comparison of online and offline modeling with wrf/chemt
DESCRIPTION
Comparison of online and offline modeling with WRF/chemT. Julius S. Chang ( 張時禹 ） Institute of Atmospheric Physics National Central University, Taiwan. Participants in WRF/chemT development: Julius S. Chang, Shu Wei Hsu, Tsun Hsien Liu, Tu fu chen, Jing Li, and - PowerPoint PPT PresentationTRANSCRIPT
Comparison of online and offline modeling
with WRF/chemT
Comparison of online and offline modeling
with WRF/chemT
Julius S. Chang ( 張時禹）Institute of Atmospheric Physics
National Central University, Taiwan
Julius S. Chang ( 張時禹）Institute of Atmospheric Physics
National Central University, Taiwan
Participants in WRF/chemT development:
Julius S. Chang, Shu Wei Hsu, Tsun Hsien Liu, Tu fu chen, Jing Li, and Chi Kang Chiang
Outline of PresentationOutline of Presentation
On the concept of off-line and online models
What is WRF/chemT? How is it different from
WRF/chem? Some preliminary findings
On the concept of off-line and online models
What is WRF/chemT? How is it different from
WRF/chem? Some preliminary findings
Coupling of Atmospheric Processes
Meteorological Model Conservation ofenergy, mass,and momentum
Emissions Model Anthropologicaland biogenic emissions
Air Quality Model Conservation ofchemical species
Off-line model:
Step 1
Step 2
Step 3
Domain 1Domain 1
Jinan
Taipei
KaohsiungHongKang
Shanghai
Guangzhou
Chongqing
Xi'an
Beijing
Shenyang
Harbin
Nanjing
Wuhan
TokyoOsaka
Seoul
Pusan
Fukuoka
Manila
Hanoi
DOMAIN 1
DOMAIN 2
DOMAIN 3
DOMAIN 4
Typical Nested Domains
Cross Section Z = 1O3; domain 2
10/31/96 14 Taiwan time
<15
>135
30
45
60
75
90
105
120
ppb
Cross Section Z = 1O3; domain 3
10/31/96 14 Taiwan time
<15
>135
30
45
60
75
90
105
120
ppb
Cross Section Z = 1O3; domain 1
10/31/96 14 Taiwan time
<15
>135
30
45
60
75
90
105
120
ppb
Cross Section Z = 1O3; domain 4
10/31/96 14 Taiwan time
<15
>135
30
45
60
75
90
105
120
ppb
Cross Section Z = 1O3
10/31/96 12 Taiwan time
<15
>135
30456075
90105120
ppb
Cross Section Z = 1O3
10/31/96 14 Taiwan time
<15
>135
30456075
90105120
ppb
Cross Section Z = 1O3
10/31/96 10 Taiwan time
<15
>135
30456075
90105120
ppb
Cross Section Z = 1O3
10/31/96 16 Taiwan time
<15
>135
30456075
90105120
ppb
Cross Section Z = 1O3
10/31/96 8 Taiwan time
<15
>135
30456075
90105120
ppb
Cross Section Z = 1O3
10/31/96 18 Taiwan time
<15
>135
30456075
90105120
ppb
(1,1) Cross Section Z = 1O3
10/31/96 21 Taiwan time
Y
<15
>135
30456075
90105120
ppb
Cross Section Z = 1O3
11/01/96 0 Taiwan time
<15
>135
30456075
90105120
ppb
Cross Section Z = 1O3
10/31/96 3 Taiwan time
<15
>135
30456075
90105120
ppb
(1,1) Cross Section Z = 1O3
10/31/96 18 Taiwan time
Y
<15
>135
30456075
90105120
ppb
Cross Section Z = 1O3
10/31/96 6 Taiwan time
<15
>135
30456075
90105120
ppb
(1,1) Cross Section Z = 1O3
10/31/96 15 Taiwan time
Y
<15
>135
30456075
90105120
ppb
Cross Section Z = 1O3
10/31/96 12 Taiwan time
<15
>135
30456075
90105120
ppb
Cross Section Z = 1O3
10/31/96 9 Taiwan time
<15
>135
30456075
90105120
ppb
O3 surface contration
domain 4
10/31/96 3 - 11/01/96 0 Taiwan time
WRF/chemTa new direct coupled
meteorology and chemistry model
WRF/chemTa new direct coupled
meteorology and chemistry model
It is derived from WRF/chem The philosophical departure is to focus
on selected options and improvements of only those options.
At NCU we assume responsibilities for correct operations of this “reduced and modified” model.
When useful, our submodels will be offered to WRF/chem working group for consideration for the “mother” model.
It is derived from WRF/chem The philosophical departure is to focus
on selected options and improvements of only those options.
At NCU we assume responsibilities for correct operations of this “reduced and modified” model.
When useful, our submodels will be offered to WRF/chem working group for consideration for the “mother” model.
WRF-chem developmentWRF-chem development
Original WRF-chem is WRF + RADM2 + . . . mostly ported from offline models.
Some of the major issues are:1. Computationally slow
It is desirable to be faster2. incomplete direct coupling of
emissions Not really “online”3. Incomplete direct coupling of other
processes more recent versions are better
Original WRF-chem is WRF + RADM2 + . . . mostly ported from offline models.
Some of the major issues are:1. Computationally slow
It is desirable to be faster2. incomplete direct coupling of
emissions Not really “online”3. Incomplete direct coupling of other
processes more recent versions are better
Computaional efficiency of WRF-chem
Computaional efficiency of WRF-chem
For a particular five day simulation over a Taiwan domain
using 128 CPUs
WRF (meteorology only): 939 sec
WRF-chem (met. and gas chemistry): 3905 sec
For a particular five day simulation over a Taiwan domain
using 128 CPUs
WRF (meteorology only): 939 sec
WRF-chem (met. and gas chemistry): 3905 sec
Not including aerosol or aq. chem.!
3-D Air Quality Model (AQM)
AQM describes atmospheric transport, transformation and deposition of airborne chemical species via a set of species conservation equations.
transport diffusion
gas-phase chemistry
source
cloud process etc.
dry deposition
Symbolically this set of partial differentialequations can be written as
Or even more briefly as
€
∂c∂t
= T +K +G +C +D( )c+E€
∂cl∂t
=∂cl∂t transp
+∂cl∂t diff
+∂cl∂t g−chem
+∂cl∂t cloud
+∂cl∂t dry
dep
+E
To simplify the symbols and afterdiscretization, we use the vector notation
€
cn+1 = I + Δt T d +K d +Gd +C d +Dd( )[ ]cn
+ ΔtEn
Apply Operator Splitting
€
cn+1 = I + ΔtT d( ) I + ΔtK d( ) I + ΔtGd( )
I + ΔtC d( ) I + ΔtDd( )cn + ΔtEn
Group No CASE No. Version Condition Time used(sec)
CASE-2 QSSA dtcmin=0.05min, dt = 45 sec, chemdt = 0.75min 1458.912
CASE-6 NCU dt = 45 sec,chemdt = 0.75min 695.490
CASE-3 QSSA dtcmin=0.05min, dt = 45 sec, chemdt = 1.5min 1419.405
CASE-7 NCU dt = 45 sec,chemdt = 1.5min 428.857
CASE-4 QSSA dtcmin=0.05min, dt = 45 sec, chemdt = 4.5min 1300.592
CASE-8 NCU dt = 45 sec,chemdt = 4.5min 316.076
1
2
3
Chemistry Computation Performance
0
200
400
600
800
1000
1200
1400
1600
1 2 3
Group No.
Time used(sec)
QSSA
NCU
4 times faster4 times faster
3 times faster3 times faster
2 times faster2 times faster
Comparison of chemical solvers for WRF/chem 2.x and WRF/chemT
To simplify the symbols and afterdiscretization, we use the vector notation
€
cn+1 = I + Δt T d +K d +Gd +C d +Dd( )[ ]cn
+ ΔtEn
Apply Operator Splitting
€
cn+1 = I + ΔtT d( ) I + ΔtK d( ) I + ΔtGd( )
I + ΔtC d( ) I + ΔtDd( )cn + ΔtEn
A most important advantage of this new approximation is the resulting computational algorithm
€
cn+α = I + ΔtDd( )cn
cn+β = I + ΔtC d( )cn+α
cn+γ = I + ΔtGd( )cn+β
cn+δ = I + ΔtK d( )cn+γ
cn+η = I + ΔtT d( )cn+δ + ΔtEn
Test CaseTest Case
No. Version Condition Time used(sec)
CASE-1 WRF-CHEM(QSSA) dt = 45 sec, Meteorology Only 939.364
CASE-2 WRF-CHEM(QSSA) dtcmin=0.05min, dt = 45 sec, chemdt = 0.75min 3904.848
CASE-3 WRF-CHEM(QSSA) dtcmin=0.05min, dt = 45 sec, chemdt = 1.5min 3865.341
CASE-4 WRF-CHEM(QSSA) dtcmin=0.05min, dt = 45 sec, chemdt = 4.5min 3746.528
CASE-5 WRF-CHEM(QSSA) dtcmin=0.01666667min, dt = 45 sec, chemdt = 4.5min 5028.567
CASE-6 WRF-CHEM(NCU) dt = 45 sec,chemdt = 0.75min 3141.426
CASE-7 WRF-CHEM(NCU) dt = 45 sec,chemdt = 1.5min 2874.793
CASE-8 WRF-CHEM(NCU) dt = 45 sec,chemdt = 4.5min 2762.012
CASE-9 WRF-CHEM(NCU) dt = 45 sec,chemdt = 0.75min, No Chem, Transport Only 2445.936
CASE-T3 WRF-CHEM(NCU) dt = 45 sec,chemdt = 4.5min,Transport=2.25 2019.926
WRF Emission Model WRF/chem 3.x
Emissions Proc.Met. Proc.
meteorology emissions chemistry
WRF/chemT
Emissions processing for WRF/chem
Emissions processing for WRF/chemT
Coupling of Atmospheric Processes
cloud, precipitation
radiation, energy balance
Meteorology
Emissions
Air Quality
Online model
point and anthrop.sources
photolysis,aerosol, cloud,wet chemistry,depositions
Asymmetric convective model (ACM)
Asymmetric convective model (ACM)
Developed as simple as the Blackadar model with a modified scheme for the downward mixing.
Strongly buoyant plumes rise from the surface layer to all leves in the CBL but downward motion is primarily a gradual compensating subsidence.
Developed as simple as the Blackadar model with a modified scheme for the downward mixing.
Strongly buoyant plumes rise from the surface layer to all leves in the CBL but downward motion is primarily a gradual compensating subsidence.
€
∂C∂t
= MuC1 −MdiCi + Mdi+1Ci+1
Δσ i+1
Δσ i
Modified Asymmetric convective model (ACM2)Modified Asymmetric convective model (ACM2)
combines the non-local convective mixing of the original ACM with local eddy diffusion to better represent the full range of turbulent transport.
combines the non-local convective mixing of the original ACM with local eddy diffusion to better represent the full range of turbulent transport.
€
∂C∂t
= MuC1 −MdiCi + Mdi+1Ci+1
Δσ i+1
Δσ i
+1
Δσ i
K i+ 12(Ci+1 −Ci)
Δσ i+ 12
+K i− 1
2(Ci −Ci−1)
Δσ i− 12
⎛
⎝ ⎜ ⎜
⎞
⎠ ⎟ ⎟
Surface ozone around Taiwan from WRF-chemT with two different sets of boundary conditions
Observed and Simulated Data
Thank you for your attention!