S ensitivity A pproximate M ethod Pedram Falsafi Amir Hakami Carleton University Canada CMAS 2015 Chapel Hill for Sensitivity Advection in CMAQ Overview Horizontal advection in CTMs ID: 729450
Download Presentation The PPT/PDF document "Implementation of L inear" is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
Slide1
Implementation of Linear Sensitivity Approximate Method
Pedram Falsafi, Amir HakamiCarleton University, Canada
CMAS 2015, Chapel Hill
for Sensitivity Advection in CMAQSlide2
OverviewHorizontal advection in CTMsIssues with current advection schemes – CMAQ and PPMLinear Sensitivity Approximate MethodPreliminary resultsFuture steps and conclusions
2Slide3
Horizontal advection in CTMsMain transport processInherently linear process3
Negative values
s
uppressed peak
d
iffused signalSlide4
Solves continuity equation for each species in two separate directions
Piecewise Parabolic Method (PPM) (with
Yamo)Monotonic and Mass conservative Therefore, Nonlinear schemeHorizontal advection in CMAQ
4Slide5
Sensitivity in AdvectionPerturbation at a single grid cellImpact at a few cells downwind
(finite-difference sensitivity) 5
?Slide6
Sensitivity in Advection
6
Negative
values
, physically
meaningless
Depend on profile (species)
There
is no “nonlinear term
” in
PPM
discontinuous operations Slide7
Sensitivity in Advection
7Slide8
Does artificial nonlinearity matter?Can we trust advection process when …There are large gradients in concentration field?
Sensitivity analysis?Derivative of an advection is not the same for nonlinear scheme (DDM and adjoint analysis)Taking differences in concentrations?We often calculate sensitivities (differences) even if we don’t call it sensitivity!
8Slide9
9 Why do we solve the continuity equation for each species?After all, it is air that is advected
; everything else is advected within the air.Linear Sensitivity Approximate M
ethodSlide10
1-D advection of
air
density fieldApply same coefficients (Jacobian
) to other species
L
inear
S
ensitivity
A
pproximate
M
ethod
10
i
jSlide11
Trace air mass of only 1 cell at a timeIndependent of concentration profile (only wind field)No negative coefficientsApproximate solutionIs it computationally efficient ?
PPM advection scheme: only 5 cells contribute11
Linear Sensitivity Approximate Method0
0
0
1
0
0
0
0
J
i-2
J
i-1
J
i
J
i
+1
J
i+2
0Slide12
How good is the approximation?
12
Using LSAM [
ppmv
]
Ozone: LSAM – PPM (8-day simulations)Slide13
Daily Maximum O313
4-day simulation8-day simulation
Using PPM [ppmv]
Using LSAM [
ppmv
]
How good is the approximation?
R-squared = 0.946
R-squared = 0.965Slide14
14
Daily Maximum CO
How good is the approximation?
Using PPM [
ppmv
]
Using LSAM [
ppmv
]
4-day simulation
8
-day simulation
R-squared
= 0.969
R-squared = 0.948Slide15
Sensitivity Tests - DDM
15
DDM Using
LSAM
DDM Using PPM
10 Time steps
120 Time steps (1 day)Slide16
ConsApproximateProspositive definite and mass conservativeTruly linear method, independent of profileSensitivities will be “exact”
Computationally less expensive Use of more expensive but more accurate scheme is feasible16Linear Sensitivity
Approximate MethodSlide17
Where to use LSAMIntended for sensitivity calculationsNot for using instead of PPM (for now!)
17Slide18
ConclusionsModeling linear advection process with a nonlinear/discontinuous scheme can entail problemsWith sensitivity calculations With concentration calculations used to estimate differencesAdvected air densities provide an approximate operator to apply on all species
More efficientMore accurate/expensive schemes are feasible 18Slide19
Future stepsPropose a reliable benchmark for validationTry less diffusive and more accurate schemesFind a smoother base profiles for estimations of the contribution matrix
19Slide20
AcknowledgementThanks to Talat Odman and Ted Russell (Georgia Tech)
for insightful discussionsSlide21
Thank You!Slide22Slide23
0
0
1000
0
0
1
0
0
0
0
0
1
0
0