞ ∂C 
∂t ∫−∞ ∫
= Cx dx D   x  − 2 x dx 
  ∂x  −∞ −∞ ∂x 
∞ ∂C  ∂C  
= −2 D ∫ x dx   ≈ 0 
−∞ ∂x  ∂x  ±∞ 
{
−2D [ xC ]−∞ − ∫ Cdx
=
∞ ∞
−∞ }
∞
2 D ∫ Cdx ( C ]∞−∞ ≈ 0)
−∞
∂ ∞ 2 ∂
∂t ∫ −∞
Cx dx
∂t
M2
∂  M2 
=
2D ∞
= =  
∂t  M 0 
∫ Cdx M0
−∞
1 ∂  M2 
D=   (b)
2 ∂t  M 0 
2-36
Now, multiply each side of Eq. (a) by x
 ∂C  
∞
∞ ∂C ∞ ∂ 2C ∞ ∂C
∫−∞ ∂t dx
= x ∫−∞ ∂x 2
=
Dx dx D  x  − ∫
 ∂x  −∞ −∞ ∂x 
dx 
∞ ∂C
−D∫
= − D[C ]∞−∞ =
dx = 0
−∞ ∂x
∂ ∞
∂t ∫−∞
∴ Cxdx =
0
∂
→ M1 = 0
∂t
∂ ∂
( M 1 /=M0 ) = (µ ) 0
∂t ∂t
→ μ is independent of time.
By the way,
M2
σ
= 2
− µ2
M0
∂ 2 ∂  M2  ∂ 2 ∂  M2 
(σ ) =   − (µ ) =   (c)
∂t ∂t  M 0  ∂t ∂t  M 0 
Combine Eq.(b) and Eq.(c)
1 ∂σ 2
D= (2.25)
2 ∂t
2-37
→ Variance of a finite distribution increases at the rate 2D no matter what its shape.
→ property of the Fickian diffusion equation: any finite initial distribution eventually
decays into Gaussian distribution.
▪ Change of moment method
Calculate diffusion coefficient from 2 concentration curves
Start from Eq.(2.25)
σ 2 2Dt + C
=
σ 2 2 2Dt2 + C
= (1)
σ 12 2Dt1 + C
= (2)
Subtract (1) from (2)
σ 2 2 − σ 12= 2 D(t2 − t1 )
Rearrange
1 σ 2 2 − σ 12
D= (2.26)
2 t2 − t1
σ 12 = variance of concentration distribution at t = t1
σ 2 2 = variance of concentration distribution at t = t2
2-38
Calculate diffusion coefficient from more than 2 curves
→ D=
1
2
( slope of σ 2 vs t curve )
σ2
[Re] Normal (Gaussian) distribution
- bell-formed distribution occur around a mean value
( t − µ )2
x 1
Distribution function: f (t ) = ∫ e 2σ 2
dt
−∞
σ 2π
2-39
Homework 2-1
Due: 1 week from today
Taking care to create as little disturbance as possible, a small sample of salt solution is
released at the center of the large tank of motionless fluid.
(a) After 24 hours have elapsed a conductivity probe is used to measure the concentration
distribution around the release location. It is found to be Gaussian with a variance of 1.53
centimeters squared. The experiment is repeated after a further 24 hours have elapsed and the
variance is found to be 3.25 centimeters squared. Determine the diffusion coefficient
indicated by the experimental data.
(b) Explain how the measured peak concentration at 24 hours and 48 hours could be used to
check the result in (a).
(c) Must the distribution be Gaussian for the method used in (a) to apply?
2-40
2.4 Solutions of the Diffusion Equation
G.E.:
∂C ∂ 2C
=D 2 (2.31)
∂t ∂x
Problem 1: Spreading of an initial slug of mass M introduced instantaneously at time zero at
the x origin
C (= t 0)= M δ ( x)
x 0, =
C ( x = ±∞, t ) = 0
Solution:
M  − x2 
C( X , t) = exp  
4π Dt  4 Dt 
2.4.1 An Initial Spatial Distribution C(x, 0)
(1) Mass M released at time t = 0 at the point x = ξ → Problem 1-1
I .C. ( x,0) M δ ( x − ξ )
C=
B.C. C (±∞, t ) = 0
Set X= x − ξ
Then, I.C. becomes
C ( X ,0) = M δ ( X )
2-41
Solution is
M −X 2 
C( X ,t) = exp  
4π Dt  4 Dt 
Convert X into x
M  −( x − ξ ) 2 
C ( x, t ) = exp   (2.28)
4π Dt  4 Dt 
0 ξ X
2-42
(2) Distributed source at time t=0 → Problem 1-2
dM ≅ f (ξ )d ξ
I .C.: C=
( x, 0) f ( x), −∞< x<∞
f(x) = arbitrary function
Assume that
- the initial input is composed from a distributed series of separate slugs, which all diffuse
independently
- motion of individual particles is independent of the concentration of other particles
Approximate f ( x ) by a series of slugs each containing mass f (ξ )d ξ
f (ξ )dξ  −( x − ξ ) 2 
dC ( x, t ) = exp  
4π Dt  4 Dt 
Then, total contribution from all slugs is the integral sum of all the individual contributions
∞ f (ξ )  −( x − ξ ) 2 
C ( x, t ) = ∫ exp   dξ (2.30)
−∞
4π Dt  4 Dt 
→ superposition integral
2-43
(3) Distributed source with step function → Problem 1-3
t0 t1
t2
C0
2
Consider a particular case where f ( x ) is given by a step function
0 x<0
I .C. C ( x,0) = 
C0 x>0
According to (2.30), solution is given as
∞ C0  −( x − ξ ) 2 
C ( x, t ) = ∫ exp   dξ (a)
0
4π Dt  4 Dt 
(x − ξ )
Set u=
4 Dt
x
=ξ 0=
: u
4 Dt
ξ = ∞ : u = −∞
−dξ
du = → dξ =
− 4 Dtdu
4 Dt (b)
2-44
x
C0
π ∫
C ( x, t ) = e − u du
2
4 Dt 2 z
π ∫
−∞ =
erf z exp(−ξ 2 )dξ
0
C  2 0 2 x

∫ π ∫
−u2
= 0 du + e − u du 
2
e 4 Dt
2  π −∞ 0

C  2 x

= 0 1 + ∫ e − u du 
2
4 Dt
2  π 0

C0   x 
= 1 + erf  
2   4 Dt 
C0   x 
C (=
x, t ) 1 + erf   (2.33)
2   4 Dt  
[Re] Error function
2 z
π ∫
=
erf z exp(−ξ 2 )dξ → Table 2.1
0
2 ∞
π ∫
=
erf ( ∞) −ξ 2 )d ξ 1
exp(=
0
[Re] Normal distribution
- Most important distribution in statistical application since many measurements have
approximate normal distributions.
- The random variable X has a normal distribution if its p.d.f. is defined by
 (x − µ 2 ) 
1
=f ( x) exp  − , − ∞ < x < ∞
σ 2π  2σ 2 
2-45
• Integral of Normal distribution
 ( x − µ )2 
∞ 1
=I ∫ exp  −  dx
−∞
σ 2π  2σ 2 
x−µ  1 
= =  dz
σ 
Let z dx
σ 
Then,
∞ 1  z2 
=I ∫−∞
2π
exp  − dz
 2
z 1  w2 
Φ ( z=
) P (−∞ < z ≤ Z=
) ∫ −∞
2π
exp  −
 2
dw

Φ (∞) =1
Φ (− z ) = 1 − Φ ( z )
2-46
(4) Distributed source with step function C0 for x < 0 → Problem 1-4
Consider instantaneous input of step function
C , x < 0
I.C. C ( x,0) =  0 
0, x>0 
Solution by line source of C0δξ is given as
C0δξ  ( x − ξ )2 
= exp  −
4 Dt 
dC
4π Dt 
Then, total contribution is
C0 0  ( x − ξ )2 
=C ( x, t )
4π Dt ∫−∞ − 4 Dt dξ
exp
x −ξ
Set η=
4 Dt
dξ
Then dη = −
4 Dt
ξ = −∞ → η = ∞
x
ξ= 0 → η=
4 Dt
Substituting this relation yields
exp ( −η= ) ( −dη ) exp ( −η 2 ) dη

C0 x / 4 Dt C0 ∞
π∫ π∫
=
C ( x, t ) 2
∞ x / 4 Dt
C0  2 
exp ( −η 2 ) dη 
∞
=
2  π ∫
x/ 4 Dt

2-47
C0  2 
exp ( −η 2 ) dη − exp ( −η 2 ) dη 
∞ 2 x / 4 Dt
=
2  π ∫0
π ∫ 0

C0   x 
= 1 − erf  
2   4 Dt  
C0  x 
= erfc  
2  4 Dt 
C0   x   C0  x 
C ( x, t ) = 1 − erf    = erfc  
2   4 Dt   2  4 Dt 
→ complementary error function
→ summing the effect of a series of line sources, each yielding an exponential of distribution
[Re] complementary error function
erfc( z ) = 1 − erf ( z )
2-48
2.4.2 Concentration Specified as a Function of Time, C(0, t)
(1) Continuous input with step function C0 = C0 (t ) → Problem 2
G.E.:
∂C ∂ 2C
=D 2 (2.31)
∂t ∂x
Consider continuous input with step input at x = 0
I .C. C ( x,=
t 0)= 0
B.C. C ( x = 0, t > 0) = C0
Solution by dimensional analysis
 x 
C = C0 f  
 Dt 
x
Set η=
Dt
∂η x  1  − 23 1 x 1
=−  t = − − η
=
∂t D  2 2t Dt 2t
∂η 1
=
∂x Dt
2-49
Then
∂C dC ∂η 1 dC
= = − η
∂t dη ∂t 2t dη
∂ 2 C 1 d 2 C  ∂ 2 C d 2 C ∂ 2η 
= =  
∂x 2 tD dη 2  ∂x 2 dη 2 ∂x 2 
Substitute these into Eq. (2.31) to obtain O.D.E.
1 df d 2 f
− η =
2 dη dη 2
2 f ''+ η f ' =
0 (a)
B.C. f (0) = 1
f ( ∞ ) =0
Solution of this is
  x   x 
C=
C0 1 − erf   =
C0erfc   x>0 (2.37)
  4 Dt   4 Dt 
2-50
[Re] Laplace transformation
For ODE, it transforms ODE into algebraic problem.
For PDE, it transforms PDE into ODE.
i) inverse transform
C ( x, t ) = L−1 ( F )
ii) linearity of Laplace transformation
L {af (t ) + bg (t )} = aL{ f (t ) + bL { g ( t )}
iii) integration of f(t)
L {∫
0
t
} 1
f (τ )dτ = L { f (t )}
s
iv) use Laplace transformation("operational calculus")
∞
Fn (=
x, s ) L=
(C ) ∫0
e− st C ( x,=
t )dt C
L(C ') =
sL(C ) − C ( x,0) =−
sC C ( x,0)
L(C '') =s 2 L(C ) − sC ( x,0) − C '( x,0)
[Re] Analytical solution by Laplace transformation
Consider advection-diffusion equation as G.E.
∂C ∂C ∂ 2C
+u =
D 2
∂t ∂x ∂x
B.C. & I.C. for continuous input are given as
C ( x = 0, t > 0) = C0 (a)
2-51
C ( x ≥ 0, t =0 ) =0 (b)
C ( x = ±∞, t ≥ 0 ) = 0 (c)
Rewrite G.E.
∂ 2C ∂C ∂C
D 2 −U − =
0
∂x ∂x ∂t
Apply Laplace transformation
∂ 2C ∂C
D 2 −U − sC − C (t =0) =0
∂x ∂x
∂ 2C ∂C C (= = 0
t 0)
D 2 −U − sC =
0
∂x ∂x from I.C.
∂C ∂ 2C
=
Set C ' = , C ''
∂x ∂x 2
U s
Then C ''− C '− C =
0
D D
λx
Assume C = e
Derive characteristic equation as
U s
λ2 − λ− =0
D D
Solution is
2-52
2
U U   s
±   + 4 
D  D  U ± U + 4 sD
2
D
=λ =
2 2D
Then, C is
=C C1e − λ1 x + C2 e − λ2 x
U + U 2 + 4 sD  U − U 2 + 4 sD 
C1 ( s ) exp  x  + C2 ( s ) exp  x  (1)
 2D   2D 
Laplace transformation of B.C. , Eq. (c) is
∞
lim C ( x, s ) = lim ∫ e − st C ( x, t )dt
x →∞ x →∞ 0
=
∞
∫= { }
e − st lim C ( x, t )dt 0
0 x →∞
If we apply this to Eq. (1)
 U + U 2 4 sD 
lim C ( x, s ) = lim C1 ( s ) exp  x 
x →∞ x →∞
  2D 
 U − U 2 4 sD 
+ lim C2 ( s ) exp  x  = 0
x →∞
  2D 
∴ C1 ( s ) should be zero
U − U 2 + 4 sD 
∴∴ C = C2 ( s ) exp  x
 2D 
2-53
Apply B.C., Eq. (a)
Laplace transformation
1
C (0, s ) = C0
s
∴ C2 ( s ) =
C0 / s
C0 U − U 2 + 4 sD 
∴∴ C = exp  x
s  2D 
 Ux  1  x U2 
= C0 exp   exp  − +s
 2D  s D 4D 
 
Get inverse Laplace transformation using Laplace transform table
2  2 2
1
 a   a 
exp −a ( s + b )  ⇔ e − ab erfc  − b t  + e ab erfc  +b t 
s   2 t  2 t 
x U
=
Set a = ,b
D 2 D
  x U   xU 
e − ab =
exp −   =
exp − 
  D 2 D   2D 
 x U   xU 
= =
e ab exp   exp  
 D 2 D  2D 
a x/ D U x − Ut
−b =
t − =
t
2 t 2 t 2 D 4 Dt
a x/ D U x + Ut
+b =
t + =
t
2 t 2 t 2 D 4 Dt
2-54
C0  Ux    −Ux   x − Ut   Ux   x + Ut  
∴C exp   exp   erfc   + exp   erfc  
2  2D    2D   4 Dt   2D   4 Dt  
C0   x − Ut   Ux   x + Ut  
= erfc   + exp  D  erfc  
2   4 Dt     4 Dt  
In case U=0
C0   x   x   x 
=C erfc   + erfc    = C0 erfc   (2.37)
2   4 Dt   4 Dt   4 Dt 
2-55
(2) Concentration specified as a function of time at fixed point → Problem 2-1
Consider the case where C0 (τ ) is a time variable concentration at x = 0
I.C. and B.C. are
C ( x,= = 0
t 0)
C ( x = 0, t > 0) = C0 (τ )
The solution is obtained by a superposition of solution, (2.37).
∂C
In each time increment δτ the concentration at x = 0 changes by an amount δτ .
∂τ
Thus, for a change occurring at time τ the result for all future times, due to the incremental
changes, is given as
∂C0  x 
δC = δτ erfc  , t >τ
∂τ  4 D(t − τ ) 
 
Then, total contribution is
t ∂C0  x 
C=∫ erfc  dτ
−∞ ∂τ  4 D(t − τ ) 
 
2-56
2.4.3 Input of mass specified as a function of time
 , −∞ < t < ∞ → Problem 3

(1) Continuous injection of mass at the rate M
 is equivalent to injecting a
We can assume that a continuous injection of mass at the rate M
 δτ after each time increment

slug of amount M δτ .
Then, the concentration resulting from the continuous injection is the sum of the
concentrations resulting from the individual slugs injected at all time prior to the time of
observation.
Thus, concentration resulting from the individual slug is
M (τ )dτ  x2 
=dC exp  − 
4π D (t − τ )  4 D (t − τ ) 
Total contribution is
t M (τ )  x2 
=C ∫
−∞
4π D(t − τ )
exp  − dτ
 4 D(t − τ ) 
(2.41)
 (τ ) = rate of input mass at time

where M τ and may vary with time = [ML-2t-1]
 ,t>0
(2) Continuous injection of mass of constant strength M
→ Problem 3-1
Eq. (2.41) gives
M t 1  x2 
4π D ∫0 t − τ
=C ( x, t ) exp  −
− τ dτ (1)
 4 D (t ) 
2-57
4 D(t − τ )
Set u=
x2
4D
→ du = − dτ
x2
x2
dτ = − du
4D
Then, substituting this into (1) yields

Mx 4 Dt
−
1
1
=C ( x, t )
4D π ∫
0
x2 u exp(− )du
2
u (2.42)
(3) Continuous injection of distributed source of mass m(x, t) → Problem 3-2
→ superposition in space and then in time to get solution
m = mass per unit length per unit time= [ML-3t-1]
t ∞ m(ξ ,τ )  ( x − ξ )2 
=C ( x, t ) ∫ ∫
−∞ −∞
4π D(t − τ )
exp  −
 4 D (t − τ )
 d ξ dτ

(2.43)
2-58
Homework #2-2
Due: 1 week from today
Consider pulse input of concentration specified as a step function
C0
C03
C02
C01
τ1 τ2 τ3 t
∂C ∂ 2C
G. E.: =D 2
∂t ∂x
I. C.: C ( x, = ) 0
t 0=
B. C.: C (= ) C01
x 0, 0<t ≤ τ 1=
C (= 2 ) C02
x 0, τ 1 <t ≤ τ =
( x 0, τ 2 <t ≤ τ=
C= 3 ) C03
C ( x 0,=
= τ 3 <t ) 0
a) Derive analytical solution using Laplace transformation.
b) Plot C vs x for various time t with assumed C0 s, for example C01 = C0 / 2 ; C02 = C0 ;
3
C03 = C0 .
2
c) Plot C vs t for various distance x.
2-59
2.4.4 Solution Accounting for Boundaries
- Consider spreading restricted by the presence of boundaries
• Principle of superposition
→ If the equation and boundary conditions are linear it is possible to superimpose any
number of individual solutions of the equation to obtain a new solution.
The method of superposition for matching the boundary condition of zero transport through
the walls (single boundary)
(1) mass input at x = 0 with nondiffusive boundary at x = -L → Problem 4
I.C. : unit mass of solute at x = 0 at t = 0
B.C.: wall through which concentration cannot diffuse located at x = -L
→ Fick’s law for the boundary condition of no transport through the wall is
∂C
q x=
−L =
−D
x=
−L → =
0 Neumann type B.C.
∂x
→ Concentration gradient must be zero at the wall.
→ This condition would be met if an additional unit mass of solute (image source) was
concentrated at the point x = -2L, and if the wall was removed so that both slugs could diffuse
to infinity in both directions.

mirror
-2L -L 0
2-60
→ Solution with the real boundary = sum of the solutions for real plus the image source w/o
the boundary
x − ( −2 L )
Unit mass;
M=1
1   x2   ( x + 2 L) 2  
=C exp  −  + exp  −  (2.44)
4π Dt   4 Dt   4 Dt  
real source image source
2-61
(2) Unit mass input x = 0 with nondiffusive boundaries at x = - L and at x = + L
→ Problem 4-1
→ put image slugs at -2L, +2L, 4L, -6L, 8L, ……
(∵ slug at x = -2L causes a positive gradient at the boundary at +L , which must be
counteracted by another slug located at x = +4L, and so on)
Then, solution is
∞
1  −( x + 2nL) 2 
C ( x, t ) = ∑ 4π Dt
exp  
n =−∞  4 Dt 
n = -2, -1, 0, +1, =2
2-62
2-63
(3) Zero concentration at x= ± L → Problem 4-2
→ C(x =
± L, t ) =
0 → Dirichlet type B.C.
→ negative image slugs at x = ±2 L
positive image slugs at x = ±4 L etc.
1 ∞
  −( x + 4nL) 2   − [ x + (4n − 2) L ]2  
C ( x, t ) ∑ exp  4 Dt  − exp 
4π Dt n = −∞ 

   4 Dt  
(4) Mass input x = 0 with nondiffusive boundaries at x = 0→ Problem 4-3
→ Solution for negative x is reflected in the plane x = 0 and superposed on the original
distribution in the region x>0.
→ reflection at a boundary x=0 means the adding of two solutions of the diffusion equation
1   x2   ( x + 0) 2  
=C exp  −  + exp  − 
4π Dt   4 Dt   4 Dt  
1  x2 
= exp  − 
π Dt  4 Dt 
2-64
2.4.5 Solutions in Two and Three Dimensions
(1) 2-D Fluid
- A mass M [M/L] deposited at t=0 at x=0, y=0 → Problem 5-1
∂C ∂ 2C ∂ 2C
G. E.:= Dx + D (2.49)
∂t ∂x 2 ∂y 2
y
I.C.: C ( x, y,0) = M δ ( x)δ ( y )
=
For molecular diffusion, Dx D=
y D
use Product rule
C ( x, y, t ) = C1 ( x, t )C2 ( y, t )
where C1 ≠ f ( y ) , C2 ≠ f ( x )
∂C ∂ ∂C ∂C
= (=
C1C2 ) C1 2 + C2 1
∂t ∂t ∂t ∂t
∂ 2C ∂ 2 ∂ 2C1
= = (C1C2 ) C2 2
∂x 2 ∂x 2 ∂x
∂ 2C ∂ 2 ∂ 2 C2
= = (C1C2 ) C1 2
∂y 2 ∂y 2 ∂y
Eq. (2.49) becomes
∂C2 ∂C1 ∂ 2 C1 ∂ 2 C2
C1 + C2 = Dx C2 + Dy C1
∂t ∂t ∂x 2 ∂y 2
2-65
Rearrange
 ∂C1 ∂ 2C1   ∂C2 ∂ 2 C2 

C2  − Dx 2  + C1  − Dy =
∂y 2 
0
 ∂t ∂x   ∂t
Whole equation = 0 if
∂C1 ∂ 2 C1 
= Dx 
∂t ∂x 2 
→ 1-D diffusion equation
∂C2 ∂ 2 C2 
= Dy
∂t ∂y 2  Mx
∞
=
∴ C1
∫ −∞
Cdx  x2 
exp  − 
4π Dxt  4 Dxt 
∞
=
∫ −∞
Cdy  y2 
exp  −
 4 D t 
C2 ∞
4π Dy t  y  M = ∫ ∫ C dxdy
−∞
 x2 y2  M
∴∴=
C C= exp  − −
1C2  4 D t 4 D t 
(2.53)
4π t Dx Dy  x y 
→ lines of constant concentration = set of concentric ellipses
▪ Iso-concentration lines
 x2 y2 
− − 
M  4 Dx t 4 Dy t 
C= e  
4π t Dx Dy
 x2 y2 
− − 
 4 Dx t 4 Dy t  4π t Dx Dy C
e  
=
M
2-66
x2 y2  4π t Dx Dy C   M 
+ −ln 
= =ln  
4 Dxt 4 Dxt  M   4π t Dx Dy C 
   
x2 y2  M 
∴ + 4tln 
= 
( ) ( )  4π t Dx Dy C 
2 2
Dx Dy  
x2 y2
+ =
A
( ) ( )
2 2
Dx Dy
x2 y2
+ =
1 → ellipses
( ) ( )
2 2
ADx ADy
If Dx = Dy
Then, x + y =
R → circle
2 2 2
Dy
Dx
2-67
2-68
2-69
(2) 3-D fluid
- A mass M [M] deposited at t=0 at x=0, y=0, z=0 → Problem 5-1
∂C ∂ 2C ∂ 2C ∂ 2C
G. E.: = D x 2 + D y 2 + Dz 2 (b)
∂t ∂x ∂y ∂z
I.C.: C ( x, y, z ,0 ) = M δ ( x ) δ ( y ) δ ( z ) → point source
Use product rule
C ( x, y, z , t ) = C1 ( x, t )C2 ( y, t )C3 ( z , t )
∂C ∂ ∂ (C2C3 ) ∂C
= = (C1C2C3 ) C1 + C2C3 1
∂t ∂t ∂t ∂t
∂C3 ∂C ∂C
= C1C2 + C1C3 2 + C2C3 1
∂t ∂t ∂t
∂ 2C ∂ 2 ∂ 2C1
= = (C1C2C3 ) C2C3 2
∂x 2 ∂x 2 ∂x
∂ 2C ∂ 2 ∂ 2 C2
= = (C1C2C3 ) C1C3
∂y 2 ∂y 2 ∂y 2
∂ 2C ∂ 2 ∂ 2C3
= = (C C C ) C C
∂z 2 ∂z 2 ∂z 2
1 2 3 1 2
Substituting these relations into (b) yields
∂C3 ∂C2 ∂C1 ∂ 2C1 ∂ 2 C2 ∂ 2C3

C1C2 + C1C3 + C2C3 = Dx C2C3 2 + D y C1C3 + Dz C1C2 2
∂t ∂t ∂t ∂x ∂y 2 ∂z
2-70
 ∂C3 ∂ 2C3   ∂C2 ∂ 2 C2   ∂C1 ∂ 2C1 

C1C2  − Dz 2 
+ C1C3  − Dy 2 
+ C2C3  − Dx 2 
=
0
 ∂t ∂ z   ∂t ∂ y   ∂t ∂ x 
=
∫ Cdx  x2 
exp  −
C1 
4π Dx t  4 Dx t 
=
∫ Cdy  y2 
exp  −
 4 D t 
C2
4π Dy t  y 
=C3 ∫ Cdz  z2 
exp  − 
4π Dz t  4 Dz t 
M  x2 y2 z2 
∴
= C C1C= exp  − − −
2 C3 3 1  4 D t 4 D t 4 D t 
(4π t ) 2 ( D x D y D z ) 2  x y z 
M = ∫∫∫ Cdxdydz
2.4.6 Advective Diffusion
(1) Governing Equation


Fluid moving with velocity u
   
u = ui + vj + wk
linear
• Advection = transport by the mean motion of the fluid
• Assume the transports by advection and by diffusion are separate and additive processes.
→ rate of mass transport through unit area (yz plane) by x component of velocity, qu
qu = uC
2-71
[Re] advective flux
mass = volume ∙ concentration
mass rate = volume rate ∙ conc.
= discharge ∙ conc.
= velocity ∙ area ∙ conc.
advective flux = mass rate /area = velocity ∙ conc.
Total rate of mass transport
 ∂C 
q= uC +  − D  (2.55)
 ∂x 
= advective flux + diffusive flux
Substitute (2.55) into mass conservation equation, (2.3)
∂C ∂q
+ =
0
∂t ∂x
∂C ∂  ∂C 
+  uC − D =0
∂t ∂x  ∂x 
∂C ∂ ∂ 2C
+ (uC ) =
D 2 → 1-D Advection-Diffusion Equation
∂t ∂x ∂x
→ Linear, 2nd order PDE
• In 3-D
i) Mass conservation equation
∂C  ∂qx ∂q y ∂qz
= −∇ ⋅ q (a) (divergence → + + )
∂t ∂x ∂y ∂z
2-72
ii) Rate of mass transport
  ∂C  ∂C  ∂C 
q = Cu − D∇C (b) (gradient → i+ j+ k)
∂x ∂y ∂z
∂C 
∂t
( )
+ ∇ ⋅ Cu − D∇C = 0
∂C 
∂t
( )
+ ∇ ⋅ Cu = D∇ 2C (2.57)
Rearrange 2nd term of LHS

  
( )
∇ ⋅ Cu = ∇C ⋅ u + C∇ ⋅ u
By the way
 ∂u ∂v ∂w
∇ ⋅=
u + + = 0 ← continuity eq. for incompressible fluid
∂x ∂y ∂z
 
∴∇ ⋅ (Cu ) = ∇C ⋅u
 ∂C  ∂C  ∂C     
= 
 ∂x
i+
∂y
j+ (
k  ui + v j + wk
∂z 
)
∂C ∂C ∂C
=u +v +w
∂x ∂y ∂z
(2.57) becomes
2-73
∂C 
+ ∇C ⋅ u= D∇ 2C (2.58)
∂t
→ 3-D Advection-Diffusion Equation
∂C ∂C ∂C ∂C  ∂ 2C ∂ 2C ∂ 2C 
+u +v + w = D 2 + 2 + 2  (2.59)
∂t ∂x ∂y ∂z  ∂x ∂y ∂z 
(2) Analytical Solutions
1) Instantaneous mass input → Problem 6-1
Assume that u is constant and gradient in y-direction is small
∂C ∂C ∂ 2C
G.E.: +u =
D 2
∂t ∂x ∂x
I.C.: C ( x,0) = M δ ( x)
B.C.: C (±∞, t ) = 0
M  ( x − ut ) 2 
=
C ( x, t ) exp  − 
4π Dt  4 Dt 
2) Instantaneous concentration input over x<0 → Problem 6-2
• Problem of pipe filled with one fluid being displaced at a mean velocity u by another fluid
with a tracer in concentration C0
I.C.: C (=
x,0) 0, x > 0
C=
( x,0) C0 , x < 0
2-74
Transform coordinate system whose origin moves at velocity u
Let x ' =
x − ut , t =
t
∂x ' ∂x '
→ =
1, =
−u
∂x ∂t
∂t ∂t
= 0,= 1
∂x ∂t
Use chain rule
∂ ∂x ' ∂ ∂t ∂ ∂
= + =
∂x ∂x ∂x ' ∂x ∂t ∂x '
∂ ∂x ' ∂ ∂t ∂ ∂ ∂
= + =
−u +
∂t ∂t ∂x ' ∂t ∂t ∂x ' ∂t
Substitute this into G.E.
∂C ∂C ∂C
=
−u +
∂t ∂x ' ∂t
∂C ∂C
u =u
∂x ∂x '
2-75
∂ 2C ∂ 2C
D 2 =D 2
∂x ∂x '
Then G.E. becomes
∂C ∂ 2C
=D 2 (a)
∂t ∂x '
→ This problem is identical to diffusion of distributed source with step function
C0 for x < 0 in a stagnant fluid (Problem 1-4).
There, solution is
C0   x' 
C (=
x ', t ) 1 − erf  
2   4 Dt 
Adjust for the moving coordinates
C0   x − ut  
C=
( x, t ) 1 − erf  
2   4 Dt   (2.63)
2-76
2-77
3) Lateral (Transverse) diffusion → Problem 6-3
- transverse mixing of two streams of different uniform concentrations flowing side by side
Start with 2-D advection-diffusion equation
∂C ∂C ∂C  ∂ 2C ∂ 2C 
+u + v = D 2 + 2 
∂t ∂x ∂y  ∂x ∂y 
Assumptions:
∂C
i) continous input: →0
∂t
∂C
ii) velocity in transverse direction is small: v →0
∂y
∂ 2C
iii) advection in x-direction is bigger than diffusion: D 2 →0
∂x
Then, G.E. becomes
∂C ∂ 2C
u =D 2
∂x ∂y
→ similar to 1-D diffusion equation
2-78
=
B. C.: C (0, y) 0 y>0
C=
(0, y ) C0 , y < 0
→ Now, this problem is similar to Problem 1-4 with t = x/u; x′ = y
Solution is
C0   x ' 
∴=
C 1 − erf  
2   4 Dt  
Convert to x-y coordinates
C0   y 
=C 1 − erf   (2.64)
2   4 Dx / u  
4) Continuous plane source → Problem 6-4
∂C ∂C ∂ 2C
G.E.: +u =
D 2
∂t ∂x ∂x
B.C.: C (0, t ) = C0 0 < t < ∞ → steady continuous input
C ( x, 0) = 0 0< x<∞
→ This problem is identical to continuous input with step function C0 = C0 (t ) (Problem 2).
The solution is
C0   x − ut   x + ut   ux  
=C ( x, t )  erfc   + erfc   exp  D   (2.65)
2   4 Dt   4 Dt   
2-79
Set
ux
Pe = Peclet number =
D
ut t
t=
R =
x x/u
Then,
 1
  1

      
(1 + t R )  exp( Pe )
C ( x, t ) 1 P 2
1 P 2
= erfc  e
 (1 − t R ) + erfc  e

 4t  2  4t 
 R   R 
C0 2
 
For advection-dominated case (large u)
C 1  x − ut 
Pe > 500; ≈ erfc  
C0 2  4 Dt 
Diffusion problem (u = 0)
C  x 
= erfc  
C0  4 Dt 
2-80
2-81
Homework Assignment #2-3
Due: Two weeks from today
a) Derive analytical solution for 1-D dispersion equation with continuous plane source
condition which is given as
C (0, t ) = C0 , 0<t <∞
C ( x,=
t 0)= 0, 0< x<∞
C ( x = ±∞, t ) = 0, 0<t <∞
C0   x − ut   x + ut   ux  
=C ( x, t ) erfc   + erfc   exp  D  
2   4 Dt   4 Dt   
b) Plot C vs. x for various values Pe of and t.
2-82
2.4.7 Maintained point source
(1) Constant point source in 3D → Problem 7-1
 at the origin (x, y, z,) in three-dimensional flow

- mass input at the rate M
∂C ∂C  ∂ 2C ∂ 2C ∂ 2C 
G.E.: + u = D 2 + 2 + 2 
∂t ∂x  ∂x ∂y ∂z 
I.C.: C ( x, y, z , 0 =) M δ ( x)δ ( y )δ ( z )
• Reduction of a three-dimensional problem to two dimensions by considering diffusion
in a moving slice
→ visualize the flow as consisting of a series of parallel slices of thickness δx

bounded by infinite parallel y-z planes
→ slices are being advected past the source, and during the passage each one receives a
 δt
slug of mass of amount M
δx
- time taken for slice to pass source; δt =
u
 δ t M δx
= M
mass collected by slice at it passes source =
u
2-83
2-D solution ← Eq. (2.53)
M  x2 y2 
= exp  − −
 4 D t 4 D t 
C
4π t Dx Dy  x y 
δx
M  ( x2 + y 2 ) 
=C u exp  − 
4π Dt  4 Dt 
 
x  M
Substitute t = and M =
u δx
M  ( x2 + y 2 ) u 
=
C ( x, y , z ) exp  −  (2.67)
4π Dx  4 Dx 
 
Eq. (2.67) was derived by neglecting diffusion in the direction of flow.
→ ut >> 2 Dt or t >> 2 D / u 2
(2) Maintained point source in 2D → Problem 7-2
M δ x / u  y2 
=C1 exp  − 
4π Dt  4 Dt 

Substitute t =
x  =M
and M
u δx
M  y 2u 
=C ( x, y ) exp  −  (2.68)
u 4π Dx / u  4 Dx 
M = strength of a line source in units of mass per unit length per unit time
2-84
2.4.8 Solutions for Mixing in Rivers
(1) 2-D Instantaneous Input
y u W
x
Assume rapid vertical mixing
∂C ∂C ∂ 2C ∂ 2C
G.E.: + u = Dx 2 + Dy 2
∂t ∂x ∂x ∂y
∂C
B.C.: = 0 → impermeable, non-diffusion boundary
∂y y = 0, w
I.C.: C ( x, y, 0) = M δ ( x)δ ( y )
i) Case A: Right-bank input
Use product rule C = C1 ( x, t ) C2 ( y, t )
 ∂C1 ∂C ∂ 2 C1   ∂C2 ∂ 2 C2 
C2  +u − Dx  + C1  − Dy =0
 ∂t ∂x ∂x 2   ∂t ∂y 2 
M1  ( x − ut ) 2 
=C1 exp  − 
4π Dx t  4 Dx t 
∞
M2   ( y + 2nW ) 2 
=C2 ∑ exp −
4π Dy t 

n = −∞  4 D y t 
 ( x − ut ) 2  ∞ 
M  ( y + 2nW ) 2 
C= exp  −  ∑ exp − 
4π t Dx Dy  4 Dx t  n = −∞   4 Dy t 
2-85
ii) Case B: Centerline input
a) For axis at right bank
M1  ( x − ut ) 2 
=C1 exp  − 
4π Dx t  4 Dx t 
  W 
2
∞   y + (2n − 1)  
exp  − 
M2 2 
=C2 ∑
n = −∞ 4π Dy t  4 Dy t 
 
 
  W 
2
 ( x − ut ) 2  ∞   y + (2n − 1)  
M  2 
C= exp  −  ∑ exp  −
4π t Dx Dy  4 Dx t  n = −∞  4 Dy t 
 
 
b) For axis at centerline
M1  ( x − ut ) 2 
=C1 exp  − 
4π Dx t  4 Dx t 
∞
M2   ( y + nW ) 2 
=C2 ∑ exp −
4π Dy t 

n = −∞  4 D y t 
M ( x − ut ) 2  ∞   ( y + nW ) 2 
C= exp  −  ∑ exp − 
4π t Dx Dy  4 Dx t  n = −∞   4 Dy t 
[Re] Decaying substance
∂C ∂C ∂ 2C ∂ 2C
G.E.: + u = Dx 2 + Dy 2 − kC
∂t ∂x ∂x ∂y
C ( x, y=
, t ) C=
(k 0)exp(− kt )
2-86
(2) 3-D Instantaneous Input
y u W d z
x y
∂C ∂C ∂ 2C ∂ 2C ∂ 2C
G.E.: + u = Dx 2 + Dy 2 + Dz 2
∂t ∂x ∂x ∂y ∂z
B.C.:
∂C
i) water surface = 0 impermeable, non-diffusive
∂z z =d
∂C
ii) solid boundary =0
∂y y =0,W
∂C
=0
∂z z =0
I.C.: C ( x, y, z , 0 )= M δ ( x)δ ( y )δ ( z )
i) Case A: Right-bank input – surface input
Use product rule C = C1 ( x, t )C2 ( y, t )C3 ( z , t )
M1  ( x − ut ) 2 
=C1 exp  − 
4π Dx t  4 Dx t 
∞
M2   ( y + 2nW ) 2 
=C2 ∑ exp −
4π Dy t 

n = −∞  4 D y t 
2-87
n
M3   ( z + (2n − 1)d ) 2 
=C3 ∑ exp −
4π Dz t 

n = −∞  4 Dy t 
M  ( x − ut ) 2 
C exp  − 
(4π t )3 / 2 Dx Dy Dz  4 Dz t 
∞   ( y + 2nW ) 2   ( z + (2n − 1)d ) 2 

⋅ ∑ exp −  exp − 
n =−∞ 
  4 D y t   4 D y t 
ii) Case B: Right-bank input – mid-depth input
  d 2 
( z + (2 n − 1) ) 
n
M3  
=C3 ∑ exp −
4π Dy t  4 D t
2

n = −∞  z 
  
M  ( x − ut ) 2 
C exp  − 
(4π t )3 / 2 Dx Dy Dz  4 Dx t 
  d 2 
∞   ( y + 2nW ) 2   ( z + (2 n − 1) ) 
⋅ ∑ exp −  exp −
2

n =−∞   4 D y t   4 Dz t 
  
iii) Case C: Right-bank input – bottom input
n
M3   ( z + 2nd ) 2 
=C3 ∑ exp −
4π Dz t 

n = −∞  4 Dz t 
M  ( x − ut ) 2 
C exp  − 
(4π t )3 / 2 Dx Dy Dz  4 Dx t 
2-88
∞   ( y + 2nW ) 2   ( z + 2nd ) 2 
⋅ ∑ exp −  exp − 
n =−∞ 
 
 4 D y t 
  4 Dz t 
iv) Case D: Centerline input – bottom input
  W 2 
∞   ( y + (2 n − 1) ) 
M2
=C2 ∑ exp −
4π Dy t  4 Dy t
2

n = −∞  
  
n
M3   ( z + 2nd ) 2 
=C3 ∑ 4π Dz t
exp − 
n =−∞   4 Dz t 
M  ( x − ut ) 2 
C exp  − 
(4π t )3 / 2 Dx Dy Dz  4 Dx t 
  W 2 
 ( y + (2 n − 1) )
 2  exp − ( z + 2nd ) 
∞ 2
⋅ ∑ exp −   
n =−∞   4 Dy t   4 Dz t 
   
2-89
(3) 2-D Analytical Solutions for continuous point injection
y u B W d z
x A y
* Governing Equation
∂C ∂C ∂ 2C ∂ 2C
+ u = Dx 2 + Dy 2
∂t ∂c ∂x ∂y
Case I: side injection
∂C
= 0,= C (0,0, t ) C0
∂y y =0,w
C ( x, y,0) = 0
Case II: centerline injection
∂C
= 0,= C (0, w / 2, t ) C0
∂y y =0,w
C ( x, y,0) = 0
* Product Rule
C = C1 ( x, t ) C2 ( y, t )
2-90
Then, the governing equation will be modified as
∂C ∂C ∂C1 ∂ 2C1 ∂ 2C2

C2 1 + C1 2 + u = C DxC2 2 + DyC1 2
∂t ∂t ∂x
2
∂x ∂y
 ∂C1 ∂C1 ∂ 2C1   ∂C2 ∂ 2 C2 

→ C2  +u − D 2  + C1  − Dy 2  =
0
 ∂t ∂x ∂ x   ∂t ∂ y 
After that, we must to solve two equations
∂C1 ∂C1 ∂ 2C1

+u −D 2 = 0 (A)
∂t ∂x ∂x
and
∂C2 ∂ 2 C2
− Dy =
0 (B)
∂t ∂y
2
i) Case I:
C1o   x − ut   ux   x + ut  
=   + 
    
(A) C1 erfc exp erfc
2   4D t   x
D 
 x   4 Dxt  
  y  ∞  y + 2nw  ∞  − ( y − 2nw )  
(B) C2 = C2 o erfc   + ∑ erfc   + ∑ erfc  
  4 D t   4 D t   4 D t 
=
 y  n 1=
 y  n1  y 
 C   x − ut   ux   x + ut  
=∴ C  1o erfc   + exp   erfc   
 2   4D t 
 x   Dx   4 Dxt  
2-91
    ∞   ∞  − y − 2nw )  
C2 o erfc  y  + ∑ erfc  y + 2nw  + ∑ erfc  (  
 =       
   4 D t
y  n 1=
 4 D y t  n1  4 D y t  
Co   x − ut   ux   x + ut 
= erfc   + exp   
  
erfc
2    x
D 
 4 Dxt   4 Dxt 
  y  ∞  y + 2nw  ∞  − ( y − 2nw )  
erfc   + ∑ erfc   + ∑ erfc  
  4 D t   4 D t   4 D t 
=
  y  n 1=
 y  n1  y 
ii) Case II:
C1o   x − ut   ux   x + ut  
= +
  4 D t     4 D t  
(A) C1 erfc exp erfc
2   x   Dx   x  
∞  y + 2nw 
(B) C2 = ∑ C 2o erfc 
 4 D t


n =−∞
 y 
 C   x − ut   ux   x + ut    ∞  y + 2nw  
=∴ C  1o erfc 
 

+ exp   erfc 
   ∑ C2 o 
 
 

2   4 Dxt   Dx   4 Dxt    n=−∞  4 Dyt  
Co   x − ut   ux   x + ut   ∞  y + 2nw  
=  erfc   + exp   erfc    ∑ erfc  
  
      n=−∞  
2  4 D t  D  4 Dxt 4 D t
x x  y
2-92
Ch 3. Turbulent Diffusion
Chapter 3 Turbulent Diffusion
Contents
3.1 Introduction
3.2 Unified View of Diffusion and Dispersion
3-1
3.1 Introduction
- Mass introduced at a point will spread much faster in turbulent flow than in laminar flow.
- Velocities and pressures measured at a point in the fluid are unsteady and possess a random
component.
Use Navier-Stokes Eq. to explain
turbulent flow
◆ Turbulent flow: Irregularity, randomness ↔ coherent structure
Diffusivity
High Reynolds number
3-D fluctuations ↔ tendency to be isotropic
Dissipation of kinetic energy
Continuum phenomenon
Feature of flow ↔ property of fluid
◆ Scale of turbulence
mean flow → large eddy → small eddy → heat

generation energy dissipation
of turbulence cascade by viscosity
In equilibrium, transfer rate = dissipation rate
3-2
◆ Reynolds experiment
3-3
◆ Kolmogorov's universal equilibrium theory of turbulence
- Behavior of the intermediate scale is governed only by the transfer of energy which,
in turn, is exactly balanced by dissipation at the very small scales.
ε = time rate of energy dissipation per unit mass
 energy 1   F L 1   ML T 1   2 −3 
2 −2
=[ε ] == = M  
LT 
 time Mass   T M   T
µ  ML−1T −1 
ν = kinematic viscosity= =  −3 
=  L2T −1 
ρ  ML 
Kolmogorov scales: Use ε and ν to represent different scales

→ length scale ∝ ε , ν
1 1
 v3  4  L6T −3  4
= =
i) length   2= −3  [ L]
ε  L T 
1 1
v LT 
2 −1 2
[T ]
2
ii) =
time =
  =
2 −3 
ε   L T 
1
v  3 4
ε  1
= = 
1
iii) velocity 1 (=
vε )4 ( L T )( L T =
2 −1
) 
2
 LT −1 
−3 4
 v 2
 
ε 
3-4
For open ocean,
=ε 0.01=
cm 2 / sec3 ; v 0.01cm 2 / s (20°C )
→ dissipation length scale ≈ 0.1cm
time scale ≈ 1sec
velocity scale ≈ 0.1cm/sec
◆ Spreading of a slug of tracer in a high Reynolds number flow
(1) Small scale fluctuations, which are different for each cloud, distort the shape of the cloud
and produce steep concentration differences over short distance.
→ These small scale irregularities are smoothed out by molecular diffusion.
(2) Large scale fluctuations transport the entire cloud.
→ The largest scale of motion is slightly larger than the largest cloud.
(3) Ensemble average = mean over many trials
- average the random motions over a long period time
- average out the effects of the largest eddies
→ The center of mass tends to return to the origin through the process of averaging.
→ Final result is the larger spread in turbulent flow than in laminar flow.
3-5
Small scale
fluctuation
1st cloud
Large scale
fluctuation
2nd cloud
3-6
3.2 Unified View of Diffusion and Dispersion
- Similarities among the various types of diffusion and dispersion are shown.
- Diffusion and dispersion are actually advective transport mechanisms.
3.2.1 Molecular Diffusion
◆ 2-D open-channel flow
um Average velocity
um of molecules
To write the mass balance equation, we need to know how many fluid molecules and how
many tracer molecules pass through and the direction and spread of each molecule.
→ molecular approach → statistical manner
3-7
◆ Continuum approach
- Assume fluid carries tracer through at a rate depending on the concentration, c, and the fluid
velocity, u.
- However, the fluid u, cannot completely represent the tracer movement because the velocity,
u, does not account for the movement of the molecules which have directions and speeds
different from u.
- Molecular diffusion accounts for the difference between the true molecular motion and the
manner chosen to represent the motion.(i.e., by u )
∆u = um − u
Thus, mass flux by this velocity difference is
j = ∆u c
Now, apply Fick' law
- transport called molecular diffusion is proportional to the concentration gradient.
∂c
jm =∆u c ∝
∂x
∂c
jm = − Dm (a)
∂x
Dm = constant of proportionality = molecular diffusivity
Now, consider advection by mean motion
∂c
j= cu − Dm (a)
∂x
x
3-8
Then, substituting (a) into mass conservation equation yields 2-D advection-diffusion
equation as
∂c ∂c ∂2c ∂2c
+ u = Dm 2 + Dm 2 (3.1)
∂t ∂x ∂x ∂y
∂c
① = time rate of change of concentration at a point
By mean ∂t
motion
∂c
② u = advection of tracer with the fluid
∂x By velocity
fluctuation
∂2c ∂2c
③ Dm , Dm = molecular diffusion
∂x 2 ∂y 2
3.2.2 Turbulent Diffusion
3-9
Decompose velocity and concentration into mean and fluctuation
u= u + u '
c= c + c ' (assume only fluctuation in y-direction) (b)
v = v'
u , c = time-averaged values of u and c
1 T
T ∫0
u≡ udt
u=' v=' c=' 0
Where T = averaging time interval
100  102 sec for open channel flow
10−1  100 sec for pipe flow
3-10
For 2-D flow, advection-diffusion equation is
∂c ∂u ∂cvc ∂2c ∂2c

+ + = Dm 2 + Dm 2 (3.2)
∂t ∂x ∂y ∂x ∂y
Substitute (b) into (3.2), then Eq. (3.2) becomes
∂ ( c + c ') ∂ ( u + u ' ) ( c + c ') ∂v '( c + c ') ∂ 2 ( c + c ') ∂ 2 ( c + c ')

+ + = Dm + Dm
∂t ∂x ∂y ∂x 2 ∂y 2
∂c ∂ ∂ 2c ∂ 2c
+ u c = Dm 2 + Dm 2
∂t ∂x ∂x ∂y
∂c ' ∂ ∂ ∂ ∂ ∂
− − (u c ') − (u ' c ) − (u ' c ') − (v ' c ) − (v ' c ')
∂t ∂x ∂x ∂x ∂y ∂y
∂2c ' ∂2c '

+ Dm 2 + Dm 2
∂x ∂y
Integrate (average) w.r.t. time
∂c ∂ ( u c ) ∂2c ∂2c
+ = Dm 2 + Dm 2
∂t ∂x ∂x ∂y
∂c ' ∂ (u c ') ∂ (u ' c ) ∂u ' c ' ∂v ' c ∂v ' c '

− + − − − −
∂t ∂x ∂x ∂x ∂y ∂y
∂2c ' ∂2c '

+ Dm 2 + Dm 2
∂x ∂y
3-11
[Re] Reynolds rules of averages (Schlichting; p460, 371)
f = f
f +g= f +g
f ⋅g = f ⋅g
∂f ∂f
=
∂s ∂s
∫ fds = ∫ fds
Drop all zero terms using Reynolds rules of averages
∂c ∂c ∂ 2c ∂ 2 c ∂ ( −u ' c ') ∂ ( − v ' c ')

+ u = Dm 2 + Dm 2 + +
∂t ∂x ∂x ∂y ∂x ∂y

advective transport
due to u ',v ',and c '
It is assumed and confirmed experimentally that transport associated with the turbulent
fluctuations is proportional to the gradient of average concentration.
∂c ∂c
u'c' ≈ → u ' c ' = −ε x
∂x ∂x
∂c
v ' c ' = −ε y
∂y
3-12
ε x , ε y = turbulent diffusion coefficient
∂ ∂  ∂c 
∂x
( )
−u ' c ' = εx 
∂x  ∂x 
∂ ∂  ∂c 
∂y
( )
−v ' c ' = εy
∂y  ∂y 
Assuming that ε x and ε y are constant, the mass balance equation for turbulent flow is
given as
∂c ∂c ∂2c ∂2c
+u = ( Dm + ε x ) 2 + ( Dm + ε y ) 2 (3.3)
∂t ∂x ∂x ∂y
Dm  ε x , ε y
Drop overbars, and neglect molecular diffusion terms
∂c ∂c ∂2c ∂2c
+ u = εx 2 + εy 2 (3.4)
∂t ∂x ∂x ∂y
For 3-D flow:
∂c ∂c ∂c ∂c ∂ ∂c ∂ ∂c ∂ ∂c
+ u + v + w= (ε x ) + (ε y ) + (ε z ) (3.5)
∂t ∂x ∂y ∂z ∂x ∂x ∂y ∂y ∂z ∂z
∂c ∂c ∂c
☞ Remember εx , εy , εz and are actually advective transport.
∂x ∂y ∂z
3-13
3.2.3 Longitudinal Dispersion
After the tracer is essentially completely mixed laterally, the primary variation of
concentration is in just longitudinal direction.
→ one-dimensional equation
Decompose velocity and concentration into cross-sectional mean and fluctuation
u=
U + u '' u '' =
0 (c)
c=
C + c '' c '' =
0
where U, C = cross-sectional average of the velocity and concentration
After substituting (c) into (3.3), integrating (average) Eq. (3.3) over the cross-sectional area
yields
3-14
3-15
∂ ( C + c '' ) ∂ ( C + c '' ) ∂ 2 ( C + c '') ∂ 2 ( C + c '')

+ (U + u '') = ( Dm + ε x ) + ( Dm + ε y )
∂t ∂x ∂x 2 ∂y 2
By Reynolds rule
∂C
+U
∂C ∂ 2C ∂ 2C
= ( Dm + ε x ) 2 + ( Dm + ε y ) 2 −
( )
∂ u '' c ''
(3.6)
∂t ∂x ∂x ∂y ∂x
∂ 2C
Neglect because after lateral mixing is completed,
∂y 2
∂C
C f ( y)
≈ 0; C =≠
∂y
Then, (3.6) becomes
∂C
+U
∂C ∂ 2C
=( Dm + ε x ) 2 +
(
∂ −u '' c '' )
∂t ∂x ∂x ∂x
''
Taylor (1953, 1954) show that advective transport associated with u is proportional to the
longitudinal gradient of C. Gradient model
∂C
−u '' c '' ∝
∂x
∂C
−u '' c '' =
K
∂x
∂
∂x
(
−u '' c '' =)
∂  ∂C 
K  → longitudinal dispersion
∂x  ∂x 
3-16
K = longitudinal dispersion coefficient
In turbulent uniform flow
∂C ∂C ∂ 2C
+U = ( Dm + ε x + K ) 2
∂t ∂x ∂x
∂C
( Dm + ε x ) << −u '' c ''
∂x
1% 99%
∂C ∂C ∂ 2C
+U =
K 2 (3.7)
∂t ∂x ∂x
→ 1-D Dispersion Equation
''
Because the velocity distribution influences u
→ Lateral diffusion plays a large role in determining the distribution of c , both velocity
''
distribution and lateral diffusion contribute to longitudinal dispersion.
- Limitation proposed by Chatwin (1970)
0.4h 2
t> → initial period
εt
0.4uh 2
x>
εt
3-17
3.2.4 Relative Importance of Dispersion
To investigate the relative importance of dispersion, use dimensionless term as
∂C
K
dispersion rate ∂x K 1=∂C K ∂ (ln C )
=H = =
advective rate UC U C ∂x U ∂x
If H < H c ≈ 0.01 → dispersive transport may be neglected
3.2.5 Conclusion
1) Diffusion
= transport associated with fluctuating components of molecular action and with turbulent
action
= transport in a given direction at a point in the flow due to the differences between the true
advection in that direction and the time average of the advection in that direction
2) Dispersion = transport associated with the variations of the velocity across the flow section
= transport in a given direction due to the difference between the true advection in that
direction and the spatial average of the advection in that direction
3-18
THE DIFFUSION AND DISPERSION SPECTRUM
10−6 -
10−5 - Molecular Diffusion in Water
10−4 -
10−3 -
10−2 -
10−1 - Molecular Diffusion Gases
Dij , cm 2 / sec 1 -
Eddy Diffusion - Pipes
10 -
100 - Eddy Diffusion - Streams
10000 -
104 - Longitudinal Dispersion - Streams
105 -
106 - Longitudinal Tidal Dispersion - Estuaries
107 -
108 -
109 -
1010 - Eddy Diffusion - Atmosphere
3-19
Ch 4. Shear Flow Dispersion
Chapter 4 Shear Flow Dispersion
Contents
4.1 Dispersion in Laminar Shear Flow
4.2 Dispersion in Turbulent Shear Flow
4.3 Dispersion in Unsteady Shear Flow
4.4 Dispersion in Two Dimensions
Taylor, Geoffrey – English

fluid mechanician
Objectives:
1) Derive shear flow dispersion equation using Taylor’ analysis (1953, 1954)
- laminar flow in pipe (1953)
- turbulent flow (1954)
→ apply Fickian model to dispersion
→ reasonably accurate estimate of the rate of longitudinal dispersion in rivers and estuaries
2) Extend dispersion analysis to unsteady flow and two-dimensional flow
4-1
4.1 Dispersion in Laminar Shear Flow
4.1.1 Introductory Remarks
Shear effect
∙ Taylor's analysis (1953) in laminar flow in pipe
Consider laminar flow in pipe with velocity profile shown below.
Assume two molecules are being carried in the flow; one in the center and one near the wall.
u0
1) Rate of separation caused by the difference in advective velocity
≫ separation by molecular motion
2) Because of molecular diffusion, given enough time, each molecule would wander
randomly throughout the cross section.
3) If a long enough averaging time was available, a single molecule’s time-averaged velocity
would be equal to the instantaneous cross-sectional average of all molecules’ velocities.
4-2
4) If we adopt a coordinate system moving at the mean velocity, the random steps are likely
to be back and forward.
→ motion of single molecule is the sum of a series of independent steps of random length.
5) Fickian diffusion equation, Eq. (2.4) can describe the spread of particles along the axis of
the pipes, except that since the step length and time increment are much different from those
of molecular diffusion. We expect to find a different value of diffusion coefficient.
→ dispersion coefficient
Now, find the rate of spreading for laminar shear flow in pipe
For turbulent flow, diffusion coefficient is given as
ε=
< U 2 > TL
where U = velocity deviation

Mean square velocity deviation of
the molecule results from the
TL = Lagrangian time scale wandering of the molecule across
the cross section.
Molecule samples velocities
ranging from zero at the wall to the
peak velocity u0 at the centerline.
For laminar flow in pipe; < U 2 >∝ u0 2

∼ time required to sample
the whole field of velocities
2
a ∼ time scale for cross-
TL ∝
D sectional mixing
4-3
where u0 = maximum velocity at the centerline of pipe
a = radius of pipe
D = molecular diffusion coefficient
Thus, longitudinal dispersion coefficient due to combined action of shear advection and
molecular diffusion is given as
a2
K=
< U > TL ∝ u0
2 2
(4.1)
D
→ K is inversely proportional
to molecular diffusion.
4-4
4.1.2 A Generalized Introduction
(a) example velocity distribution (b) transformed coordinate system moving at the
mean velocity
Consider the 2-D laminar flow with velocity variation u(y) between walls
Define the cross-sectional mean velocity as
1 h
h ∫0
u= udy (4.2)
Then, velocity deviation is
u' u ( y ) − u
= (4.3)
Let flow carry a solute with concentration C(x, y) and molecular diffusion coefficient D.
Define the mean concentration at any cross section as
1 h
C f ( x) ≠ f ( y )
h ∫0
=
C Cdy, = (4.4)
4-5
Then, concentration deviation is
C ' =C ( y ) − C , C ' =C ' ( x, y ) (4.4a)
Now, use 2-D diffusion equation with only flow in x-direction (v =0)
∂C ∂C ∂C ∂ 2C ∂ 2C
+u +v = D 2 +D 2 (1)
∂t ∂x ∂y ∂x ∂y
∂C
=0
∂y
Substitute (4.2)~(4.4) into (1)
∂ ∂  ∂2 ∂2 
(C + C ' ) + ( u + u ' ) (C +
= C ' ) D  2 (C + C ' ) + 2 (C + C ' )  (4.5)
∂t ∂x  ∂x ∂y 
Now, simplify (4.5) by a transformation of coordinate system whose origin moves at the
mean flow velocity
∂ξ ∂ξ
ξ=
x − ut → =
1 =
−u
∂x ∂t
∂τ ∂τ
τ t
= → = 0 = 1
∂x ∂t
Chain rule
∂ ∂ξ ∂ ∂τ ∂ ∂
= + = (b)
∂x ∂x ∂ξ ∂x ∂τ ∂ξ
4-6
∂ ∂ξ ∂ ∂τ ∂ ∂ ∂
= + =
−u + (c)
∂t ∂t ∂ξ ∂t ∂τ ∂ξ ∂τ
Substitute Eq. (b)-(c) into Eq. (4.5)
∂ ∂ ∂  ∂2 ∂ 2C ' 
−u (C + C ) +
'
(C + C ) + ( u + u ) (C =
' '
+ C ) D  2 (C + C ) + 2 
' '
∂ξ ∂τ ∂ξ  ∂ξ ∂y 
∂ ∂  ∂2 ∂ 2C ' 
(C + C ' ) + u ' (C =
+ C ' ) D  2 (C + C ' ) + 2  (4.8)
∂τ ∂ξ  ∂ξ ∂y 
→ view the flow as an observer moving at the mean velocity
→ u is only observable velocity

'
Now, neglect longitudinal diffusion because rate of spreading along the flow direction due to
velocity difference greatly exceed that due to molecular diffusion.
∂ ∂2
u'
(C + C )  D 2 (C + C ' )
'
∂ξ ∂ξ
∂C ∂C ' ∂C ∂C ' ∂ 2C '

+ + u' + u' =
D 2 (4.9)
∂τ ∂τ ∂ξ ∂ξ ∂y
→ This equation is still intractable because u varies with y.

'
→ General solution cannot be found because a general procedure for dealing with differential
equations with variable coefficients is not available.
4-7
Now introduce Taylor's assumption
→ discard three terms to leave the easily solvable equation for C ( y )

'
∂C ∂ 2C '
u '
=D 2 (4.10)
∂ξ ∂y
[Re] Derivation of Eq. (4.10) using order of magnitude analysis
Take average over the cross section of Eq. (4.9)
1 h
h ∫0
→ apply the operator ( )dy
∂C ∂C ' ∂C ∂C ' ∂ 2C '

+ + u' + u' =
D 2
∂τ ∂τ ∂ξ ∂ξ ∂y
Apply Reynolds rule of average
∂C ' ∂C
'
+u =
0 (4.11)
∂τ ∂ξ
Subtract Eq.(4.11) from Eq.(4.9)
∂C ' ' ∂C ' ∂C

'
' ∂C
'
∂ 2C '
+u +u −u =
D 2
∂τ ∂ξ ∂ξ ∂ξ ∂y
Assume C , C are well behaved, slowly varying functions and C >> C

' '
∂C ∂C ' ' ∂C '

Then u' >> u ' ,u
∂ξ ∂ξ ∂ξ
4-8
∂C ' ' ∂C '

'
Thus we can drop u ,u
∂ξ ∂ξ
∂C ' ∂ 2C ' ∂C
= D 2 − u' (d)
∂τ ∂y ∂ξ
∂C
−u ' = source term of variable strength
∂ξ
→ Net addition by source term is zero because the average of u is zero.

'
∂C
Assume that remains constant for a long time, so that the source is constant.
∂ξ
Then, Eq. (a) can be assumed as steady state.
∂C '
→ =0
∂τ
Then (a) becomes
∂C ∂ 2C '
longitudinal u '
=D 2 Cross-sectional
advective ∂ξ ∂y diffusive transport
transport
(A) (B)
→ same as Eq. (4.10)
→ cross sectional concentration profile C ( y ) is established by a balance between

'
longitudinal advective transport and cross sectional diffusive transport.
4-9
<Fig. 4.3> The balance of advective flux versus diffusive flux
In balance, net transport = 0
 ' ∂ '   ∂C '  ∂C ' ∂  ∂C '   

u Cdy − u Cdy + ( u C ) dxdy  +  − D
'
dx −  − D dx +  − D  dydx  = 0
 ∂x   ∂y  ∂y ∂y  ∂y   
∂ ' ∂  ∂C ' 
− ( u C ) dxdy +  D  dydx =
0
∂x ∂y  ∂y 
∂ ' ∂  ∂C ' 
∂x
( ) ∂y  D ∂y 
u C =
 
Now, let’s find solution of Eq. (4.10)
∂ 2C ' 1 ∂C ' 1 ∂C '

= = u u (e)
∂y 2 D ∂ξ D ∂x
Integrate (e) twice w.r.t. y
1 ∂C y y '
=C ' ( y) ∫ ∫ u dydy + C ' ( 0 ) (4.14)
D ∂x 0 0
4-10
Consider mass transport in the streamwise direction
 ' '  ∂C '  

M
h h
= ∫
0
q=
x dy ∫
0
u C +  − D
 
  dy
∂x  
(f)
Substitute (4.14) in (f)
1 ∂C h ' y y '
M
h
= ∫=
0
u 'C 'dy
D ∂x ∫0 ∫0 ∫0
u u dydydy (4.15)
∫ u {C ( 0 )} dy = 0
h
' '
since
0
constant
→ Eq. (4.15) means that total mass transport in the streamwise direction is proportional to
the concentration gradient in that direction.
∂C
M ∝ (g)
∂x
→This is exactly the same result that we found for molecular diffusion.
∂C
q = −D
∂x
But this is diffusion due to whole field of flow.
4-11
Let q = rate of mass transport per unit area per unit time
Then, (g) becomes
M ∂C
q= = −K (h)
h ×1 ∂x
where h = depth = area per unit width of flow
K = longitudinal dispersion coefficient (= bulk transport coefficient)
→ express as the diffusive property of the velocity distribution (shear flow)
Then, (h) becomes
∂C
M = −hK (4.16)
∂x
Comparing Eq. (4.15) and Eq. (4.16) we see that
1 h ' y y '
hD ∫0 ∫0 ∫0
K= − u u dydydy (4.17)
1
K∝
D
Now, we can express this transport process due to velocity distribution as a one-dimensional
Fickian-type diffusion equation in moving coordinate system.
∂C ∂ 2C
=K 2 (4.18)
∂τ ∂ξ
4-12
Return to fixed coordinate system
∂C ∂C ∂ 2C
+u =
K 2 (4.19)
∂t ∂x ∂x
→ 1-D advection-dispersion equation
C , u = cross-sectional average values
▪ Balance of advection and diffusion in Eq. (4.10)
Suppose that at some initial time t = 0 a line source of tracer is deposited in the flow.
→ Initially the line source is advected and distorted by the velocity profile (Fig. 4.4).
At the same time the distorted source begins to diffuse across the cross section.
→ Shortly we see a smeared cloud with trailing stringers along the boundaries (Fig. 4.4b).
During this period, advection and diffusion are by no means in balance.
→ Cross-sectional average concentration is skewed distribution (Fig. 4.4c).
4-13
→ Taylor’s assumption does not apply.
If we wait much longer time, the cloud of tracer extends over a long distance in the x
direction.
∂C
→ C varies slowly along the channel, and is essentially constant over a long period
∂x
of time.
'
→ C becomes small because cross-sectional diffusion evens out cross-sectional
concentration gradient.
Chatwin (1970) suggested
h2
i) Initial period: t < 0.4
D
→ advection > diffusion
h2
ii) Taylor period: t > 0.4
D
→ advection ≈ diffusion
→ can use Eq. (4.19) → The initial skew degenerates into the normal distribution
∂σ 2
= 2K
∂t
4-14
4-15
4-16
4-17
4-18
4.1.3 A Simple Example
Consider laminar flow between two plates → Couette flow
0.042
Fig. 4.5 Velocity profile and the resulting concentration profile
y
u( y) = U
h
1 h2 y
h ∫− 2 h
=u = hU dy 0
∴u ' =
u
h2
Suppose t > → tracer is well distributed
D
→ Taylor’s analysis can be applied
From Eq. (4.14)
1 ∂C y y '
=C' ( y ) ∫ ∫ u dydy + C ' (0)
D ∂x 0 0
1 ∂C y y Uy h
= ∫ h∫ h
D ∂x − 2 − 2 h
dydy + C ' (− )
2
(a)
4-19
1 ∂C y  U 2 
y
h
D ∂x ∫− 2  2h 
= h y dy + C ' (− )
−
h 2
2
1 ∂C y Uy 2 Uh  h
= ∫ h
D ∂x − 2  2h
− 
8 
dy + C ' (− )
2
y
1 ∂C Uy 3 Uh  h
= − + C ' (− )
D ∂x  6h 8 
y
−
h 2
2
1 ∂C Uy 3 Uh Uh 2 Uh 2  h
= − + − + '
−
D ∂x  6h 16 
y C ( )
8 48 2
1 ∂C U  y 3 h 2 h3   h
=  − y −  + C'  − 
D ∂x 2h  3 4 12   2
=
By symmetry C 0= '
@ y 0
1 ∂C U  h3  ' h
= − +  − 
D ∂x 2h  12 
0 C
 2
 h  1 ∂C Uh
2
C −  =
'
 2  D ∂x 24
1 ∂C U  y 3 h 2 
∴= '
−
D ∂x 2h  3
C ( y) y (4.21)
4 
h ' 1 ∂C  h 2 
→= = U −
D ∂x  24 
@y ;C
2
C'D 1
→ =
− =
−0.042
∂C 2 24
Uh
∂x
4-20
From (a):
Dispersion coefficient K
DC ' ( y )  '  h 
=( A) ∂C   C ( y ) − C '  − 
h  2 
1 2 y y
∂x
hD ∫− 2 ∫2 ∫2
K= − h u ' h h u ' dydydy

( A)
1 2 ' D  '  h 
h
=
− ∫ hu
hD − 2 ∂C  C ( y ) − C '  −   dy
 2 
∂x
1  h2 ' ' ' h  h2 ' 

∂C  ∫− h2
=
− u C dy +C  −  ∫ h u dy 
h  2  −2 
∂x
1 h
Uy  1 ∂C U  y 3 h 2  
∂C ∫− h2 h  D ∂x 2h  3
=
− 2
( ) − y  dy
4 
h
∂x
U 2 h2  y 4 h 2 y 2 
− 3 ∫ h −
=  dy
2h D − 2  3 4 
h
U y 2
h y  5 2 3 2
=
− 3  − 
2h D  15 12  − h
2
U 2 h2
=
120 D
1
Note that K ∝
D
→ Larger lateral mixing coefficient makes C to be decreased.

'
4-21
4.1.4 Taylor's Analysis of Laminar Flow in a Tube
Consider axial symmetrical flow in a tube → Poiseuille flow
Tracer is well distributed over the cross section.
 r2 
( r ) u0 1 − 2  → paraboloid
u= (a)
 a 
Integrate u to obtain mean velocity
dQ ≅ u 2π rdr
4-22
a   r 2  
=∴Q ∫0
2π r u0 1 − 2  dr
  a  
r r2   r  1 1
= 2π u0 a ∫ 1 −= π ∫ z (1 − z 2 )dz
2 2
 d   2 u 0 a
 a  a
0 a 2 0
1
 z2 z2  π 2
= 2π u0 a  −  = a u0 2
2 4 0 2
By the way, Q= u ⋅ π a
2
u
∴ u =0
2
2-D advection-dispersion equation in cylindrical coordinate is
∂C  r 2  ∂C  ∂ 2C 1 ∂C ∂ 2C 
+ u0  1 − 2  = D  2 + +  (b)
∂t  a  ∂x  ∂r r ∂r ∂x 2 
u0
Shift to a coordinate system moving at velocity
2
∂C ∂ 2C
Neglect and as before
∂t ∂x 2
r
Let z =, ξ =
x − ut ,τ =
t
a
Decompose C, then (b) becomes
u0 a 2 1 2 ∂C ∂ 2C ' 1 ∂C '
( −z ) = 2 +
D 2 ∂ξ ∂z z ∂z
4-23
∂C '
= 0 at z =1
∂z
Integrate twice w.r.t. z
u0 a 2  2 1 4  ∂C
C =
'
z − z  + const (c)
8D  2  ∂x
M 1
K=
−
∂C
=
− ∫
∂C A
u 'C 'dA (d)
A A
∂x ∂x
Substitute (a), (c) into (d), and then perform integration
a 2 u0 2
K=
192 D
[Example] Salt in water flowing in a tube
D = 10−5 cm 2 / sec
u0 = 1 cm / sec
a = 2mm
ud (0.01) ( 0.004 )
R= = = 40 << 2000 → laminar flow
1 × 10−6
e
v
a 2 u0 2 ( 0.2 ) (1)
2 2
=K = = 21cm 2 / sec ≈ 106 D

192 D 192 (10 ) −5
4-24
☞ Initial period
a 2 0.4 ( 0.2 )
2
= =
t0 0.4 = = 27 min
1600sec
D ( )
10 −5
u0
=
x0 ut=
0 t0
2
= (=
0.5 )(1600 ) 800cm
800
= = 4000a
0.2
x > x0 → 1-D dispersion model can be applied
4-25
Homework Assignment No. 4-1
A hypothetical river is 30 m wide and consists of three "lanes", each 10m in width.
The two outside lanes move at 0.2 m/sec and the middle lane at 0.4m/sec. Every tm seconds
complete mixing across the cross section of the river (but not longitudinally) occurs. An
instantaneous injection of a conservative tracer results in a uniform of 100mg/ℓ in the water 2
m upstream and downstream of the injection point. The concentration is initially zero
elsewhere. As the tracer is carried downstream and is mixed across the cross-section of the
stream, it also becomes mixed longitudinally, due to the velocity difference between lanes,
even though there is no longitudinal diffusion within lanes. We call this type of mixing
"dispersion".
1) Mathematically simulate the tracer concentration profile (concentration vs.
longitudinal distance) as a function of time for several (at least four) values of tm
including 10 sec.
4-26
2) Compare the profiles and decide whether you think the effective longitudinal
mixing increases or decrease as tm increases.
This "scenario" represents the one-dimensional unsteady-state advection and
longitudinal dispersion of an instantaneous impulse of tracer for which the concentration
profile follow the Gaussian plume equation
M  ( x − Ut )2 
=C exp − 
4π Kt  4 Kt 
in which x = distance downstream of the injection point, M = mass injected width of the
stream, K = longitudinal dispersion coefficient, U = bulk velocity of the stream
(flowrate/cross-sectional area), t = elapsed time since injection.
3) Using your best guess of a value for U, find a best-fit value for K for each and for
which you calculated a concentration profile. Tabulate of plot the effective K as a
function tm of and make a guess of what you think the functional form is.
4-27
◆ Dispersion mechanism in a hypothetical river
1) 3 lanes of different velocities
2) Every tm seconds complete mixing occurs across the cross section of the river (but not
longitudinally) occurs, after shear advection is completed.
→ sequential mixing model
∂  ∂C 
εx →0
∂x  ∂x 
W2
tm ≅
εy
3) Instantaneous injection
tm = 10 s; ua =0.2 m/s; ∆x =2 m
4-28
ii) t=tm -
t=tm +: After lateral mixing
0 67 100 33 0
0 67 100 33 0
0 67 100 33 0
(iii) t= 2 tm -: After shear advection t= 2 tm +: After lateral mixing
0 0 67 100 33 0 0 0 45 89 55 11 0
0 0 0 67 100 33 0 0 45 89 55 11 0
0 0 67 100 33 0 0 0 45 89 55 11 0
4-29
Initial period Taylor period
4-30
[Re] Longitudinal Dispersion in 2-lane river
u + ∆u α= Area fraction of river

Fast 1−α
occupied by slow lane
u
Slow α 0 ≤α ≤1
uS = u
u F = u + ∆u
u = cross-sectional mean velocity
= α u + (1 − α )( u + ∆u )
Consider deviations:
uS′ = uS − u = u − α u − (1 − α )( u + ∆u )
= u − α u − u − ∆u + α u + α∆u = − (1 − α ) ∆u
u′F = u F − u = u + ∆u − u = u + ∆u − α u − (1 − α )( u + ∆u )
= α∆u
(i) Before any processes
F Cu Cd 1−α
S Cu Cd α
∆x ∆x
∆x =∆u ⋅ tm
4-31
(ii) Just before mixing (JBM) …. after advection only
F 0 Cu Cd 1−α
S Cu Cd
0 α
C= α Cd + (1 − α ) Cu
CS′ = Cd − C = Cd − α Cd − (1 − α ) Cu
(1 − α ) ( Cd − Cu )
=
CF′ = Cu − C = Cu − α Cd − (1 − α ) Cu
−α ( Cd − Cu )
=
(iii) Just after mixing (JAM)
F Cu2 Cd 2
S Cu2 Cd 2
∆x
C = Cd 2
CS′ = 0
CF′ = 0
4-32
1
A ∫A
u′C ′ = u′C ′ d A
1
2
u′C ′ ≅
u′C ′
JBM
{( )
+ u′C ′
JAM
( ) }
= {α ( u′C ′ ) S + (1 − α )( u′C ′ ) F }
1
2
=
1
{ }
α  − (1 − α ) ∆u  (1 − α ) ( Cd − Cu )  + (1 − α ) [α∆u ] ( −α ) ( Cd − Cu ) 
2 
= (α 2 − α ) ∆u ( Cd − Cu )
1
2
∂C Cd − Cu
≈
∂x ∆u tm
u′C ′ 2 (α − α ) ∆u ( Cd − Cu )
1 2
K=
− =
∂C ( Cd − Cu )
∂x ∆u tm
=
1
2
( α − α 2 ) ( ∆u ) tm
2
<Example>
2
α = ; ∆u =0.2 ; tm = 10 sec
3
1 2  2  
2
K=  −   ( 0.2 ) tm =
2
0.0044 tm
2  3  3  
tm = 5 10 20 30
K = 0.0222 0.0444 0.0889 0.1333
4-33
4.1.5 Aris's Analysis
Aris (1956) proposed the concentration moment method in which he obtain Taylor’s main
results without stipulating the feature of the concentration distribution.
Begin with 2-D advective-diffusion equation in the moving coordinate system to analyze the
flow between two plates (Fig. 4.5)
∂C ' ∂C  ∂ 2C ∂ 2C 
+u = D 2 + 2  (4.29)
∂τ ∂ξ  ∂ξ ∂y 
(1) (2) (3) (4)
Now, define the pth moments of the concentration distribution
∞
CP ( y ) = ∫ ξ p C (ξ , y ) d ξ (4.30)
−∞
Define cross-sectional average of pth moment
1
A ∫A
M=
p C=
P CP ( y )dA
∞
Take the moment of Eq. (4.29) by applying the operator ∫ ξ P ( ) dξ
−∞
∞
p ∂C ∂ ∞ p ∂C p
=(1) ∫−∞ ∂τ
= ξ d ξ
∂τ ∫−∞
= ξ Cd ξ
∂τ
← Leibnitz rule
[Re] Leibnitz formula
u1 ∂f d u1
∫u0 ∂α
dx =
dα ∫u0
fdx
4-34
u v’
p ' ∂C p ∂C
∞ ∞
= ∫−∞ ∂ξ
= ξ ξ ∫−∞ ∂ξ dξ
ξ ← integral by parts
'
( 2) u d u
C ξ = ±∞ = 0
{
= u ' ξ p C  − ∫ pξ p−1C d ξ
−∞ −∞
∞ ∞
}
∞
− pu ' ∫ ξ p −1Cdξ =
= − pu 'C p −1
−∞
u v’
∂ 2C ∞ ∂  ∂C 
∞
=( 3) ∫= ξ D 2 dξ D ∫ ξ p  dξ ← integral by parts
p
−∞ ∂ξ −∞ ∂ξ  ∂ξ 
  p ∂C  ∞ ∞ ∂C 
= D  ξ  − ∫−∞ ∂ξ pξ d ξ 
p −1
  ∂ξ  −∞ 
∞ ∂C
= − Dp ∫ ξ p −1 dξ
−∞ ∂ξ
{ ∞ ∞
− Dp ξ p −1C  − ∫ C ( p − 1)ξ p −2 d ξ
=
−∞ −∞ }
∞
Dp ( p − 1) ∫ ξ p −2Cdξ =
= Dp ( p − 1)C p −2
−∞
∂ 2C ∂2 ∞ p∞ ∂ 2C P
( 4) ∫=
∂y 2 ∫−∞
= ξ D 2 d ξ D= ξ Cdξ D 2
p
−∞ ∂y ∂y
Therefore Eq. (4.29) becomes
∂C p  ∂ 2C P 
− pu C p−1 = D  p( p − 1)C p−2 +
'
 (4.33)
∂τ  ∂y 2 
4-35
B.C. gives
∂CP
D= 0=
at y 0, h ← impermeable boundary
∂y
Take cross-sectional average of Eq. (4.33) ∂ 2CP ∂ 2CP ∂  ∂CP 

= = = 0
∂C p  ∂ 2CP  ∂y 2 ∂y 2 ∂y  ∂y 
− pu C p −1 = D  p ( p − 1)C p −2 +
'

∂τ  ∂y 2 
dM p
− pu 'C p −1 =
p ( p − 1) DM p −2 (4.34)
dτ
Eq. (4.34) can be solved sequentially for p = 0, 1, 2, …
Equation Consequences as t → ∞
p=0 dM 0 / dτ = 0 Mass is conserved
1 1 ∞
A ∫A A ∫A ∫−∞
M0 C0 ( y ) dA = Cd ξ dA
∂C0 ∂ 2 C0
(4.33) → =D 2
∂τ ∂y
dM 1
p =1 = u 'C0 M 1 → consant
dt
∂C1 ∂ 2C1
(4.33) → − u C0 =
'
D 2
∂τ ∂y
dM 2 dσ 2
p=2 = 2u 'C1 + 2 DC0 = 2K + 2D
dt dt
→ molecular diffusion and shear flow dispersion are additive
Aris’ analysis is more general than Taylor’s analysis in that it applies for low values of time.
4-36
4.2 Dispersion in Turbulent Shear Flow
4.2.1 Extension of Taylor’s analysis to turbulent flow
Cross-sectional velocity profile in turbulent motion in the channel is different than in a
laminar flow.
Consider unidirectional turbulent flow between parallel plates
Begin with 2-D turbulent diffusion equation
∂C ∂C ∂C ∂  ∂C  ∂  ∂C 
+u + v= εx + εy
∂y ∂x  ∂x  ∂y  ∂y 
(a)
∂t ∂x
Here, the cross-sectional mixing coefficient ε ( y ) is function of cross-sectional position.
1 T
T ∫0
C , u , v = time mean values; C= C= cdt
Let v = 0 , turbulent fluctuation v ≠ 0

'
4-37
∂ ∂C ∂C
Assume εx << u
∂x ∂x ∂x
Then (a) becomes
∂C ∂C ∂  ∂C 
+u = εy
∂x ∂y  ∂y 
(b)
∂t
Now, decompose C and u into cross-sectional mean and deviation
∂ (C + C ' ) ∂ ∂ ∂
+ ( u + u ' ) (C=+ C') ε y (C + C ' ) (c)
∂t ∂x ∂y ∂y
Transform coordinate system into moving coordinate according to u
∂C ∂C ' ∂C ∂C ' ∂ ∂C '

+ + u' + u' = εy
∂τ ∂τ ∂ξ ∂ξ ∂y ∂y
Now, introduce Taylor's assumptions (discard three terms)
∂C ∂ ∂C '
u'
= εy (4.35)
∂ξ ∂y ∂y
Solution of Eq. (4.35) can be derived by integrating twice w.r.t. y
∂C y 1 y '
=C' ∫
∂ξ ε y
0 ∫0
u dydy + C ' (0)
4-38
Mass transport in streamwise direction is
∂C h ' y 1 y '
M
h
= ∫=
u 'C 'dy
0 ∂ξ ∫0 ∫0 ε y ∫0
u u dydydy
M ∂C
q= = −K
h ∂ξ
1 h ' y 1 y '
h ∫0 ∫0 ε y ∫0
K= − u u dydydy (4.36)
4-39
4.2.2 Taylor's analysis of turbulent flow in pipe (1954)
r dz 1
Set z = → =
a dr a
Then, velocity profile is
u ( z=
) u0 − u * f ( z ) (a)
τ0
in which u = shear velocity =
*
f(z) = logarithmic function
[Re] velocity defect law [Eq. (1.27)]
3 u* 2.30 * ζ
u =+
u + u log10
2κ κ a
in which κ = von Karman's constant ≈ 0.4

ς = distance from the wall
ζ
u=
u + 3.75u * + 5.75u * log10
a
4-40
u −u ζ
=*
3.75 + 2.5ln
u a
The cross-sectional mixing coefficient can be obtained from Reynolds analogy.
→ The mixing coefficients for momentum and mass transports are the same.
i) momentum flux through a surface
τ ∂u
= −ε ☜ Daily & Harleman (p. 56)
ρ ∂r
Kinematic eddy
viscosity
ii) mass flux - Fickian behavior
∂C
q = −ε
∂r
q τ
ε
∴= = (b)
∂C ∂u
− −ρ
∂r ∂r
For turbulent flow in pipe, shear stress is given
r
τ τ=
= 0 zτ 0 (c)
a
Differentiate (a) w.r.t. r
∂u df ( z ) dz df 1
=
−u * =
−u * (d)
∂r dz dr dz a
4-41
Divide (c) by (d)
τ zτ 0
= (e)
∂u df 1
−u *
∂r dz a
Substitute (e) into (b) (u )

* 2
τ zτ 0 az (τ 0 / ρ ) azu*
∴ε =− = = =
∂u * df 1 * df df
ρ ρu u
∂r dz a dz dz
Now, it is possible to tabulate u =

( r ) u( r ) − u ,
'
ε (r) (f)
And, numerically integrate Eq. (4.39) [Taylor’s equation in radial coordinates] to obtain
C ' ( r ) using ε ( r ) obtained in (f)
∂C  ∂ 2C ' 1 ∂C ' 
=
u' ε 2 +
r ∂r 
(4.39)
∂ξ  ∂r
Again, numerically integrate Eq. (4.36) to find K
K = 10.1au * (4.40)
au0 2
in which a = pipe radius u ∝
*
a 2 u0 2
[Cf] For laminar flow in a tube, K =
192 D
4-42
4.2.3 Elder's application of Taylor's method (1959)
Consider turbulent flow down an infinitely wide inclined plane assuming von Karman
logarithmic velocity profile
u*
u=
( y)'
(1 + ln y ' ) (a)
κ
du u * 1 1
where u =u −u → =
'
(b)
dy κ y ' d
y' = y / d
du
=0
d = depth of channel dy
For open channel flow, shear stress is given

Parabolic
du
τ ρε = τ 0 (1 − y ' )
= profile (c)
dy
τ 0 (1 − y ' ) τ 0 (1 − y ' )
ε=
( y) = = κ y '(1 − y ' )du* (d)
ρ du ρ u 1 1
*
dy κ y' d
Substitute Eq. (a) and Eq. (d) into Eq. (4.36) and integrate
∂C d ∞ 1 d − y n
=C' (∑ ( ) − 0.648) (4.44)
∂x κ 2 n=1 n 2 d
0.404
K= du * (4.45)
κ 3
4-43
Input κ = 0.41
K = 5.93du * (4.46)
4-44
▪General form for the longitudinal dispersion coefficient
Introduce dimensionless quantities
y
y' = → y = hy ' , dy = hdy ' (a)
h
u'
u=
''
→ u=' u '' u '2 (b)
2
u'
ε
ε' = → ε = ε 'E (c)
E
Where E = cross-sectional average of ε
u ' = velocity deviation from cross-sectional mean velocity
1
1 h  2
u '2 =  ∫ (u ' ) 2 dy 
h 0 
= intensity of the velocity deviation (different from turbulent intensity)
= measure of how much the turbulent averaged velocity deviates throughout the cross section
from its cross-sectional mean
Substitute (a) ~ (c) into Eq. (4.36)
1 1 '' '2 y ' 1 y ' '' '2 3 ' ' '

h ∫0
K= − u u ∫ ' ∫ u u h dy dy dy
0 ε E 0
1 '2 1 '2 3 1 '' y ' 1 y ' '' ' ' '

= − u u h ∫ u ∫ ' ∫ u dy dy dy
h E 0 0 ε 0
u '2 h 2  1 '' y ' 1 y ' '' ' ' ' 

E  ∫0 ∫0 ε ' ∫0
= − u u dy dy dy  (d)

4-45
1 y' 1 y'
Set I = − ∫ u '' ∫ ∫ u ''dy 'dy 'dy ' (4.48)
0 0 ε ' 0
Then (d) becomes
h 2 u '2
K= I (4.47)
E
▪Range of values of I for flows of practical interest
I = 0.054 ~ 0.10 → I ≅ 0.10
Charac.
Flow Velocity profile I K
length, h
r2 a 2 u0 2
(i)Laminar flow in a tube =
u u0 (1 − 2 ) a 0.0625
a 192 D
(ii)Laminar flow at depth   y  y2  8 d 2 u0 2

=u u0  2   − 2  d 0.0952
down on inclined plane  d  d  945 D
(iii)Laminar flow with a linear

y U 2 h2
velocity profile across a u =U h 0.10
h 120 D
spacing
(iv)Turbulent flow in a pipe empirical a 0.054 10.1 au *
(v)Turbulent flow at depth

u* y
u =+
u (1 + ln ) d 0.067 5.93du*
down an inclined plane κ d
4-46
4.3 Dispersion in Unsteady Shear Flow
Real environmental flows are often unsteady flow.
- reversing flow in a tidal estuary; wind driven flow in a lake caused by a passing storm
∙ unsteady flow = steady component + oscillatory component
Application of Taylor's analysis to an oscillatory shear flow
(i) Linear velocity profile with a sinusoidal oscillation
y  2π t 
u =U sin   (4.49)
h  T 
where T = period of oscillation
▪'flip-flop' sort of flow
4-47
y y
- reversing instantaneously between u = U and −u =U after every time interval
h h
T
2
→ after each reversal the concentration profile has to be reversed
→ substitute – y for y in Eq. (4.21)
→ but enough time bigger than mixing time ( Tc ≈ h / D ) is required before the
2
concentration profile is completely adopted to a new velocity profile.
(1) T >> TC
- concentration profile will have sufficient time to adopt itself to the velocity profile in each
direction
- time required for to reach the profile given by Eq.(4.21) is short compared to the time
during which has that profile.
→ dispersion coefficient will be the same as that in a steady flow
→ dispersion as if flow were steady in either direction
(2) T << TC
- period of reversal is very short compared to the cross-sectional mixing time
- concentration profile does not have time to respond to the velocity profile
' '
- C will oscillate around the mean of the symmetric limiting profiles, which is C =0.
→ dispersion coefficient tends toward zero
→ no dispersion due to the velocity profile
4-48
4-49
▪ Fate of an instantaneous line source when T << TC
Solution of Eq. (4.13) by Carslaw and Jaeger (1959)
unsteady
∂C ∂C'
∂C
2 '
source term
−D 2 =
−u '
∂τ ∂y ∂ξ
Taylor ’s
y 2π t
equation for u= u=' U sin ( u= 0)
unsteady flow h T
∂C ' h
B.C. = 0 at y = ±
∂y 2
I.C. C '( y,0) = 0
y
−u =U
h
4-50
∂C
by a source of constant strength by setting t = t0
'
Replace unsteady source term u
∂ξ
∂C * ∂ 2C * y ∂C 2π t0
−D 2 = −U sin( )
∂τ ∂y h ∂x T
∂C * h
= 0 at y = ±
∂y 2
C * ( y, 0) = 0
*
where C = distribution resulting from a suddenly imposed source distribution of constant
strength
As diagrammed in Fig. 2.8, the solution for a series of sources of variable strength, can be
obtained by
∂ *
t
=
C ' ( y, t ) ∫ 0 ∂t
C ( y, t − t0 ; t0 )d 0t
For large t
t∂ *
=
C ' ( y, t ) ∫ −∞ ∂t
C ( y, t − t0 ; t0 )d 0t
C * can be expressed by the sum
C * ( y=
, t ) u ( y ) + w( y , t )
w( y , t ) can be solved by separation of variables and Fourier expansion.
4-51
Further integration of the result leads to
2Uh 2 T ∂C ∞ (−1) n y
C '
∑
π D Tc ∂x n=1 ( 2n − 1)
3 2
sin( 2n − 1)π
h
1
−
 π  T  2
 2π t
2

×  ( 2n − 1)  + 1 sin  + θ 2 n −1 
 2  Tc   T 
 − 
1
 2
 2 
−1    1 2 T 
where θ 2 n −1 =sin  −   π ( 2n − 1)  + 1 
   2 Tc   
 
Average over the period of oscillation of K
1 T 2 ' ' ∂C 
h
T ∫0  ∫− 2
=K  − u C dy / h  dt
∂x 
h
−1
 2 2 2
 
U h T  ∞ 2
−2  π
2
2 T  
=   ∑ ( 2n − 1)   (2n − 1)    + 1
π D  Tc  n=1
4
  2  Tc   
T << Tc , K → 0
→ 2 2
T >> T , K = 1 U h
 c 0
240 D
[Re] Case of T >> Tc
y
For a linear steady velocity profile, u = U sin α
h
4-52
1 U 2h2 2 α
K st = sin
120 D D
1 U 2h2
→ K0 = is an ensemble average of K st over all values of α
240 D
Intermediate behavior → Fig.4.7
T
= 0.1→ K ≈ 0.03K 0
Tc
T
=1→ K ≈ 0.8 K 0
Tc
T
= 10 → K = K 0
Tc
4-53
(ii) Flow including oscillating and a steady component
→ pulsating flow found in blood vessel
=u ( y ) u1 ( y ) sin 2π t / T + u2 ( y )
u=
1 u=
2 Uy / h
Assume that the results by separate velocity profile are additive.
∂C ' ∂C ∂ 2C '
Let C= C + C2
' ' '
is solution to + u (t ) ε 2
=
∂t ∂x ∂y
1
Then C1 ' is solution to the equation
∂C1' ∂C ∂ 2C '
+ u1 sin(2π t / T ) ε 21
=
∂t ∂x ∂y
C2 ' is solution to the equation
∂C2 ' ∂C ∂ 2 C2 '

+ u2 ε
=
∂t ∂x ∂y 2
cycle-averaged dispersion coefficient
1 T 1 h
 2π t 
K =∫ − ∫ 2
 u1 sin + u2 (C1' + C2 ' )dydt
T 0 ∂C −
h
2  T 
h
∂x
1  1 T h2 2π t h

∂C  T ∫0 ∫− h2 1 1 ∫− h2 2 2 
=
− u C '
sin dydt + 2
u C '
dy
T
h
∂x
= K1 + K 2
4-54
where K1 = result of oscillatory profile = f (T / Tc ) → Fig. 4.7
K 2 = result of steady profile
▪ Application to tidal rivers and estuaries
Consider shear effects in estuaries and tidal rivers
Flow oscillation - flow goes back and forth.
Consider effect of oscillation on the longitudinal dispersion coeff.
K = K 0 f (T ′) (7.1)
where f (T ′) is plotted in Fig. 4. 7.
T ′ = T / Tc = dimensionless time scale for cross-sectional mixing
T = tidal period ∼12 hrs
TC = cross-sectional mixing time = W 2 / ε t
K 0 = dispersion coefficient if T  Tc
• For wide and shallow cross section with no density effects
K 0 = I u′2TC (5.17)
where I = dimensionless triple integral ≈ 0.1 (Table 4.1)
Combine Eq. (7.1) and Eq. (5.17)
4-55
K = 0.1u′2T (1/ T ′ ) f (T ′ )  (7.2)
Function (1/ T ′ ) f (T ′ )  is plotted in Fig.7.4
Max = 0.8
W2
i) TC is small (narrow estuary) TC =
εt
T
T=′ >> 1 → K is small
TC
ii) TC is very large (very wide estuary)
T
T=′ << 1 → K is smallest
TC
T
iii) T ′ =
C
≈ 1 : (1/ T ′ ) f (T ′ )  ≈ 0.8
T
0.08u′2T
∴ K max =
= =
[Ex] T 12.5 =
hrs, u 0.3 m/s, u′2 0.2u 2 Ch. 5
K max =0.08 × 0.2(0.3) 2 × (12.5 × 3600) ≈ 60 m 2 /s
4-56
4.4 Dispersion in Two Dimensions
In many environmental flows velocity vector rotates with depth
  
=u iu ( z ) + jv ( z )

where u = component of velocity u in the x direction

v = component of velocity u in the y direction
c' ( z )
▽ x
u(z)
y v(z)
Fig. 4.8 skewed shear flow in the surface layer of Lake Huron
• Taylor’s analysis applied to a skewed shear low with velocity profiles
The 2-D form of Eq. (4.10) for turbulent flow is
∂C ∂C ∂ ∂C '
u' + v' = ε (4.61)
∂x ∂y ∂z ∂z
∂C '
= 0 at z = 0, h (water surface & bottom)
∂z
4-57
Integrate (4.61) w.r.t. z twice
1  ' ∂C ∂C 
C' ( z)
z z
= ∫ ε∫
0 0  u
 ∂x
+ v'
∂y 
dzdz (4.62)
Bulk dispersion tensor can be defined by
∂C ∂C
M x =
h
∫0 =
− −
' '
u C dz hK hK
∂x ∂y
xx xy
∂C ∂C
M y =
h
∫0 =
− −
' '
v C dz hK hK (4.63)
∂x ∂y
yx yy
Substitute (4.62) into (4.63)
h z 1 z ' ∂C ∂C   ∂C ∂C 
(a): ∫0
u' ∫
0 ε∫ 0  u
 ∂x
+ v' 
∂y 
dzdzdz =
h  − K xx
 ∂x
− K xy
∂y 
1 h ' z1 z '
h ∫0 ∫0 ε ∫0
K xx = − u u dzdzdz (4.64a)
1 h ' z1 z '
h ∫0 ∫0 ε ∫0
K xy = − u v dzdzdz (4.64b)
depend on the interaction of the x

and y velocity profiles
h z 1  ' ∂C
z
' ∂C   ∂C ∂C 
(b): ∫0
v' ∫
0 ε ∫0  ∂x
 u + v 
∂y 
dzdzdz =
h 

− K yx
∂x
− K yy
∂y 
1 h ' z1 z '
h ∫0 ∫0 ε ∫0
K yx = − v u dzdzdz (4.64c)
1 h ' z1 z '
h ∫0 ∫0 ε ∫0
K yy = − v v dzdzdz (4.64d)
4-58
The velocity gradient in the x direction can produce mass transport in the y direction and vice
versa.
K xy = x-dispersion coefficient due to velocity gradient in the y direction
K yx = y-dispersion coefficient due to velocity gradient in the x direction
▪ Mean flow on a continental shelf (Fischer, 1978)
y (offshore)
v = V0
x (alongshare)
d u
2
z=d
U0
v = −V0
d 2  U 0 / 120 5U 0V0 / 192 

2
K=   (4.65)
ε  5U 0V0 / 192 U 0 2 / 120 
Distribution of
concentrated slug of
dye after 5 days
1.2km
y
t = 5days
u = 5cm / s
x=22km U 0 = 5cm / s
( = ut ) 72.6° V0 = 5cm / s
x
28 km
Source
4-59
[Re] Derivation of 2-D dispersion equation
qy x
qx dx
∂qx
dy qx + ∆x
∂x
∂q y
y qy + ∆y
∂y
(i) Conservation of mass
∂C   ∂q    ∂q 
∆x∆=
y qx −  qx + x ∆x   ∆y + q y −  q y + y ∆y   ∆x
∂t   ∂x    ∂y 
∂C ∂q ∂q
∴ =
− x− y (1)
∂t ∂x ∂y
(ii) Apply Taylor’s Analysis on 2-D shear flow

q x M
= = x ( )
u 'C ' =
h ∫ 0
h
u 'C '=
dz ∫ u ∫ ε ∫  u
' 1  ' ∂C
∂x
+ v'
∂C 
∂y 
dzdzdz
∂C ∂C
=
− K xx − K xy (2)
∂x ∂y
= 
=
qy M y ( )
v ' c '=
h ∫
h
0
v ' c=
'
dz ' 1 
∫ v ∫ ε ∫  u
' ∂C
∂x
+ v'
∂C 
∂y 
dzdzdz
∂C ∂C
=
− K yx − K yy (3)
∂x ∂y
(iii) Substitute (2) & (3) into (1)
∂C ∂ ∂C ∂C  ∂  ∂C ∂C 
=
−  − K xx − K xy  −  − K yx − K yy
∂t ∂x  ∂x ∂y  ∂y  ∂x ∂y 
4-60
(iv) Return to fixed coordinate system containing mean advective velocities
∂C ∂C ∂C ∂  ∂C ∂C  ∂  ∂C ∂C 
+u + v=  + K xy  +  K yx + K yy
∂y 
K xx
∂t ∂x ∂y ∂x  ∂x ∂y  ∂y  ∂x
In general K xy and K yx are small compared with K xx and K yy . Thus, those two terms
are often neglected. Then, 2-D depth-averaged transport equation becomes
∂C ∂C ∂C ∂  ∂C  ∂  ∂C 
+u +=  K xx  +  K yy
∂y 
v
∂t ∂x ∂y ∂x  ∂x  ∂y 
4-61
[Cf] 2-D depth-averaged models (ASCE, 1988; vol.114, No.9)

∙ Scalar transport equation for Φ
∂( HΦ) ∂ ( HU Φ ) ∂ ( HV Φ ) 1 ∂ 1 ∂
∂t
+
∂x
+
∂y
=
ρ ∂x
( HJ x ) +
ρ ∂y
( HJ y )
1 ∂ 1 ∂
+ ∫ ρU 'Φ 'dz +
ρ ∂x  ∫ ρV 'Φ 'dz
ρ ∂y

dispersion dispersion
Jx= ∫ − ρ u φ dz
' '
where turbulent diffusion in x-dir
Jy= ∫ − ρ u φ dz
' '
turbulent diffusion in y-dir
u '= u − U → time fluctuation
φ'= φ − Φ
U=' U − U → depth deviation
Φ' = Φ − Φ
If dispersion >> turbulent diffusion
→ neglect turbulent diffusion or incorporate turbulent diffusion into dispersion.
4-62
Ch 5. Mixing in Rivers
Chapter 5 Mixing in Rivers
Contents
5.1 Turbulent Mixing in Rivers
5.2 Longitudinal Dispersion in Rivers
5.3 Dispersion in Real Streams
5.4 Estimation of Dispersion Coefficient in Real Streams
Objectives:
- Discuss turbulent diffusion
- Study transverse mixing in the 2nd stage mixing
- Discuss process of longitudinal dispersion for the analysis of final stage
- Study determination methods of dispersion coefficient
5-1
Consider a stream of effluent discharged into a river.
Near-field:
3D model
Far-field:
1D model
Far-field:
2D model
What happens can be divided into three stages:
Stage I: Near field mixing
~ Initial momentum and buoyancy determine mixing near the outlet.
~ Vertical mixing is usually completed at the end of this region.
→ Ch.9 Turbulent jets and plumes; Ch.10 Design of ocean wastewater discharge system
Stage II: Two-dimensional mixing (longitudinal + lateral mixing)
~ Waste is mixed across the receiving channel primarily by turbulence in the receiving stream.
→ dye mixing across the Columbia River (Fig. 5.6)
5-2
[Confluence of Nakdong and Keumho Rivers]
Stage III: Longitudinal dispersion
~ Process of longitudinal shear flow dispersion erases any longitudinal concentration
variations.
~ We could apply Taylor's analysis of longitudinal dispersion.
[Re] Two phases of hydrodynamics mixing processes
1) Near field: mixing is controlled by the initial jet characteristics of momentum flux,
buoyancy flux, and outfall geometry
2) Far field: source characteristics are less important, mixing is controlled by buoyant
spreading motions and passive mixing due to ambient turbulence and currents
 Far field = Stage II+ Stage III
→ deal with a source of tracer without its own momentum or buoyancy
5-3
[Re] Analysis of near field mixing
jet
▪ Multiport diffuser
- linear structure consisting of many closely spaced ports, or nozzles, through which
wastewater effluent is discharged at high velocity into the receiving water body
- attractive engineering solution to the problem of managing wastewater discharge in an
environmentally sound way
- offer high degree of initial dilution
- optimally adapted to the assimilative characteristic of the water body
• Thermal diffuser: heated water discharge from the once-through cooling systems of
nuclear power plant and fossil fuel power plant
S ~ 10
5-4
• Wastewater diffuser: wastewater discharge from the sewage treatment plants
S ~ 100
▪ Three groups of parameters for jet analysis
1) Receiving water flow patterns: ambient water depth, velocity, density stratification
2) Pollutant discharge flow characteristics: discharge velocity (momentum), flow rate,
density of pollutant (buoyancy)
3) Diffuser characteristics: single/multi ports, submerged/surface discharge,
alignment of port
▪ Jet analysis model: CORMIX (Cornell Mixing Zone Expert System)
VISJET
▪ Water-quality policy in USA
Office of Water (1991) "Technical support document for water quality-based toxics control,"
Washington, DC.
~ regulations on toxic control with higher initial mixing requirements
Regulatory mixing zone (RMZ)
= limited area or volume of water where initial dilution of an aqueous pollutant discharge
occurs
- regulator = U.S. Environmental Protection Agency
5-5
- should predict the initial dilution of a discharge and extent of its mixing zone
- toxic dilution zone (TDZ) for toxic substances
- regularly mixing zone (RMZ) for conventional pollutants
streams, rivers lakes, estuaries
Florida: RMZ ≤ 800m ≤ 125,600 m²
and ≤ 10% total length and ≤ 10% surface area
Michigan: RMZ ≤ 1/4 cross-sectional area ≤ 1000 ft radius
West Virginia: RMZ ≤ 20∼33% cross-

≤ 300 ft any direction
sectional area and ≤ 5∼10 times width
5-6
5.1 Turbulent Mixing in Rivers
5.1.1 Idealized Case of a Uniform, Straight, Infinitely Wide Channel of Constant Depth
Consider mixing of source of tracer without its own momentum or buoyancy in a straight
channel of constant depth and great width
The turbulence is homogeneous, stationary because the channel is uniform.
If the sidewalls are very far apart the width of the flow should play no role.
→ The important length scale is depth.
From Eq. (3.40), turbulent mixing coefficient is given as
ε =  L u  '2 2
(1)
 
where ε = turbulent mixing coefficient
 L = Lagrangian length scale ≈ d (a)
1
u '2  2 = intensity of turbulence
 
1 1
=
u '2 ∫ = ∫ (u − u ) 2 dt
2
u ' dt
T T
∙ Experiments (Lauffer, 1950) show that in any wall shear flow
1
1
u '2  ∝ τ 0
2
(b)
τ=
τ0 =
− ρ u'v ' =∫ (u − u )(v − v )dt
T
 
5-7
For dimensional reasons use shear velocity
τ
=
u* =0
gdS (5.1)
ρ
where τ 0 = shear stress on the channel bottom
[Re] shear stress (Henderson, 1966)
~ bottom shear stress is evaluated by a force balance
τ 0 = ρ gdS
where S = slope of the channel
Substitute (a) & (b) into (1)
ε ∝ d u*
ε = α d u*
→ turbulence will not be isotropic
i) vertical mixing, εv
~ influence of surface and bottom boundaries
ii) transverse and longitudinal mixing, εt ,εl
~ no boundaries to influence flow
5-8
[Re] Shear stress and shear velocity
F1
F2
Apply Newton’s 2nd law of motion to uniform flow

 
 a =0
ΣF =ma F1 = F2
F1 − bottom shear + W sin θ − F2 =

0
−τ 0 Pdx + ρ gAdx sin θ =

0
A
τ0 = ρg sin θ
P
where P = wetted perimeter
Set =S tan θ ≈ sin θ
A
R = hydraulic radius =
P
Then τ 0 = γ RS
For very wide channel (b>>d)
bd d
=R = ≈d
b + 2d 1 + 2 d
b
τ 0 = γ dS
5-9
5.1.1.1 Vertical Mixing
Vertical mixing coefficient is needed for 3D model
→ there is no dispersion effect by shear flow
i) The vertically varying coefficient
The vertical mixing coefficient for momentum (eddy viscosity) can be derived from
logarithmic law velocity profile (Eq. 4.43).
z z
=ε v κ du* 1 −  (5.2)
d d
[Re] Derivation of (5.2)
u* z u*
u( z ) =
u+ (1 + ln ) =
u + (1 + ln z ' ) (1)
κ d κ
du u* 1 1
= (2)
dz κ z ' d Reynolds
analogy
 z du
τ = τ 0  1 −  = ρε v (3)
 d dz
Substitute (2) into (3)
u* 1 1
τ 0 (1 − z ) =
'
ρε v (4)
κ z' d
Rearrange (4)
τ0 '
εv κ d
= z (1 −=
z ' ) κ d * z ' (1u− z ' ) (5)
ρ
→ parabolic distribution
5-10
The Reynolds analogy states that the same coefficient can be used for transports of mass and
momentum.
→ verified by Jobson and Sayre (1970)
[Re] Relation between eddy viscosity ( t ) and turbulent diffusion coefficient ( t )
→ use turbulent Prandtl (heat) or Schmidt number (mass), t
t
t 
t
where t ~ is assumed to be constant, and usually less than unity
[Re] Velocity profiles:
- vertical profile of u-velocity ∼ logarithmic
- vertical profile of v-velocity ∼ linear/cubic → might be neglected because v-
velocity is relatively small compared to u-velocity
ii) The depth-averaged coefficient
Average Eq. (5.2) over the depth, taking κ = 0.4
1 d *  z   z  κ *
=εv
d ∫0
κ du   1 − 
 d    d 
=
  dz
6
=du 0.067 du * (5.3)
[Cf] For atmospheric boundary layer: ε v = 0.05du *
*
where d = depth of boundary layer; u = shear velocity at the earth surface
5-11
5.1.1.2 Transverse Mixing
(1) Transverse mixing coefficient in 3D model
∼ no dispersion effect by shear flow, turbulence effect only
→ vertically varying coefficient
For infinitely wide uniform channel, there is no transverse profile of velocity.
∼ not possible to establish a transverse analogy of Eq. (5.2)
→ need to know velocity profiles:
- transverse profile of u-velocity ∼ parabolic/beta function
- transverse profile of w-velocity ∼ might be neglected because w-velocity is usually very
small
[Re] Turbulent diffusion coefficient for 3-D flow
∂c ∂c ∂c ∂c ∂ ∂c ∂ ∂c ∂ ∂c
+ u + v + w= (ε x ) + (ε y ) + (ε z )
∂t ∂x ∂y ∂z ∂x ∂x ∂y ∂y ∂z ∂z
Consider shear stress tensor
 σ xx τ xy τ xz 
τ σ yy τ yz 
 yx
τ σ zz 
 zx τ zy
Now, consider velocity gradients for each turbulent diffusion coefficient
5-12
du dv dw
τ xz = ρε v τ yz = ρε v σ zz = ρε v
dz dz dz
du dv dw
τ xy = ρε t σ yy = ρε t τ zy = ρε t
dy dy dy
du dv dw
σ xx = ρε l τ yx = ρε l τ zx = ρε l
dx dx dx
1) vertical mixing
- vertical profile of u-velocity ∼ logarithmic
- vertical profile of v-velocity ∼ linear/cubic → might be neglected because v-velocity is
relatively small compared to u-velocity
2) transverse mixing
- transverse profile of u-velocity ∼ parabolic/beta function
- transverse profile of w-velocity → might be neglected because w-velocity is usually very
small
3) longitudinal mixing
- longitudinal profile of v-velocity ∼ linear/cubic
- longitudinal profile of w-velocity → might be neglected because w-velocity is usually very
small
5-13
(2) Transverse mixing coefficient in 2D model
Depth-averaged 2D model is
∂c ∂c ∂c ∂ 2c ∂ 2c
+u + v = DL 2 + DT 2
∂t ∂x ∂z ∂x ∂z
Include dispersion effect by shear flow due to vertical variation of v-velocity
v= v ( z )= v + v '
▪ Depth-averaged coefficient for rectangular open channels
→ rely on experiments (Table 5.1 for results of 75 separate experiments)
DT= ε t ≅ 0.15du* (5.4)
Researchers (Okoye, 1970; Lau and Krishnappan, 1977) proposed that
DT = α du*
α = f (W / d )
5-14
5.1.1.3 Longitudinal Mixing
(1) Longitudinal mixing coefficient in 3D model
∼ no dispersion effect by shear flow, turbulence effect only
→ ε l ~ longitudinal analogy of Eq. (5.2)
(2) Longitudinal mixing coefficient in 2D model
∼longitudinal turbulent mixing is the same rate as transverse mixing because there is an
equal lack of boundaries to inhibit motion
∼ However, longitudinal mixing by turbulent motion is unimportant because shear flow
dispersion coefficient caused by the velocity gradient (vertical variation of u-velocity) is
much bigger than mixing coefficient caused by turbulence alone
Use Elder’s result for depth-averaged longitudinal dispersion coefficient
=Dl 5.93du* ≈ 40ε t
[Re] Aris (1956) showed that coefficients due to turbulent mixing and shear flow are additive.
Dl + ε  → DL → can neglect the longitudinal turbulent diffusion coefficient
(3) Longitudinal dispersion coefficient in 1D model
See Section 5.2
→ K1l + K 2l + ε  → K
where K1l ~ due to lateral variation of u-velocity;
K 2l ~ due to vertical variation of u-velocity

5-15
5.1.2 Mixing in Irregular Channels and Natural Streams
5.1.2.1 Mixing in natural channels
Natural streams differ from uniform rectangular channels:
- depth may vary irregularly → pool and riffle sequences
- the channel is likely to curve → meandering rivers
- there may be large sidewall irregularities → groins, dikes
5-16
5-17
5-18
(1) Vertical mixing coefficient
These have not much influence on vertical mixing since scale of vertical motion is limited by
the local depth, d
ε v = 0.067du*
(2) Transverse mixing coefficient
Transverse mixing is strongly affected by the channel irregularities because they are capable
of generating a wide variety of transverse motions
→ transverse dispersion enhanced by vertical variation of v-velocity
Transverse mixing in open channels with curves and irregular sides
→ see Table 5.2
εt
0.3 < < 0.7
du *
1) Effect of channel irregularity
∼ the bigger the irregularity, the faster the transverse mixing
5-19
2) Effect of channel curvature
∼ secondary flow causes transverse dispersion due to shear flow
∼ when a flow rounds a bend, the centrifugal forces induce a flow towards the outside bank
at the surface, and a compensating reverse flow near the bottom.
→ secondary flow generates
Fischer (1969) predict a transverse dispersion coefficient based on the transverse shear flow
2 2
DT  u  d 
= 25  *   (5.5)
du* u  R
where R = radius of curvature
Yotsukura and Sayre (1976) revised Eq.
(5.5) (Fig.5.3)
2 2
DT  u   W 
∝   
du*  u*   R 
where W = channel width
For straight, uniform channels, DT = 0.15du

*
For natural channels with side irregularities, DT = 0.4du

*
For meandering channels with side irregularities, =

DT ( 0.6 ± 50% ) du*
5-20
5-21
5-22
[Re] 2D depth-averaged model
Depth-averaged 2D model is
∂c ∂c ∂c ∂ 2c ∂ 2c
+u + v = DL 2 + DT 2
∂t ∂x ∂z ∂x ∂z
D=
T Dt + ε t
Dt = transverse dispersion coefficient due to vertical profile of v-velocity
ε t = transverse dispersion coefficient due to transverse profile of u-velocity
• Transverse dispersion coefficient in meandering channels
- Baek et al. (2006) - observation
- Baek and Seo (2008), Baek and Seo (2011) – prediction
• Transverse dispersion coefficient in natural streams
- Seo et al. (2006) - observation
- Jeon et al. (2007) - prediction
- Baek and Seo (2010) – observation
- Jeon et al. (2007)
d
 u  W   d  e
b c
DT
= a *      Sn
 u   d   RC 
*
du
a=0.029; b=0.463; c=0.299; d=0; e=0.733
5-23
- Baek and Seo (2008)
2
 u  W   x 1
2 2
DT x
0.04  *     sin(2π )+  I
du*  u   R   2L C LC 2
[Re] Determination of dispersion coefficients for 2-D modeling
1) Observation – calculation of observed concentration curves from field data
2) Prediction – estimation of dispersion coefficient using theoretical or empirical
equations
Observation Method
Moment Simple moment method
method Stream-tube moment method
Routing 2-D routing method
procedure 2-D stream-tube routing method
Prediction Method
Theoretical Use vertical profile of v-velocity
equation for DT Baek & Seo (2008), Baek and Seo (2011)
Empirical Use mean hydraulic data
equation for DT
Fischer (1969)
Yotsukura & Sayre (1976)
Jeon et al. (2007)
5-24
▪ 3D model
∂c ∂c ∂c ∂c ∂ ∂c ∂ ∂c ∂ ∂c
+ u + v + w= (ε x ) + (ε y ) + (ε z )
∂t ∂x ∂y ∂z ∂x ∂x ∂y ∂y ∂z ∂z
Turbulent diffusion coefficients
ε=
x ε=l 0.15du*
ε=
y ε=t 0.15du*
ε=
z ε=
v 0.067du*
▪ 2D model
Depth-averaged 2D model
∂c ∂c ∂c ∂ 2c ∂ 2c
+u + v = DL 2 + DT 2
∂t ∂x ∂z ∂x ∂z
Mixing coefficients = dispersion coefficient + turbulent diffusion coefficients
D=
L Dl + ε 
=Dl 5.93du* ≈ 40ε l
ε l = 0.15du*
D=
T Dt + ε t
0.463 0.299
Dt u  W 
= 0.029  *    Sn 0.733
u  d 
*
du
ε t = 0.15du*
5-25
▪ Numerical model
· In numerical calculations of large water bodies, additional processes are represented by the
diffusivity.
1) Sub-grid advection
Owing to computer limitations, the numerical grid of the numerical calculations cannot be
made so fine as to obtain grid-independent solutions.
→ All advective motions smaller than the mesh size, such as in small recirculation zones,
cannot be resolved. Thus, their contribution to the transport must be accounted for by the
diffusivity.
2) Numerical diffusion
The approximation of the differential equations by difference equations introduces errors
which act to smooth out variations of the dependent variables and thus effectively increase
the diffusivity.
→ This numerical diffusion is larger for coarser grids.
· An effective diffusivity accounts for turbulent transport, numerical diffusion, sub-grid scale
motions, and dispersion (in the case of depth-average calculations).
→ The choice of a suitable mixing coefficient ( DMT ) is usually not a turbulence model
problem but a matter of numerical model calibration.
For 2D model,
DMT = Dt + ε t + ε sgm − ε nd
5-26
5.1.3 2D concentration distributions
Compute the distribution of concentration downstream from a continuous effluent discharge
in a flowing stream
In most of the natural streams the flow is much wider than it is deep; a typical channel
dimension might be 30 m wide by 1 m deep, for example.
Recall that the mixing time is proportional to the square of the length divided by the mixing
coefficient,
( length )
2
T∝
ε
W 30
≅ =
30
d 1
ε 0.6du *
=t
≈ 10
εv 0.067 du *
(W ) ( d )
2 2
 W  ε v  30   1 
2 2
T
=
∴ t /=  =     = 90 ≈ 10
2
Tv εt εv  d  ε t  1   10 
∴ Tt ≈ 102 Tv
→ vertical mixing is instantaneous compared to transverse mixing
Thus, in most practical problems, we can start assuming that the effluent is uniformly
distributed over the vertical.
→ analyze the two-dimensional spread from a uniform line source
5-27
⋅
Now consider the case of a rectangular channel of depth d into which is discharged M
units of mass (per time) in the form of line source.
⋅
~ is equivalent to a point source of strength M / d in a two-dimensional flow
→ maintained source in 2D
Line source
Recall Eq. (2.68)
⋅
M/d  y 2u 
=C exp  −  (5.7)
x  4ε t x 
u 4πε t
u
i) For very wide channel, when t >> 2ε t / u

2
→ use Eq. (5.7)
5-28
ii) For narrow channel, consider effect of boundaries
∂C
= 0= =
at y 0 and y W
∂y
→ method of superposition
Define dimensionless quantities by setting
⋅
M
C0 = = mass rate / volume of ambient water
udW
∼ concentration after cross-sectional mixing is completed
xε t
x' =
uW 2
y' = y /W
Then Eq. (5.7) becomes
⋅
M  y 
 ( )2 
= udW exp − W
C  4ε x 
4πε t x  t

uW 2  uW 2

5-29
C0  y '2 
= exp  − ' 
4π x '  4x 
C 1  y '2 
= exp  − ' 
C0 (4π x ' )1/ 2  4x 
=
If the source is located at y y=
0(y
'
y '0 )
Consider real and image sources, then superposition gives the downstream concentration
distribution as
real I1 I2
C 1   ( y ' − y '0 )2   ( y ' + y '0 )2   ( y ' − 2 + y '0 )2  

= 1  exp( −   ) + exp( −   ) + exp( −   ) + • • • 
C0
(4π x ' ) 2   4x '   4x '   4x '  
∑ {exp  −( y − 2n + y }
∞
) 2 / 4 x '  + exp  − ( y ' − 2n + y ' 0 ) / 4 x ' 
1 ' ' 2
(5.9)
1
' 2 n =−∞
0
 
(4π x )
Sum for n = 0, ± 1, ± 2
I2 I1
5-30
Continuous centerline discharge: y 0 = 1/ 2

'
Centerline
injection
From this figure, for x’ greater than about 0.1 the concentration is within 5 % of its mean
value everywhere on the cross section.
Thus, the longitudinal distance for complete transverse mixing for centerline injection is
Lc = 0.1uW 2 / ε t (5.10)
C xε t
[Re] = 0.95 at = =
x ' 0.1
C0 uW 2
Lc= x= 0.1uW 2 / ε t
5-31
For side injection, the width over which mixing must take place is twice that for a centerline
injection
= =
L 0.1u (2W ) 2 / ε t 0.4uW 2 / ε t (5.10a)
∂C
=0
∂y
5-32
[Ex 5.1] Spread of a plume from a point source
An industry discharges effluent;

Continuous
C = 200ppm injection
3 × 106 GPD
=
Q = 3MGPD = 4.64 CFS
7.48G / ft 3 × 24 × 3600
Thus, rate of mass input is
•
M = QC= 4.64(200ppm) = 928 CFSppm
Centerline injection in very wide, slowly meandering stream
d = 30 ft ; u = 2 fps ; u * = 0.2 fps
Determine the width of the plume, and maximum concentration 1000 ft downstream from
discharge assuming that the effluent is completely mixed over the vertical.
[Sol]
For meandering stream,
ε t = 0.6du * 0.6
= = ( 30 )( 0.2 ) 3.6 ft 2 / s
Use Eq.(5.7) for line source
Exponential
Peak
decay
concentration
⋅
M  y 2u 
=C ( x, y ) exp  −  (5.7)
1
 4ε x 
 4πε t x  2 t
ud  
 u 
5-33
1  y2 
=
Compare with normal distribution; C exp  − 2 
σ 2π  2σ 
 
 y2   y2 
exp  − =exp  − 2 
4ε t x   2σ 
 
 u 
2ε t x
σ2 =
u
2ε t x
σ=
u
a) width of plume can be approximate by 4σ (includes 95% of total mass)
2ε t x 2 ( 3.6 )(1000 )
= σ 4
b 4= = 4 = 240 ft
u 2
b) maximum concentration
⋅
M 928CFSppm
=Cmax = 1 1
 4πε t x  2  4π × 3.6 ft / s × 1000 ft 
2 2
ud   ( 2 fps )( 30 ft )  
 u   2 ft / s 
1000 ft
5-34
[Ex 5.2] Mixing across a stream
→ consider boundary effect
Given:
Find: length of channel required for "complete mixing" as defined to mean that the
concentration of the substance varies by no more than 5% over the cross section
[Sol]
Shear velocity
=u* =
gdS 32.2 ( 5 )( 0.0002
= ) 0.18 ft / s
For uniform, straight channel
ε t = 0.15du * 0.15
= = ( 5)( 0.18) 0.135 ft 3 / s
For complete mixing from a side discharge Very long distance

for a real channel
L = 0.4uW 2 / ε t
= 0.4 ( 2 )( 200 ) / 0.135

= 237, 000 ft ≈ 45mile ≈ 72km
2
5-35
[Ex 5.3] Blending of two streams
Compute the mixing of two streams which flow together at a smooth junction so that the
streams flow side by side until turbulence accomplishes the mixing.
Given:
=Q 50 ft 3= =
/ s;W 20 =
ft; S 0.001; n 0.030
Find:
a) length of channel required for complete mixing for uniform straight channel
b) length of channel required for complete mixing for curved channel with a radius of 100 ft
[Sol]
The velocity and depth of flow can be found by solving Manning's formula
5-36
1.49 23 12
u= R S
n
R = hydraulic radius = A/P
1.49 1.49 A5/3 1/2

= =
Q Au AR 2/3 S=
1/2
S
n n P 2/3
1.49 ( 20d )
5/3
d 5/3
= ( 0.001) 145.41
1/2
100
0.030 ( 20 + 2d )2/3 (10 + d )
2/3
d 5 / 3 0.688 (10 + d )
=
2/3
=d 0.799 (10 + d )
2/5
By trial-error method, d = 2.2 ft
2.2 ( 20 )
=R = 1.8 ft
( 20 + 4.4 )
1.49  2.2 × 20 
2/3
=  ( 0.001)
1/ 2
u  2.32 ft / s
0.030  20 + 4.4 
∴=
u* =
gRS 32.2 (1.8 )( 0.001
=) 0.24 fps
ε t = 0.15du * = 0.15(2.2)(0.24)=0.079 ft2/s
5-37
For the case of blending of two streams, there is a tracer whose concentration is C0 in one
stream and zero in the other.
If the steams were mixed completely the concentration would be 1/2 C0 everywhere on the
cross section.
The initial condition may be considered to consist of a uniform distribution of unit inputs in
one-half of the channel.
→ The exact solution can be obtained by superposition of solutions for the step function in
an unbounded system [Eq. (2.33)].
Consider sources ranging y 0 = 0 ~ 1/ 2

'
Method of images gives
5-38
C 1 ∞  y ' + 1/ 2 + 2n y ' − 1/ 2 + 2n 
= ∑  erf
C0 2 n=−∞ 
− erf 
4x ' 4 x' 
xε t
=
where y '
y=
/ W ; x'
uW 2
From Fig. 5.9, maximum deviation in concentration is 5% of the mean when x ' ≈ 0.3 .
Lε t
=x' = 0.3
uW 2
(2.32) ( 20 )
2
uW 2
=L 0.3= 0.3 = 3515 ft < 4687 ft
εt 0.15 ( 2.2 )( 0.24 )
[Re] For side injection only
( 2.32 )( 20 )
2
uW 2
= = 0.4
L 0.4 = 4687 ft
εt 0.15 ( 2.2 )( 0.24 )
For curved channel
εt
2 2
 u  d
= 25  *  
du * u   R
2 2
 2.23   2.2 
∴ε t 25 =  ( 2.2 )( 0.24 ) 0.60 ft / s
2
 
 0.24   100 
uW 2 0.3 ( 2.32 )( 20 )
2
= =
L 0.3 = 464 ft
εt 0.60
5-39
5.1.4 Cumulative Discharge Method for 2D Mixing
Previous analysis was presented assuming a uniform flow of constant velocity everywhere in
the channel.
However, in real rivers, the downstream velocity varies across the cross section, and there are
irregularities along the channel.

u
Use cumulative discharge method (Stream-tube method) by Yotsukura and Sayre (1976)
Define velocity averaged over depth at some value of y as
 1 0
d ( y ) ∫− d ( y )
u= udz (a)
Then, cumulative discharge is given as
y y 
q=
( y) ∫=
dq
0 ∫ 0
d ( y )udy (b)
q( y)
= 0=
at y 0 (c)
q ( y ) Q=
= at y W
5-40
[Cf] u = cross-sectional average velocity
Now, derive depth-averaged 2D equation for transverse diffusion assuming steady-state
concentration distribution and neglecting longitudinal mixing and v-velocity
∂C ∂C ∂C ∂  ∂C  ∂  ∂C 
+u + v=  εl + εt
∂y ∂x  ∂x  ∂y  ∂y 
(d)
∂t ∂x
Integrate (d) over depth
0 ∂C 0 ∂ ∂C
∫ −d
u
∂x
dz = ∫
− d ∂y
(ε t
∂y
)dz (e)
From Eq.(a)

udz = d ( y ) u
0
∫ −d
Eq. (e) becomes
 ∂C ∂  ∂C 
d ( y)u =  d ( y )εt 
∂x ∂y  ∂y 
∂C 1 ∂  ∂C 
=   d ( y )εt
∂y 
(f)
∂x d ( y ) u ∂y 
∂q ∂  y   
∂y ∂y  ∫0
= = d ( y ) ud y d ( y ) u
Transformation from y to q gives 
∂ ∂q ∂  ∂
= = d ( y)u (g)
∂y ∂y ∂q ∂q
5-41
Substituting Eq. (g) into Eq.(f) yields
∂C 1  ∂    ∂C   ∂  2  ∂C 
=  d ( y ) u  d ( y )εt  d ( y ) u    d ( y )εtu
∂x d ( y)u ∂q   ∂q   ∂q  ∂q 

=
If we set εq d 2
ε t u ≅ constant diffusivity, then equation becomes
∂C ∂ 2C
= εq 2
∂x ∂q
→ Fickian Diffusion equation; Gaussian solution in the x-q coordinate system
5-42
▪ Advantage of x-q coordinate system
- A fixed value of q is attached to a fixed streamline, so that the coordinate system shifts back
and forth within the cross section along with the flow.
→ simplifies interpretation of tracer measurements in meandering streams
→ Transformation from transverse distance to cumulative discharge as the independent
variable essentially transforms meandering river into an equivalent straight river.
Thalweg
line The peak
remains at the
injection
location.
The peak of the Gaussian

concentration distribution
curves moves
from side to side
as the river
meanders.
5-43
5.2 Longitudinal Dispersion in Rivers
After a tracer has mixed across the cross section, the final stage in the mixing process is the
reduction of longitudinal gradients by longitudinal dispersion.
The longitudinal dispersion may be neglected when effluent is discharged at a constant rate
→ Streeter-Phelps equation for BOD-DO analysis
There are, however, practical cases where longitudinal dispersion is important.
→ accidental spill of a quantity of pollutant; output from a STP which has a daily cyclic
variation
• 1D dispersion equation
∂C ∂C ∂ 2C
+u =
K 2
∂t ∂x ∂x
→ apply shear flow dispersion theory to evaluate the longitudinal dispersion coefficient K
5.2.1 Theoretical Derivation of Longitudinal Dispersion Coefficient
Elder's analysis
- dispersion due to vertical variation of u-velocity (logarithmic profile)
u*
u ( z ) =+
u
κ
{1 + ln [ z + d / d ]}
K = 5.93du *
5-44
Elder’s equation does not describe longitudinal dispersion in real streams (1D model).
Experimental results shows K >> 5.93du → Table 5.3

*
1) Fischer (1967) - Laboratory channel
K
= 150 ~ 392
du *
2) Fischer (1968) - Green-Duwamish River
K
= 120 ~ 160
du *
3) Godfrey and Frederick (1970)
– natural streams in which radioactive tracer Gold-198 was used
K
= 140 ~ 500
du *
4) Yotsukura et al. (1970) - Missouri river
K
= 7500
du *
5-45
▪ Fischer’s model (1966, 1967)
He showed that the reason that Elder's result does not apply to 1D model is because
of transverse variation of across the stream.
Vertical velocity profile, u ( z ) is approximately logarithmic.
Now, consider transverse variation of depth-averaged velocity
 1 0
u ( y) = u ( y, z ) dz
d ( y ) ∫− d ( y )
u'
Transverse velocity profile would be approximated by parabolic, polynomial, or beta function.
5-46
5-47

u ( y ) is a shear flow velocity profile extending over the stream width W, whereas u( z ) ,
the profile used in Elder’s analysis, extends only over the depth of flow d.
Remember that longitudinal dispersion coefficient is proportional to the square of the
distance over which the shear flow profile extends.
h 2 u '2
Eq. (4.26): K= I
E
K ∝ h2
where h = characteristic length, W or d
Say that W / d ≈ 10
Therefore,
KW ≈ 100 K d
→ Transverse profile u(y) is 100 or more times as important in producing longitudinal
dispersion as the vertical profile.
→ The dispersion coefficient in a real stream (1D model) should be obtained by neglecting
the vertical profile entirely and applying Taylor's analysis to the transverse velocity profile.
5-48
Consider balance of diffusion and advection

Let u '=
( y) u( y) − u

C=
'
( y) C ( y) − C
u = cross-sectional average velocity = U
Equivalent of Eq. (4.35) is
∂C ∂ ∂C '
u' ( y ) = εt (a)
∂x ∂y ∂y
Shear
advection Transverse
diffusion
Integrate Eq. (a) over the depth
∂C 0 ∂ ∂C '
∫− d u ( y ) ∂x dz = ∫− d ∂yε t ∂y dz
0
'
(b)
∂C ∂ ∂C '
u ( y)d ( y)
'
= d ( y )εt (c)
∂x ∂y ∂y
5-49
Integrate Eq. (c) w.r.t. y (in the transverse direction)
∂C ∂C '
∫0 u ( y ) d ( y ) ∂x dy = dε t ∂y
y
'
(5.15)
∂C ' 1 ∂C
∫ u ( y ) d ( y ) ∂x dy
y
= '
(d)
∂y d ε t 0
Integrate again Eq. (d) w.r.t. y (in the transverse direction)
1 ∂C
u' ( y ) d ( y )
y y
C' = ∫ ∫ dydy (e)
0 dε t 0 ∂x
Eq. (4.27)
1
K= − ∫ u C dA
' '
(f)
∂C A
A
∂x
Substitute Eq. (e) into Eq. (f)
1 1 ∂C 1
K= − ∫ ∫ ∫
'
u du 'dydydA
A ∂C ∂x A dε t
∂x
Substitute dA = dy d
1 W ' y 1 y '
A ∫0
ud∫
0 ε d ∫0
K= − u d dydydy (5.16)
t
5-50
This result is only an estimate because it is based on the concept of a uniform flow in a
constant cross section.
[Re] K1l + K 2l + ε  → K
where K1l ~ due to lateral variation of u-velocity;
K 2l ~ due to vertical variation of u-velocity
▪ Simplified equation
u' εt y
=
Let d d / d=
'
; u ''
; ε 't
= = ; y'
u '2 εt W
Overbars mean cross-sectional average; d = cross-sectional average depth
Then
W 2 u '2
K= I (5.17)
εt
where I is dimensionless integral given as
1 y' 1 y' '' ' ' '

I = − ∫ u ''d ' ∫
ε 't d ' ∫0
u dy dy dy
0 0
Compare with Eq. (4.26)
h 2 u '2
K= I
E
5-51
[Example 5.4] cross-sectional distribution of velocity (Fig.5.11) of Green-Duwamish at
Renton Junction
ε t = 0.133 ft 2 / sec
Estimate longitudinal dispersion coefficient
Solution: divide whole cross section into 8 subarea
42
42 12.6
1 W ' y 1 y '
A ∫0
u d∫
0 ε d ∫0
K= − du dydydy
t
→ perform inner integral first
Column 2: transverse distance to the end of subarea
Column 4: ∆A = d ∆y

Column 46: ∆Q = u ∆A
Column 8: Relative ∆Q = u ∆A
'
Column 9: Cumulative of Relative ∆Q = u ∆A

'
y 1 y ' ∆y
Column 11: ∫0 ε t d ∫0
du dydy = ∑ Col (10)
εtd
5-52
W y 1 y '
Column 13: ∫
0
u 'd ∫
0 ε t d ∫0
= Col (8) × Col (12)
du dydydy
1
K= − Cumulativeof Col (13)
A
5-53
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
û
Rel. ∆Q
u′
y
y
d
∆A= d × ∆y
Stream
mean
∆Q
= û × ∆A ∫0
u ′ dA
Average
subarea (측정치) = uˆ × ∆A = uˆ − u of
velocity (CFS) (8)을
(ft) (ft2) (9)
(ft) (측정치) (CFS) (fps)
(4)*(7) 누가한 값
(ft/s)
63 0
1 1.8 12.6 0.105 1.323 -0.796 -10.026 -5.013
=0.105(12.
70 =1.8(7) -10.026
6)
2 4.2 42 0.526 22.092 -0.375 -15.738 -17.895
80 -25.764
3 4.2 42 0.986 41.412 0.085 3.582 -23.973
90 -22.182
4 4.8 48 1.091 52.368 0.190 9.134 -17.616
100 -13.049
5 5.2 52 1.196 62.192 0.295 15.355 -5.371
110 2.306
6 6.6 66 1.148 75.768 0.247 16.321 10.466
120 18.627
7 6.4 64 0.766 49.024 -0.135 -8.622 14.316
130 10.005
8 2 12 0.067 0.804 -0.834 -10.005 5.002
136 0.000
Sum A= 338.6 Q= 304.98 0.000
εt = 0.133
=u Q=
/A 0.90 fps K= -(-26254)/A = 77.54 ft2/s
ft2/s
∑ du ′∆=y ∑ u ′∆A
y
(5) given in p.128 (9) ∫0
du ′ dy
= (  d ∆y =∆A )
(5.16) : Inner integral first (11)

y 1 ∆y
dy dy ∑ ∫ du ′ dy ∑ (10 ) × ∆y / ε d
y y
∫0 ε t d ∫0 du ′= 0
=
εt d
t (11):
( −5.013)( 7 ) / ( 0.133)(1.8) =
−146.6
1 W y 1 1 y  
∑ u ′ d ∑ (8) × (12 )
y
∫ u ′d ∫ ∫
y
K= − du ′ dy dy dy (14) ∆y  ∫ ∫ du ′ dy dy  =
εt d    εt d
 
A 0 0 0 0 0
(9) ∆
A
   (8)
Re l. ∆Q = (12 )
(11)
  
(14 )
−146.6 + ( −17.895 )( 7 ) / ( 0.133 × 4.2 ) =−467.0
5-54

1. Estimate the longitudinal dispersion coefficient using the cross-sectional distribution of
velocity measured in the field using Eq. (5.16). Take S (channel slope) = 0.00025 for natural
streams.
2. Compare this result with Elder's analysis and Fischer's approximate formula, Eq. (5.19).
Table 1 Cross-sectional Velocity Distribution at Ottawa in the Fox River, Illinois
Y from left bank Depth, d Mean Velocity

Station
(ft) (ft) (ft/sec)
1 0.00 0.0 0.00
2 4.17 1.4 0.45
3 7.83 3.0 0.68
4 11.50 3.7 1.05
5 15.70 4.7 0.98
6 22.50 5.3 1.50
7 29.83 6.2 1.65
8 40.83 6.7 2.10
9 55.50 7.0 1.80
10 70.17 6.5 2.40
11 84.83 6.3 2.55
12 99.50 6.8 2.45
5-55
13 114.17 7.4 2.20
14 132.50 7.3 2.65
15 150.83 7.1 2.70
16 169.16 7.4 2.35
17 187.49 7.8 2.65
18 205.82 7.8 2.80
19 224.15 7.8 2.60
20 242.48 6.6 2.50
21 260.81 6.3 2.30
22 279.14 6.2 2.35
23 297.47 6.6 2.30
24 315.80 6.0 2.65
25 334.13 5.5 2.50
26 352.46 5.4 2.10
27 370.79 5.2 2.25
28 389.12 5.5 2.30
29 407.45 5.7 1.50
30 416.62 3.2 1.30
31 422.00 0.0 0.00
5-56
5.2.2 Dispersion in Real Streams
So far the analyses have been limited to uniform channels because Taylor’s analysis assumes
that everywhere along the stream the cross section is the same.
Real streams have bends, sandbars, side pockets, pools and riffles, bridge piers, man-made
revetments.
→ Every irregularities contribute to dispersion.
→ It is not suitable to apply Taylor’s analysis to real streams with these irregularities.
5.2.2. 1 Limitation of Taylor's analysis
Taylor’s analysis cannot be applied until after the initial period.
Numerical experiments showed that in a uniform channel the variance of dispersing cloud
behaves as a line as shown in Fig. 5.14.
5-57
x
=
A) generation of skewed distribution: x( '
) < 0.4 ( initial period)
uW 2 / ε t
B) decay of the skewed distribution: 0.4 < x < 1.0

'
C) approach to Gaussian distribution: 1.0 < x

'
∂σ 2
D) zone of linear growth of the variance: 0.2 < x ; = 2D
'
∂t
E) zone where use of the routing procedure is acceptable: 0.4 < x

'
Analytical solution of 1D
advection-dispersion model
5.2.2.2 Two-zone Models
Irregularities in real streams increase the length of the initial period, and produce long tail on
the observed concentration distribution due to detention of small amounts of effluent cloud
and release slowly after the main cloud has passed.
Pockets of dye are retained in small irregularities along the side of the channel. The dye is
released slowly from these pockets, and causes measurable concentrations of dye to be
observed after the main portion of the cloud has passed.
5-58
Flow zone
Storage zone
Mass
exchange
5-59
• Field studies
Godfray and Frederick (1974)
Nordin and Savol (1974)
Day (1975)
Legrand-Marcq and Laudelot (1985)
showed nonlinear behavior of variance for times beyond the initial period. (increased faster
than linearly with time)
σ 2 = f ( t1.4 )
→ skewed concentration distribution
→ cannot apply Taylor's analysis
• Effect of storage zones (dead zones)
1) increases the length of the initial period
2) increases the magnitude of the longitudinal dispersion coefficient
• Two zone models
~ divide stream area into two zones
5-60
Flow zone: advection, dispersion, reaction, mass exchange
∂CF ∂CF ∂  ∂C 
AF + U F AF =  KAF F +F
∂t ∂x ∂x  ∂y 
Storage zone: vortex, dispersion, reaction, mass exchange
∂CS
AS = −F
∂t
Introduce auxiliary equation for mass exchange term F
Exchange model: =F k (CF − CS ) P
∂CS
Diffusion model: F = −ε y
∂y y =0
Flow zone
Storage zone
▪ Dead zone model
Hays et al (1967)
Valentine and Wood (1977, 1979), Valentine (1978)
Tsai and Holley (1979)
Bencala and Waters (1983), Jackman et al (1984)

5-61
▪ Storage zone model
Seo (1990), Seo and Maxwell (1991, 1992)
Seo and Yu (1993)
Seo & Cheong (2001), Cheong & Seo (2003)
∙ Effect of bends
1) Bends increase the rate of transverse mixing.
2) Transverse velocity profile induced by meandering flow increase longitudinal dispersion
coefficient significantly because the velocity differences across the stream are accentuated.
(3) Effect of alternating series of bends depends on the ratio of the cross-sectional diffusion
time to the time required for flow round the bend.
W 2 / εt
γ= (5.18)
L/u
where L= length of the curve
γ ≤ 25 = γ 0 → K = K 0 → no effect due to alternating direction
γ0
γ > 25 → K =
K0
γ
K 0 = dispersion coefficient for the steady-state concentration profile, Eq. (5.16)
5-62
5.2.3 Estimating and Using the Dispersion Coefficient
∙Observation – calculation of observed values from field data
∙Prediction – estimation of dispersion coefficient by theoretical or empirical equations
5.2.3.1 Observation of dispersion coefficient
1) Change of moment method
σ x 2 2 − σ x12 σ t 2 2 − σ t12
=K = U 2
(2.30a)
2 ( t2 − t1 ) 2 ( t2 − t1 )
where σ x 2 = variance of C-x curve;
σ t 2 = variance of C-t curve;
t1 = centroid of C-t curve at x=x1
It is difficult to compute a meaningful value of variance when concentration distributions are
skewed because of long tails on the observed distributions.
2) Routing procedure (Fischer, 1968)
→ match a downstream observation of passage of a tracer cloud to the prediction based on
the upstream observation using an analytical solution.
We can use this procedure only when x ' >0.4
5-63
Predicted concentration distribution at downstream station is obtained according to the
solution of one-dimensional Fickian dispersion model.
 u 2 ( t2 − t1 − t + τ ) 2 
exp  −
∞
 4 K ( t2 − t1 ) 
C ( x2 , t ) = ∫ C ( x1 ,τ )
p
u dτ (5.20)
−∞
4π K ( t2 − t1 )
where
t 1 = mean time of passage at the upstream station (x1)
t 2 = mean time of passage at downstream station (x2)
τ = timelike variable of integration
C ( x 1 ,τ ) = upstream observed concentration-time curve
p
Compare C ( x2 , t ) with C ( x2 , t ) [= downstream observed concentration curve] until it
fits together with varying dispersion coefficient K.
Then, the best fit value is regarded as the observed dispersion coefficient
5-64
5.2.3.2 Prediction of dispersion coefficient
1) Theoretical equation
1 W ' y 1 y '
A ∫0
u d∫
0 ε d ∫0
K= − du dydydy (5.16)
t
∙ Elder (1959) use vertical profile
∞
∙ Seo and Baek (2004) Γ(α )
= ∫ 0
xα −1e − x dx, α > 0
~ use beta function for transverse profile of u-velocity
α −1
u Γ(α + β )  y  y β −1
=   (1 − )
U Γ(α )Γ( β )  W  W
U 2W 2
K =γ
du *
2) Empirical equation
∙ Fischer (1975)
Iu '2 h 2
K =
'
(4.26)
E
Select I = 0.07(0.054 ~ 0.10)
5-65
h = 0.7W ( 0.5 ~ 1.0W )
u '2 = 0.2u 2 (0.17 ~ 0.25)
E= ε=
t 0.6du *
Then (4.26) becomes
U 2W 2
K = 0.01 (5.19)
du *
∙ Seo and Cheong (1998)
Use dimensional analysis to find significant factors
Include dispersion by shear flow and mixing by storage effects
b c
K  U  W 
= a *   
u   d 
*
du
Fischer (1975): a=0.011; b=2.0; c=2.0
Liu (1979): a=0.18; b=0.5; c=2.0
Iwasa and Aya (1991): a=2.0; b=0; c=1.5
Koussis and Rodrguez-Mirasol (1998): a=0.6; b=0; c=2.0
Seo and Cheong (1998): a=5.92; b=1.43; c=0.62
5-66
[Re] Empirical methods
1) Data driven methods
Dimensional analysis → regression method
2) Soft computing methods
Artificial Neural Network (ANN)
Adaptive Neuro-Fuzzy Inference System technique
Expert System
Fuzzy Logic
Genetic Algorithm (GA)
Machine Learning Approach
Model Tree: M5 vs M5’
Neural Networks
Support Vector Machine
5-67
[Ex 5.5] Dispersion of slug (instantaneous input)
Given:
M = 10lb (Rhodamine WT dye); u = 0.90 ft / s ; W = 73 ft ; A = 338.6
d = 4.46 ft , (weighted average)
ε t = 0.133 ft 2 / s
εt 0.133
=
u* = = 0.072 ft / s
0.4d 0.4 ( 4.64 )
Find:
(a) K by Eq. (5.19)
(b) length of initial zone in which Taylor's analysis does not apply
(c) length of dye cloud at the time that peak passes =20,000 ft
(d) C peak at x =20,000ft
[Solution]
(a) Eq. (5.19)
K = 0.011u 2W 2 / du *
= 0.011( 0.90 ) ( 73) / ( 4.46 )( 0.072 )

2 2
= 142.1 ft 2 / s
K ( 5.19 )=
/ K ( 5.16 ) 142.1/
= 77.5 1.83
5-68
[Cf] K by Seo & Cheong (1998)
1.43 0.62
K  U  W 
= =
5.92  *   294 ft 2 / s
u   d 
*
du
→ include dispersion by shear flow and storage effects
(b) initial period
= = uW 2 / ε t 0.4 ( 0.90 )( 73)=

/ ( 0.133) 14,424 ft
2
x 0.4
(c) length of cloud
= ε t / uW 2
x ' x=
( 20,000 =)( 0.133) 0.55
( 0.90 )( 73)
2
- decay of skewed concentration distribution
→ assume Gaussian distribution
dσ 2
= 2K
dt
From Fig.5.14
σ 2ε t
=
2 KW 2
( x − 0.07 )
'
=σ 2 2 K (W 2 / ε t )( x ' − 0.07)
=2 (142 )( 73) / 0.133 ( 0.55 − 0.07 ) =5.46 × 10−6 ft 2

2
∴σ =2.337
5-69
length of cloud σ 4(2,337)

= 4= = 9,348 ft
(d) peak concentration
M 10
C= = = 4.69 × 10−6 lb / ft 3
A 4π Kx / u ( 338.6 ) 4π (142 )( 20,000 ) / ( 0.90 )
max
453.6 g
= 4.69 × 10−6 × 3
= 75.1 × 10−3 g / m3 (= mg / l = ppm)
0.0283m
= 75.1ppb
5-70
Concentration-time data listed in Table 2 are obtained from dispersion study by Godfrey and
Fredrick (1970).
1) Plot concentration vs. time
2) Calculate time to centroid, variance, skew coefficient.
3) Calculate dispersion coefficient using the change of moment method and routing
procedure.
4) Compare and discuss the results.
Test reach of the stream is straight and necessary data for the calculation of dispersion
coefficient are
u =
1.70 ft / s; W 60 ft ;
= =
d 2.77 ft ; u * 0.33 ft / s
Table-2 Time-concentration data for Copper Creek, Virginia
Section 1 Section 2 Section 3 Section 4 Section 5 Section 6
x=630ft x =3310ft x =5670ft x =7870ft x =11000ft x =13550ft
T (hr) C/C0 T (hr) C/C0 T (hr) C/C0 T (hr) C/C0 T (hr) C/C0 T (hr) C/C0
5-71
1111.5 0.00 1125.0 0.00 1138.0 0.00 1149.0 0.00 1210.0 0.00 1226.0 0.00
1112.5 2.00 1126.0 0.15 1139.0 0.12 1152.0 0.26 1215.0 0.05 1231.0 0.07
1112.5 16.50 1127.0 1.13 1140.0 0.30 1155.0 0.67 1220.0 0.25 1236.0 0.22
1113.0 13.45 1128.0 2.30 1143.0 1.21 1158.0 0.95 1225.0 0.52 1241.0 0.40
1113.5 7.26 1128.5 2.74 1145.0 1.61 1200.0 1.09 1228.0 0.64 1245.0 0.50
1114.0 5.29 1129.0 2.91 1147.0 1.64 1202.0 1.13 1231.0 0.70 1249.0 0.58
1115.0 3.37 1129.5 2.91 1149.0 1.56 1204.0 1.10 1234.0 0.72 1251.0 0.59
1116.0 2.29 1130.0 2.80 1153.0 1.26 1206.0 1.04 1237.0 0.71 1253.0 0.59
1117.0 1.54 1131.0 2.59 1158.0 0.86 1208.0 0.95 1240.0 0.65 1257.0 0.54
1118.0 1.03 1133.0 2.18 1203.0 0.53 1213.0 0.72 1244.0 0.55 1304.0 0.44
1120.0 0.40 1137.0 1.34 1208.0 0.30 1218.0 0.50 1248.0 0.45 1313.0 0.27
1124.0 0.10 1143.0 0.60 1213.0 0.17 1223.0 0.31 1258.0 0.24 1323.0 0.14
1128.0 0.04 1149.0 0.23 1218.0 0.10 1228.0 0.21 1308.0 0.12 1333.0 0.06
1133.0 0.02 1158.0 0.08 1228.0 0.04 1238.0 0.08 1318.0 0.06 1343.0 0.03
1138.0 0.00 1208.0 0.03 1238.0 0.01 1248.0 0.02 1333.0 0.03 1403.0 0.02
- - 1218.0 0.00 1248.0 0.00 1300.0 0.00 1353.0 0.00 1423.0 0.00
5-72
Ch 6. Dispersion of Decaying Substances
Chapter 6 Mixing of Decaying Substances
Contents
6.1 Non-Conservative Pollutants
6.2 Analytical Solutions for Non-Conservative Pollutants
6.3 Modeling BOD-DO Coupled System
6.4 Modeling Heat Transport
6.5 Modeling Suspended Solids
6.6 Modeling Toxic Substances
6-1
6.1 Non-Conservative Pollutants
6.1.1 Category of Non-Conservative Pollutants
1) Toxic Substance
- Metals: mercury, cadmium, lead
- Industrial chemicals: toluene, benzenes, phenols, PCB
- Hydrocarbons: PAH (polycyclic aromatic hydrocarbons)
- Agricultural chemicals: pesticides, herbicides, DDT
- Radioactive substances
2) BOD-DO coupled system

3) Temperature
4) Suspended Solids
5) Bacteria and pathogens
6) Nutrients
7) Oil
[Cf] Conservative pollutants
- one which does not undergo any chemical or biochemical changes in transport
- no loss due to chemical reactions or biochemical degradation
- salt, chloride, total dissolved solids, some metals
6-2
6.1.2 Transport of Non-Conservative Pollutants
(1) Toxic Substance
Physio-chemical phases of the transport of toxic substances:
- loss of the chemical due to biodegradation, volatilization, photolysis, and other
chemical and bio-chemical reactions
- sorption and desorption between dissolved and particulate forms in the water
column and bed sediment
- settling and resuspension mechanisms of particulates between water column
and bed sediment
Assume only loss of the chemical
∂ (hC ) ∂ ∂
+ (uCh) + (vCh) = ∇ ⋅ (hD∇C ) + hS
∂t ∂x ∂y
where S = sink/source term

Assume first-order decay
- decay rate is proportional to the amount of material present
dC
=
−kC =
S
dt
where C = mass/volume; S = mass/(volume‧time); k = 1/time = decay rate
6-3
- Rate of disappearance of BOD due to biodegradation
- Radioactive substance also decay in strength in this way
- Coliform bacteria and pathogens die away with a rate of first-order decay
∂ (hC ) ∂ ∂
+ (uCh) + (vCh) = ∇ ⋅ (hD∇C ) − khC
∂t ∂x ∂y
6-4
(2) BOD-DO
- Linked materials
- Behavior of one material depends upon the amount of another- Conc. of

dissolved oxygen depends not only on transport of DO but also on the amount
of BOD present
- Biodegradable substances undergo biochemical reactions
- Oxygen is used up in aerobic decomposition
6-5
(3) Heat transport

∂hT ∂ ∂
+ (uTh ) + ( vTh ) = ∇ ⋅ ( hD∇T ) + hS
∂t ∂x ∂y
dT qnet K (T − Te )
=
S = =
d t ρ cph ρ cph
K = atmospheric heat exchange coefficient ( W / m 2 °C )

c p = specific heat of water
Te = equilibrium temperature
= temperature that a body of water would reach if all meteorological

conditions were constant in time
Assume that u , v , h satisfy the continuity eq.

∂T ∂T ∂T 1 A
+u +v = ∇ ⋅ ( hD∇T ) + (T − Te )
∂t ∂x ∂y h ρ cp
qsw
T=
e Td +
K
Te = dew point temperature
6-6
(4) Suspended Solids
Suspended solids= suspended sediments + particulate forms in the water column
∂ (hC ) ∂ ∂
+ (uCh) + (vCh) = ∇ ⋅ (hD∇C ) + hS
∂t ∂x ∂y
Ws (Ceq − C )
S=
h
Ggr D ρ s
Ceq =
ρh
6-7
6.2 Analytical Solutions for Non-Conservative Pollutants
6.2.1 Steady state equations and solutions
- Pollutant discharges are nearly constant over time (continuous discharge)

- Concentration in the stream will reach a steady state condition soon after the
discharge starts.
∂C ∂C ∂ 2C
=
−u +E 2 +S
∂t ∂x ∂x
I. Conservative Pollutant
Item W/O Dispersion (Case 1) W/ Dispersion (Case 2)
∂C ∂C ∂ 2C
G.E. 0=u 0=
−u +E 2
∂x ∂x ∂x
C ( 0 ) = C0 C ( 0 ) = C0
B.C.'s
- continuous discharge - continuous discharge
C ( x ) = C0 , x ≥ 0 u
= C C0 exp( x ), x < 0
Solutions Qs Cws + QwCww E
C0 = =C C0 , ( x ≥ 0)
Qs + Qw
Qs = flowrate of the stream; Qw = flowrate of waste discharge
Cws = conc. of waste material in the stream flow
Cww = conc. of waste material in the waste discharge
Graphics
6-8
II. Non-conservative Pollutant

Item W/O Dispersion (Case 3) W/ Dispersion (Case 4)
∂C ∂C ∂ 2C
G.E. 0=
−u − kC 0=
−u + E 2 − kC
∂x ∂x ∂x
C ( −∞ ) = 0
C ( 0 ) = C0 C ( ∞ ) =0
B.C.'s W Qs Cws + QwCww C ( 0 ) = C0
C= =
Qs + Qw
0
Q
W 1
C0 =
Qα
( x ) C0 exp  (1 + α ) x  , x ≤ 0
u
C=
 2E 
( x ) C0 exp  (1 − α ) x  , x ≥ 0
k u
Solutions C ( x ) = C0 exp( − x ), x ≥ 0 C=
u  2E 
4kE
α= 1+
u2
Graphics
6-9
▪ Boundary condition at the outfall
1) Conservative substance
Consider mass balance at discharge point (outfall)
Mass rate of substance upstream + mass rate added by outfall

= mass rate of substance immediately downstream from outfall assuming
complete mixing (using midstream discharge of diffuser)
Qu su + Qe se =
Qs (a)
=
W Q=
e se input waste load [M/T] (b)
[Re] mass rate = flow rate × concentration
Flow continuity
Qu + Qe =
Q (c)
Combine (a) ~ (c)

Qu su + Qe se Qu su + W
=s =
Q Q
6-10
If su = 0
W
Then s =
Q
Thus, boundary condition at the outfall is give as

W + CuQu
=
C C=
0 at=
x 0
Q
2) Non-conservative substance
Derive general solution to the governing equation below

∂C ∂ 2C
0=
−u + E 2 − kC (6.32)
∂x ∂x
This G.E. can be regarded as second order ordinary differential equation since
C is a function of x .
It is a kind of characteristic equation.
∂C ∂ 2C
Thus, we try a solution of the form C = e , then
mx
= me and
mx
= m 2e mx .
∂x ∂x 2
Applying these equations into G.E. gives characteristic equation as follow.

0=
−u (me mx ) + E (m 2e mx ) − k (e mx ) =
e mx ( Em 2 − um − k )
u ± u 2 + 4kE u  4kE 
∴ m1 , m=
2 = 1 ± 1 + 
2E 2 E  u 2 
∴∴ C1 = e m1x and C2 = e m2 x
6-11
4kE u
Let α = 1 + 2
> 1 , then m1=
, m2 (1 ± α )
u 2E
These functions are linearly independent and hence form a fundamental set of
solution. Therefore, general solution of the form of
=C K1 exp(m1 x) + K 2 exp ( m2 x )
K1 and K 2 can be evaluated from B.C.'s.

Boundary conditions are
C = 0 at x = ±∞
=C C=
0 at x 0
For x ≤ 0
m2 ( −∞ )
0 K1e m1 ( −∞ ) + K 2 e
C == =K1e −∞ + K 2 e +∞ → K 2 =0
C = C0 = K1 + K 2 = K1 ∴ C0 = K1
For x ≥ 0
C=0 =K1e m1∞ + K 2e m2∞ =K1e +∞ + K 2e −∞ → K1 =0
C = C0 = K1 + K 2 = K 2 ∴ C0 = K 2
Solutions are
=C C0e m1x , x ≤ 0
=C C0e m2 x , x ≥ 0
To get C0 , Consider flux
6-12
 ∂C   ∂C 
= ( − JA ) x
W= −+ ( − JA ) x +=
−  uC0 − E  A +  uC0 − E A
0= 0
=  ∂x x 
0−=  ∂x x 0+ 
W  ∂C   ∂C 
∴=   − 
 ∂x  x
EA= 0−= ∂x  x 0+
x ≤ 0 : C = C0e m1x
∂C − ∂C
= C0 m1e m1x ; =
a t x 0= , C0 m1
∂x ∂x
x ≥ 0 : C = C0e m2 x
∂C + ∂C
= C0 m2e m2 x ; =
a t x 0= , C0 m2
∂x ∂x
W
∴ C0 m2 C0 ( m1 − m2 )
= C0 m1 − =
EA
u u u
= C0 (1 + α − (1 −=
α )) C0 [=
2α ] C0 α
2E 2E E
W 1 W 1
=
C0 =
uA α Q α
Note that
W 4kE
If k =
0 or E =→
0 C0 = ; α =1 + 2 =1
Q u
6-13
▪Comparison of initial mixing distance X m with decay distance X d
Initial mixing distance X m for steady flow (state)
X m = 0.4 uW 2 / ε t (a)
Distance required for decay of substance to a factor of e −1 (0.3678)

Xd = u / k (b)
Therefore, ratio is given as

Xd u /k ε
= = 2.5 t 2 (c)
X m 0.4 uW / ε t
2
kW
4Ek
Set α1 = (d)
u2
Eq. (5.19): E = 0.011 u 2W 2 / du * (e)
Eq.(5.6): ε t = 0.6 du * (f)
Thus, combining these equations gives

α1 = 0.0264 W 2 k / ε t (g)
Equating (c) and (g) yields

X d 0.06
=
Xm α1
6-14
Consider analytical solution for non-conservative substance with dispersion and

decay terms
 kx  2 
( x ) C0 exp  − 
C= ( ) 
α1 + 1 − 1  
 u  α1  (6.33)
 ux
= C0 exp  −
 2E
( )
α1 + 1 − 1  , x ≥ 0
Xd
If >1 ( Xd > Xm )
Xm
Then,
0.06  2  
> 1 → α1 < 0.06 →   ( α1 + 1 − 1 ≈ 0.985 ≈ 1
α1  α1  
From Eq. (6.33), this means that

 kx 
C ≅ C0 exp −  → solution for case w/o dispersion
 u
Therefore, there are two possibilities

(i) Xd > Xm
→ The longitudinal dispersion term has a negligible effect.
→ We may drop the dispersion term.
 kx 
→ We may use an analytical solution.
= C C0 exp − 
 u
(ii) X d < X m
→ The material decays before it mixes across the cross section
→ Eq. (6.32) is not a suitable model.
→ The concentration distribution must be computed numerically.
6-15
6.2.2 Time Variable Analysis
Time variable behavior of water quality in river downstream of an outfall to

describe
1) the downstream transport of peak in a waste-water discharge load;
2) an accidential spill of a chemical;
3) the day-to-day variation in water quality due to day-to-day changes in waste

load inputs.
◆ Three types
W/O Dispersion W/ Dispersion
Instantaneous Input Continuous Input
▪ No mixing in ▪ Stream dispersion spreads ▪ Rectangular input

the longitudinal the response temporally kE
▪ η = 2 →dimensionless
direction u
and spatially.
▪ Relative effects of
▪ Plug flow longitudinal mixing for a

▪ used to estimate the
nonconservative substance
transient effects due to
▪ Impulse input ① upland stream
inputs such as BOD from
remains its storm water overflow (by η ≈< 0.01 → not significant
shape as it pass from combined ② Main drainage rivers
progress sewer system) and batch η ≈ 0.01 ~ 0.5 → significant
downstream discharges of municipal ③ Large rivers
& industrial wastes η ≈ 0.5 ~ 1.0
6-16
Time variable equations & solutions

∂C ∂C ∂ 2C
=
−u +E 2 +S
∂t ∂x ∂x
S = −kC → first-order decay
Item W/O Dispersion W/ Dispersion
∂C ∂C ∂C ∂C ∂ 2C
G.E. =
−u − kC =
−u + E 2 − kC
∂t ∂x ∂t ∂x ∂x
Case Case 5 Case 6 Case 7
Continuous Input over
Instantaneous Input
=
C (=
0) C0 an interval of time
±Rectangular Input
B.C.'s W (t )
C0 =
Q M = Mδ ( x) C ( x = 0, t < τ ) = C0
C ( x = 0, t > τ ) = 0
C  kx 
C ( x, t ) = 0 ⋅ exp  − 
2  u 
M
C ( x, t ) =   x − u (t − τ ) (1 + η )
A 4π Et erf 
 
 4E (t − τ )
 kx 
W ( t − t *) exp  − 
 − ( x − ut )2 
Sol.  u  exp  − kt  x − nt (1 + η )  
C ( x, t ) =
Q  4 Et  −erf 
4 Et 
mean = ut
τ = time interval of input
variance σ = 4Et2
kE
η=
u2
6-17
6.3 Modeling BOD-DO Coupled System
6.3.1 Solutions of BOD-DO Coupled System
◆ Coupled system (BOD, DO)
- determination of dissolved oxygen concentrations downstream of a discharge

of BOD
◆ Oxygen Demand
= indirect measure of organic materials (= organic pollutants) in terms of the

amount of oxygen required to (completely) oxidize it.
COD - Chemical Oxygen Demand

BOD: CBOD - Carbonaceous BOD
NBOD - Nitrogeneous BOD
Organic matter + O2 → CO2 + H 2O
◆ Importance of dissolved oxygen (DO)
- Anaerobic conditions in a stream are indicative of extreme pollution

- Low dissolved oxygen concentrations have severe effects on the kind of biota
which inhabit the stream
◆ Sources and sinks of DO
(a) Sources
- Reaeration from the atmosphere
6-18
- Photosynthetic oxygen production

- DO in incoming tributaries
(b) Sinks
- Oxidation of BOD
- Oxygen demand of sediments of water body
- Use of oxygen for respiration by aquatic plants
dC
∴ = reaeration + (Photosynthesis-respiration) − oxydation of BOD
dt
− sediment oxygen demand ± oxygen transport (into and out of segment)
Let C = concentration of DO
L = concentration of BOD
(1) rate of utilization of DO by BOD
dL
= −k1 L → exertion of BOD
dt
= utilization of DO
= depletion of DO
k1 = deoxygenation coefficient
(2) reaeration from the atmosphere
= diffuse of oxygen into the stream rate of reaeration
∝ degree to which the water is unsaturated with oxygen
6-19
Let Cs = DO saturation concentration
then oxygen deficit, DOD = Cs − C
∴ rate of reaeration
dC
=
+ k2 (Cs − C )
dt
k2 = reaeration coefficient (1/T)
∴ Conservation equation for C
∂C ∂C ∂ 2C
=
−u + E 2 − k1 L + k2 (Cs − C )
∂t ∂x ∂x
= Cs − C
Let D
Then dD = −dC
∂D ∂D ∂2D
∴− = +u − E 2 − k1L + k2 D
∂t ∂x ∂x
∂D ∂D ∂2D
=
−u + E 2 + k1L − k2 D ⇒ G.E. for DO Deficit
∂t ∂x ∂x
For BOD
∂L ∂L ∂2 L
=
−u + E 2 − k1 L ⇒ G.E. for BOD concentration
∂t ∂x ∂x
6-20
Let reaction terms
S L = − kr L
=
S D kd L − ka D
in which
L = concentration of remaining BOD
D = dissolved oxygen deficit (DOD) = Cs − C
Cs = DO saturation concentration
C = actual DO concentraion
kr = BOD removal coefficient = kd + ks
ks = settling coefficient
kd = deoxygenation coefficient = biochemical degradation
ka = reaeration coefficient
G.E.: unsteady state
∂L ∂L ∂2 L
BOD =
−u + E 2 − kr L
∂t ∂x ∂x
∂D ∂D ∂2 D
DOD =
−u + E 2 + kd L − ka D
∂t ∂x ∂x
6-21
◆ Steady state, W/ Dispersion (Estuary)

∂L ∂2L
(i) BOD: 0 =
−u + E 2 − kr L → same as case 4
∂x ∂x
 u 
L= L0 exp  (1 + α r ) x  , x ≤ 0
 2E 
 u 
= L0 exp  (1 − α r ) x  , x≤0
 2E 
in which
W
L0 =
Qα r
4kr E
α=
r 1+
u2
∂D ∂2 D
(ii) DOD: 0 =
−u + E 2 + kd L − ka D
∂x ∂x
  u   u 
 exp  (1 + α r ) x  exp  (1 − α a ) x  
W kd   2E −  2E  , x ≤ 0
=
D  
Q ka − kr  αr αa 
 
  u   u 
 exp  (1 − α r ) x  exp  (1 + α a ) x  
W kd   2E −  2E  , x ≥ 0
=
D  
Q ka − kr  αr αa 
 
in which
4ka E
α=
a 1+
u2
6-22
Thomann & Mueller (1987)
6-23
6.3.2 Streeter-Phelps Equation
◆ Streeter-Phelps Equation (1925)
- no dispersion (river) E = 0
- steady state
∂L
BOD: 0 =
−u − k1 L
∂x
∂D
DOD: 0 =
−u + k1 L − k2 D
∂x
For BOD we have solution (Case 3)

k1 W k
∴
= L L0 exp( − = x) exp( − 1 x )
u Q u
B.C.: = D=
D (0) 0 Cs − C0
Solution:
k1  − ku1  x − ku2  x  k 
− 2  x
D( x)
= L0 e  
−e  
 + D0e   , x ≥ 0
u
(1)
k2 − k1  
6-24
◆ Critical deficit Dc (@ tc uptake of oxygen by BOD is just balanced by the

input of oxygen from atmosphere)
let x/u =t (= time of flow) time of travel
→ Then Eq. (1) becomes

k1
D (t )
= L0 e − k1t − e − k2t  + D0e − k2t , x ≥ 0 (2)
k2 − k1
→ tc may be found as
∂D 1  k  ( k − k ) D  
= = 0 ; tc ln  2 1 − 2 1 0  
∂t k2 − k1  k1  k1 L0  
k1
Dc = L0 e − k1tc ← from Eq.(2)
k2
◆ Modified Streeter-Phelps equation
- account for other processes
(1) BOD removal due to sedimentation
- non-oxygen-demanding
- reduces BOD w/o changing oxygen concentration
∂L
= − k3 L
∂t
6-25
(2) BOD addition due to scour from the bottom and surface runoff from the land
- rate of addition is assumed to be constant
(3) Oxygen use other than by aerobic biochemical oxygen demand in the water,
and oxygen addition other than through reaeration.
- net of these processes is constant rate A
◆ Solution :
 P  −( k1 + k2 ) t
Dt =
k1
 L0 −
k2 − ( k1 + k3 ) 
 e
k1 + k2 
{ − e − k 2t }
k1  P A
+  −  (1 − e − k2t )
k2  k1 + k2 k1 
Consider only k3 neglect A & P

Let k=
r k1 + k3 , kd = k1 , ks = k3 , ka = k2
Then
{ }
kd
=Dt L0 e − kr t − e − ka t + D0 e − ka t
ka − kr
[Remark]
∂L
G.E.: 0 =
−u − kr L BOD
∂x
∂D
0=
−u + kd L − ka D DOD
∂x
6-26
Ch 7. Mixing in Estuaries
Ch.7 Mixing in Estuaries
Contents
7.1 Introduction and Classification
7.2 The Causes of Mixing in Estuaries
7.3 Cross-sectional Mixing in Estuaries
7.4 Longitudinal Dispersion and Salinity Intrusion
7.5 One-Dimensional Analysis of Dispersion of Wastes
7-1
7.1 Introduction and Classification
○ Estuary: semienclosed coastal body of water which has a free connection with the open
sea and within which sea water is measureably diluted with fresh water derived from land
drainage (Pritchard, 1967)
- Mixing in estuaries: mixing in a flow driven by the slope of the tidal wave, by wind
stresses and by internal density variations + flow oscillations ՜ complex, unsteady, spatially
varying flow
[Cf]
Mixing in rivers: mixing in a flow driven by the slope of the water surface
Mixing in reservoirs: mixing in a flow driven by wind stresses and by internal density
variations
○ Hydrodynamic categories (Bowden, 1967)
1) Sharply stratified estuary: salt-wedge estuary
2) Partially stratified estuary: significant vertical density gradient
3) Well mixed estuary ← mixed by strong tidal action
7-2
○ Geomorphological categories
1) Coastal-plain estuary: formed by gradual drowning of a river system
2) Fjord type estuary: formed by glacial action
3) Bar-built estuary: formed by the closing off of an embayment by a sand bar
4) Rest
○ Analytical classification (Hansen& Rattray, 1966)
- method based on the vertical variation of salinity and strength of the internal density-
driven circulation
○ Geometrical classification
- method based on the ratios of length L, width W, and mean depth D
7-3
7.2 The Causes of Mixing in Estuaries
○ Mixing in estuary
- combination of small-scale turbulent diffusion and a larger scale variation of advective
mean velocities
[Cf] Mixing in rivers
•advective mean velocities: a set of steady stream lines
•small-scale turbulent diffusion: mass transfer between stream lines
•dispersion: caused by flow along different stream lines with different speeds
1) Turbulent fluctuations: fluctuations with a period of less than a few minutes
2) Diffusive transport: transport whose scale is larger than a few minutes
3) Advective transport: varying in time, space, and direction
- semidiurnal and diurnal tidal variations
- wind-induced variations of almost any period
- inertial frequency caused by earth’s rotation
- fluctuations of longer periods caused by monthly and longer term variation of tidal
cycle and by seasonal variations of meteorological influences and tributary inflows
- Flow is going in different directions at different depths
○ Separate mechanisms
① mixing caused by wind
② mixing caused by tide
③ mixing caused by river
7-4
7.2.1 Mixing Caused by the Wind
• Wind: - dominant source of energy in large lakes, the open ocean and coastal areas
- may or may not play a major role in estuary
• Breaking waves caused by wind have little to do with large scale dispersion
1) long, narrow estuary: - flow may be predominantly tidal
- wind has little chance to generate much current
2) wide estuary: - wind stresses can generate currents of considerable importance
• Effect of wind
① drag on the water surface: dispersion of oil spill is directly affected by the local wind
② wind induced currents: dispersion of dissolved substances
• Rotational current caused by a uniform wind blowing over a basin of variable depth
Line of action of the wind-

induced force is through the
centroid of the water surface.
Hence the line of action of the wind-induced

force passes on the shallow side of the center
of mass of the water, and a torque is induced
causing the water mass to rotate.
7-5
• Prediction of wind-induced current
- Depth-integrated 2D equations of motion
- Numerical solution (Leendertse, 1967): 2HT model (2D depth-averaged tidally varying
model) Circulation
by wind
7-6
7.2.2 Mixing Caused by the Tide
• Tide generates mixing in two ways:
① turbulent mixing due to turbulence generated by friction of tidal flow running over
channel bottom → Section 7.3
② large scale currents generated by the interaction of the tidal wave with the bathymetry
→ Shear flow dispersion → Ch. 4
Tidal pumping
Tidal trapping
7.2.2.1 Shear Effects in Estuaries and Tidal Rivers
Flow oscillation - Flow goes back and forth.
• Effect of oscillation on the longitudinal dispersion coeff. → Section 4.3
K  K 0 f (T ) (7.1)
where f (T ) is plotted in Fig. 4. 7.
T   T / Tc = dimensionless time scale for cross-sectional mixing
T  tidal period ∼12 hrs
TC  cross-sectional mixing time = W 2 /  t
K 0  dispersion coefficient if T  Tc
• For wide and shallow cross section with no density effects
K 0  Iu2TC (5.17)
where I = dimensionless triple integral  0.1 (Table 4.1)

7-7
K  0.1u2T 1/ T   f T    (7.2)
where the function 1/ T   f T    is plotted in Fig.7.4
Max = 0.8
In general T is fixed  12.5 hrs
W2
i) TC is small (narrow estuary) TC 
t
T
T   1
TC
 K is small
ii) TC is very large (very wide estuary)
T
T   1
TC
7-8
 K is smallest
TC
iii) T    1 : 1/ T   f T     0.8
T
 K  0.08u2T
Ch. 5
[Ex] T  12.5 hrs, u  0.3 m/s, u  0.2u

2 2
Maximum K  0.08  0.2(0.3)  (12.5  3600)

2
 60 m 2 /s
• Limitations of Eq. (7.2)
1) The channel must be relatively uniform over a long reach.
2) The channel must be much wider than it is deep.
3) The water must be of uniform density.
4) The shear flow appears to be the dominant mechanism for dispersion.← neglect all
other mechanism which can increase the dispersion coefficient
5) It gives a first estimate of the dispersion coefficient in constant-density portions of an
estuary.
→ Eq. (7.2) predicts much smaller dispersion coefficients in tidal flows than in similar
steady river flows.
2
- Potomac River (tidal reach): K = 6 ~ 20 m /s
2
- Missouri River (steady flow reach): K = 1,500 m /s
7-9
[Cf] Revisit of Eq. (4.55) - linear velocity profile in vertical direction
U 2h2
i) T  0.01TC K  0.0002( )  0.0002 K 0
240 D
ii) T  1.0TC K  0.8K 0
iii) T  10TC K  K0
[Example 7.1] Dispersion coefficient in a tidal slough
L= 20 km ; W=100 m ; h = 3 m; T = 12.5 hr
tidal range = 1 m; saline water all the way in the slough ← no fresh water inflow
1
(Sol) Since tidal range =1/3 of water depth, tidal excursion length   20 km  7 km
3
mean tidal velocity u  7 km / 6.25 hr  0.31 m/s
Assume u  0.1u  0.031 m/s ;  t  0.6hu   0.6  3 0.031  0.059 m2 /s
TC  W 2 /  t  100  / 0.059  170,000sec  47.22hrs

2
T   T / TC  0.27 ; From Fig.7.4, 1 / T   f T     0.5
K  0.1u2T (0.5)  0.1(0.2u 2 )(12.5  3,600)(0.5)  44 m 2 /s
7-10
7.2.2.2. Tidal Pumping
• Residual circulation:
- net, steady circulation
- resulting velocity field obtained by averaging the velocity over the tidal cycle
- no tidal cycle is identical
- cause of the residual circulation:
the earth’s rotation
interaction of the tidal flow with the irregular bathymetry
- circulation driven by the tide → additional to circulations driven by the wind
and the river → tidal pumping
- important part of the flow distribution that increases longitudinal dispersion
- usually calculated using the 2D numerical model
[Ex 1] Mersey Estuary (Bowden and Gilligan, 1971)
7-11
Flood channel: flood current is stronger than ebb current
Ebb channel: ebb current is stronger than flood current
[Ex 2] residual circulation at an inlet of estuary (Stommel and Farmer, 1952)
Inflow from river

S0
S0
i) flood (inward) flow: confined jet-like flow having width equal to width of the mouth, a
ii) ebb (outward) flow: sink flow coming uniformly from a semicircle of radius b
Assume volume of ebb flow equal to the volume of flood flow plus fresh water inflows from
river
1
V   b 2 d  aLd  Q f T (7.3)
2
where Q f = river inflow ; d = water depth; T = duration of tidal cycle
Consider salinity transport for steady state
1
aLdS 0   b 2 dS  ( S 0  S )abd (7.4)
2
7-12
where S = salinity of ebb flow ; S0 = ocean salinity

S / S0  1  Q f T / V  a 2Vd /   (7.5)
[Ex 3] Pumped circulation at South San Francisco Bay
- net flow around islands or in braided channels
- oscillatory tidal current flowing over an irregular bottom topography, such as a series
of shoals, induces residual vortices
- counterclockwise gyre predicted by a 2D model
[Ex4] Dutch Wadden Sea (Zimmerman, 1978)
Longitudinal dispersion coefficient resulting from the residual vortices ~ 800 m2/s
7-13
7.2.2.3 Tidal Trapping

[Cf] effects of
• Trapping of low velocity water along the sides of an estuary dead zones in
rivers
~ due to effects of side embayments and small branching channels
- The storage effects are enhanced by tidal action.
[Ex] Dutch estuaries analyzed by Schijf and Schonfeld (1953)
- They called “Storing basin” mechanism
- responsible for all diffusive salt flux
• Trapping mechanism Tracer cloud
Separation
of cloud
- The propagation of the tide in an estuary represents a balance between the inertia of the
water mass, the pressure force due to the slope of the tidal waves, and the retarding
force of bottom friction.
- In the main channel, tidal elevations and velocities are usually not in phase; high water
occurs before high slack tide and low water before lower slack tide.
- This is because of the momentum of the flow in the main channel, which causes the
current to continue to flow against an opposing pressure gradient.
7-14
- The side channel, in contrast, has less momentum and the current direction changes
when the water level begins to drop.
High slack
High water
tide
0 T/2 T
• Longitudinal diffusivity by trapping mechanism (Okubo, 1973)
Assume uniform velocity in the main channel; u  u0 cos t
Assume uniform distribution of traps along the sides
r = ratio of trap volume to channel volume
1/ k = characteristic exchange time between trap and main flow
K ru0 2
K  (7.6)
1  r 2k (1  r ) 2 (1  r   )
k
where K = effective longitudinal diffusivity; K  = longitudinal diffusivity in the main
chnnel
[Ex] Mersey estuary: u0  1.5 m/s ;   1.4  104 sec1; r  0.1 ; k 1  104 sec
 K  0.9 K   360 m 2 /s
[Cf] longitudinal dispersion coefficient
Shear flow dispersion: ~ 60 m2/s
Tidal pumping: 800 m2/s
Tidal trapping: > 360 m2/s
7-15
7.2.3 Mixing Caused by the River
River delivers a discharge of fresh water Qf.
Assume all the fresh water comes from a single upstream source.
i) Salt wedge estuary: Fig. 7.1 a)
If a river discharges into an estuary connected to a nearly tideless sea
→ the fresh water overrides the salt water
→ flows as a nearly undiluted layer into the sea
→ salt water intrudes underneath the fresh water layer in the form of a wedge
ii) Partially stratified estuary: Fig. 7.1 b)
If there is some tide
→ the wedge moves back and forth
→ the more the wedge motion the more kinetic energy is available to break down the
interface and turbulently mix the fresh and saline layers
→ ① The river is a source of buoyancy,
 gQ f
② The tide is a source of kinetic energy to overcome the buoyancy,
WU t3
where  = the difference in density between the river and ocean water;
Q f = discharge of fresh water; U t = the rms tidal velocity; W = the channel width
7-16
○ Estuarine Richardson Number, R

( ) gQ f

R
WU t3
i) R large → the estuary is strongly stratified and flow is dominated by density currents
ii) R small → the estuary is well mixed, density effects are negligible.
iii) 0.08  R  0.8 → transition from a well mixed to a strongly stratified estuary occurs
○ Isohalines for partially stratified estuary
Tidal elevation
- Isohaline = lines of constant salinity
- tends to become horizontal because that is the condition of a stratified water body at rest
- the sloping isohalines imply pressure gradients which will drive a current tending to
bring the isohalines to the horizontal.
→ The necessary currents are a flow landward along the bottom and seaward at the surface
→ Estuarine circulation (gravitational circulation)
baroclinic circulation: internal flows driven by density variations; river-driven flows
barotropic circulation: flows of constant density; tide-driven flows
7-17
• Well mixed estuary: Fig. 7.1 c)
~ isohalines are vertical
→ still has a horizontal density gradient
→ internal baroclinic circulation is driven by a longitudinal density gradient
Tidal elevation
◎ Analysis of density-driven circulation
- Hansen and Rattray (1965, 1966)
- circulation in a vertical two-dimensional plane assuming no variation across the
channel
- Assume steady flow
- tidal effect was to induce vertical and longitudinal turbulent mixing
→ Fig. 7.12:
vertical distribution of velocity and salinity ~ f (river flow, depth, width of channel)
- densimetric Froude number
Qf
Fm  1
(7.7)
A   /   gd  2
where A = cross-sectional area
7-18
S
 salinity difference between surface and bottom divided by the mean salinity
S
Qf
U S  residual velocity at the surface; U f 
A
7-19
  fraction of landward transport of salinity caused by all dispersion mechanisms other

than the density-driven circulation (Fischer, 1972)
[Ex] San Francisco Bay

Q f  100 m 3 /s, d  8 m, W  3,125 m,  0.025,

U f  0.75 m/s
 R  0.019, Fm  0.0029
From Fig. 7.12;
 S / S  0.08
→ salinity difference = 2 ppt; mean salinity = 25 ppt
  0.7
→ 70% of the salt balance in this bay is maintained by mechanisms other than density
driven circulations
→ Experimental results by Fischer & Dudley (1975), Fig. 8.16, give similar salinity
intrusion
◎ Transverse gravitational circulation: Fischer (1972)
- Limitations of Hansen and Rattray's analysis because of assumption of two
dimensionality
- The actual baroclinic circulation is complicated by the variation of depth across the
channel.
- The channel geometry turns a vertically 2-D circulation cell into a horizontal
circulation cell as shown in Fig. 7. 11b).
7-20
- The horizontal velocity gradients generally lead to much larger dispersion coefficients
than do vertical ones, because in real channels the widths are so much greater than the
depths.
○ Longitudinal dispersion coefficient for triangular channel of depth d and width W
2
5 g   d 6W 2
K  1.9  10   (7.8)
  x  E0  t
2
where E0  vertical mixing coefficient for momentum
[Ex] Mersey estuary:
K = 360 m2/s
◎ Laboratory studies of baroclinic circulation
① Waterways Experiment station in Vicksburg, MI
- conducted by Ippen and Harleman (1961)
② Delft Hydraulics Lab, The Netherlands
- Rigter (1973)
③ UC Berkeley
- By Abraham et al. (1975)
width of flume = 11 ft
length of flume = 600 ft
7-21
○ Relation of observed intrusion of salinity to estuarine Richardson number
• Modified estuarine Richardson number
 gQ f
R  (7.9)
 Wu *3
where u* = shear velocity

*
• Ordinate: K / du
→ shear velocity is used to include the effect of varying bottom friction
i f  K
LU
where Li = length of salinity intrusion
→ most results are from narrow rectangular flumes, where tide- and wind-driven mixing
mechanisms are absent. → minimum length
7-22
7.2.4. Synthesis and Summary
○ Three main causes of mixing
wind
tide
river
- Motions resulting from all three causes are superposed.
- The main cause may change from season to season, or even from week to week.
- Many estuaries change from partially stratified or salt wedge in the wet season to well
mixed in the dry season.
- A flood of a week or so may stratify a well-mixed estuary.
- A passing hurricane changes a stratified estuary into a well-mixed one.
→ The simple steady-state analysis is not adequate to explain these seasonal and
catastrophic events.
○ Practical engineering studies of estuaries
① Physical modeling
neglect certain mechanisms
② Numerical modeling
③ Analytical model (1-D approach)
→lump all of mixing mechanisms into a single longitudinal dispersion coefficient
7-23
▪ Longitudinal dispersion coefficient
- Based on an analysis of one mechanism at the neglect of others
Mixing mechanism K Range (m2/s)
Mixing Shear flow K  0.1u2T 1/ T   f T    < 60 (Maximum)
Caused by dispersion
the Tide Tidal 800 (Dutch Wadden
pumping Sea)
Tidal K ru0 2 >360 (Mersey

K 
trapping 1  r 2k (1  r ) 2 (1  r   ) estuary)
k
Mixing Caused by the 2 360 (Mersey

 g   d 6W 2
5
K  1.9  10  
  x  E0  t
2
River estuary)
7-24
7.3 Cross-sectional Mixing in Estuaries
7.3.1 Vertical Mixing
i) Constant-density tidal flow:
- Vertical mixing is caused by turbulence generated by bottom shear stresses.
 v  0.067du*
- u* varies from nearly zero at slack tide to a maximum at the time of highest velocity.
→ Use the average value of u*

○ Bowden (1967)
 v  0.0025dU a (7.10)
where U a  depth mean amplitude of the current
ii) Stratified flow
→ Turbulent mixing requires that some of the tidal energy be used to raise the potential
energy of the water column.
→ The most of the energy for mixing is extracted from the bottom and internal shear.
○ Munk and Anderson (1948)
3
 v   0 (1  3.33Ri ) 2
(7.11)
7-25
where  0 = value of  v for neutral stability
 u
Ri = gradient Richardson number  g ( ) /  ( )2
z z
[Cf] Richardson number = buoyancy/kinetic energy
○ Pritchard (1960)
 v  8.59  103 U t  z 2 (d  z ) 2 / d 3  1  0.276 Ri 

2
(7.12)
where U t = the rms tidal velocity
◎ Field data for v
○ Mersey estuary (Bowden, 1963)
 5 cm 2 /s at surface
v  2
 71 cm /s at middepth
→ These fit Munk & Anderson's formula, Eq. (7.11) reasonably well.
For Mersey estuary,
Ri = 0.1~1.0
 0 = 500 cm2/s ← Eq. (7.10)
v
 0.064 at Ri  0.5
0
v 1
  1
0 100 10
7-26
○ Duwamish waterway near Seattle (Partch & Smith, 1978)

Higher rate of mixing during
- saltwedge estuary ebb was caused by an
internal hydraulic jump.
 0.5 cm 2 /s during most of the cycle
v  2
 5 cm /s during the period of maximum ebb flow
 0 = 55 cm2/s ← Eq. (7.10)
v 1
  1
0 100 10
○ Blumberg (1975)
- numerical modeling
- suggest relationship between  v and Ri different in form from (7.11) and (7.12)
7.3.2 Transverse Mixing
i) For rivers, bottom-generated turbulence is a main cause of transverse mixing.
- For straight, rectangular section
  t  0.15du*
- For irregular section, much larger values are induced by sidwall irregularities and channel
curvature.
  t  0.6du*
ii) For estuaries
- Mixing mechanisms are very complex.
- All mixing mechanisms are in part transverse mixing mechanisms:
7-27
→ Flow into and out of traps
Pumped circulation in bays
Wind-driven gyre (Fig. 7.3)
Transverse baroclinic circulation (Fig. 11b)
○ Transverse mixing coefficient
= small-scale turbulent fluctuations
+ transverse shear flow dispersion due to whatever transverse velocity profile that is
caused by the superposition of all the mechanisms
◎ Measured values for t
i) Well-mixed estuary (constant density reaches)
○ Ward (1974, 1976)
- San Francisco Bay:  t  1.0du*
- Cordova Bay, British Columbia:  t  0.42du*
- Gironde estuary, France:  t  1.03du*
 0.44du*  slack tide

- Fraser estuary, British Columbia:  t  
 1.61du  ebb tide
*
○ Fischer (1974)
- Delaware estuary :  t  1.2du*
ii) Stratified-flow
○ Experiments by Sumer & Fischer (1977)
7-28
- Laboratory channel: W  3.5 m , trapezoidal section
- 2 sets of experiments uniform side
wavy side
→ The sidewall waviness generated vertical mixing on the wavy side, which in turn
established transverse density gradients which drove a transverse baroclinic circulation
as shown in Fig. 7.14.
→ The rate of transverse mixing was greatly enhanced in the case of the wavy wall.
→ Stratification in estuaries may greatly enhance the rate of transverse mixing by
driving transverse circulations.
7-29
7.4 Longitudinal Dispersion and Salinity Intrusion
○ 1D longitudinal dispersion:
- longitudinal dispersion of pollutants along the channel axis
- intrusion of ocean-derived salinity up the channel axis by dispersive mechanisms
→ combine the result of all mechanisms into a single dispersion coefficient K
→ applied to relatively long, narrow estuary: Delaware, US; Thames, UK
○ Salt balance in an estuary in steady state
S
UfS  K (7.13)
x
where U f  Q f / A  net downstream velocity caused by the freshwater discharge;
K = dispersion coefficient in which result of all mechanisms is combined
→ Eq. (7.13) states that the downstream advection of salt by the mean flow is in balance
with upstream transport by all other mechanisms.
[Re] Total rate of mass transport by advective flux and dispersive flux
 S 
q  U f S   K 
 x 
S q
Mass conservation equation:  0
t x
Combine (A) and (B)
S   S  S S   S 
 U f S  K   0 → U f  K 
t x  x  t x x  x 
7-30
7.4.1 Decomposition of the Salinity and Velocity Profiles Longer period

storm and seasonal
○ Divide the observed fluctuations in velocity and salinity into components fluctuations are
also important.
timewise variations: periodic variations at the frequency of tidal cycle
spacewise variations: caused by the variations of depth across a cross section, and by the
variation of cross sectional shape along the axis of the estuary
○ Decomposition by Fischer (1972)
7-31
The velocity and salinity at any point can be divided into four components
u ( x, y, z , t )  ua  uc ( x, t )  us ( x, y, z )  u( x, y, z, t ) (7.14)
S ( x, y, z, t )  Sa  Sc ( x, t )  S s ( x, y, z )  S ( x, y, z , t ) (7.15)
1) ua , Sa  average over the cress section and tidal cycle
1 T1 W d
T 0 A 0 0
ua  u  u dzdydt
Sa  S 
where < > = tidal cycle average; overbar = cross-sectional average
2) uc , Sc  cross-sectional averages at any point during the tidal cycle, minus the tidal
cycle averages
uc  u  ua
Sc  S  S a (7.16)
7-32
3) us , S s  the tidal cycle averages at any point, minus cross-sectionally averaged tidal
cycle averages
us  u  ua
S s  S   S a (7.17)
4) u, S   the remainder that are left over when the various averages are subtracted
from the observed velocity
○ Total transport of salinity through a cross section during a tidal cycle

M  A  uS  Q f S a  A  uc Sc  us S s   uS    (7.18)
uc Ss  0
where Q f = tributary discharge   uA 
 
→ M

 

uA Sa  A  uc Sc   us S s   u S  
 

mean advection dispersion
○ If the estuary is in steady state
– the salinity distribution is the same at the beginning and end of the tidal cycle
S
K  uc Sc  us S s  uS  (7.19)
x
i)  uc Sc  → tidal cycle correlation of the cross-sectional averages
7-33
uc  ucm sint
Thus, some researchers
assumed it was zero.
Sc  Scm cost
where   2 /T
T
 uc Sc  ucm Scm  sin t cos t dt  0
0
However, in real estuaries, peak salinity occurs before high slack water and the minimum
salinity before low slack water. → the correlation gives a net landward transport of salt →
trapping mechanism
ii) us Ss → residual circulation
- We still have no adequate way of predicting its effects in general
- The difficulty is that wind-driven gyres, bathymetric tidal pumping, and density-driven
currents all contribute, and observations may result from any combination thereof.
[Ex] Fig. 11b
residual baroclinic circulation – seaward in the shallow portions of a channel
tidal pumping/wind driven currents – opposite circulation
iii) u S  → the result of
oscillatory shear flow
random motions on time scales shorter than the tidal cycle
short-term variations in the wind
trapping mechanisms due to dead zones along the side of a channel
7-34
7.4.2 The Relative Magnitudes of the Terms
○ Some observations in real estuaries
→ The relative magnitude of the terms in Eq. (7.19) can be measured by establishing an
observation transect and measuring the velocity at all points in the cross section throughout a
typical tidal cycle.
① Dyer's data (1974)
Vellar estuary in India
Southampton Water and Mersey estuary in UK
② Murray et al. (1975)
Guayas estuary in Equador
○ us and u should be separated into transverse and vertical variations
us  ust  usv
u  ut  uv (7.20)
where ust = transverse variation of the vertical mean
u sv = vertical deviation from the vertical mean
Then, Eq. (7.19) becomes
us S s  uS   ust S st  usv Ssv   utSt    uv Sv  (7.21)
○ Table 7.1
- Vellar estuary: a strongly stratified estuary
→ vertical residual circulation dominates transverse or shear effects
7-35
→ trapping term  uc Sc  is also important
- Southampton Water: Partially stratified estuary
→ transverse and vertical residual circulation dominates
○ Guayas estuary: Fig. 7.17 ~ 7.19
- velocity and salinity distributions/residual velocity and salinity distributions
- relative contribution to upstream advective transport
transverse variation: 53%
vertical variation: 35%
cross-product terms: 12%
◈ General conclusion
Gravitational
current
1) Strongly stratified estuary
→ vertical residual circulation dominates the other cross-sectional variations
→ trapping mechanism is also important

Shape-induced
2) Partially stratified estuary current
→ transverse residual circulation becomes more important as stratification decreases
→ trapping mechanism is also important
7-36
7.4.3 Observed values of the Longitudinal Dispersion Coefficient
○ K are obtained by observing a longitudinal salinity gradient and fresh water outflow.
UfS
K
S / x
- The result depends on whether S is observed at high slack water, low slack water, or is an
average over the tidal cycle.
○ Table 7.2
- Many of the values are in the range of 100 ~ 300 m 2 /s which is notably smaller than
the values observed in moderately sized rivers (Table 5.3).
→ The reason is that the shear flow mechanism in rivers is limited in estuaries as
discussed in Sec. 7.2.2.1.
- In the constant density portions of estuaries, lower values of K are reported.
→ K are in the range of 10 ~ 50 m 2 /s
→ resulting from shear flow dispersion only; maximum ~ 60 m 2 /s
7-37
7-38
7.5 One-Dimensional Analysis of Dispersion of Wastes
○ For long, narrow, unstratified estuaries, 1-D analysis is enough.
→ 1-D analysis is a firmly established engineering tool because it is convenient, relatively
simple, and capable of giving practical answers.
→ In the 1-D model, the effect of cross-sectional variations and all the mixing mechanisms
are lumped into the longitudinal dispersion coefficient K .
7.5.1 Tidal Exchange at the Mouth
The longshore current deflects the ebb flow downcoast and

delivers the supply of new ocean water for the flood.
○ Volume of water entering an estuary during flood tide
= water that left the estuary on the previous ebbs + new ocean water
V f  V fe  V0 (7.25)
○ Tidal exchange ratio R
V0
R (7.26)
Vf
where V0 = volume of new ocean water entering the estuary during the flood tide;
7-39
V f = total water volume entering the estuary on the flood tide;
V fe  part of V f which flowed out of the estuary on the previous ebb
○ Total salt and water contents of the estuary are to remain constant
S f V f  SeVe (7.22)
V f  VQ  Ve (7.23)
S f V f  SeV fe  S0V0 (7.24)
Combine the above equations
S f  Se
R (7.27)
S0  Se
○ Accurate determination of average flood and ebb salinities requires complete cross-
sectional measurements of both salinity and velocity throughout the tidal cycle.
Te Te
Se    Su dAdt   udAdt (7.28)
0 A 0 A
where S and u are point values of salinity and velocity;
Te  duration of the ebb flow; A = time-varying cross-sectional area
Se VQ
R (7.29)
S0  S e V f
7-40
[Proof] Eq. (7.27) & (7.29)
1) Subtract (7.25)ⅹSe from (7.24)
S f V f  SeV fe  S0V0 (7.24)
SeV f  SeV fe  SeV0 (7.25)
S f  Se V f   S0  Se V0
V0 S  Se
 R f
Vf S0  S e
2) Substitute (7.22) into (7.27)
Ve
Se  Se
Vf Se  Ve 
R    1
S0  S e S0  S e  V f 
Substitute (7.23)
Se  Ve  V f  Se VQ
R   
S0  Se  V f  S0  Se V f
○ Measurements of tidal exchange at San Francisco Bay
- Nelson and Lerseth (1972)
- Salinity and velocity were measured throughout the tidal cycle at a number of points
on a transect at the Golden Gate.
- Fig. 7. 21: tidal exchange ratio ∝ flood tide range
- Numerical modeling showed that increasing the model values of R from 0.20 to 0.30
decreased the pollutant concentrations near the Golden Gate by 30%.
7-41
7-42
7.5.2 Tidal Exchange within the Estuary
○ A prediction of concentration at points up and downstream is needed when a given
loading of effluent is discharged at a given point.
→ use the 1-D analysis using the distribution of ambient salinity as a guide.
○ Salt balance in the estuary
Q0 S0  (Q0  Qe  Q f ) S (7.30)
where Q0 = circulating flow of ocean water;
Q f = tributary discharge from all tributaries upstream of the effluent discharge point;
Qe = effluent flow;
S0 =ocean salinity; S = salinity in the estuary
Rearranging Eq. (7.30) yields
(Qe  Q f ) S
Q0  (7.31)
S0  S
7-43
The total flow available for diluting the effluent is
(Qe  Q f ) So
Qd  Q0  Qe  Q f  (7.32)
S0  S
Assume complete mixing

where Qd = return discharge near the discharge point
in short time
○ Mean concentration of effluent near the point of discharge
M
Cd  (7.33)
Qd
where M = discharge rate of material (mass/time)
○ Tidal exchange analysis of Sec. 7.5.1 gives
Q0T  RV f
RV f
 Q0  (A)
T
Se VQ
R
Introducing Eq. (7.29) into (A) yields S0  S e V f
Se VQ Se
Q0   (Qe  Q f ) (7.34)
S0  Se T S0  Se
VQ
 QQ  Qe  Q f
T
→ discharge of river water
→ the same as Eq. (7.31) if S is taken to be the average ebb flow salinity, S e .
7-44
○ In bays with low tributary inflow, the salinity may be near oceanic
RP
Q0  (7.35)
T
where P = tidal prism  V f
○ R decreases rapidly as one moves upstream from the mouth of the estuary:
Ex: South San Francisco Bay
R = 0.076 → 0.039 → 0.031
[Ex. 7.2] Discharge to an estuary with tributary inflow
Qe  100 cfs (industrial STP)

Parts per thousand
Ce  10 ppt (toxic material)
Q f  1000 cfs (tributary inflow)
Se  19 ppt
S0  33 ppt
Estimate the average concentration of the toxic material in the estuary in the vicinity of the
discharge point.
Solution: First, use salinity data to obtain Qd
Qd  Q0  Qe  Q f  (Qe  Q f ) S0 / ( S0  S ) (7.32)
7-45
33
 (100  1,000)  2596 cfs
(33  19)
M
 Cd  (7.33)
Qd
  mass / time  Q C
where M e e
100(10)
 Cd   0.385 ppt
2596
[Example 7.3] Discharge to an estuary with no inflow; use of a dye study
Q f  0 (no tributary discharge)
Qe  100 cfs
Ce  10 ppt
Qdye  33 cc / min (continuous dye release for a period of 15 tidal cycle)

Parts per
Cdye  200,000,000 ppb billion
Cd dye  8.5 ppb
Assuming no dye decay and that 15 tidal cycle is sufficient to reach equilibrium, estimate
the concentration in the bay.
Solution:
First, for dilution of the dye, the dilution discharge is found by solving for Qd in Eq. (7.33)
7-46
M
Qd 
Cd
33 cc / min  2  108 ppb

  7.76  108 cc / min  459 cfs  Q0
8.5 ppb
→ This discharge can be used for effluent analysis.
M  M  10 ppt  100 cfs

 Cd  effluent  eff
 eff
  1.79 ppt
Qd Q0  Qe 459 cfs  100 cfs
○ For a conservative material,
i) Downstream of the outfall
→ the effluent continues to be diluted as it approaches the ocean, just as fresh water is
diluted.
S0  S x
C x  Cd , downstream (7.36)
S0  S d
ii)Upstream of the discharge point
→ concentration of the effluent is reduced in the same way that the salinity is reduced by
nixing with the tributary fresh water.
Sx
C x  Cd , upstream (7.37)
Sd
◎ Assumptions
(i) The salinity observed at some instant is representative of steady-state conditions.
7-47
(ii) The results are valid only for conservative substances
(iii) The results give cross-sectional averages, not peak concentrations.
[Example 7.4] Distribution along the estuary
The same conditions as for worked Example 7.2. Find the concentration downstream of the
outfall at a point where the salinity is 24 ppt and upstream of the outfall at a point where the
salinity is 5 ppt.
Sx up
 5 ppt; S x down
 24 ppt
33  24
Cx  0.385  0.248 ppt
down
33  19
5
Cx  0.385  0.101 ppt
up
19
7-48
7.5.3 Dispersion of Decaying Substances
○ 1-D analysis for dispersion of pollutants based on the time and space averaged equation
C C  C
A  Qf  ( KA )  (source / sink) (7.38)
t x x x
where the time derivative means the change per tidal cycle; K expresses the result of all
the mixing processes that occur within the tidal cycle; A = cross-sectional area at mean
tide.
→ Eq. (7.38) is a postulated model (empirical model) subject to verification.
○ Numerical modeling of Eq. (7.38)
- seven-segment model of San Francisco Bay by Cox and Macola (1967)
- Fig. 7.23
○ Analytical solution of Eq. (7.38)
- steady-state distribution
- a tracer undergoing first order decay
- a channel with constant cross section and dispersion coefficient
C  2C
Uf  K 2  kC (7.39)
x x
where k = rate coefficient
This equation has the two solutions: Sec. 5.5
 
C  C0 exp  x 1    1 
 (7.40)
7-49
Ufx 4 Kk
where x  ; 
2K Uf2
○ Suppose that a pollutant is introduced into an estuary at a rate M (mass/time) at a point
x  L , where x  0 is the mouth of the estuary, and x positive landward.
→ Since U f is a seaward velocity, x is negative everywhere in the estuary.
I.Case 1: estuary mouth is far from the discharge point
Boundary conditions:
C 0 at x  
(i) Solution for upstream of the pollutant source
 
C  C0 exp  x  L  1  1   
  (7.41)
where L  U f L / 2 K
(ii) Solution for downstream reach
 
C  C0 exp  x  L  1  1   
  (7.42)
Concentration at the injection point is given as
M 1
C0 
Qf 1
7-50

because M   kCAdx

○ In estuary, U f is often small and  often large. For case of tidal inlet with no fresh
water inflow, U f  0;  
M
C exp    x  L  k / K  (7.43)
A 4 Kk
II. Case 2: complete removal of the pollutant at the estuary mouth
Boundary condition: c  0 at x  0
c  0 at x  
(i) upstream solution is the same as Eq. (7.40)
(ii) downstream solution
C  C0   
  
exp  1  1   x  exp  1  1   x
  (7.44)
  
exp  1  1   L  exp  1  1   L
    
C0 (concentration at the injection point) is also affected by the loss at the mouth, and
is given by
C0  M

1  exp 2 L 1    (7.45)
Qf 1
7-51
Homework Assignment # 1
Due: 2 weeks from today
1. Derive Eq. (7.40).
2. Derive Eq. (7.44) & (7.45).
3. a) Solve Example 7.5.
b) Plot C-x curve for both cases with various values of K and k.
For example, K = 30, 60, 120, 240 m2/s; k = 0.1, 0.2, 0.4, 0.8 1/day.
7-52
[Example 7.5] Waste distribution in an estuary
A sewage treatment plant discharges BOD into an estuary.
M  2 kg/s of BOD from STP, k  0.2 /day
A  600 m2 , Q f  10 m3 /s (fresh water discharge), K  60 m2 /s
Plot the longitudinal distribution of mean concentration for two cases:
(a) L  30 km from the mouth; and (b) L  5 km from the mouth
Solution:
U f  Q f / A  10 / 600  1 / 60 m/s
1
x  U f x / 2 K    x / 2(60)  1.39  104 x (m)
60
 0.69 Caseb
L  U f L / 2 K  
 4.14 Case a
4 Kk 4  60m / s   0.2 / 86,400s 

2
 2  2
 1.99
Uf  1 
 m / s
 60 
Uf
 1.39  104
2K
1  1    1  1  1.99  2.729
1  1    1  1  1.99  0.729
7-53
i) Case (a) :
L is large enough that the BOD will almost completely decay before reaching the mouth.
→ use Eqs. (7.40) - (7.42)
•Upstream
U
  
C  C0 exp  x   L 1  1     C0 exp  f 1  1  
   2K
  x  L 
 C0 exp  3.79  104  x  L 
•Downstream
U
  
C  C0 exp  x   L 1  1     C0 exp  f 1  1  
   2K
  x  L 
 C0 exp 1.01  104  x  L 
M 1 2 kg / s 1
C0    0.116 kg/m 3  116 mg/l (ppm)
1   10 m / s
3
Qf 2.99
ii) Case (b):
Solution is affected by the nearness of the mouth.
→ use Eqs. (7.44) - (7.45)
•Upstream
 
C  C0 exp  x   L 1  1     C0 exp  3.79  104  x  L  
 
7-54
•Downstream
C  C0      
exp  1  1   x   exp  1  1   x 
 
  
exp  1  1   L  exp  1  1   L
    
exp 1.01  104 x   exp  3.79  104 x 
 C0
1.66  0.15
 
C0  116 mg/l 1  exp 2  0.69  2.99   105 mg/l
 
[Re]
1  
1   L  ( 0.729)( 0.69)  0.503
1  
1   L  (2.729)( 0.69)  1.883
 
exp  1  1   L    exp(0.503)  1.654
 
 
exp  1  1   L    exp( 1.883)  0.152
 
Case a)
Case b)
7-55
○ Concentration distributions obtained from Eqs. (7.40) - (7.45) are averaged
concentrations over the tidal cycle.
○ Actually, high concentrations build up near the outfall at slack tide because there is little
current to provide dilution.
→ Then, during the following flood or ebb, the peak formed at slack tide is advected and
dispersed.
○ 1-D model using tidally varying quantities (Li, 1974)
   c
t x
x 

 AC   uAC   Kt A   source/sink terms
x 
(7.46)
where K t = dispersion coefficient appropriate to a tidally varying analysis;
u = tidally varying cross-sectional mean velocity.
→ Li’s analysis is limited to oscillating flow in a uniform channel sufficiently narrow that
cross-sectional mixing is immediate.
○ In wide and irregular estuaries,
the cross-sectional mixing time is much longer than the tidal cycle.
→ use either tidally averaged model, Eq. (7.38) or 2-D numerical model
[Ex] Delaware Estuary near Marcus Hook
W = 1,200 m
Time for transverse mixing ( Tc  W /  t ) ~ 10 days

2
7-56
7.5.4 Calculation of the Flushing Time
○ Flushing time = mean detention time
~ mean time that a particle of tracer remains inside an estuary
○ Freshness = fraction of any sample of water that is pure fresh water
S0  S
f  (7.47)
S0
where S0 = salinity of pure ocean water
f= 1 for pure fresh water
0 for pure ocean water
○ Total amount of fresh water in an estuary between x  L and x  0 (mouth)
L
V   f dAdx (7.48)
0 A
○ Flushing time between x  L and x  0
V
Tf  (7.49)
Qf
T f = time required for the freshwater discharge (river), Qf to completely replace the
fresh water in the estuarine volume
○ Replacement time
~ time required for replacement of a tracer throughout a distance L landward from the
mouth in an estuary with dispersion coefficient K
7-57
~ time required for complete mixing between x  L and x  0
→ similar to cross-sectional mixing time in Sec. 5.1.3
L2
T  0.4 (7.50)
K
where T = time required for a slug of material initially concentrated at one end of basin
to reach an approximately uniform concentration throughout the basin;
L = length of the basin; K = dispersion coeff.
[Example 7.6] An estuary has a constant cross-sectional area.
A  10,000 m2
K  100 m2 /s (longitudinal dispersion coeff.)
Q f  30 m3 /s (fresh water inflow)
L  30 km  30,000 m
Find the mean detention time by Eqs. (7.49) and (7.50)
Solution:
○ Analytical solution for 1D equation of conservative pollutant for steady state (Ch. 6)
C  2C
0  U f K 2
x x
Uf
C  C0 exp( x)
K
7-58
○ Salinity distribution
 Q   x 
S  S0 exp    f   
  A   K 
  30   x   5
 S0 exp         S0 exp(3  10 x)
  10,000   100  
○ Freshness distribution
( S0  S )
f   1  exp(3  105 x)
S0
○ Volume of fresh water in the bounded estuary volume (between x  0 and L )
L
V f   fAdx
0
1  exp  3  105 x  dx

30,000
 A  
0
 1.02 108 m3
○ Flushing time by Eq. (7.49)
Vf 1.02  108 m3
Tf   2
 3.41  106 s  39.4 day
Qf 30 m /s
○ Replacement time by Eq. (7.50)
 30,000 m 
2
L2
T  0.4  0.4  3.6  106 s  41.7 day
K 100 m 2 /s
7-59
7.5.5 Uses and Limitations of One-Dimensional Analysis
○ Conditions required for 1-D model application
(1) The time scale for mixing across the estuary is significantly less than the time required
for the effluent to pass out of the estuary, or for the substance to decay.
W2
Tc  0.4
t
L
Tt 
Uf
Tc
 1
Tt
(2) The estuary is not significantly stratified, so that the effluent can be expected to mix
uniformly over the depth. If the estuary is strongly stratified, it is possible to use
separate one-dimensional analysis in each layer.
(3) Allowance is made in the analysis for the higher concentrations expected near the
source, before cross-sectional mixing takes place, and for distributed sources and
sinks in the case of a naturally occurring substances such as nutrients, dissolved
oxygen, etc.
[Example 7.7] Use of 1-D analysis for effluent discharge
An industry plans to discharge conservative constituent into San Pablo Bay (Fig. 7.23).
Ce  1,000 ppm (conservative pollutant from STP)
Qe  0.5 m3 /s
7-60
A  25,000 m2 (cross-sectional area)
d  8 m (mean depth)
W  3.125 m (mean width)
U t  0.75 m/s (rms tidal velocity)
u*  0.075 m/s (rms shear velocity)
Q f  100 m3 /s (tributary net outflow)
 24 ppt at the surface

S
 26 ppt at the bottom
S0  33 ppt
Find the resulting concentration in the bay.
Solution:
○ Dilution discharge computed by Eq. (7.32)
Qd 
Q e  Q f  S0
S0  S
33
 (0.5  100 m3 /s)  415 m3 /s
33  25
○ Tidal exchange ratio
T
tidal prism, P  U t A
2
Q0T 2(Qd  Qe  Q f ) 2(415  0.5  100)

R    0.034
P Ut A 0.75  25,000
7-61
○ Mean concentration in San Pablo Bay is computed by Eq. (7.33)
M
Cd 
Qd
0.5 m3 /s  1000 ppm

  1.2 ppm
415 m3 /s
○ Influence of stratification on vertical mixing
 gd
Ri 
 U t2
9.8 m/s 2  8 m
 0.0014   0.2
(0.75 m/s) 2
→ Ri is large enough to suggest to reduce vertical turbulent mixing.
 v  0.005du*  cf   v  0.067du*
 0.005  8 0.075 
 0.003 m 2 /s
○ Time for vertical mixing
d2
Tv  0.4
v
82
 0.4  8,500 s  2.5 hr
0.0003
7-62
○ Travel distance for effluent
Lt  U t Tv
 0.75  8,500   6,400 m  6.4 km
○ Transverse mixing
 t  0.6du *
 0.6  8 0.075  0.36 m2 /s
○ Transverse extent of effluent after 2.5hr
4 t  4 2 t Tv
 4 2  0.36  8500  313 m
○ Peak concentration computed by Eq. (5.7)
M
Cp  -line source of M
dU t 4 t  x / U t 
 / d in 2-D flow
-point source of M
0.5 m3 /s  1000 ppm


8 m  0.75 m/s 4  0.36 m 2 /s  8500 s
 0.42 ppm
7-63
○ Comparison of C d and C p
Cd  C p
where Cd = background buildup concentration of effluent over many tidal cycles
C p = peak concentration by instant source
○ Time required for complete transverse mixing.
W2
Tt  0.4
t
 3125
2
 0.4  125 day

0.36
→ Actual transverse mixing time is much less than 125 days because of larger scale
circulations.
7-64
Ch 8. Numerical Modeling
Ch.8 Numerical Modeling
Contents
8.1 River and Estuary Models
8.2 Model Building and Use
8.3 Finite Difference Method
8.4 Finite Particle (“Random Walk”) Model
8.5 Finite Element Method
8.6 Kinetic Models
8-1
8.1 River and Estuary Models
8.1.1 Considerations in Choosing a Model
Table 8.1 Types of Transport Models
Code Name Description
A numerical solution of 1-D tidally averaged dispersion

equation [Eq. (7.38)]
One-dimensional
1A ①steady state model: coefficients are constant in time.
tidally averaged
② unsteady model: flow parameters and dispersion
coefficient vary between tidal cycles.
One-dimensional A numerical solution of Eq. (7. 46)

1T Tidal evaluation, velocity and dispersion coefficient vary
tidally varying
during tidal cycle.
Branching 1-D
1TB A network of 1T models connected at junctions.
tidally averaged

2VA Two-dimensional
equation.
2HA tidally averaged 2V : horizontally averaged model
2H : vertically averaged model
2VT Two-dimensional
A numerical solution of 2-D tidally varying dispersion
2HT tidally varying equation
Three-dimensional
3A
tidally averaged equation
Three-dimensional
A numerical solution of 3-D tidally varying dispersion
3T
tidally varying equation
A small-scale physical replica of the prototype geometry

P Physical model
with provisions of generating tidal and river flows
8-2
Hybrid numerical
A combination of a physical and a numerical model, using
NP
physical one model to generate input information for the other
◎ Dispersion mechanisms to be replicated by models
Mixing Appropriate
Description
mechanism Model
2HT
Well verified for simulation of trapping mechanism
Trapping physical model
1TB Branches represent traps.
In case transverse gravitational circulation is

2VA
2VT not important.
Density-driven
If density-driven currents are important, the equations
circulation
3A
determining the flow and the salinity distribution
3T
are coupled.
2HT
Tidal pumping Accuracy of 2HT may be difficult to establish.
physical model
Shear flow 2HT

dispersion 2VT
2HT
Wind effects Fig. 8.1
3T
Rotational Easily modeled in 2HT models.

2HT
effects
Catastrophic/ 1A
Long term simulation for a period of a year of more
seasonal changes physical model
8-3
8.1.2 Numerical Models
8.1.2.1 One-Dimensional Models
• 1-D model is accurate in any case where the time scale of the process being studied is
substantially greater than the time scale for cross sectional mixing
→ In practical use of 1-D models, instantaneous complete cross sectional mixing is assumed.
(1) Finite Difference Models
• 1A (tidally averaged model)
∂C ∂C ∂  ∂C 
A + Qf =  KA  + source/sink (7. 38)
∂t ∂x ∂x  ∂x 
• 1T (tidally varying model)
∂ ∂ ∂ ∂C 
( AC ) + (uAC=
)  Kt A  + source/sink (7. 46)
∂t ∂x ∂x  ∂x 
• Numerical solution of 1-D Eq. of motion
→ method of characteristics solution by Streeter and Wylie (1967)
• Finite-difference representation of derivatives
(i) Explicit technique
- all the derivatives are expressed in terms of known values
(values of C at time level n)
- easier to program
8-4
① backward difference operator
∂C C j , n − C j −1, n
≈ (8.1)
∂x ∆x
② forward difference operator
∂C C j +1, n − C j , n
≈ (8.2)
∂x ∆x
③ central difference operator
∂C C j +1, n − C j −1, n
≈ (8.3)
∂x 2∆x
(ii) Implicit technique
- use some of unknown values of C at time level n + 1
- a set of simultaneous equations must be solved to obtain all the values at the new time level
at the same time
- implicit schemes are more stable and a longer time step can be used.
• central difference operator
∂C 1  C j +1, n +1 − C j −1, n +1 C j +1, n − C j −1, n 

≈  +  (8.4)
∂x 2  2∆x 2∆x 
○ Numerical diffusion
∙ Most numerical schemes induce unwanted numerical spreading
8-5
∙ Numerical diffusion is an apparent diffusivity caused by the numerical process.
∙ Numerical diffusion in Pure Advection Problem
① The mass represented by the concentration at a grid point is advected forward during a
time step a distance u∆t / ∆x grid points.
② Then, mass is divided between the two nearest grid points proportionally according to the
distance from each.
③ Division between the grid points is necessary because the numerical scheme has no way of
representing a concentration except at grid point.
④ A mass originally concentrated at one point is now spread numerically two points.
• Variance at the end of time step
2 2
∞  2  1 1   2 2
σ =
=∫−∞ ( x − x) cdx =∆ ( x)
x    +  ∆x    =∆
2 2 2

 3  3 3   3 9
1 2 1 2  1 ∆x 2
K′
= dσ /= ( ) 
∆ − =
∆
2
dt x 0 / x
2 2  9  9 ∆t
→ numerical diffusivity
8-6
○ How to control numerical diffusion
i) The numerical diffusion represented by K ′ must be kept much smaller than the actual
(real) dispersion represented by K .
ii) Bella and Grenny (1970) suggested that if the K ′ is forecast accurately and K ′ is less
than K , the value of K can be reduced accordingly.
K new= K − K '
• K ′ can be estimated by setting K = 0 in the numerical program and observing the results.
iii) Higher order scheme by Stone and Brian (1963)
- spread form forward difference for time derivative
∂C 1 
=  ( C j −1, n +1 − C j −1, n ) + ( C j , n +1 − C j , n ) + ( C j +1, n +1 − C j +1, n )  / ∆t
2 1
∂t x = j  6 3 6 
(8.6)
- Crank -Nicholson approximation for diffusive term
∂ 2C 1  C j +1,n+1 − 2C j ,n+1 + C j −1,n+1 C j +1,n − 2C j ,n + C j −1,n 

≈ +
∂x 2 2   (8.7)
∆x 2 ∆x 2 
• This scheme is the most accurate for the problems for which the diffusion coefficient is
relatively small.
• Stone and Brian's method can be used equally well for a tidally averaged or a tidally varying
analysis.
8-7
8.1.2.2 Multidimensional Models
• In multidimensional models, it is mostly important to understand and express
properly physics of flow and exchange.
• The mixing coefficients used in the numerical models express the net results of all processes
whose scale is less than the grid size of the model.
• In river and estuary models, turbulent mixing is smaller than the mixing caused by the
skewed shear flow of the velocity profile (Fig. 4.8)
• For numerical models which are averaged over at least one spatial dimension, over the tidal
cycle, or over both, the mixing coefficient represents what has been averaged.
◎ Two-dimensional models
in section models : horizontally averaged
in plan models : vertically averaged
2HA : tidally averaged model
2HT : tidally varying model
◎ 2HA models
∂C ∂C ∂C 1  ∂  ∂C  ∂  ∂C  
+U = +V   dK x  +  dK y  (8.11)
∂t ∂x ∂y d  ∂x  ∂x  ∂y  ∂y  
where ∂ / ∂t means a change per tidal cycle ; U , V are tidal averages of the vertical
averaged x- and y- direction velocities ; d is the local depth; K x , K y express the results of
all the mechanisms (shear flow mixing, pumping, trapping) that cause mixing within a tidal
cycle
8-8
∂ ∂C 
• Terms like 
∂x 
dK xy
∂y 
ought to be included, but usually not because it is difficult to
evaluate K xy and K yx .
• 2HA models should be used only in conjunction with extensive field data to define the
magnitude of the dispersion coefficients.
◎ 2HT models
∂C ∂C ∂C 1  ∂  ∂C  ∂  ∂C  
+U = +V   x dK  +  y
dK 
∂t ∂x ∂y d  ∂x  ∂x  ∂y  ∂y  
• 2HT models are in common use and have the advantage that they represent the important
dispersion mechanisms of trapping, pumping, and wind and Coriolis driven circulations.
• 2HT models appear to be practical for smaller bodies of water.
→ Leendertse model (1970) applied to Jamaica Bay
• If coarse spatial grid is used for large water bodies, advantages of replicating the tidal cycle
may be lost.
◎ Limitations of 2HT models
① The model should be operated to simulate at least as much real time as is needed to reach
an equilibrium distribution of tracer.
<Example> Tequil ≈ 100 days in large estuary
∆t =1 min
N (no. of time step) = 144,000
8-9
② The water column must not be sufficiently stratified to inhibit vertical mixing.
∂C ∂C ∂C 1  ∂  ∂C  ∂  ∂C  
Ⓐ: +U = +V   dK x  +  dK yt 
∂t ∂x ∂y d  ∂x  ∂x  ∂y  ∂y  
Where U t ,Vt = tidal velocity ; K xt , K yt = dispersion coefficient which represent only the
effect of the vertical velocity profile. (shear flow dispersion)
K xt , K yt << K x , K y
• Bigger mixing by tidal pumping and tidal trapping are now represented by the time-variable
advection
i. e., ut and vt
③ Since a time-varying flow field must be obtained from a first-stage 2HT flow model, the
flow model must produce the complex flows which leads to trapping, pumping, and other
dispersion mechanisms.
→ The residual circulations are caused by the nonlinear frictional and inertial terms in the
equations of motion.
→ Leendertse's (1967) model included nonlinear friction and inertial terms; however the no-
slip boundary condition was not imposed.
→ Tee's (1976) model incorporated the no-slip boundary condition.
→ Tee computed a residual circulation from boundary layer separation
8-10
8-11
8.1.3 Physical models
8.1.3.1 Introduction
• Physical model - physical representation at small scale
• Useful to the engineer, lay public, and political decision maker
◎ Distorted model to contain the model in a building of reasonable size
horizontal scale for river and estuary 1/1000
vertical scale 1/100
typical estuarine depth 5∼30 m
model depth 5∼30 cm
• Flow would be dominated by viscous and surface tension effects if model depth is less than
5cm.
→ The vertical exaggeration converts a typically wide and shallow cross section into the
more canyonlike cross section
→ The conversion serves the essential purpose of making the model flow turbulent, but it
also changes the longitudinal slope of the channels and distorts rates of vertical and
transverse mixing.
→ The tendency of the model flow to be too fast, because of the increased slope, must be
resisted by adding friction to the channels.
8-12
→ The vertical copper strips are arrayed over the entire channel to provide the extra friction
needed to counteract the distorted channel and water surface slopes.
→ During the construction process most models are calibrated against prototype observations
of tidal elevation and currents in the main channels.
→ These calibrations do not assure that mixing will be modeled correctly.
8.1.3.2 Model Laws and Scaling Ratios
• Fixed bed estuary models
Lp
Lr = length ratio =
Lm
dp
d r = depth ratio =
dm
Once the length and depth ratios have been selected, the ratios of all other quantities are
established by physical laws.
(1) Froude law
A frictionless small amplitude wave propagates at the correct velocity.
wave velocity c = gd
cm gd m dm
= =
cp gd p dp
cr = d r1/ 2 (8. 12)
time required for wave to propagate a distance L
t = L/c
8-13
Lr
=
tr = Lr d r −1/ 2 (8. 13)
cr
Lr
velocity ratio, u=
r Lr / t=
r = c=
r d r1/ 2 (8. 14)
Lr / cr
ur
=
Froude number, Fr = 1
gd r1/ 2
inertial force
Fr =
gravitational force
• The propagation of tidal and flood waves depends on gravitational, inertial, and frictionless
forces.
→ Froude law scaling assures the proper ratio of gravitational forces.
→ The copper strips are used to obtain the proper ratio of frictional forces.
• Density stratified flows
- The internal Froude number should be the same in the model and in the prototype to obtain
the correct ratio for internal wave velocities.
1
−
 ∆ρ  2
Fi = u  gd 
 ρ 
 ∆ρ 
∴  =1
 ρ r
• Other important ratios
slope ratio sr = d r / Lr
8-14
width ratio Wr = Lr
cross sectional area ratio=

Ar W=
r dr Lr d r
discharge ratio = A= Lr d r d= 1/ 2
Qr r ur r Lr d r 3 / 2
• Shear velocity ratio
=ur* (=
gds ) r1/2 d r / Lr1/2
• Mixing coefficient ratio
ε r = σ r 2 / tr
ε vr d=
= 2
r / tr d r 5 / 2 / Lr (8.15 a)
ε tr L=
= 2
r / tr Lr d r1/ 2 (8.15 b)
cf) ε= ∝ du *
εr
∴= ( du =
) *
r
dr 2 / Lr1/ 2 (8.15 c)
Eqs. (8.15 a, b) and (8.15 c) are quite different.
→ The turbulent mixing may not be modeled correctly.
<Example> The model of San Francisco Bay :
= =
Lr 1000 d r 100
= =
ur 10 Qr 1,000,000
= =
tr 100 ur * 3.13
8-15
= =
Sr 1/10 ( du* ) 313 r
• The model operator controls only the discharge ratio for the tributary inflows, the height
ratio and time ratio for the ocean tides, and the salinity ratio in the ocean.
• The actual elevations, currents, and salinities occurring throughout the model are
determined by the frictional characteristics of the model channels and the distribution of the
copper strips.
8-16
8.2 Model Building and Use
8.2.1 Definition of Model
○ What is a model?
Model = a deliberate misrepresentation of reality
(simplification) (real system)
approximation
Reason = convenience
Purpose = understanding (gain understanding)
prediction (predict an outcome)
Constraints = degree of simplification
degree of accuracy
8-17
• Real System
Simplification
Conceptual Model = set of assumptions
Physical Model Mathematical Model = Compact form of a set of equations
Governing Equation (Eq. of mass balance,
flux eqs, kinetic eq.)
+ Initial & Boundary conditions
+ Domain Geometry
+ Model Parameters
Analytical model Numerical model
① irregular shape of the domain's boundaries
② heterogeneity of the domain (coefficient)
③ irregular temporal and spatial distributions of various inputs
④ source and sink
8-18
8.2.2 Model Mechanisticity (Karplus (1976))
← Empirical Air Pollution Mechanistic →
(Black Box) (Gray Box) (White Box)
Social Econometric Standard Deterministic Process Models of
Model Watershed Water Quality Control Computers
Model Model
* Karplus, W.J. (1976) The future of mathematical models of water resources systems.
In “System Simulation in Water Resources”, pp11-18 North Holland Publishing Company
~ No model is either completely mechanistic or empirical.
~ All models are somewhere in between.
8.2.3 Modeling Procedure
◎ Models and Parameters
→ Model Parameters exist only in context of model. (Model Coefficients)
◎ Model Calibration = parameter tuning to fit observed data to predicted data.
Model becomes less mechanistic (more empirical)
= Parameter Identification Problem
Inputs → Model → Outputs
(Excitation) (Parameter) (Responses)
8-19
Prediction Problem
√ √ ?
(Forecasting)
√ ? √ Parameter Identification
(Estimation or Inverse)
? √ √
Signal Identification
○ Verification
= Establishment of the model validity by comparison between observed and predicted data.
○ Calibration and verification should be done with two separate (different) data sets
○ Procedure of Calibration and Verification
Data Set I Data Set II
Input I1 I2
Output O1 O1
Parameter ? P
(i) Calibration
I1 → Model(?) → O1 Fit O1 to O1
Find P (set of values of parameters (coefficients))
(ii) Verification
I 2 → Model (P) → O 2 Predict O 2 with calibrated parameter P
Compare O 2 to O2 to see if O 2 = O2
8-20
◎ Dimensionless Coefficients
E (Dispersion Coefficient) p126 Fischer et al
E / du *
= 0.1 ~ 0.4 m 2 /s Lab channel 180~400
0.76~1500 8.6~7500
See paper by Seo (1991)
◎ Best fit
~ techniques for determining the "best", or "optimal" values of the model coefficients,
i.e., values that make the predicted values and the measured ones sufficiently close to each
other.
◎ Calibration
~ to estimate parameters of model from available information
Measured Real System Measured END

Inputs → (unknown)→ → Outputs
O.K.
Comparison
NO
Measured Computed
Model
Input → → Outputs
Parameter
Corrector
Initial Parameter
Estimates
8-21
8.3 Finite Difference Method
8.3.1 Errors
Analytical Solution
= closed-form algebraic expression for temporal and spatial distribution of the constituent
~ easier to use than a numerical model
Numerical Solutions
Complex water body geometry and flow fields
Nonlinearities of the source / sink terms
→ make it impossible to obtain analytical solutions to the differential equation
→ solve using numerical techniques
○ Numerical techniques
~ simultaneous solution of a series of mass balances on a number of small fluid elements
~ matrix-inversion methods
8-22
◎ Source of error
Conceptual model
Conceptual errors
Physical Mathematical = G.E. + B.C.’s

model model
Parameter errors
(most frequent source)
Truncation & Roundoff
errors
Analytical Numerical
Solution Solution (=numerical model)
FDM FEM Special

Techniques
• Truncation error = discretization error
• Round-off error = error occurred in the arithmetic operations needed to solve FDE
8-23
8.3.2 Finite-Difference Methods
• Basic Relationships
①Break x ( y & z ) into finite segments of ∆x in length
②Subscript all variables and constants, Ci , U i , Ai , Ei ,… etc.,
such that i subscript indicates the value of variable or parameter at point i
③Apply Taylor Series expansions
∂Ci ∆x 2 ∂ 2Ci ∆x3 ∂ 3Ci

Ci += Ci + ∆x + + + Ο∆x 4 (a)
∂x 2 ∂x 3! ∂x
1 2 3
∂Ci ∆x 2 ∂ 2Ci ∆x3 ∂ 3Ci

Ci −= Ci − ∆x + − + Ο∆x 4 (b)
∂x 2 ∂x 3! ∂x
1 2 3
∂Ci ∂C
≡
∂x ∂x x =i
∆x 2 =( ∆x )
2
Ο∆x 4 = order of ( ∆x 4 ) and smaller
8-24
(i) Forward-difference
∂Ci Ci +1 − Ci ∆x ∂ 2Ci ∆x 2 ∂ 3Ci

(a): ≅ − − − Ο∆x3
∂x ∆x 2 ∂x

2
3!
 ∂x

3

Ο∆x ~ first-order error
(ii) Backward-difference = upwind difference
∂Ci Ci − Ci −1 ∆x ∂ 2Ci ∆x 2 ∂ 3Ci

(b): ≅ + − − Ο∆x 3
∂x ∆x 2 ∂x

2
3!
 ∂x

3

Ο∆x
(iii) Central-difference
Subtract (b) from (a)
∂Ci Ci +1 − Ci 1 2 ∂ 3Ci
≅ + ∆x − Ο(∆x 4 )
∂x 2∆x 3
 ∂ 3
x
Ο∆x 2 ~ 2nd-order error
(iv) Central-difference for 2nd derivative
Add (a) and (b)
∂ 2Ci Ci +1 − 2Ci + Ci −1
≅ − Q(∆x 2 )
∂x 2
∆x 2
Assembling a model
A. 1-D transient transport w/ dispersion, Conservative
∂C ∂C 1  ∂ ∂C 
+u =  EA 
∂t ∂x A  ∂x ∂x 
A, E ,U = f n ( x)
8-25
(1) Explicit Solutions
Superscript n - time step
Subscript i - distance step
(a) Formulation a
Forward-difference for time derivative → explicit
Forward-difference for 1st derivative in x
∂C Cin +1 − Cin
≈
∂t ∆t
∂C Cin+1 − Cin
≈
∂x ∆t
1 ∂ ∂C  Ei Ai (Cin+1 − Cin ) − Ei −1 Ai −1 (Cin − Cin−1 )

 EA ≈
A  ∂x ∂x  Ai ∆x 2
Substituting & rearranging
∆t E ∆t E A ∆t
Cin +1 =Cin − ui (Cin+1 − Cin ) + i 2 (Cin+1 − Cin ) − i −1 i −1 2 ( Cin − Cin−1 )
∆x ∆x Ai ∆x
Rearranging Further
 u ∆t E ∆t E A ∆t 
Cin+1 = 1 + i − i 2 − i −1 i −1 2  Cin
 ∆x ∆x Ai ∆x 
 E ∆t u ∆t  E A ∆t
+  i 2 − i  Cin+1 + i −1 i −1 2 Cin−1
 ∆x ∆x  Ai ∆x
8-26
Let
ui ∆t
= ai → Courant No. dimensionless numbers
∆x
Ei ∆t
= bi → Peclet No.
∆x 2
Ei −1 Ai −1 ∆t
= di
Ai ∆x 2
Then
+1
Cin= di Cin−1 + (1 + ai − bi − di ) Cin + ( bi − ai ) Cin+1
Note that
di + (1 + a i − b i − d i ) + ( b i − ai ) = 1
∴ Cin +1 is weighted average of Cin−1 ,
Cin and Cin+1
Solution
Boundary conditions : ① C known for all x @ t = 0
② C known for all t @ x = 0
Procedure : ① Use equation to get C11 , C21 , C31 , etc.
8-27
② Then get C12 , C22 , C32 on the basis of C11 , C21 , etc.
③ Continue as far in time as desired
(b) Formulation b
Forward difference for time derivative
Backward difference for spatial derivative
∂C Cin +1 − Cin
≈
∂t ∆t
∂C Cin − Cin−1
≈
∂x ∆x
1 ∂ ∂C  Ei +1 Ai +1 (Cin+1 − Cin ) − Ei Ai (Cin − Cin−1 )

 EA ≈
A  ∂x ∂x  Ai ∆x 2
Let's include the source/sink term in this time
Si = f ( c )
Substituting and rearranging
ui ∆t n E A ∆t
Cin +1 =Cin − (Ci − Cin−1 ) + i i +1 2 (Cin+1 − Cin )
∆x A i ∆x
Ei ∆t n
− (Ci − Cin−1 ) + f ( Cin ) ∆t
∆x 2
-
 ui ∆t Ei Ai +1 ∆t Ei ∆t  n  ui ∆t Ei ∆t  n
=
1 − − − C +
2  i
+ 2  i −1
C
 ∆x A i ∆x 2
∆x   ∆x ∆x 
8-28
Ei ∆t Ai +1 n
+ Ci +1 + f (Cin )∆t
∆x Ai
2
Ei +1 Ai +1 ∆t
Let = di
Ai ∆x 2
Cin+1 = (ai + bi )Cin−1 + diCin+1 + (1 − ai − bi − di ) Cin + f (Cin )∆t
Assume first-order decay
S = f ( c ) = −kc
∴ f (Cin )∆t =−ki ∆tCin
∴ Cin +1 = ( ai + bi ) Cin−1 + (1 − ai − bi − di − ki ∆t ) Cin + di Cin+1

Note that now ∑ Coeffs ≠ 1
Stability Problem → cause problem to explicit scheme only
Convergence
The numerical scheme is convergent
If for any fixed time T = n∆t and fixed location
X=
i∆t , C ( X ,T ) → C ( X ,T ) (o rC ( X , T ) − C ( X , T ) =
0)
as ∆t → 0 and ∆t → 0
in which C ( X , T ) = computed value at the fixed point X , T of the FDE
8-29
C ( X , T ) = exact solution to the PDE
Consistency
The FDE is consistent with the PDE if the local truncation error goes to zero as ∆x → 0
and ∆t → 0
Stability
The numerical scheme is stable if Ein remains bounded as n → ∞ for fixed ∆t
(t → ∞ or as computation proceeds)
in which Ein = roundoff errors

~
= C ( x, t ) − C ( x, t )
C ( x, t ) = computed value of FDE by computer
~
C ( x, t ) = exact solution to the FDE
◎ Lax Equivalence Theorem
Consistency + Stability → Convergence
◎ Analysis of Stability
Von Neumann Method
Matrix Method
8-30
◎ Source of Errors
Mathematical Model PDE C

↓ ↓ ← Truncation Error
Numerical Model FDE C
↓ ← Roundoff Error
Solution to FDE C
◎ Errors in machine computations Kuo (1970) p 313
Roundoff error = stem from a finite number of digits in a computer word
or from initial data
Truncation error = due to finite approximations of limiting processes
◎ Roundoff Errors
(i) Decimal-binary conversion error
~ computer converts decimal number to its binary equivalent
~ conversion error may be introduced because of
finite word length of computer particularly
if there is not exact binary equivalent
(ii) Non decimal-binary conversion error
~ if calculation requires more digits than available digits through a machine
(decimal computer)
8-31
(Examples)
1 3
1 1
(i) 0.625 =   +   = .101
2 2
1 3 10 16 17 21
1 1 1 1 1 1
=   +   +   +   +   +   + ⋅ ⋅ ⋅ (infinite series)
0.626
2 2 2 2 2 2
= .101000....
if binary machine has 20 bits available binary-decimal reconversion to a decimal equivalent
with 8-digit accuracy
→ 0.62599945 without rounding
0.62600040 with rounding
(ii) if decimal computer of capacity of 8 significant digits
0.33333333 0.33333333
+ 0.33333333 +0.33333333
add 3000 times +0.33333333
Expected value = 999.99999 +0.33333333
Rounded-off value= 999.99091 1.33333332 → 1.3333333
Roundoff error = 0.00908 8digit
True value Truncated to
Error = 0.00000002
8-32
◎ Stability Problem
○ Explicit Solutions
~ accurate & easy
~ may be unstable
→ need a stability criterion
○ Let C=
i
n
Ti n + ein (1)
Cin = computed value of FDE by computer
Ti n = true (exact) value to FDE at the x and t associated with i and n = C
ein = error at that point
Substitute this into formulation a)
ein+1 = (1 + ai − bi − di )ein + ( bi − ai ) ein+1 + di ein+1 ①
− Ti n+1 + (1 + ai − bi − ai ) Ti n + ( bi − ai ) Ti +n1 + diTi −n1 ② (2)
→ error for newly-calculated concentration depends not only on true concentration
(exact solution to FDE) (T-terms) but also on other errors (e-terms)
~ Part ② may not be zero because of truncated terms in formulating FDE out of PDE
PDE FDE Solution
Truncation error roundoff error
T-terms e-terms
8-33
But assume truncation error is zero, and worry only about the e-terms or propagation
(magnification) of roundoff errors.
To insure stability (to prevent magnification of errors)
ein +1 ≤ max  ein−1 , ein , ein+1  (3)
For Formulation a)
ein +1 ≤ { 1 + ai − bi − di + bi − ai + di } ein
1 + ai − bi − di + bi − ai + di ≤ 1 (4)
Absolute values of the coefficients should add to less than one.
Now, since ai , bi , di ≥ 0 , there are 4 possibilities.
α ) if 1 + ai − bi − di > 0 & bi − ai > 0
then 1 + a i − bi − di + bi − ai + d=
i 1 ≤1
which is satisfied for all values of ai , bi , and di
which meet these conditions.
β ) if 1 + ai − bi − di > 0 & bi − ai ≤ 0
1 + ai − bi − di − bi + ai + d i ≤ 1
2ai ≤ 2bi
ai ≤ bi
8-34
γ ) if 1 + ai − bi − di < 0 & bi − ai > 0
−1 − ai + bi + di + bi − ai + di ≤ 1
2di + 2bi − 2ai ≤ 2
di + bi − ai ≤ 1
if di = bi
then 2bi ≤ 1 + ai
δ ) if 1 + ai − bi − di < 0 & bi − ai ≤ 0
−1 − a/ i + b/ i + di − b/ i + a/ i + di ≤ 1
2d i ≤ 2
di ≤ 1 or bi ≤ 1 if bi = di
○ Restrictions on ∆x and ∆t
a≤b
8-35
u∆t E ∆t E
≤ → ∆x ≤ ①
∆x ∆x 2 u
2b ≤ 1 + a
E ∆t u∆t 1 2E u
2 ≤ 1 + → ≥ −
∆x 2 ∆x ∆t ∆x 2 ∆x
1
→ ∆t ≤ ②
2E u
−
∆x 2
∆x
Substitute ① into ②
1
∆t <
2E u
2 −
E E
   u 
u
1 1 E
∆t < 2
=
2 2
= 2
2u u u u
−
E E E
E
∆t < ③
u2
E
∴ ∆x ≤
u
E
∆t <
u2
1
∆t ≤
2E u
−
∆x 2 ∆x
8-36
○ Stability Criterion for Formulation (b) w/o decay term
1 − ai − bi − di + ai + bi + d i ≤ 1
there are only 2 cases to be considered
α ) if 1 − ai − bi − di ≥ 0
1 − a i − bi − di + ai + b i + di ≤ 1
1≤1
Satisfied for any values of ai , bi , and di
β ) if 1 − ai − bi − di ≤ 0
− 1 + ai + bi + di + ai + bi + di ≤ 1
2 ( ai + bi + di ) ≤ 2
ai + bi + di ≤ 1
if bi = di
ai + 2bi ≤ 1 ai ≤ 1 − 2bi
8-37
a + 2b ≤ 1
u ∆t E ∆t
+2 2 ≤ 1
∆x ∆x
1
∆t ≤
u 2E
+
∆x ∆x 2
γ ) Formulation EXCD
Central difference for spatial derivative
Forward difference for time derivative
∂C Cin+1 − Cin−1
≈
∂x 2∆x
∂C Cin+1 − Cin
≈
∂x ∆t
∂ 2 C Cin+1 − 2Cin + Cin−1

≈
∂x 2 ∆x 2
Substitute these into 1-D transport equation
∂C ∂C ∂ 2C
+u =
E 2
∂t ∂x ∂x
Cin +1 − Cin C n − Cin−1 C n − 2Cin + Cin−1

+ u i +1 =
E i +1
∆x 2∆x ∆x 2
 E ∆t  n  E ∆t u ∆t  n  E ∆t u ∆t  n
Cin +1 =
1 − 2 2  Ci +  2 −  Ci +1 +  2 +  Ci −1
 ∆x   ∆x 2∆x   ∆x 2∆x 
8-38
a   a
Cin +1 =  + b  Cin−1 + (1 − 2b ) Cin+1 +  b −  Cin+1
2   2
u ∆t
a=
∆x
E ∆t
b=
∆x 2
• Stability Criterion
a a
1 − 2b + + b + b − ≤1
2 2
a
α ) if 1 − 2b > 0 & b− >0
2
a a
1 − 2b + + b + b − ≤1
2 2
1≤1
a
β ) if 1 − 2b > 0 & b− <0
2
a a
1 − 2b + + b/ − b + ≤ 1
2 2
a − 2b ≤ 0
a ≤ 2b
a
γ ) if 1 − 2b < 0 & b− >0
2
8-39
a a
− 1 − 2b + + b + b − ≤1
2 2
1
4b ≤ 2 b ≤
2
a
δ ) if 1 − 2b < 0 & b− <0
2
a a
−1 + 2b + + b − b + ≤1
2 2
2b + a ≤ 2
a
b+ ≤1
2
a ≤ 2b
u ∆t E ∆t
≤2 2
∆x ∆x
E
∆x ≤ 2
u
1
b≤
2
E ∆t 1
≤
∆x 2 2
∆x 2
∆t ≤
2E
◎ Numerical Dispersion
Taylor series expansion
8-40
Ο∆
 
2
x
∂C Cin+1 − Cin−1 ∆x ∂ C
2 3
= + + Ο∆x3 (α ) cential
∂x 2∆x 3! ∂x 3
∂C Cin+1 − Cin ∆t ∂ 2C
= − −Ο∆x 2 ( β ) forward
∂t ∆t 2 ∂t 2
By the way think about 1-D transport equation w/o dispersion term (pure advection)
∂C ∂C
= −u ①
∂t ∂x
differentiate w.r.t x
∂ 2C ∂ 2C
= −u 2 ②
∂x∂t ∂x
differentiate ① w.r.t. t
∂ 2C ∂ 2C
= −u ③
∂t 2 ∂t ∂x
∂ 2C ∂ 2C
② = −u 2
∂x∂t ∂x
∂ 2C 1 ∂ 2C
③ = −
∂t∂x u ∂t 2
∂ 2C 1 ∂ 2C
∴ − u 2 =−
∂x u ∂t 2
∂ 2C 2 ∂ C
2
∴∴ =u ④
∂t 2 ∂x 2
8-41
Formulate ① with (α ) & ( β )
Cin+1 − Cin ∆t ∂ 2C Cin+1 − Cin−1

− =−u ( ) + Ο(∆t 2 , ∆x 2 )
∆t 2 ∂t 2
2∆t
Substitute ④
Cin+1 − Cin Cin+1 − Cin−1 ∆t 2 ∂ 2C

= −u ( )+ u +Ο(∆t 2 , ∆x 2 )
∆t 2∆t 2
 ∂x2
numerical dispersion term
∆t 2 u ∆x u∆t
Let =
En = u a =
(a = Courant No)
2 2 ∆x
= numerical dispersion coeff
Then
∂C ∂C ∂ 2C
+u − En 2 + Ο(∆t 2 , ∆x 2 ) =0
∂t ∂x ∂x
So, add En to physical dispersion coeff. E
E f= E + En
for numerical dispersion correction
2. Implicit Solutions
(1) Formulation (c)
8-42
∂C
Backward difference for
∂t
∂C
Forward difference for
∂x
∂C Cin − Cin −1
≈
∂t ∆t
∂C Cin+1 − Cin
≈
∂x ∆x
1 ∂ ∂C
2 { i i
E A (Cin+1 − Cin ) − Ei −1 Ai −1 (Cin − Cin−1 )}
1
( EA ) ≈
A ∂x ∂x Ai ∆x
u ∆t E ∆t E A ∆t
Cin − Cin−1 =
− i (Cin+1 − Cin ) + i 2 (Cin+1 − Cin ) − i −1 i −12 (Cin − Cin−1 )
∆x ∆x Ai ∆x
ui ∆t Ei ∆t Ei −1 Ai −1∆t n ui ∆t Ei ∆t n
(1 − + + )Ci + ( − )Ci +1
∆x ∆x 2 Ai ∆x 2 ∆x ∆x 2
Ei −1 Ai −1∆t n
− Cin −1
Ci −1 =
Ai ∆x 2
ui ∆t
let ai =
∆x
Ei ∆t
bi =
∆x 2
Ei −1 Ai −1∆t
di =
Ai ∆x 2
8-43
Cin−1
(1 − ai + bi + di )Cin + (ai − bi )Cin+1 − diCin−1 =
→ Cin−1 = n n n
weighted average of Ci , Ci +1 , and Ci −1
We need I.C., UBC, and DBC to solve system of algebraic equation
→ See Fig. in next page.
◎ let Li =−di ; M i =−
1 ai + bi + di , U i =ai − bi
then Cin−1
LiCin−1 + M iCin + U iCin+1 =
= x 0=(i 0)
(i) If C known @ 
x =∞ (i =
m + 1) C (0), C (m + 1)
→ Dirichlet (first kind) type B.C.
=i 1 : L1C0n + M 1C1n + U1C2n= C1n−1
→ M 1C1n + U1C2n =C1n−1 − L1C0n


Known
i= 2 : L2 C1n + M 2 C2n + U 2 C3n= C2n −1
=i m : LmCmn −1 + M mCmn + U mCmn=

+1 Cmn−1
Cmn−1 − U mCmn +1
→ LmCmn −1 + M mCmn =
  
Known
8-44
 M1 U1 0 . ... . 0  C1n−1 − L1C0n 

C  
n

L M2 U2 0 ... 0  1
 n   C2
n −1

 2  C2  
 0  n −1 
L3 M 3 U3 0 ... 0 C n   C3 
   3 
 0   =   ⋅ 
     ⋅ 
  C n   
 Lm−1 M m−1 U m−1   m −1
  Cm−1 n −1 
  Cm 
n  
 Lm Mm  Cmn−1 − U mCmn +1 
IC
UBC
DBC
→ All concentrations for one value of n are solved for simultaneously, and the solution
marches in time.
→ Implicit Solution
• Tridiagonal matrix → Gaussian elimination
Thomas Algorithm
(ii) If C known @ x=0 (i=0) → Dirichet
∂C
And known @ x =∞ (i =m + 1) → Neumann(2nd kind)
∂x
8-45
∂C
n
◎ No flux @ boundary → =0
∂x m +1
∂C Cmn +1 − Cmn
n
≈ =
0 Backward difference
∂x m +1 ∆ x
∴ Cmn +1 =
Cmn
=i m : LmCmn −1 + M mCmn + U mCm=

n
+1 Cmn−1
Cmn −1
∴ Lm Cmn −1 + ( M m + U m )Cmn =
 M1 U1 0 
C1  C1 − L1C0 
n n −1 n
L 
 2
M 2 U2 0
 C   C n−1 
   2   2 
 ⋅   ⋅ 
   
    =
 ⋅   ⋅ 
   
  Cm−1  n
Cm−1 −1
 Lm−1 M m−1 U m−1     

 Cm   Cm n −1 
 0 Lm M m + U m  
8-46
8.3.3 Boundary Conditions
Dirichlet (1st Type) C ( x, y ) = f1 ( x, y ) on T1
∂C
Neumann (2nd Type) = f 2 ( x, y ) on T2
∂n
∂C ∂C ∂C
= derivative normal to a boundary = or
∂n ∂x ∂y
∂C
Mixed (3rd Type) a + bC =
f 3 ( x, y ) on T3
∂n
(2) Formulation d ~ most commonly used formulation
∂C
∂t
∂C
∂x
≈
∂t ∆t
≈
∂x ∆x
1 ∂ ∂C 1
( EA ) ≈  E A (Cnn+1 − Cin ) − Ei Ai (Cin − Cin−1 ) 
2  i +1 i +1
A ∂x ∂x Ai ∆x
8-47
Cin −1
(1 + ai + bi + di )Cin − (ai + bi )Cin−1 − di Cin+1 =
E i +1 Ai +1 ∆t
where di =
Ai ∆x 2
let Li =
−(ai + bi )
M i = (1 + ai + bi + di )
U i = −di
then Cin −1
Li Cin−1 + M i Cin + U i Cin+1 =
(3) Formulation Im-Cd
∂C
∂t
∂C
Central difference for
∂x
≈
∂t ∆t
∂C Cin+1 − Cin−1
≈
∂x 2∆x
8-48
Final discretized equation results in
Cin −1
U i Cin−1 + M i Cin + U i Cin+1 =
ai
where U i =− − bi
2
M i = 1 + 2bi
ai
U=
i − bi
2
Ei ∆t
bi =
∆x 2
= artificial viscosity, numerical dissipation
= smearing of concentration fronts due to excessive damping
= Taylor's series truncation error
<Ref>
Lantz, R.B., "Quantitative evaluation of numerical diffusion (truncation error)," Soc. Pet.
Engr. J., pp.315-320, Sept., 1971.
◎ Formulation b
Taylor series expansion in x direction
8-49
∂Ci ∆x 2 ∂ 2Ci
C = C − ∆x
n
i −1
n
+ − Ο(∆x3 )
∂x 2 ∂x
i 2
∂C ∆x 2 ∂ 2Ci
(C = C + ∆x
n
i +1
n
+ + Ο(∆x3 ))
∂x 2 ∂x
i 2
Cin − Cin−1 ∂Ci ∆x ∂ 2Ci

= − − Ο(∆x 2 ) ①
∆x ∂x 2 ∂x 2
n +1 ∂C ∆t 2 ∂ 2C
C = C + ∆t n
+ + Ο(∆t 3 )
∂t 2 ∂t
i i 2
Cin+1 − Cin ∂C ∆t ∂ 2C
= + + Ο(∆t 2 ) ②
∆t ∂t 2 ∂x 2
Consider the 1-D transport equation with no dispersion term (pure advection)
∂C ∂C
= −u
∂t ∂x
Cin+1 − Cin Cin − Cin−1

Formulation b = −u ③
∆t ∆x
Cin+1 − Cin ∂Ci ∆x ∂ 2Ci

Ex a = + + Ο(∆x 2 )
∆x ∂x 2 ∂x 2
Cin − Cin−1 ∂C0 ∆x ∂ 2C0

Ex b = − − Ο(∆x 2 )
∆x ∂x 2 ∂x 2
Cin+1 − Cin−1 ∂Ci ∆x 2 ∂ 3C

Ex c = − − Ο(∆x 4 )
2∆x ∂x 3 ∂x 3
8-50
Im a
∂C ∆t 2 ∂ 2Ci
C = C − ∆t
n
i −1
n
+ − Ο(∆t 3 )
∂t 2 ∂t
i 2
Cin − Cin−1 ∂C ∆t ∂ 2C
= − + Ο(∆t 3 )
∆t ∂t 2 ∂t 2
differentiating ③ w.r.t. t
∂ 2C ∂ 2C
= −u ④
∂t 2 ∂t ∂x
differentiating ③ w.r,t x
∂ 2C ∂ 2C
= −u 2 ⑤
∂x∂t ∂x
∂ 2C 1 ∂ 2C 
= −  1 ∂ 2C
∂t∂x u ∂t 2
∂ 2C
− =
−u 2
∂ 2C ∂ 2C ∂ 2C  u ∂t ∂x
2
= = −u 2
∂x∂t ∂t∂x ∂x 
∂ 2C 2 ∂ C
2
∴ 2 = u ⑥
∂t ∂x 2
Substituting ① and ② into ③
∂C ∆t ∂ 2C  ∂C ∆x ∂ 2C 
* + = −u  − 2 
+ Ο(∆t 2 + ∆x 2 )
∂t 2 ∂t 2
 ∂x 2 ∂x 
8-51
Substituting ⑥
∂C ∂C  u∆x ∆t 2  ∂ 2C
= −u + − u  2 + Ο(∆t 2 + ∆x 2 )
∂t
 ∂x  2  ∂x 
2   
Truncation error
PDE
 
Numerical dispersion

Total error
Define numerical dispersion coefficient
u ∆x u ∆t u ∆x
En = (1 − )= (1 − a )
2 ∆x 2
u ∆t
=a = Courant No
∆x
∂C ∂C ∂ 2C
Then =
−u + En 2
∂t ∂x ∂x
If we include real dispersion term
∂C ∂C ∂ 2C
+u = ( E + En )
∂t ∂x    ∂x 2
Ec = Computed dispersion
◎ How to remove En
(i) Choose and ∆t and ∆x such that En = 0
u ∆t
E=
n (1 − a=
) 0
2
u ∆t
∴ a= = 1 ①
∆x
8-52
However, stability criterion for Formulation b is
u ∆t 2 E ∆t
+ ≤1 ②
∆x ∆x 2
u ∆t
If we make =1
∆x
then ② becomes
E ∆t
≤0 ③
∆x 2
Therefore we have to choose and ∆t and ∆x
Satisfying both ① & ③ → impossible
<Example>
u =1
∆x = 2 , ∆t = 1
1 ⋅1 1 1(2)  1  1
a= = En = 1 −  =
2 2 2  2 2
1 2 (1) 1 E
+ E = + ≤1
2 ( 2 )2 2 2
E 1
≤
2 2
1
E≤
4
8-53
(ii) Dispersion correction technique
→ make Ec = E
For Formulation b, subtract En from Ec
∴ Ec′ =E + En − En = E
(iii) make ∆x and ∆t small
◎ Ec= E + En
Formulation Numerical dispersion, En Effective solution to ND
u∆x
Ex a − (1 + a ) Add (− En )
2
u∆x
Ex b (1 − a ) Subtract En ( Be careful when E < En )
2
u2
Ex cd − ∆t ( No numerical Add (− En )
2
dispersion due to advection)
u∆x
Im c (1 − a ) Make a =
1
2
u∆x
Im d (1 + a ) Subtract En
2
u2
Im cd ∆t Subtract En
2
◎ Lagrangian Formulations
∂C ∂C 1 ∂ ∂C
=
+u ( EA ) + S
∂t ∂x A ∂x ∂x
(1) Formulation e -explicit
8-54
∂C
Forward difference formula at the i+1 grid point for
∂t
∂C Cin++11 − Cin+1
≈
∂t ∆t
∂C
Forward difference formula for
∂x
∂C Cin+1 − Cin
≈
∂x ∆x
Eulerian formulation for second derivative
1 ∂ ∂C
2  i i ( i +1
 E A C n − Cin ) − Ei −1 Ai −1 ( Cin − Cin−1 ) 
1
( EA ) ≈ 
A ∂x ∂x A; ∆x
Substituting into Governing eq.
ui ∆t n E ∆t
Cin++11 = Cin+1 −
∆x
( Ci +1 − Cin ) + i 2 ( Cin+1 − Cin )
∆x
−
Ai ∆x 2 ( Ci − Cin−1 ) + Si ∆t
Rearranging further
ui ∆t E i ∆t n ui ∆t E i ∆t Ei −1 Ai −1∆t n
Cin++11 =
(1 − − )Ci +1 + ( − − )Ci
∆x ∆x 2 ∆x ∆x 2 Ai ∆x 2
+ Ci −1 + Si ∆t
Ai ∆x 2
8-55
ui ∆t
Let ai =
∆x
Ei ∆t
bi =
∆x 2
Ei −1 Ai −1∆t
di =
Ai ∆x 2
then Cin++11 = (1 − ai + bi )Cin+1 + (ai − bi − di )Cin + diCin−1 + Si ∆t
• We need 2 UBC and IC, need no DBC
Pure Advection Problem
Cin++11 − Cin+1 Cin+1 − Cin

= −u ① Formulation e
∆t ∆x
Taylor Series Expansion in t direction
+1 ∂C ∆t 2 ∂ 2C
Cin=
+1 Cin+1 + ∆t + + Ο(∆t 3 )
∂t 2 ∂t 2
Cin++11 − Cin+1 ∂C ∆t ∂ 2C
→ = + + Ο(∆t 2 ) ②
∆t ∂t 2 ∂t 2
Taylor Series Expansion in x direction
8-56
∂C ∆x 2 ∂ 2C
C = C + ∆x
n
i +1 +
n
+ Ο(∆x3 )
∂x 2 ∂x
i 2
Cin+1 − Cin ∂C ∆x ∂ 2C
→ = + + Ο(∆x 2 ) ③
∆x ∂x 2 ∂x 2
Substitute ② & ③ into ①
∂C ∆t ∂ 2C  ∂C ∆x ∂ 2C 
+ + Ο(∆t ) = −u 
2
+ + Ο(∆x 2 ) 
∂t 2 ∂t 2
 ∂x 2 ∂x 2

∂C ∂C ∆t ∂ 2C ∆x ∂ 2C
= −u − −u + Ο(∆x 2 + ∆t 2 )
∂t ∂x 2 ∂t 2
2 ∂x 2
∂ 2C
u2
∂x 2
∂C ∂C u ∂ 2C
∴ = −u − ∆x(1 + a) 2 + Ο(∆x 2 + ∆t 2 )
∂t ∂x 
2  ∂x
En
u∆x
∴ En = − (1 + a )
2
◎ Stability Criteria
1 − ai + bi ≥ 0 & a − bi − di ≥ 0
↓ ↓
∆x 2 2E
∆t ≤ ∆x ≥
u ∆x − E u
(2) Formulation f - Implicit
8-57
∂C
Backward difference for at the i-l grid point
∂t
∂C
∂x
∂C Cin−1 − Cin−−11
≈
∂t ∆t
≈
∂x ∆x
( ai + di − 1) Cin−1 + ( −ai − bi − di ) Cin + biCin+1 + Si ∆t =Cin−−11
n We need 1 IC &
2 DBC
n +1 No UBC
i −1 i i +1
u ∆x
=
En (1 + a ) ~ Im d
2
◎ "Two-Step" techniques
~ Advection is "tracked" to a new set of grid points and dispersion follows separately
◎ Lagrangian approach (Bella & Dobbins, 1968)
~ Observer is traveling at the same speed as the parcel of water under observation
8-58
• Two-step explicit method → Two processes are assumed to occur sequentially rather than
simultaneously as in the prototype.
(1) 1st step (advection process) : to advect the pollutant downstream for one-time step
→ Eulerian Frame Cin = Cin +1
C0n = C0n +1
(2) 2nd step (dispersion process) : to calculate new values on the n+1 row using only the
dispersion
Cin +1 − Cin E
→ Lagrangian Frame = (Cin−1 − 2Cin + Cin+1 )
∆t ∆x 2
E ∆t n
Cin +1 =Cin + (Ci −1 − 2Cin + Cin+1 )
∆x 2
◎ Crank-Nicholson Scheme
(1) Upwind (Backward) → Formulation CN-b
Cin+1 − Cin u E
− (Ciε − Ciε−1 ) + 2 (Ciε+1 − 2Ciε + Ciε−1 )
=
∆t ∆x  ∆x
B. D.
ε =n → Explicit 4Point scheme
ε= n + 1 → Implicit
1
ε= n + → Crank-Nicholson
2
8-59
1
n+ 1 n
=
Ci
2
(Ci + Cin +1 )
2
Cin +1 − Cin u 1 1 
∴ −  (Cin + Cin +1 ) − (Cin−1 + Cin−+11 ) 
=
∆t ∆x  2 2 
E 1 n 1 1 
+ 2 
(Ci +1 + Cin++11 ) − ⋅ 2(Cin + Cin+1 ) + (Cin−1 + Cin−+11 ) 
∆x  2 2 2 
 E ∆t u∆t  n+1 u∆t E ∆t n+1 E ∆t n+1

 −  Ci −1 + (1 + + )Ci − Ci +1
 2∆x 2∆x  2∆x ∆x 2 2∆x 2
2
 E ∆t u∆t  n u∆t E ∆t n E ∆t n
=  +  Ci −1 + (1 − − )Ci + Ci +1
 2∆x 2∆x  2∆x ∆x 2 2∆x 2
2
[ A=
]{C} [ B ]{C}n + {b}
n +1
[ A] , [ B ] → Tridiagonal Method
u u u 6 Point scheme
n +1
3 knowns @ time level n
k k k 3 unknowns @ time level n+1

n
C-N method → Ο(∆x + ∆t 2 )
Fully Implicit → Ο ( ∆x + ∆t )
8-60
∂C
(2) Central difference for → Formulation CN-cd
∂x
u∆t ε E ∆t
Cin+1 − Cin =
− (Ci +1 − Ciε−1 ) + 2 (Ciε+1 − 2Ciε + Ciε−1 )
2∆x   ∆x
C .D
ε =n Explicit
ε= n + 1 Implicit
1
ε= n + C-N
2
u ∆t  1 n n +1 
 (Ci +1 + Ci +1 ) − ( Ci −1 + Ci −1 ) 
1 n
Cin +1 − Cin =
− n +1
2∆x  2 2 
E 1 n 
+ 
∆x  2
2 ( Ci +1 + C n +1
i +1 ) −
1
2
⋅ 2 ( Ci
n
+ Ci
n +1
) +
2
(
1 n
Ci −1 + Cin−+11 ) 

 E ∆t u∆t  n+1  E ∆t  n+1  u∆t E ∆t  n+1

 −  Ci −1 + 1 + 2  Ci +  − 2  i +1
C
 2∆x 4∆x   ∆x   4∆x 2∆x 
2
 E ∆t u ∆t  n  E ∆t  n  E ∆t u ∆t  n
=  +  Ci −1 + 1 − 2  Ci +  −  Ci +1
 2∆x 4∆x   ∆x   2∆x 4∆x 
2 2
◎ Models based on Solutions to Ordinary Differential Equations
Consider transient zero dimensional problems (Box model)

(Q, C )in
↓
dc
=
S + Q(Cin − Cout )
dt
C (t )
→ (Q, C )out
8-61
→ Initial value problem
→ Solution marches forward in time
(1) Euler Method (Explicit method)
dc C n +1 − C n
≈ σ (∆t )
dt ∆t
then
C n +=
1
C n + ∆t [ S + Q(Cin − Cout ) ]
in which S = S (C n )
∴ Choose ∆t , march forward in time
8-62
(2) Runge-Kutta Method
2nd order RK
4th order RK → most popular
5th order RK
Ci +1
C **
C i +1/2
Ci dc
*
Given = f ( x, c )
C i +1/2 dx
Ci +1= C + ∆Ci
∆x / 2 ∆x / 2
x
i 1 i +1
i+ C
2
Calculate in order
∆x
C∗ =
1 Ci + f ( Ci , X i )
i+ 2
2
∆x  ∗ 
C ∗∗=
1 Ci + f  Ci +1/ 2 , X 1 
i+ 2  i+
2 2 
∆x  ∗∗ 
Ci∗+=
1 Ci + f  Ci +1/ 2 , X 1 
2  i+
2 
Then
1   1   1 
f  Ci∗∗+1/ 2 , X 1  + f ( Ci∗+1 + X i +1 ) 
1
Ci +1 =
Ci + ∆x  f (Ci , X i ) + f  Ci∗+1/ 2 , X 1  +
 6 
i+ i+
3  2  3  2  6
~ Ο ( ∆x 4 )
8-63
◎ 4th R-K with Runge’s coefficient
○ Euler's Method
dc
= f ( x, c ) → Ci +=
1 Ci + ∆xf ( x, c) + Ο ( ∆x )
dx
Ci +1= Ci + ∆Ci
∆x
=
∆Ci [ K 0 + 2 K1 + 2 K 2 + K3 ] ~ Ο ( ∆x 4 )
6
~ weighted average of slopes
in which
K 0 = f ( xi , ci )
 ∆x K 
K1= f  xi + , Ci + 0 ∆x 
 2 2 
 ∆x K 
K 2= f  xi + , Ci + 1 ∆x 
 2 2 
K 3 f ( xi + ∆x, Ci + K 2 ∆x )
=
8-64
○ Derivation by Taylor series expansion
→ see "Computer Applications of Numerical Methods", S. Kuo (1972) p.137
◎ R-K formula with Kutta coefficient
Ci +1= Ci + ∆Ci
∆x
∆=
Ci ( K 0 + 3K1 + 3K 2 + K3 )
8
K 0 = f ( X i , Ci )
 ∆x K 
K1= f  xi + , Ci + 0 ∆x 
 3 3 
 2∆x − K + K1 
K2 =f  xi + , Ci + ( 0 )∆x 
 3 3 
K 3 f [ xi + ∆x, Ci + ( K 0 − K1 + K 2 )∆x ]
=
◎ Simpson rule
8-65
~ Special case of R-K with Runge coefficients
dc
If = f (only x) independent of C
dx
Then Ci +1= Ci + ∆Ci
∆x
∆=
Ci ( K 0 + 2 K1 + 2 K 2 + K3 )
6
K 0 = f ( xi ) ,
 ∆x 
=
K1 f  xi + ,
 2 
 ∆x 
=
K 2 f  xi + ,
 2 
K 3 f ( xi + ∆x )
=
∆x   ∆x  
=
∴∆Ci f ( xi ) + 4 f  xi +  + f ( xi + ∆x ) 
6   2  
◎ 1-D Steady-state Problem
dc d 2c
=
u E 2 − kC ① 2nd order ODE
dx dx
dc
Let Z=
dx
then ① becomes coupled 1st-order ODE
dz
=
uz E − kC ②
dx
8-66
dc
Z= ③
dx
Boundary conditions (Initial Conditions)
(i) Initial Value Problems
=c ( 0 ) c=
0 @ x 0
dc
=0 z=
0 @ x 0
dx
Solve simultaneously by using either Euler or R-K method
dz u k
② : = z + e = f1 ( z , c, x)
dx E E
dc
③ : = z= f 2 ( z , c, x)
dx
∆x
c∗ 1= ci + zi
i+ 2
2
∆x  u k 
z∗ 1 =zi +  zi + ci 
i+
2 2 E E 
∆x ∗ ∆x  ∗ ∗ 
c∗∗1 =ci + z 1= ci + f2  z 1 , c 1 , x 1 
i+ 2 i+ 2 2  i+ 2 i+ 2 i+ 2 
2
8-67
∆x  ∗ ∗  ∆x  u ∗ k ∗ 
z ∗∗1 =
zi + f1  z 1 , c 1 , x 1  = zi +  z 1 + c 1
i+ i+ i+ i+ i+ i+
2 2  2 2 2  2  E 2 E 2 
 
ci*+1= ci + ∆xf 2  z ∗∗1 , c∗∗1 , x∗∗1 
 i+ 2 i+ 2 i+ 2 
= ci + ∆x z ∗∗1
i
2
u k 
zi∗∗+1= zi + ∆x  z ∗∗1 + c∗∗1 
 E i+ 2 E i+ 2 
∆x  * 
∴ ci +1 =
ci +  zi + 2 zi + 1 + 2 zi + 1 + zi +1 
* **
6  2 2 
∆x  u k  u ∗ k ∗ 
zi +1 =
zi +  zi + ci  + 2  zi + 1 + ci + 1 
6  E E  E 2 E 2
u k  u k 
+ 2  z ∗∗1 + c∗∗1  +  zi∗+1 + ci∗+1  
 E i+ 2 E i+ 2   E E  
(ii) Boundary value problems
=c c=
0 @ x 0
=c c=
L @ x L
→ use "Shooting method"
Guess z0 @ x=0
8-68
Solve ② and ③ simultaneously by using R − K
check cLj = cL
Vary z0 such that target = cL is hit
→ "Shooting method"
• Iteration rules for Shooting method
dc
let z0j = jth estimate of x =0
dx
dc
z0j +1= j + 1th estimate of x =0
dx
By interpolation
j +1 cLj − cL
z =
z − j
j
j −1
( z0j − z0j −1 )
cL − cL
0 0
<Ref.>
1. Pinder, G. and W. Gray, Finite Element Simulation in Surface and Subsurface Hydrology,
Academic Press, New York, 1977
2. Ames, W.F. , Numerical Methods for Partial Differential Equations, Academic Press,
New York, 1977
3. Richtmyer, R.D. and K.W. Morton, Difference Methods for Initial-Value Problems,
Interscience Publishers, 1967.
4. Lapidus, L. and G.F. Pinder, Numerical Solution of Partial Differential Equations in
Science and Engineering, A Wiley-Interscience Publication, 1982.
8-69
8.4 Finite Particle ("Random Walk") Model
~ based an the concept that dispersion is a random process
New position = Old position + Advection + Dispersion
~ In the computer code, enough particles are included (released) so that their locations and
density are adequate to describe the distribution of the dissolved constituent of interest
→ "Giant Molecule" method
~ release a number of particles, each representing a finite mass of solute, at a rate
proportional to the strength of each source.
The particles are then "tracked" in space and time. → "Particle Tracking" method
Ref. Prickett et al. (1981)
○ Distribution of concentration of solute
~ represented by the distribution of a finite number of discrete particles
~ each particle is moved by flow and is assigned a mass which represents a fraction of the
total mass of chemical constituent.
∂c ∂c ∂2c
+ u= E 2 (−kc) (1)
∂t ∂x ∂x
8-70
If a unit slug of solute placed initially at x = 0
then analytical solution is
1  ( x − ut ) 2 
=
c ( x, t ) exp  −  (2)
4π Et  4 Et 
C↑
◎ Statistics
○ Random variable x if said to be normally distributed if its density function, n( x) is given
by
1  ( x − µ )2 
=n( x ) exp  −  (3)
2πσ  2σ 2 
σ = standard deviation
µ = mean
Now, if we let
σ = 2Et (4)
µ = ut (5)
n ( x ) = c ( x, t ) (6)
8-71
Then Eqs (2) and (3) are equivalent.
So, the key to solute transport is the realization that dispersion can be considered a random
process, tending to the normal distribution.
○ Random walk modeling
Advective distance = u ∆t (7)
∆t = time increment
Dispersive distance = ±6σ
= 2 E ∆t ANORM(0) (8)
in which ±6σ =Probable locations of particles out to 6 standard deviations either side of
the mean ( > 99.9% )
ANORM (0) = a random number between -6 and +6, drawn from a normal distribution of
numbers having a standard deviation of 1 and a mean of zero.
8-72
∴ New position of the particle
= Old position + u ∆t + 2 E ∆t ANORM(0) (9)
◎ Repeat for numerous particles, all having the same initial position and advection term.
→ Create a map of the new positions of the particles having the discrete density function.
N
c ( x, t ) → n ( x ) →
∆x
N0  ( x − u∆t ) 2 
= exp  −  (10)
2π 2∆xE ∆t  4∆xE ∆t 
in which ∆x = incremental distance over which N particles are found
N o = total number of particles in the experiment
◎ The distribution of particles around the mean position, u ∆t , is made to be normally
distributed via the function ANORM(0)
○ Generation of ANORM(0) in computer code.
(1) Summation of Random function
12
ANORM(0) = ∑
RF (0,1) − 6

i =1

In LOTUS or EXCEL (RAND())
use @RAND function to generate a uniform random number between 0 and 1 = U (0,1)
8-73
(2) Multiply Random function
ANORM(0) = RF (0,1) × 12 − 6



RAND ()
◎ Numerical Recipes
ANORM(0)= GASDEV (IDUM)
RAND() = RAN1(IDUM)
◎ Advantages of the Random-walk technique
1. There is no numerical dispersion, despite the use of an Eulerian framework.
2. Computer CPU time is drastically reduced. Solutions are additive. If not enough particles
are included for adequate definition in one run, subsequent runs may be made and the results
of these may be superimposed upon the first.
3. This method is particularly suited to time-sharing systems where velocity fields can be
stored.
◎ Disadvantages
1. It may require a large number of particles to obtain meaningful results.
2. It doesn't easily accommodate nonlinear kinetic expressions.
8-74
◎ 2-D Model : Depth-averaged
∂c ∂c ∂c ∂  ∂c  ∂  ∂c 
+ u ( x, y ) + υ ( x, y ) = εx  + εy +S
∂t ∂x ∂y ∂x  ∂x  ∂y  ∂y 
in which
ε x = 5.93du*
ε y = 0.6du*
u* = gds
cf) 1-D Model : Depth & Width-averaged
(Cross-sectional)
∂c ∂c ∂ ∂c
+U = ( DL )
∂t ∂x ∂x ∂x
8-75
◎ 2-D Advection-Dispersion Model
○ Longitudinal and transverse dispersion take place simultaneously
x= x0 + u ∆t + 2 Ex ∆t ANORM (0)
y= y0 + υ∆t + 2 E y ∆t ANORM (0)
○ In natural rivers
EY = 5.93 dU *
E y = 0.6 dU *
8-76
8.5 Finite Element Method
∂c ∂  ∂c 
G.E=
.:  E − uc  PDE ①
∂t ∂x  ∂x 
○ Numerical solutions to PDE
FDM
Approximating Derivatives
PDE System of Linear
Algebraic Eq.
FEM
Approximating Solution → Integral Eq.
Matrix Solver
Numerical Solution
(Approximate Solution)
8.5.1 Procedure (Summary) of FEM
1. Discretize domain into elements.
2. Select Basis Functions.
3. Derive an Integral equation based on. Method of Weighted Residuals (MWR).
4. Compute element matrix and vectors.
5. Assemble global matrix and vectors.
6. Incorporate boundary conditions.
7. Use a finite difference for time discretization.
8. Solve a system of simultaneous linear algebraic eg.
8-77
A. Domain Discretization
c( x) = true(and unknown) solution to PDE
continuous function of x
cˆ( x) = approximate solution
piecewise continuous function
We may approximate the true solution by a polynomial
m
cˆ e ( x) = ∑ c jφ j e ( x ) ②
j =1
in which
φ j = basis function (shape, approximate) functions
Now, we are seeking the "best" value of the c j to give us the best values for cê ( x)
8-78
B. Basis functions
(1) Lagrangian Interpolating Polynomials
m x − xk
φ j ( x) = Π
k =1 x − x
k≠ j j k
(i) linear ; m=2
x − x2
φ1 ( x ) =
x1 − x2
x − x1
φ2 ( x ) =
x2 − x1
∴ cê ( x) = c1φ1 ( x) + c2φ2 ( x)
(ii) quadratic ; m=3
φ1 ( x ) =
( x − x2 ) ( x − x3 )
( x1 − x2 ) ( x1 − x3 )
φ2 ( x ) =
( x − x1 ) ( x − x3 )
( x2 − x1 ) ( x2 − x3 )
φ3 ( x ) =
( x − x1 )( x − x2 )
( x3 − x1 )( x3 − x2 )
∴ cê ( x) = c1φ1 ( x) + c2φ2 ( x) + c3φ3 ( x)
8-79
(2) Hermitian Interpolating Polynomials
dc
~ interpolate c( xi ) and (function and slope)
dx xi
m 
 dc  1 
c( x) ≈ ∑ c jφ j ( ) +   φ j ( ) 
0

j =1   dx  j 
C. Method of Weighted Residuals
◎ Formulation of approximating integral equation
Variational method
Method of Weighted Residuals (MWR)
MWR :
Substitute ② into ①
∂cˆˆ ∂  ∂c 
−  E − ucˆ  ≠ 0 =R ( x, t ) ……residual ③
∂t ∂x  ∂x 
If ĉ = c then R ( x, t ) = 0
But ĉ ≠ c R ( x, t ) ≠ 0
So, in the MWR, an attempt is made force this residual to zero through selection of the
constant c j ( j = 1, 2,.....M ) .
Let’s set the weighted integrals of the residual to zero, i.e., →MWR
∫ R ( x, t )ωi ( x ) d Ω = 0, i = 1,2,....M ④
Ωe
8-80
→ Integral Eq.
∂cˆˆ ∂  ∂c 
∫  ∂t − ∂x  E ∂x − ucˆ  ⋅ ωi ( x ) dx =
0 ⑤
There are several MWRs which is distinguished by the choice of weighting function ωi
(1) Galerkin method : ωi = φi ( x )
(2) Subdomain method
divide domain B into M subdomains Bi
 1, x in B
ωj =  i
 0, x not in Bi
(3) Collocation method
M point xi (collocation points) are specified in B and weighting functions are Dirac
delta functions
ωi δ ( x − xi )
=
which have the property that
∫ R ( x )=
B
ω dx
i ( xi ) 0
R=
(4) Least Squares Method
∂R
ωi = p ( x )
∂ai
p(x) = arbitrary positive function
8-81
minimize the integrated square residual w.r.t ai
I = ∫ p ( x ) R 2 ( x ) dx
∂I
∴ = 0 (i= 1, 2,......., M )
∂ai
FDM ~ domain of interest is replaced by a set of discrete points
FEM ~ domain is divided into subdomains (finite elements) unknown function C is
represented by an interpolating polynomials within each element
N
uˆ ( ⋅)
u (⋅) ≈= ∑ u φ (=
j =1
⋅), j
j j 1,2,....., N
u j = undetermined coefficient
φ j ( ⋅) = function over both time and space
φ j ( ⋅) = Basis (shape, interpolation) function
~ chosen to be polynomials that satisfy certain boundary conditions imposed
on the problem
PDE : Lu − f =
0
Luˆ ( ⋅) − f = R ( ⋅)= residual
The objective is to select the undetermined coefficients u j such that this residual is
minimized in some sense.
∫ ∫ R (⋅)ω (⋅) dvdt=

t v i 0, =
i 1, 2,....., N
8-82
(1) Galerkin Method
~ Weighting function is chosen to be the basis function
∫ ∫ R (⋅)φ (⋅) dvdt

t v i = 0, =i 1,2,....., N
(2) Subdomain Method
∫ R ( x=
v
) ω dv
i 0,=i 1, 2,......, N
where
1, ( x, y, z ) in vi*
ωi = 
0, ( x, y, z ) not in vi
*
~ integrations are less tedious than those in Galerkin's method

8-83
(3) Collocation Method
~ Weighting function is chosen to be the Dirac delta
ωi δ ( x − xi )
=
∫ ∫ R (⋅)δ (⋅) dvdt=

t v i 0, =
i 1, 2,3,....., N
~ Calculate the value of residual at the selected points
∫ ∫ a (⋅)δ ( x − x , y − y , z − z , t − t ) dvdt ≡ a
t v i i i i i xi , yi , zi ,ti
<Example>
dT
+ k (T − Te ) =
0
dt
0 ≤ t ≤1
T (= = 1
t 0)
1
=k 2=
; Te
2
8-84
 t − ti −1
 t − t , ti −1 ≤ t ≤ ti
φi =  i i −1
 ti +1 − t
 t − t , ti ≤ t ≤ ti +1
 i +1 i
ｉ = nodal points
3
1. ∑ T jφ j ( t )
T ≈ Tˆ =
j =1
2. ∫t
R(t ) wi=
(t )dt 0,=i 1, 2,3
 ∧ 
d T ∧

∫ 
dt
+ k (T − Te ) wi (t )dt =
0
 
t
 3 dφ j 
∫ ∑ T j ( + kφ j ) − kTe  wi ( t ) dt = 0, i = 1, 2,3
 j =1 dt 
t
3. Galerkin
3 1  dφ j 
∑
1
T j ∫0  + kφ j =
φi dt ∫ kT φ dt , =i
e i 1,2,3
j =1  dt  0
3 1  dφ j 
∑
1
=i 1 : T j ∫0  + kφ=
j  φ1 dt ∫ kT φ dt
e 1
j =1  dt  0
1  dφ  1  dφ  1  dφ  1
=i 1 T1 ∫  1 + kφ1 φ1dt + T2 ∫  2 + kφ2 φ1dt + T3 ∫  3 + kφ3 φ=
1dt ∫ kT φ dt
e 1
0
 dt  0
 dt  0
 dt  0
8-85
1  dφ  1  dφ  1  dφ  1
=i 2 T1 ∫  1 + kφ1 φ2 dt + T2 ∫  2 + kφ2 φ2 dt + T3 ∫  3 + kφ3 φ=
2 dt ∫ kT φ dt
e 2
0
 dt  0
 dt  0
 dt  0
1  dφ  1  dφ  1  dφ  1
=i 3 T1 ∫  1 + kφ1 φ3 dt + T2 ∫  2 + kφ2 φ3 dt + T3 ∫  3 + kφ3 φ=
3 dt ∫ kT φ dt
e 3
0
 dt  0
 dt  0
 dt  0
φ1φ3 = 0
Expansion yields the following matrix equation
 12  dφ1 
1
 dφ2  
 ∫0  φ1 + kφ1φ1  dt ∫  φ1 + kφ2φ1 dt  T1   12 
2
0
  dt  0
 dt      ∫0 kTeφ1dt 
  dφ1
1
  dφ2
1  1  dφ      1 
 ∫02  φ2 + kφ1φ2  dt ∫0  dt φ2 + kφ2φ2 dt ∫2  dt
1 
2
φ2 + kφ 3φ 2  dt  T2  =  ∫ kTeφ2 dt `
  dt       0 
    1 kT φ dt 
1  dφ  1  dφ   T3   ∫12 e 3 
 0 ∫2  dt2 φ3 + kφ2φ3  dt
1 ∫2  dt
1 
3
φ3 + k φ3 φ3 dt 
 

 k k  1 
 −1 + 3 1+ 0   
 Te 

6
 k 2 
1 k 2k k   1 
−1 + 1+ T2  =  kTe 
2 6 3 6   2
 kT 
0 k k     e
−1+ 1+ T3   2 
 6 3 
○ Basis functions (Interpolation)
∑ c j (t )φ ej ( x, t )
c( x, t ) ≈ cˆˆ( x, t ) =
j =1
m
= ∑ cˆ j (t )φ ej ( x )
j =1
∧
∂C m ∂φ j
= ∑Cj
∂t j =1 ∂x
8-86
∧
∂ C m dc
= ∑ φ j ( x)
∂t j =1 dt
○ Natural Coordinate System for element basis function
(Dimensionless ξ coordinate system where − 1 ≤ ξ ≤ 1 )
(1)Linear
1
φ1e (=
ξ) (1 − ξ )
2
1
φ2e (=
ξ) (1 + ξ )
2
dφ1e 1
= −
dξ 2
dφ2e 1
=
dξ 2
(2) Quadratic
1
φ−1 (ξ ) =
− ξ (1 − ξ )
2
φ0 (ξ ) = 1 − ξ 2
1
φ=
1 (ξ ) ξ (1 + ξ )
2
8-87
◎ Galerkin method
Select the basis functions as the weighting functions
ωi = φi
Thus the weighted integral equation of the residuals becomes
 ∧ ∧
^ 
∂ c ∂ ∂c 
∫Ωe ∂t ∂x ∂t
 − ( E − u c ) φi dx =
0 ⑥
 
∧ ∧
∂c ∂ ∂e ∧
∫Ω
e
∂
φi dx
t  Ω 
− ∫ e
( E
∂x 
− u
∂x  
e ) φie dx =
0
u
A
 d
υ

B
Term A ※ (See 7-1 for basis functions) – use Linear, Basis function
∂cˆ m dc j
∫Ωe ∂t φi dx = ∫ ∑ dt φ ( x)φie dx
e
Ω e j
j =1
m dc j (t )
=∑ ∫ φ j ( x)φi ( x ) dx
dt Ωe
j =1
Term B : Integration by parts ∫ = uv − ∫ ν du

udv
∂φie ( x) ∂cˆˆ
x
 e ∂c 
m
−∫ e ( E − uc) + φi ( x)( E − uc) 

ˆˆ
Ω ∂x ∂x  ∂x  x1
∂φie ( x)  m ∂φ ej ( x) m   e ∂cˆ 
xm
=
−  E ∑ C j (t ) − u ∑ C j (t )φ j ( x)  dx + φi ( x)( E − ucˆ) 
e
= ∂x  j 1 = ∂x j 1   ∂x  x1
8-88
m
 ∂φ ej ( x) ∂φ ej ∂φi   e ∂cˆ 
xm
+ ∑ C j (t )  ∫ e E
= dx − ∫Ω uφ (e
x ) dx  − φi ( x)( E ∂x − ucˆ) 
∂ ∂ ∂x 
Ω j

e
j =1 x x x1
M dcˆ j (t )  ∂φ ej ( x) ∂φie ( x)
m
∴ Eq ⑥ : ∑ φ j ( x)φi ( x ) dx + ∑ c j (t )  ∫ e E
dt ∫Ωe
dx
= J 1= j 1  Ω ∂x ∂x
∂φ ej ( x) ∂φie ( x)   e ∂cˆ
x

m
−∫ e E dx  − φi ( x)( E ∂x − ucˆ)  = 0 ⑦

Ω ∂x ∂x  x1
∂φ ej ( x) ∂φie ∂φ
Let a =∫ e E
e
dx − ∫ e uφ ej ( x) i dx , i =1,...., m
∂x ∂x ∂x
ij Ω Ω
 e ∂cˆ 
φ
 1 ( x )( E − ucˆ) x1 
∂x
 
e
 . 
mije ∫= φ ( x)φi ( x)dx ,
Ωe j
{B}  
 . 
 e ∂cˆ 
φm ( x)( E − ucˆ) xm 
 ∂x 
D. Element matrix equation
~ Element matrix equation results in
 dcˆ 
[ A] {cˆ} + [ M ]   + { B} =
e e e
0 ⑧
 dt 
Use Linear Basis function
∂φ ej ( x) ∂φie ∂φ
=a ∫ eE e
dx − ∫ e uφ ej ( x) i dx
∂x ∂x ∂x
Ω ij Ω
8-89
1 ∂φ j ∂ξ ∂φ j ∂ξ dx 1 ∂φ j ∂ξ dx
∫ −1
E
∂ξ ∂x ∂ξ ∂x
dξ
dξ
− ∫ uφ j
−1 ∂ξ ∂x
dξ
dξ
1 ∂φ j ∂φi dξ 1 ∂φ
= ∫ −1
E
∂ξ ∂ξ
dξ
dx
− ∫ uφ j i d ξ
−1 ∂ξ
By the way
2 dφ e (ξ )
dx dφ1e dφ e
= xj ∑ = j
x1 + x2 2
dξ j =1 dξ dξ dξ
1 1 1 ∆x
= x1 (− ) + x2 ( ) = ( x2 − x1 ) =
2 2 2 2
1 ∂φ j ∂φi 2 1 ∂φi
=aije ∫ −1
E
∂ξ ∂ξ
dξ ( ) −
∆x ∫
−1
uφ j
∂ξ
dξ
dx
φ j ( x )φi ( x ) dx φ j (ξ )φi (ξ ) d ξ
1
=mije ∫=
e
Ω ∫−1 dξ
1 ∆x
= ∫ φ jφi dξ
−1 2
x2
  ∂cˆ 
− φie ( x)  E − ucˆ  
B =e
 ∂x
i
   x1
E u E u
 ∆x − 2 − +
∆x 2 
[ A] = 
e

− E − u E u 
+
 ∆x 2 ∆x 2 
∆x  2 1 
[M ] =
e
6 1 2 
8-90
 ∂cˆ  
 E ∂x − ucˆ  
  x1 
{B} = 
e

−  E ∂cˆ − ucˆ  
  ∂x  
 x2 

E. Assemble global matrix equations
Combining element equations
For each element, apply
 dc 
[M ]   + [ A] {cˆ} + {B} =
e e e
0
 dt 
[M ] & [ A]
e e
~ 2×2 matrices
○ Numbering Systems
Local
1 2 1 2 1 2 ...
Element No 1 2 3 4 5
Global 1 2 3 4 5 6 ...
Let N (= number of element) = 30
number of nodes = 31
Then
x1 x2 x3 xN xN +1
8-91
e=1 2 3
 a11(1) a12(1) 
 (1) (1) 
 a21 a22 + a11
(2)

 a12(2) 0 
 
 a21 (2) (2)
a22 + a11(3) 
30
 
[ A] ∑
= = [ A]  
e
31×31 e =1

 
 
 0 
 
 (29)
a22 + a11(30) a12 
( 30)
 
 
(30) (30)
a21 a22
30
[M ] = ∑[M ] [ A]
e
~ same as
31×31 e =1
 B1(1)   B1(1) 
 (1) (2)   
 B2 + B1  0 
 B (2) + B (3)  0 
 2 1   
30
 .  . 
[ B ] ∑=
= {B}  =
e
  
31×1 e =1  .  . 
 .  . 
 (29)   
 B2 + B1  0 
(30)
 (30)   B (30) 
 2
B   2 
Since que + qLe +1 =

0
8-92
◎ Boundary Conditions
∂cˆ
At x= x1 ; ucˆ − E = uco
∂x
∂cˆ
At x= xN ; ucˆ − E = ucN
∂x
F. Time discretization
(1) Fully Implicit
{}
k +1
− {eˆ}
k
Cˆ
{}
k +1
[M ] + [ A] Cˆ + { B}
k +1
=
0 ⑨
∆t
 [M ]
{} [M ]
{}
k +1
 ~ ∧
k
[ ] − { B}
k +1
 + A  C = C
 ∆t  ∆t
{} {}
k +1 k
∧ ∧
[ R=
]C [S ] C − { B} ⑩
[M ] +
In which [ R]
= [ A]
∆t
[M ]
[S ] =
∆t
(2) Crank - Nicholson scheme
{C} − {C}
k +1 k
1 1
[M ] + [ A]{C} + { B}
k+ k+
2 2 =
0
∆t
{C} − {C}  {C}k +1 − {C}k 

k +1 k
1
[M ] + [ A]   + { B} 2 =
k+
0
∆t  2 
 
8-93
 [ M ] [ A]   [ M ] [ A]   { B}k +1 + { B}k 
{C} =  {C} − 
k +1
∴ + −
k
 
 ∆t 2   ∆t 2  2 
     
[ P] [Q ]
∴ [ P ]{C} =[Q ]{C} − { B}

k +1 k
⑪
8-94
8.6 Kinetic Models
○ Si term
~ effects of biochemical changes → Non-conservative Solutes
chemical
physical sources and sinks → lateral input
storage zone solute
○ Fundamental differential form of conservation equation
∂Ci
= − ∇ J i + Si
∂t
↑ ↑ ↑
accumulation flux generation
Ci = concentration of substance i scalar, mass/vol
○ Fickian Diffusion
∂Ci
+ ∇(Ci u i ) = ∇( Di ∇Ci ) + Si
∂t
○ Turbulent Diffusion
∂Ci _
+ ∇(Ci u ) = ∇ {Di + ε i } ∇Ci  + Si
∂t
εi =
8-95
1 T
u=
T
= ∫0
udt
○ Dispersion
∂C ∂C ∂C 1 ∂ ∂C 1 ∂ ∂C
2-D : + ux = + uy ( H Ex )+ (H Ey ) + Si
∂t ∂x ∂y H ∂x ∂x H ∂y ∂y
∂C ∂C 1 ∂ ∂C
1-D : =
+u ( AE ) + Si
∂t ∂x A ∂x ∂x
E= longitudinal dispersion coefficient
○ Solute Transport
∂C ∂ Q 1 ∂ ∂C
=
− ( C) + ( AD )
∂t ∂x A A ∂x ∂t
Q, A, D = const.
∂C ∂C ∂ 2C
=
− +D 2
∂t ∂x ∂x
○ Groundwater and tributary inputs QL , CL
transient storage zones
∂C Q ∂C 1 ∂ ∂C Q
=
− + ( AD ) + L (CL − C ) + α (CS − C )
∂t A ∂x A ∂x ∂x A
∂CS A
−α
= (CS − C )
∂t AS
8-96
QL = lateral inflow per unit length of stream
CL = solute concentration in lateral inputs (assumed to be const)
CS = solute concentration in transient storage zones
AS = cross-sectional area of the storage zone
α = coefficient for storage zone exchange
◎ Non-conservative solutes
~ terms to simulate solute transfers
~ variety of forms and complexity
~ depends on type of solute and use of model
○ Uptake of a first-order function
(loss of solute from the water column)
∂C ∂ Q 1 ∂ ∂C
Sol) =
− ( C) + ( AD ) − kC C
∂t ∂x A A ∂x ∂t
M  ( x − Ut ) 2 
C (=
x, t ) exp  − − ke t 
2 A π KT  4kt 
ke = overall uptake rate coefficient
○ Single benthic compartment with first-order release
∂C ∂ Q 1 ∂ ∂C 1
=
− ( C) + ( AD ) − kC C + k B CB
∂t ∂x A A ∂x ∂t h
∂CB
= hke C − k B CB
∂t
8-97
C = water column solute concentration
CB = benthic concentration
h = depth
kC , k B = first-order exchange rate coefficients
V f = hkc mass transfer coefficient
∂C ∂ Q 1 ∂ ∂C V 1
=
− ( C) + ( AD ) − f C + k BCB
∂t ∂x A A ∂x ∂t h h
∂CB
= V f C − k B CB
∂t
○ Non-conservative solute
= nutrients -nitrogen
-phosphorus
-sulfer
-dissolved organic carbon
-trace metals copper
○ Stream transport of copper by incorporating first order mass transfer equations for
periphyton and sediment reactions
∂C ∂ Q 1 ∂ ∂C ∂C ∂C
=
− ( C) + ( AD ) − PB B − PS S
∂t ∂x A A ∂x ∂x ∂t ∂t
∂CB
= λB (CB − Kd B C )
∂t
8-98
∂CS
= λS (CS − Kd S C )
∂t
in which CB = periphyton solute concentration
CS = sediment solute concentration
PB , PS = mass of periphyton and sediment
λB , λS = first-order exchange rate coefficients
kdB , kdS = partition coefficients, expressing equilibrium ratios of periphyton
to water column and sediment to water column solute conc.
○ Non-linear uptake function (Kuwabara & Heliker, 1988)
~ Monod (or Michaelis-Menten) → algal uptake
U 
S= − 
h ∂C 1 U max C
S= = −
U max C  ∂t h KS + C
U=
K S + C 
in which U max = maximum uptake per unit area of stream bottom
K S = half-saturation constant
~ solute concentration at which uptake is one-half the maximum
◎ Concepts and method for assessing solute dynamics in stream ecosystems (1990)
◎ Solute dynamics
8-99
transport = advection, dispersion
transfer (transformation) chemical transformations
changes in physical state(phase) = sorption, desorption
biological processes
◎ Chemical processes in natural water bodies (Orlob p70)
oxidation-reduction reactions
acid-base reactions
gas-solution processes and outgassing
coordination reactions of metal ions and ligands
precipitation and dissolution of solid phases
adsorption-desorption processes at interfaces
○ Aquatic organisms influence the concentrations of many substances by metabolic uptake,
transformation, storage, and release.
◎ Reaction Rates and Reaction Equation (Orlob p79)
○ In simplified water quality modeling, chemical, biochemical, and processes are described
by
dc
(1) = KC n ~ irreversible diminish (decay)
dt
n = 0
linear
=n 1 → first-order reaction

nonlinear n ≠ 0, n ≠ 1
8-100
dc KC
(2) = 1 (Michaelis - Menten)
dt K 2 + C
~ non-linear
~ uptake rate of external nutrient concentration
dc1
(3) = KC2n
dt
dc1
= K (C2,0 − C2 ) n
dt
C2,0 = saturation conc
~ reaeration , adsorption
dc1
(4) =
K1C2n + K 2C1 , n≠0
dt
~ Streeter-Phelps Eq
~ adsorption - desorption processes
dc1
(5) = K (C1,0 − C1 )C2n − K 2C1
dt
~ adsorption
dc1 C2
(6) = K1 C1 (Monod)
dt K 2 + C2
~ change of the biomass C1 of a primary producer
8-101
dc dc2
(7)=1
K1 − K 2 C1
dt dt
~ very rapid reactions
8-102

Advanced Environmental Hydraulics - SNU

Diunggah oleh

Informasi Dokumen

Deskripsi Asli:

Judul Asli

Hak Cipta

Format Tersedia

Bagikan dokumen Ini

Bagikan atau Tanam Dokumen

Opsi Berbagi

Apakah menurut Anda dokumen ini bermanfaat?

Apakah konten ini tidak pantas?

Hak Cipta:

Format Tersedia

Advanced Environmental Hydraulics - SNU

Diunggah oleh

Hak Cipta:

Format Tersedia

Ch 1.

Concepts and Definitions

Chapter 1 Concepts and Definitions

1.1 The Role of Hydrology and Hydraulic Engineering in Environmental Management

1.2 Environmental Hydraulics

1.3 Strategies and Approaches for Problems Solving

1.4 Basic Definitions and Concepts

1.1 The Role of Hydrology and Hydraulic Engineering in Environmental Management

• Water quality = water quantity

+ hydrodynamics of transport and mixing

+ chemistry and biology of natural water systems

atmosphere - air motion

hydrosphere - water motion

Fluid motions transport and disperse disposal of residuals

- massive discharges of wastewater into rivers and oceans

Without fluid motion, there is no transport and dispersion

1.1.1 Overall Framework for Environmental Management

•Environmental Fluid Mechanics

that relate to problems connected to human activities within the environment.

connected to human activities within the environment.

•Transport - Hydraulic engineer

• Role of hydraulic engineer in environmental management

1) design of hydraulic structure (outfalls, diffusers)

-> jets and plumes

2) analyze water quality processes -> diffusion and dispersion

Conceptual View of Pollutant Inputs

Agricultural Urban Area

Agricultural Combined Sewer Separate

Point Conservative Miscible

• Environmental Control System

(a) control of pollutants at the source (pre-treatment, clean technology)

(b) wastewater treatment (primary, secondary treatment)

(c) disposal in the environment (post-treatment, wastewater outfalls)

• Role of Hydraulic Engineer (Environmental Hydraulics)

- make interface between man's activities and the natural environment

- draw water supply from natural water bodies

- develop technology how wastewater is returned → design of outfalls

Flow Diagram of Water Quality Engineering

Desirable Water Use: Water Quality

Pollutant Inputs: Aquatic

Actual > Desired

1.1.2 Using the Water Environment for Waste Assimilation

• Types of pollutant (from the least dangerous to the most hazardous)

(1) Natural inorganic salts and sediments

- not toxic unless in excessive doses

(2) Waste heat or heated water discharges

- cooling water for electric generating plants

- decrease water's assimilative capacity for oxygen

(3) Organic wastes

- domestic sewage → biochemical oxygen demand (BOD)

- carbon, nitrogen, phosphorous: nutrients → eutrophication

Organic waste  O2  CO2  H 2O

- amount of dissolved oxygen to oxidize the organic wastes using chemicals

COD > BOD

Cooling system for nuclear power plant

LNG plant (Manzanillo, Mexico)

• Drinking water quality

• ppm = parts per million = mg/ℓ

1ppm = 1g of BOD/1,000,000g of water

BOD-DO coupled system

(4) Trace metals

- industrial wastewater – electro-planting, battery manufacturing, mining, smelting, refining