Development and evaluation of the Weather Research and Forecasting WRF air quality model

1
Development and evaluation of the Weather
Research and Forecasting (WRF) air quality model

• Georg Grell
• Directly involved in WRF/CHEM development at
  NOAA/Boulder
• Steven Peckham (NOAA/ESRL/GSD), and Stu McKeen
  (NOAA/ESRL/CSD)
• South America Rainer Schmitz (U. of Chile)
• Other major developers Jerome Fast, Bill
  Gustafson (PNNL)
• And other contributions from Saulo Freitas, and
• Many more national and international collaborators

2
Structure of talk
John Michalakes

• What is WRF, status of WRF/chem
• First shot at evaluation Comparison of WRFV2
  with WRFV1/chem, other models, and observations
  during NEAQS2004 (Steven Peckham, Tuesday talk)
• Future WRF/chem plans at NOAA/ESRL and WG11
• Online/offline comparisons and other research
  applications

3
Weather Research and Forecast (WRF) Model

• Collaborative effort in the US
• For research and operations
• Direct path from research to operations

John Michalakess talk tomorrow
4
                                                
                                                  
          WORKING GROUP 11 ATMOSPHERIC
CHEMISTRY                                       
                                                  
                              
http//www.wrf-model.org/WG11
Mission The mission of the atmospheric chemistry
working group is to guide the development of the
capability to simulate chemistry and aerosols ?
online as well as offline ? within the WRF
model.  The resulting WRF-chem model will have
the option to simulate the coupling between
dynamics, radiation and chemistry. Uses include
forecasting chemical-weather, testing air
pollution abatement strategies, planning and
forecasting for field campaigns, analyzing
measurements from field campaigns and the
assimilation of satellite and in-situ chemical
measurements. Interaction with other WRF
Groups The initial development of WRF-chem is
involved with the Numerics and Model Dynamics
(WG1), Model Physics (WG11), and  Land Surface
Modeling (WG14).   Current Status of
WRF-chem Anthropogenic Emissions Available for
WRF-chem Model Evaluation Future Plans WRF/Chem
FAQs Real-time Air Quality Forecasts using
WRF-chem
This page originally developed by Bill Moninger
and Randy Collander and later by Steven
PeckhamModel questions should be directed to
Georg Grell and Steven Peckham.Last modified
Monday February 7, 2005 0700 AM
5
A few things about the meteorological side of
WRF/Chem
6
Nonhydrostatic Model Solvers within WRFV2.1
Common Infrastructure

• Eulerian flux-form mass coordinate (ARW core)
• NMM model (NCEP core), to be released with
  chemistry soon

Many different physics options (MM5-ETA-RUC.),
at least for ARW core
Also available 3DVAR systems (WRF, GSI, RUC) for
meteorological analysis
7
Basic WRF 3DVAR (NCAR, ARW) Observations

• Conventional
• Surface (SYNOP, METAR, SHIPS).
• Upper air (radiosondes, pilot balloons,
  aircraft).
• Remotely sensed retrievals
• Cloud-track winds (SATOBS).
• ATOVS thicknesses (SATEMs).
• Ground-based GPS TPW.
• SSM/I oceanic surface wind speed and TPW.
• SSM/T1 temperature retrievals.
• SSM/T2 relative humidity retrievals.
• Radiances
• SSM/I brightness temperatures.
• Observation errors assumed uncorrelated.

Contact Dale Barker for questions about WRF-Var
(also check WRF-Var WEB-Page
8
WRF/chem (similar to MM5/chem )

• As of now Online, sometimes also called
  inline
• Completely embedded within WRF CI
• Consistent all transport done by meteorology
  model
• Same vertical and horizontal coordinates (no
  horizontal and vertical interpolation)
• Same physics parameterization for subgrid scale
  transport
• No interpolation in time
• Easy handling (Data management)
• Least amount of computing time if only doing one
  run

9
Chemistry package - Transport

• WRF grid-scale transport of all species
  (currently mass core, mass and scalar conserving
  5th order in space, 3rd order in time)
• a version of PPM (positive definite, but less
  accurate) is now also available within ARW mass
  core
• Subgrid-scale transport by turbulence
• Subgrid-scale transport by convection

10
Current Chemistry Package V2.1

• Chemical mechanism from RADM2 (Quasi Steady State
  Approximation method with 22 diagnosed, 3
  constant, and 38 (!) predicted species is used
  for the numerical solution
• Photolysis (Madronich), coupled with hydrometeors
  and aerosols

11
Aerosols

• Based on Modal Aerosol Dynamics Model for Europe
  (MADE, Ackermann et al. 1998)
• Modified to include Secondary Organic Aerosols
  (SOA), (Schell et al. 2001)
• Extra transport total number of aerosol
  particles within each mode as well as all primary
  and secondary species for Aitken as well as
  Accumulation mode
• Diagnostic 3D variables PM2.5, PM10, 3 variables
  for interaction with photolysis and atmospheric
  radiation

12
WRF/ChemV2.1 Aerosol/radiation feedback through
three variables

• Initially (V2.1) only with simple Dudhia SW
  radiation
• Dry scattering aerosol mass (organic and
  inorganic mass without soot)
• Dry absorbing aerosol mass, soot only
• Aerosol liquid water content

13
Current Chemistry Package V2.1

• Dry deposition (coupled with soil/veg scheme,
  flux-resistance analogy)
• Wet deposition by convective parameterization
• Biogenic emissions (as in Simpson et al. 1995 and
  Guenther et al. 1994), include temperature and
  radiation dependent emissions of isoprene,
  monoterpenes, also nitrogen emissions by soil
• May be calculated online based on USGS landuse
• May be input

14
Implementation of BEIS3 in WRFV2-Chem
Based on EPA BEIS3 v11 for SMOKE processor
Off-line
On-line
BELD3 1km gridded vegetation
WRF T, P, shortwave flux
Speciated gridded emissions at each emissions
timestep
module_bioemi_beis311
Reference emission factors
normbeis311
RADM/RACM speciation factors
Speciated reference emissions on WRF-Chem grid
RACM species emitted by BEIS3 ISO, API, LIM,
XYL, ETH, HC3, ETE, OLT, OLI, HCHO, ALD, KET,
ORA2, CO, NO
15
Example of WRF-Chem BEIS3 Processing 27 km grid,
040609 0Z forecast
Reference Isoprene Emissions
Isoprene Emissions on 040609 at 17Z
16
WRFV2.1 FLOW CHART
Other physics drivers
specific packages
emissions
radiation driver
solve_em
solve_interface
TUV
default
photolysis driver
advection
future package?
dry deposition
microphysics driver
RADM2
default
convective transport
future package?
gas-phase driver
MADE/SORGAM
default
chemistry driver
aerosol driver
future package?
17
New Stuff in next release for WRFV2.1.1/Chem (end
of January)

• MOSAIC sectional aerosols (with 4 or 8 bins)
• CBMZ chemical mechanism
• Radiation schemes coupled to aerosols

Many of these additions because of contributions
from PNNL (Jerome Fast, Bill Gustafson,)
For linux compilers V2.1.1 will only work with
at least pgi6.x or ifort9.x
18
WRFV2.1.1 FLOW CHART
Other physics drivers
specific packages
emissions
radiation driver
solve_em
solve_interface
TUV
photolysis driver
Fast-j
advection
future package?
dry deposition
microphysics driver
RADM2
convective transport
CBMZ
future pckge? - KPP
gas-phase driver
MADE/SORGAM
chemistry driver
aerosol driver
MOSAIC
future package?
19
MOSAIC

• Model for Simulating Aerosol Interactions and
  Chemistry (MOSAIC) Fast et al., submitted to
  JGR
• sectional size distribution moving-center or
  two-moment approach for the dynamic equations for
  mass and number 112 prognostic species
• MTEM mixing rule for activity coefficients of
  various electrolytes in multi-component aqueous
  solutions Zaveri et al., JGR, 2005
• MESA thermodynamic equilibrium solver for solid,
  liquid, or mixed phase state of aerosols Zaveri
  et al., In Press JGR, 2006
• ASTEEM dynamic integration of the coupled
  gas-aerosol partitioning differential equations
  Zaveri et al., In preparation

MADE / SORGAM - modal approach
MOSAIC - sectional approach
composition sulfate nitrate ammonium chloride carb
onate sodium calcium other inorganics organic
carbon elemental carbon
mass
mass
0.01
0.1
1
10
100
0.01
0.1
1
10
100
particle diameter (mm)
particle diameter (mm)
Accumulation Mode
Coarse Mode
Aiken Mode
size bins are internally mixed
20
Radiative Forcing

• Direct Effect (contributions from James Barnard
  and Rahul Zaveri)
• In-Direct Effect (contributions from Steven Ghan
  and Richard Easter)

size and number distribution, composition,
aerosol water
scattering and absorption of shortwave radiation
refractive indices
3-D ?? , ?o , and g
Mie theory
cloud albedo, precipitation, cloud lifetime
prognostic cloud droplet number, aqueous chemistry
aerosol activation
wet removal
aerosol number
Nk - grid cell mean droplet number mixing ratio
in layer k Dk - vertical diffusion Ck - droplet
loss due to collision/coalescence collection.
Ek - droplet loss due to evaporation Sk -
droplet source due to nucleation
21
TexAQS 2000
Average Aerosol Radiative Forcing August 28 - 1
September, 2000
Instantaneous Aerosol Radiative Forcing Noon,
August 31, 2000
NW wind
W m-2
W m-2
36 33 30 27 24 21 18 15
27 26 25 24 23 22 21 20
largest impact north of industrial corridor
Houston
Houston
Galveston Bay
Galveston Bay
lines major highways dots major industrial
sources
?x 1.3 km
?x 1.3 km
120 km
impact of organic and elemental carbon from urban
and industrial sources
impact of sulfate from power plant
(typical GCM ?x 100 km)

• Impact of anthropogenic particulates are a major
  uncertainty in GCMs
• Large spatial variations in particulates and the
  resulting radiative forcing over urban areas are
  not resolved by Global Climate Models (GCMs)

22
New Stuff after WRFV2.1.1 Chemical Mechanism,
Chemical Solver

• NMM core (other meteorological dynamics, from
  EMC/NCEP)
• 2-way nesting capabilities
• Aerosol indirect effects (microphysics
  interactions)
• MADRID sectional aerosol module
• More CMAQ modules
• PPM-type advection
• Aqueous-phase chemistry within convective
  transport routine (maybe released with V2.1.1)

23
New Stuff after WRFV2.1.1 Chemical Mechanism,
Chemical Solver

• RACM chemistry with implicit Seulex solver
• Seulex solver generated using KPP (mass
  conserving)
• 73 species (49 transported species)
• 237 reactions
• Much better biogenics (important for SOA
  formation, also for Ozone)
• First step towards generalization (split of
  mechanism/ solver, KPP generation of solver)
• KPP offers adjoint generation
• Complete KPP implementation (Mainz MPI, IOWA)

24
Cloud-Aerosol Interactions

• Prognostic droplet nucleation scheme diagnoses
  the maximum supersaturation (Smax) in updrafts
• The and mass of aerosol particles activated is
  determined by assuming particles with critical S
  lt Smax are activated
• Aerosol phase transition
• Aqueous chemistry can increase cloud-phase
  sulfate mass resulting in transfer between size
  bins or modes
• Wet aerosol scavenging
• Cloud-phase aerosol scavenging by precipitation
  at same rate as cloud water
• Interstitial aerosol scavenging by impact with
  precipitating particles instantaneous rain out
  assumed
• Change in cloud droplet number can alter cloud
  albedo

25
Cloud-Aerosols Interactions
Explicit Convective Cloud Simulation
CCN at 1 supersaturation
PM2.5
venting
aqueous phase aerosols
cloud boundary
scavenging
rain boundary
surface PM2.5
PM2.5 aloft
surface PM2.5 difference with and without
scavenging
scavenging
160 km
26
More new Stuff after V2.1.1

• Global version (ESRL/GSD, NCAR, NCEP)
• Chemical data assimilation (3dvar4dvar in
  particular for aerosols)
• Real-time forecasting NAM grid, NMM-WRF/Chem,
  Houston 2006
• GOCART aerosols (ESRL, NASA)
• More on ensemble methods (ESRL)
• Clean offline version (ESRL, CDAC/India)
• Tracer transport by shallow convection (ESRL,
  CPTEC/Brazil)
• Implementation of real-time Smoke emissions from
  wild fires
• Offline version

27
More new Stuff after V2.1.1 some applications
relating to Mega Cities

• LES simulations for Houstone area (dx 200m),
  requires connection of ABL boundary
  parameterization, 3d turbulence with real
  surface fluxes, also PPM-type advection. Results
  are begin compared to coarse resolution
  simulations with dx 1km and dx 4km
• Connection of urban parameterizations to the
  available boundary layer and surface
  parameterizations

28
WRF/Chem ongoing stuff and future

• Keep up-to-date through checking, or
• better join the WRF-Chem users group

http//www.wrf-model.org/WG11
29
How to look at model output?

• RIP (developed at U of Washington for MM5 and
  WRF, uses NCAR graphics)
• VIS5D
• FX-Net (for real-time forecasting, based on NWS
  AWIPS workstations)
• Ncview or other netcdf viewing packages

30
Possible applications of current modeling system

• Prediction and simulation of weather, or regional
  or local climate
• Coupled weather prediction/dispersion model to
  simulate release and transport of constituents
• Coupled weather/dispersion/air quality model with
  full interaction of chemical species with
  prediction of O3, UV radiation, as well as PM

31
WRF/Chem, some applicationsverification and
evaluation

• NEAQS 2004 WRFV2/Chem, WRFV1/chem, also 5 other
  AQ forecast models, Ozone and PM

32
WRFV2/chem evaluation has started

• NEAQS2004 field experiment
• WRFV2/Chem was run with 27km and 12km horizontal
  resolution, WRFV1/Chem with 27km
• 8 models were used to produce ensemble forecast
• Aircraft, Radiosondes, Boundary layer profilers,
  and other observations will allow verification
  and evaluation in 3d

33
Effect of Convection on pollutant transport
34
More convection and warm frontal lifting
35
Statistics for 8 Air Quality Forecast Modelswith
342 AIRNOW O3 monitors(7/14/04 through 8/17/04 -
34 days) Statistics for maximum 8-hr averages,
based on 00Z forecasts only. Medians of 342
monitor comparisons
36
Statistics for 5 Air Quality Forecast Modelswith
118 AIRNOW PM2.5 monitors(7/14/04 through
8/17/04 - 34 days) Statistics for 14Z to 22Z 8-hr
averages, based on 00Z forecasts only. Medians of
118 monitor comparisons
Ensemble methods even better when using in
combination with other techniques such as Kalman
Filter or linear regression
37
WRF/chem Current Work at NOAA/ESRL

• Completing verification runs with Summer of 2004
  data and assisting AL with evaluation
• Online/offline
• NMM core
• Lots of work with aerosol/radiation/pbl
  interaction
• Add new modules into repository
• Real-time runs will continue
• Our scientists will work with scientists
  nationally and internationally to implement
  further improvements

38
Collaborations outside of NOAA

• NCAR (WRFV2.X, Repository, further developments
  on photolysis scheme, chemical mechanisms,
  biogenic emissions)
• PNNL (various aspects of WRF/Chem)
• University of Chile (RACM mechanism, Rosenbrock
  solver for chemistry, aerosols)
• CDAC in India (offline version of WRF/Chem)
• CPTEC in Brazil (convective transport, wet
  deposition)
• MPI Mainz, University of Iowa (KPP)

39
Some research online/offline comparisons

40
What do I mean with online

• Online chemistry is integrated simultaneously
  with meteorology
• One model, no interpolation in time or space
• Physical parameterizations are the same
• Feedback to meteorology can be considered
  (aerosols/radiation or aerosols/clouds)
• Cheapest in overall CPU costs for one set of runs
• Easiest to manage

41
What do I mean with offline

• Offline A chemical transport model is run using
  output from meteorological model
• Single or two different numerical models
• Weather forecast completed, then chemistry
• Wind fields and thermodynamic fields are
  interpolated
• Space different computational grids
• Time often using weather from hourly output
• Different physical parameterizations
• No feedback to meteorology
• Computationally cheaper only if running chemistry
  repeatedly with same meteorology

for WRF/Chem everything consistent, only time
interpolation necessary
42
Online/offline

• For real-time forecasting, using parallel
  computers, there should be NO advantage of
  offline approach!

43
Online/offline comparisons WRF/Chem

• D01 (Domain 1)
• 171x181 _at_ 12 km horiz. res.
• 35 Vertical levels
• Vertical stretched
• 7m _at_ lowest level
• 300 m _at_ 2 km AGL

Not cloud resolving resolution, but compatible to
resolutions used by current operational models
44
Online/offline why on this resolution?

• For MM5/Chem we found large differences in one
  extreme case (cloud resolving simulation), but
  does that matter for
• AQ forecasts of peak ozone, PM
• Averaged over many cases
• Different model, different resolution

45
Methodology

• 1 online simulation for one case study
• Winds (u,v,w) output every time interval, look
  at frequency analysis in time
• Real-time forecasts over longer time-period in
  online mode
• several offline real-time forecasts over same
  longer time period
• Meteorology and chemistry coupled at different
  time intervals
• Meteorological fields are time averaged,
  therefore mass consistent
• Linearly interpolated meteorology between
  coupling times

46
MM5, COAMPS and WRF-ARW Spectra
MM5 AMPS /Antarctica 20 Sept 2003, dx 10
km COAMPS BAMEX 2 June 2003, dx 10
km WRF-ARW BAMEX 1 3 June 2003, dx 10 km
47
Fraction of captured variability, WRF/Chem,
dx12km, centered at 14Z
60 min
30 min
10 min
Level 10, normal day, no severe convection
48
What changes when looking at longer time period?
Peak 8-hr average ozone - mean bias difference
for 60min versus online
60min
10min
Bias is increased by almost 4ppb!
49
Look at box averaged time series
60min
10min
Bias is increased by almost 4ppb!
50
Time series, averaged over box A and the lowest
600m
51
Time series, averaged over box B and the lowest
600m
52
Vertical profile over box B time snap shot for
day with convection
53
upper level CO and 1-hr precipitation
1-hr Precip
Upper level CO difference between online and
60min offline
54
lower and upper vertical levels O3 differences
LOW
HIGH
55
Conclusions?

• Vertical motion fields are very important
• Most important where tracer gradients are largest
  (top of PBL, tropopause), and vertical velocity
  variability largest (low levels, or when
  convective)
• Ascent descent both reduce chemical
  concentration in PBL (venting and clean air
  fluxes)
• Time averaging winds and/or other fields does not
  change results much
• Very similar results with two different models

56
Conclusions ?
Physics No Mans Land
Offline gray scales
Offline gray and red scales may be moved by using
different coupling interval