The WRF-Chem model Implementation on ARIS HPC Theodore M. Giannaros Post-doc Researcher National Observatory of Athens Institute for Environmental Research and Sustainable Development Email: [email protected] VI-SEEM REG-CL: Training Event, 11 October 2017, Belgrade, Serbia What is it? A version of the WRF model that can be used for the simulation of trace gases and particulate matter simultaneously with the meteorology (Grell et al., 2005) Online chemistry, fully embedded within the WRF Consistency: same grid structure (vertical, horizontal), same physics for sub-grid scale transport, no time interpolations Perfectly suited for examining meteorology-chemistry feedbacks on local to global scales (climate change)
Suitable for operational air quality forecasting on regional to cloudresolving scales VI-SEEM REG-CL: Training Event, 11 October 2017, Belgrade, Serbia Why online is important? Eventually though, a migration to an integrated modeling system will provide new opportunities for weather prediction modelers as well. The simulation of chemical species will allow identification of shortcomings in currently used forecast models as well as lead to better use of meteorological data assimilation More realistic representation of the atmosphere: the offline approach introduces errors that increase with increasing grid resolution Numerically more consistent: same grid structure (horizontal, vertical), no time interpolation Proven improvements in medium-range weather forecasts Nevertheless:
Computationally very expensive Less flexibility for conducting ensemble modeling VI-SEEM REG-CL: Training Event, 11 October 2017, Belgrade, Serbia How does the coupling work? Advection and diffusion provided by WRF Sub-grid scale transport carried out by WRF physics parameterizations, PBL schemes and convective parameterization schemes Chemical processes carried out by WRF-Chem, dependent on meteorological input: emissions (anthropogenic, biogenic, fire, dust, sea salt, volcanic), dry deposition, wet scavenging Chemical reactions carried out by WRF-Chem: aqueous and gas phase chemistry, aerosols Chemical radiation processes carried out by WRF-Chem: computation of photolysis rates Chemistry-meteorology feedback carried out by coupling interface: radiation, microphysics, convection VI-SEEM REG-CL: Training Event, 11 October 2017, Belgrade, Serbia
Installation Similar procedure and dependencies as for installing the WRF model export WRF_CHEM=1 ./configure ./compile em_real 50+ compilation options: Serial, DM, SM, Hybrid (DM+SM), numerous compilers and architectures VI-SEEM REG-CL: Training Event, 11 October 2017, Belgrade, Serbia Quick how-to Procedure similar to a meteorological WRF simulation 1. Setup your modeling domain as you do for a typical WRF simulation ./geogrid.exe 2. Decode forcing data ./ungrib.exe 3. Horizontally interpolate forcing data on your modeling domain ./metgrid.exe
Chemistry part 1.Select chemistry option chem_opt = ?? 2.Prepare emissions (users job from scratch) ** will not be covered in this talk 3.Generate initial and boundary conditions (optionally also lower boundaries) for both meteorology and chemistry ./real.exe 4.Run your simulation! ./wrf.exe VI-SEEM REG-CL: Training Event, 11 October 2017, Belgrade, Serbia Getting started: Read - Think - Design Define your objectives What are your scientific and/or practical objectives? Why do you need to run WRF-Chem? How will you know that your simulations are successful? You and your scientific problem: to know us better Review literature! What are the atmospheric and chemical processes
involved? Which are the most important (clouds, radiation, convection, aerosols, etc.)? What is known? Is anything missing? Judge the efficacy of your simulations-to-do. Determine available observational datasets What observations are available? Again, become familiar with the processes that you want to study. How will the observations be used for verifying and/or complementing your simulations? Judge the adequacy of your simulations-to-do. Prepare your strategy Are you going to focus on a case study? If yes, which one and why? Are there adequate observations for verifying your simulations-to-do? Will you set up an operational weather and air quality forecasting service? What are the practical requirements? VI-SEEM REG-CL: Training Event, 11 October 2017, Belgrade, Serbia Setting up domains First things, first Target horizontal grid spacing Resolution of initialization data
Most often, you will need to adopt a nesting strategy. Hints Place domain boundaries away from each other, and away from steep topography Odd parent-child ratios are preferred (e.g. 3:1, 5:1) Higher horizontal resolution will also require higher vertical resolution Use at least 30-35 vertical levels; larger density closer to the ground and to the model top to avoid numerical instability (aka CFL violation) Lambert: mid-latitudes, Mercator: low-latitudes, Lat-Lon: global, Rotated lat-lon: regional Start inside-out (first the nest, then move up) Never forget! Avoid the grey zone (4-10 km) What about computational requirements? VI-SEEM REG-CL: Training Event, 11 October 2017, Belgrade, Serbia Forcing data Garbage in, garbage out
(GIGO) Read - Think - Design General questions: Do the static data (topography, land use, etc.) represent my study area adequately? How good are the meteorological forcing data? Does their resolution (temporal, spatial) fit my domain setup? Do I need lower boundary conditions (e.g. SST)? WRF-Chem specific questions: Do I need an emission inventory? Do I have one? Does it represent well my study area? Does it have adequate resolution (temporal and spatial) and fit the purposes of my simulation? What kind of chemistry do I need? Gas phase only? Do I need to include aerosols? VI-SEEM REG-CL: Training Event, 11 October 2017, Belgrade, Serbia Physics Too many options! Where to start from?
Back to basics: Which processes are important? Review literature. What others did? Consider first well documented (extensively tested) schemes Hints Convective schemes are generally not required at dx<4 km Sophisticated microphysics schemes (double-moment, detailed species) may not be necessary at dx>>10 km Try to have consistent physics between the domains or use 1-way nesting If your simulation spans more than 5 days, you could start thinking to adopt the SST update option What about the meteorology-chemistry feedback? Certain physics parameterizations may be required to account for e.g. aerosol-radiation and aerosol-microphysics interactions VI-SEEM REG-CL: Training Event, 11 October 2017, Belgrade, Serbia Chemistry (1)
Chemistry option Timesteps Input emissions options Anthro/volcanic emissions Input emissions options Photolysis option Dry deposition Biogenic, dust, sea salt VI-SEEM REG-CL: Training Event, 11 October 2017, Belgrade, Serbia Chemistry (2) Gas/aerosols ICBC Gas/aerosols chemistry switches Wet scavenging Aerosol effects
Sub-grid scale processes Biomass burning options Gas/aerosols ICBC Aerosol effects VI-SEEM REG-CL: Training Event, 11 October 2017, Belgrade, Serbia Dust in WRF-Chem Why dust? Large uncertainty in estimating global dust emissions: 514 - 4313 Tg/yr Emissions depend heavily on surface wind speed and soil properties: high spatial and temporal variability Incomplete understanding of the processes that lead to dust emission: threshold friction velocity, horizontal saltation flux, vertical flux Understanding changes in dust emissions is of
paramount importance for both interpreting past and predicting future climate change VI-SEEM REG-CL: Training Event, 11 October 2017, Belgrade, Serbia Dust emissions options Available dust emission schemes (version 3.9) All schemes are founded on the GOCART mechanism: Always, examine 1.GOCART (dust_opt = 1) code! module_gocart_dust.F 2. AFWA (dust_opt = 3) module_gocart_dust_afwa.F 3. UoC (dust_opt = 4) with either dust_schme=1 (Shao, 2001) or dust_schme = 2 (Shao, 2004) or dust_schme = 4 (Shao, 2011) module_uoc_dust.F
module_uoc_dustwd.F module_qf03.F Required input Dust source function: Ginoux erodibility (default, provided via WPS) Dust emissions are computed online using surface wind speed and soil properties; no need for any emission inventory Advice: Adopt a sufficiently long spin-up period for building up emissions VI-SEEM REG-CL: Training Event, 11 October 2017, Belgrade, Serbia (Dust) aerosol models Available aerosol models (version 3.9) 1.Modal (MADE/SORGAM, MADE/VBS, MAM) Size distribution of aerosols represented by several overlapping intervals (modes), assuming a log-normal distribution within each mode Computationally efficient Less accurate 2. Sectional (MOSAIC) Size distribution of aerosols represented by several discrete size bins,
specified by the upper and lower dry particle diameters Computationally expensive More accurate 3. Bulk (chem_opt = 401) Provides only dust concentration, assuming 10 ash size bins Computationally fast Dust is treated as a passive tracer When do I need to use an aerosol model for dust? Necessary to account for aerosol direct and indirect effects VI-SEEM REG-CL: Training Event, 11 October 2017, Belgrade, Serbia (Dust) aerosol direct (radiative) effects (1) Direct effect: absorption Solar/IR
scattering & Semi-direct effect: Modification of heating (sensible, latent), static stability VI-SEEM REG-CL: Training Event, 11 October 2017, Belgrade, Serbia Modification of radiative balance the rates, surface fluxes (Dust) aerosol direct (radiative) effects (2)
Aerosol optical properties (AOD, asymmetry factor, SSA) computed for 4 SW and 16 LW wavelengths: large uncertainty in determining dust refractive indices! Compatible aerosol models: Bulk, MADE/SORGAM, MAM, MOSAIC Compatible radiation schemes: Goddard, RRTMG Setting up your namelist ra_sw_physics = 2 or 4 ra_lw_physics = 4 aer_ra_feedback = 1 aer_op_opt > 0 (select mixing rule for Mie calculations) cu_rad_feedback = .true. (account for sub-grid scale cloud effects) chem_opt = any (except for 401 - dust-only) Evaluate direct effects Compare simulations with aer_ra_feedback = 0 (OFF) and 1 (ON) but consider semi-direct effects (changes in clouds induced by radiative effects) VI-SEEM REG-CL: Training Event, 11 October 2017, Belgrade, Serbia
(Dust) aerosol indirect effects (1) When activated, they are effective cloud condensation nuclei (CCN) and ice nuclei (IN), thus modifying the microphysical properties of clouds. VI-SEEM REG-CL: Training Event, 11 October 2017, Belgrade, Serbia (Dust) aerosol indirect effects (2)
Prognostic aerosol number concentrations influence cloud albedo and rain mixing ratio Compatible chemistry options: any with aqueous reactions Compatible aerosol models: Bulk, MADE/SORGAM, MAM, MOSAIC Compatible radiation schemes: Goddard, RRTMG Compatible microphysics schemes: Lin, Morrison Setting up your namelist progn = 1 (enable prognostic number concentrations for microphysics species) mp_physics = 2 or 10 cldchem_onff = 1 (turn on cloud chemistry) wetscav_onff = 1 (turn on wet scavenging) chem_opt = any that supports aqueous chemistry (suggested 10MOSAIC, 11-MADE/SORGAM) Evaluate indirect effects 1 control simulation (CNTL) 1 radiative experiment (RAD) 1 experiment with radiative and indirect effects (TOT) indirect effects = TOT - RAD VI-SEEM REG-CL: Training Event, 11 October 2017, Belgrade, Serbia
WRF-Chem application examples (1) Operational dust forecasting system for Europe WRF-Chem v3.6.1 Dust-only Modified GOCART scheme (chem_opt=401) to use 8-bin size distribution following DREAM (Flaounas et al., 2017 GMD) Daily, 84-h forecasts Part of WMO SDS-WAS for NA-ME-E http://meteo.gr/meteomaps/
wrf_dust.cfm VI-SEEM REG-CL: Training Event, 11 October 2017, Belgrade, Serbia WRF-Chem application examples (2) Work in progress! Dust-cyclones interactions Application on ARIS HPC in the frame of VISEEM WRF-Chem v3.7.1 Study direct/indirect effects of dust Medicane GOCART/MOSAIC (Zhao et al., 2010)
chem_opt=10, dust_opt=2 Nested modeling: 10, 3.333 km on a VI-SEEM REG-CL: Training Event, 11 October 2017, Belgrade, Serbia WRF-Chem application examples (3) Work in Dust storms - Haboob in Iran progress! Application on ARIS HPC in the frame of
VISEEM WRF-Chem v3.7.1 Numerical simulation of an intense, short-lived dust storm (haboob) in Iran GOCART/MOSAIC (Zhao et al., 2010) chem_opt=10, dust_opt=2 Nested modeling: 9, 3 km VI-SEEM REG-CL: Training Event, 11 October 2017, Belgrade, Serbia WRF-Chem application examples (4) Work in Dust outbreak in the Mediterranean progress! Application on ARIS HPC in the frame
of VISEEM WRF-Chem v3.7.1 Numerical simulation of an intense dust outbreak in SE Med GOCART/MOSAIC (Zhao et al., 2010) chem_opt=10, dust_opt=2 1 domain: 10 km VI-SEEM REG-CL: Training Event, 11 October 2017, Belgrade, Serbia Summary WRF-Chem is a powerful integrated modeling system that can support a wide range of applications including: air quality forecasting services case study modeling of chemistry-meteorology feedbacks (e.g. aerosols) climate change assessment studies focusing on chemistry (e.g.
aerosols) Take-aways & Hints Read - think - design Get your hands dirty - dig into the code! There are some hidden schemes and options, and bugs as well WRF-Chem is a really heavy model Consider carefully the scope of your application, its adequacy and its efficiency Resources (not the wealth you may be used to for WRF) https://ruc.noaa.gov/wrf/wrf-chem/ (main WRF-Chem page) VI-SEEM REG-CL: Training Event, 11 October 2017, Belgrade, Serbia Thank you!
Questions? Comments? Contact: Theodore [email protected] M. VI-SEEM REG-CL: Training Event, 11 October 2017, Belgrade, Serbia Giannaros,
