1.72, Groundwat er Hydrology Prof. Charles Harvey
Le ct u r e Pa ck e t # 9 : N u m e r ica l M ode lin g of Gr oun dw a t e r Flow
Sim ulat ion: The predict ion of quant it ies of int erest ( dependent variables) based upon an equat ion or series of equat ions t hat describe syst em behavior under a set of assum ed sim plificat ions.
Gr ou n dw a t e r Flow Sim u la t ion
• Predict hydraulic heads ( 1D, 2D, 3D)
• For part icular condit ions – confined, unconfined, isot ropic, anisot ropic, hom ogeneous, het erogeneous, infinit e, finit e, st eady, t ransient .
• Varying levels of com plexit y:
Analyst ic solut ions – Theim , Theis, et c.
Advant ages: exact , sim ple, cheap, can provide sufficient insight Analog sim ulat ion
Physical m odels – scale m odels of aquifers Num erical Sim ulat ion
N u m e r ica l Sim u la t ion
• Given a PDE and appropriat e I Cs and BCs
• Discret ize t he syst em
• Approxim at e t he PDE corresponding t o t he discret izat ion
• Solve t he approxim at ed PDE on a com put er
• Com m only finit e differences or finit e elem ent s for GW flow
• Advant ages: can handle com plex geom et ries, I Cs, and C condit ions; can be used for nonlinear syst em s.
Be w a r e of t h e t e r m “M ode l” a n d h ow it is u se d
• “ A m odel should be used as sim ple as possible, but not sim pler”
• Mat hem at ical “ m odel” – a PDE
• Num erical “ m odel” – a part icular t echnique is applied
• Com put er or sim ulat ion “ m odel” – a code
Fin it e D iffe r e n ce s
A num erical m et hod t hat approxim at es t he governing PDE by replacing t he derivat ives in t he equat ion w it h t heir respect ive difference represent at ions. Procedure involves: Grid ( or Mesh) and Equat ion
Equ a t ion – D iffe r e n ce Appr ox im a t ion of D e r iva t ive s
2 2
∂
h
∂
h
S
∂
h
∂
x
2+
∂
y
2=
T
∂
t
2D flow equat ionApproxim at ing t he Tim e Derivat ive:
Back w ar d Differ en ce:
⎛ ∂
h
i , j⎟⎟
⎞
≈
∆
h
≈
h
i , j ,n−
h
i , j ,n − 1⎜⎜
⎝ ∂
t
⎠
∆∆
t
t
n
∆
t
∆
h∆
t hn- 1hn head
t im e tn- 1
tn End of
w here, t im e st ep
n = t he current t im e st ep index
∆ t = t im e st ep
i,j = x and y coordinat e indices Approxim at ing t he Space Derivat ives:
Consider a 2D discret izat ion, if w e assum e t hat t he grid spacing in t he x- direct ion and y- direct ion are t he sam e, our discret ized grid for an int ernal node w ill be:
h h
h
h
h
∆ x
∆ y y
x i+ 1,j
i,j i- 1,j
i,j + 1
i,j - 1
i- 1/ 2 i+ 1/ 2
h
i − ,1 j+
h
i + ,1 j+
h
i , j − 1+
h
i , j + 1−
4
h
i , jS h
i , j , n−
h
i , j , n − 1=
∆
x
2T
∆
t
What does t he finit e- difference equat ion indicat e for st eady st at e condit ions?
S
∂
h
≈
S
⎜⎜
⎛
h
i , j , n−
h
i , j , n −1⎞
⎟⎟
=
0
∂
t
⎝
∆
t
h
⎠
i −1 ,j
+
h
i +1 ,j+
h
i , j −1+
h
i , j +1−
4
h
i , j=
0
∆
x
2h
h
i −1 ,j+
h
i +1 ,j+
h
i , j − 1+
h
i , j + 1−
4
h
i , j=
0
i −1 ,j
+
h
i +1 ,j+
h
i , j −1+
h
i , j +1=
h
i , j4
h h
h
h
h 10 10 10
9 11
100 65
i+ 1,j i,j
i- 1,j
i,j + 1
i,j - 1
Flow
Value of head at node is t he average of t he surrounding nodes ( for SS isot ropic hom ogeneous case)
Consider t he 1D st eady- st at e flow equat ion w it h a sink is:
2
H
d
T
−
w
'
=
0
ordx
22
H
d
w
'
=
dx
2T
∆ xh1 h2 h3
H= 10 H= 9
N ode
0 1 2 3 4
The nodal finit e difference equat ion is:
h
i , j −1−
2
h
i , j+
h
i , j +1w
'
=
∆
x
2T
Or
( )
2∆
x
w
'
=
h
i , j −1−
2
h
i , j+
h
i , j +1T
Node 1:
1H0 + - 2h1 + 1h2 = 0 1( 10) + - 2h1 + 1h2 = 0 - 2h1 + 1h2 = - 10 Node 2:
1h1 + - 2h2 + 1h3 = 0 Node 3:
Node 3 has a forcing t erm due t o t he pum ping of t he w ell. This t ranslat es int o an init ial r ight hand side of t he equat ion:
2
∆
x
w
'
( 1
0 )
.
21
Given ∆x = 0.1,
( )
=
=
1
T
.
01
1h2 + - 2h3 + 1H4 = 1 1h2 + - 2h3 + 1( 9) = 1
1h2 + - 2h3 = 1- 9 = - 8
The syst em of finit e- difference equat ions consist s of 3 equat ions and 3 unknow ns
−
2
h
11
h
2=
−
10
h
1−
2
h
21
h
3=
0
1
h
2−
2
h
3=
−
8
Un k n ow n s Kn ow n s
Or in coeffiecient m at rix form
⎛−
2
1
⎞⎛
h
1⎞
⎛−
10
⎞
⎜
⎟⎜ ⎟ ⎜
⎟
⎜
1
−
2
1
⎟⎜
h
2⎟ = ⎜
0
⎟
⎜
⎟⎜
⎜
⎝
1
−
2
⎠⎝
h
3⎠
⎟
⎝ −
8
⎠
⎟
FD Coeff. Unknow n RHS cont aining Mat rix Heads boundary
Or in m at rix not at ion it can be w rit t en as Ah = b’
A is t he m at rix of difference coefficient s H is t he vect or of unknow n heads b’ is t he RHS vect or of know n quant it ies
A com put at ional linear solve yields t he vect or h:
2
3
8.5
1 2 3 4
Tr a n sie n t Sim u la t ion Usin g Fin it e D iffe r e n ce s
Procedure: March Through Tim e
h
1h
h
10
5
.
9
⎛
⎜
⎜
⎜
⎛
⎜
⎜
⎜
⎞
⎟
⎟
⎟
⎞
⎟
⎟
⎟
9
9.5=
5
.
8
⎝
⎠
⎝
⎠
9.0• St art w it h init ial condit ions ( t hese are known)
• Solve for heads at end of first t im e st ep ∆t ; t his give t he spat ial dist ribut ion of head ( a m ap) aft er a sm all t im e increm ent .
•
•
Given known heads at end of first t im e st ep solve for heads at t he end of t he second t im e st ep.
Wit h known value at t he end of t im e st ep solve for next t im e st ep – t his is called m arching t hrough t im e
Ahn = b* w here b* = b’ + hn- 1
The right - hand side alw ays cont ains know ns. The m at rix A is t he m at rix of finit e difference coefficient s reflect ing t he syst em param et ers and discret izat ion.
So if you can solve spat ial equat ions for one t im e st ep. Then you can solve it for as m any t im e st eps as you like.
The t im e st ep m ust be sm all w hen changes in heads are rapid – such as, w hen you st art t o pum p a w ell, ∆t m ust be seconds or m inut es. I t can be increased as changes in head becom e sm aller.
FD Sim u la t ion M ode ls: codes t hat solve t he above syst em of linear algebraic equat ions – fairly robust .
Tim e St e ppin g
⎧
h
i − 1−
2
h
i+
h
i + 1⎫
S h
i ,n−
h
i ,n − 1⎨
⎬
=
T
∆
t
⎩
∆
x
2⎭
n? n- 1?Som et hing else?
head
Ex plicit Cr a n k
-N ich olson I m plicit
∆
h∆
t hn- 1hn
tn- 1 tn
∆
t
T
−
2
h
i+
h
i + 1)
=
h
, ,(
h
i − 1 i n−
h
i n − 1s
∆
x
2Ex plicit : hi,n = chi- 1,n- 1 + ( 1- 2C) hi,n- 1 + chi+ 1,n- 1
t
T
∆
Where:
c
=
s
∆
x
2Easy t o calculat e – no linear algebra, but unst able if t im e- st eps are t oo large. I m plicit : chi- 1,n - ( 1+ 2C) hi,n + chi+ 1,n = - hi,n- 1
Result s in syst em of linear equat ions t hat m ust be solved sim ult aneously. St able, but issues of accuracy.
c
h
Cr a n k N ich olson :
2
i − ,1 n−
(1
+
c
)
h
i ,n+
c
h
i + ,1 n=
(
c
−
1)
h
i − ,1 n−
c
(
h
i −1 ,n − 1+
h
i − ,1 n − 1)
2
2
Not m uch m ore num erical expense t han fully im plicit , but m ore accurat e Bou n da r y Con dit ion s:
Const ant Head – Don’t w rit e equat ion for const ant head node. Use value of const ant head in equat ions for neighboring nodes.
Flux Boundary – replace first derivat ive t erm w it h const ant
h
⎛
∂
h
⎛
i −1 ,j−
h
i , j−
h
i , j−
h
i + ,1 j⎞
⎟⎟
2
2
i , j