Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
Finite Difference
Differential eq. => difference eqn. Choices of grids in horizontal and vertical (orthogonal) Different orders of approximation in space and time Large matrices, solved interatively
MWRA, 1996
Example Codes
3-D
Princeton Ocean Model Regional Ocean Modeling System (ROMS) GLLVHT Model EFDC WIFM-SAL
LARM
1-D Cross-sectionalaveraged
QUAL2E
1-D Horizontallyaveraged
Grids
Horizontal
Vertical
ui+1
u (c c ) ci c 2ci + ci 1 = i i i 1 + E i +1 2 t x x
u i ci u i 1ci 1 ci +1 2ci + ci 1 = + E 2 x x
Time stepping
Explicit (evaluate RHS at time n)
A ci
n +1
ui t 2 Et Et n ui t n Et =c i [1 ] + ci 1 [ + 2 ] + ci +1 [ 2 ] 2 x x x x x
n
Finite Element
Information stored at element nodes Approx soln to differential eqn. Large matrices, solved iteratively More flexible than FD Somewhat more overhead
Example Codes
3-D
real c
discrete c
j+1 j+1 x
j+1
2c c c = +u E 2 x x t
W = weighted residual
L
= wRdx = 0
0
PSEG, 2000
Harmonic Models
Periodic motion outside => periodic motion inside Plus harmonics Transient problem => steady problem Best for tidallydominated flows
,u,v t
Example Codes
3-D
2-D Horizontal
Harmonic Decomposition
i t u ( x, y, t ) = Ax e i t v ( x, y , t ) = A y e
Ex.
* (i )e i t = A t * Ax u =h e i t h x x * A v y i t =h e h y y
i t iA* e +h n
A* x x
e i t + h
A* y y
e i t = 0
TEA-Basic Equations
+ (uh ) + (vh ) = y t x
(u ) (v ) x y
b u xs u fv + = u u v u x ,nl +g x t h h x y h
s b v v y v v +g + fu + = u v y ,nl t h h y x y h
Linear Terms
Non-Linear Terms
Westerink et al. (1985)
Non-linear Terms
Products of sine/cosine functions produce new sine/cosine functions with sums and differences of frequencies Ex:
Method of Characteristics
C1 C2
Time n +
U
C1 C2
Backward tracking of characteristic lines Interpolation among nodes at feet of characteristics Avoids difficulties with advection-dominated flows
Time n
Diffusion
Diffusion/simple reaction uses implicit Galerkin FEM under stationary conditions No stability limit on t Not intrinsically mass conserving Linearity facilitates source/receptor calculations
Baptista et al. (1984)
C1
C2
Time n + 1
ELA-Basic Equation
c c 1 + ui = t xi h xi
advection
c +Q hDij x j
reaction
dispersion
2 c c c +Q = Dij + ui xi x j xi t
1 (hDij ) u = ui h x j
i
Operator Splitting
2 c c c = Dij + ui +Q xi t xi x j
c c c + ui =0 t xi n
n
n+
n +1
c t
n+
2 c = Dij + {Q}n +1 n +1 x i x j
Puff Algorithm
Gaussian puffs distributed backwards in time over near field Advected/diffused over intermediate field Projected to grid after sufficient diffusion (hybrid model) Or, self-contained model (Transient Plume Model)
& m
Lagrangian Models
Y Location (km) 20 0
Particle Models
Forward Puffs
Backward Puffs
-20 0
20
40
60
80
100
X Location (km)
2-D Simulations
E E h xi = u + x + x t x hx + 2 E x t p i + S / S E y E y h + y i = v + t y hy + 2 E y t p i + S / S
Larvae introduced
Concentration (ppb)
0 0
Time (h)
Figure by MIT OCW.
100
200
Conclusion: most larvae imported (Connecticut and Thames Rivers) Supported by studies using Mitochondrial DNA and trace metal accumulation
Dee, D.P., A pragmatic approach to model validation, in Quantitative Skills Assessment for Coastal Models (D.R. Lynch and A. M. Davies, ed), AGU, 1-13, 1995. Ditmars, J.D., Adams, E.E., Bedford, K.W., Ford, D.E., Performance Evaluation of Surface Water Transport and Dispersion Models, J. Hydraulic Engrg, 113: 961-980, 1987. Oreskes, N., Shrader-Frechette, K., Belitz, K, Verification, Validation and Confirmation of Numerical Models in the Earth Sciences, Science, 263: 641-646, 1994. GESAMP (IMO/FAO/UNESCO/WMO/WHO/IAEA/UN/UNEP Joint Group of Experts on the Scientific Aspects of Marine Pollution), Coastal Modeling, GESAMP Reports and Studies, No 43, International Atomic Energy Agency, Vienna, 1991.
Who is evaluating?
Model Developer
Evaluates whether simulated processes matches real world behavior Output-oriented Ability to accurately simulate conditions at specific location(s) under variety of extreme and design conditions Reliability, cost-effectiveness
Model User
Decision makers
Problem Identification
What are the important processes and what are their space and time scales? Ex: If biogeochemical transformations are quicker than the hydraulic residence time, then perhaps steady state is OK
Model calibration
Best model fit against a known data set. Make sure output is appropriate
physically relevant those requiring least change relative to expected range of variation.
Model validation
Comparison against independent data set (or a different period of time) without changing model parameters/coefficients. Choice of appropriate metrics (mean error, rms error, etc). Perfect agreement not possible; but are results believable? (Validity connotes legitimacy) Oreskes et al. (1994) refers to model confirmation
Additional Comments
Absolute vs Relative accuracy
Latter is easier as uncertainties may cancel when comparing options under same conditions
Usually underestimated because of unknown unknowns Will model be used at different site?
Additional Comments
Purpose of models is insight