Weather Research and Forecasting Model Goals: Develop an advanced mesoscale forecast and assimilation system, and accelerate research advances into operations 36h WRF Precip Forecast Collaborative partnership, principally among NCAR, NOAA, DoD, OU/CAPS, FAA, and university community WRF governance through multi-agency Oversight and Science Boards; development conducted by 15 WRF Working Groups Software framework provides portable, scalable code with plug-compatible modules Ongoing active testing and rapidly growing community use Over 1,200 registered community users, annual workshops and tutorials for research community Daily experimental real-time forecasting at NCAR, NSSL, FSL, AFWA, U. of Illinois Operational implementation at NCEP and AFWA in FY04 Analyzed Precip 27 Sept. 2002
WRF Software Design Performance-Portable Scaling on foreseeable parallel platforms Architecture independence No specification of external packages Run-Time Configurable Domain size, nest configurations, parallelism Physics, numerics, data, and I/O options Maintainability & Extensibility Single source code Modular, hierarchical design, coding standards Plug compatible physics, dynamical cores Registry to describe and manage data and I/O
Software Architecture Driver Layer Driver Mediation Layer Config Inquiry Solve DM comm OMP I/O API Package Independent Model Layer Config Module WRF Tile-callable Subroutines Threads Message Passing Data formats, Parallel I/O Package Dependent Driver: I/O, communication, multi-nests, state data Model routines computational, tile-callable, thread-safe Mediation layer: interface between model and driver Interfaces to external packagese External Packages
WRF Multi-Layer Domain Decomposition Logical domain 1 Patch, divided into multiple tiles Single version of code enabled for efficient execution on: Shared-memory multiprocessors Distributed-memory multiprocessors Distributed clusters of SMPs Vector and scalar processors Model domains are decomposed for parallelism on two-levels Patch: section of model domain allocated to a distributed memory node Tile: section of a patch allocated to a shared-memory processor within a node Inter-processor communication Distributed memory parallelism is over patches; shared memory parallelism is over tiles within patches
Scaling Performance WRF EM Core, 425x300x35, DX=12km, DT=72s Gflop/s 170 160 150 140 130 120 110 100 90 80 70 60 50 40 30 20 10 ijet IBM Regatta TCS 1-rail IBM Winterhawk II TCS 2-rails 1400 1300 1200 1100 1000 900 800 700 600 500 400 300 200 100 simulation speed (hours/hour) 0 0 0 100 200 300 400 500 600 700 800 900 1000 1100 processors
Benefit of Higher Order Model Numerics Error versus resolution Error versus cost solution error low-order method high-order method solution error low-order method high-order method resolution cost
Eulerian Nonhydrostatic Model Solvers Full conservation of variables in flux form Prognostic equations for conserved quantities Pressure and temperature diagnosed from thermodynamics High order numerics Two level, 3 rd order Runge-Kutta split-explicit time integration 2 nd -6 th order centered or upwind advection Alternative vertical coordinates Terrain-following height coordinate Terrain-following mass coordinate
WRF Model Applications Basic research Idealized simulations Atmospheric process studies Other geophysical fluid dynamical applications Numerical weather research and prediction Regional NWP Storm-scale forecasting Hurricane forecasting (ocean coupling) Global weather modeling Applied meteorological applications Air quality studies (chemistry coupling) Fire weather research (combustion coupling) Regional climate studies
Gravity Current Simulations x = z = 100 m Height Coordinate 5 min 10 min 15 min Mass Coordinate
2-D Mountain Wave Simulation a = 1 km, dx = 200 m a = 100 km, dx = 20 km Height Coordinate Mass Coordinate
2D Squall Lines
Supercell Thunderstorm Simulation Surface temperature, surface winds and cloud field at 2 hours (dx = 2 km, dz = 500 m, dt = 12 s, 80 x 80 x 20 km domain )
WRF Real-Time Forecasting NCAR: 10 and 22 km Continental US, 4 km Central US (BAMEX) NCEP: 8 km Mass and NMM, West, Central, and Eastern US NSSL: 12 km Continental US 3 km Regional FSL: 10 km Northeastern US AFWA: 15 km Continental US U. Of Illinois: 25 km, Midwestern US (http://wrf-model.org)
36 h Forecast Valid 12Z 27 Sept 02 24 h Precipitation Verification 175 150 125 100 75 50 35 30 25 20 15 10 5 0 (mm) 24 h RFC Analysis 12 km Opnl ETA 10 km WRF 22 km WRF
3-6 h Accumulated Precip Forecasts Valid 18Z 4 June 2002 precip (mm) 4 km Analysis 10 km WRF 22 km WRF 50 20 12 8 4 0 (From Mike Baldwin and Matt Wandishin, NOAA/NSSL)
3-6 h Accumulated Precip Forecasts Valid 18Z 4 June 2002 precip (mm) 50 4 km Analysis 8 km NMM 12 km opnl ETA 20 12 8 4 0 (From Mike Baldwin and Matt Wandishin, NOAA/NSSL)
Power Spectra for 3 h Precipitation 12Z forecasts, 15-18 Z accum precip, valid 4 June 2002 (From Mike Baldwin and Matt Wandishin, NOAA/NSSL)
Model Physics in High Resolution NWP Physics No Man s Land 1 10 100 km Resolved Convection Cumulus Parameterization 3-D Radiation Two Stream Radiation LES PBL Parameterization
Convection Resolving NWP using WRF Questions to address: Is there any increased skill in convection-resolving forecasts, measured objectively or subjectively? Is there increased value in these forecasts? What can we expect given that the small spatial and temporal scales we are now resolving are inherently unpredictable at forecast times of O(day)? If the forecasts are more valuable, are they worth the cost?
Realtime 4 km BAMEX Forecast 24-25 May 2003 Radar reflectivity 00Z 24 May initialization 36 h forecast
Reflectivity, 12 Z 24 May 2003 Observed WRF 12 h 4 km forecast
Reflectivity, 06 Z 25 May 2003 Observed WRF 30 h 4 km forecast
Realtime 4 km BAMEX Forecasts Valid 6/8/03 12Z 4 km BAMEX forecast 36 h Reflectivity 4 km BAMEX forecast 12 h Reflectivity Composite NEXRAD Radar
Realtime 4 km BAMEX Forecasts Valid 6/10/03 12Z 4 km BAMEX forecast 36 h Reflectivity 4 km BAMEX forecast 12 h Reflectivity Composite NEXRAD Radar
Realtime 4 km BAMEX Forecasts Valid 6/10/03 12Z 10 km BAMEX forecast 36 h Reflectivity 10 km BAMEX forecast 12 h Reflectivity Composite NEXRAD Radar
Realtime 4 km BAMEX Forecasts Valid 6/10/03 12Z 22 km CONUS forecast 36 h Reflectivity 22 km CONUS forecast 12 h Reflectivity Composite NEXRAD Radar
Problems with Traditional Verification Schemes truth forecast 1 forecast 2 Issue: the obviously poorer forecast has better skill scores From Mike Baldwin NOAA/NSSL
Ensemble Forecasting Advantages Ensemble mean is generally superior Ensembles provide a measure of expected skill or confidence a quantitative basis for probabilistic forecasting a rational framework for forecast verification information for targeted observations Initial State Uncertainty Mean Truth Limitations/Challenges Not clear how to optimally specify the initial conditions (singular vectors, breeding, perturbed observations) Requires more computer resources Deterministic Forecast t critical Probabilistic Forecast
Coupled Systems (Source: Rick Allard, NRL)
Model Coupling Adapting WRF framework for model coupling Extension of WRF I/O API specification Use of Model Coupling Environment Library, and Model Coupling Toolkit 27km WRF 10m wind vel. Nov. 7-8 2002 Applications Atmosphere/ocean coupling (Hurricane-WRF) Atmosphere/chemistry coupling (WRF-Chem) SWAN Wave Heights (Mobile Bay)
WRF-Chem Based on EPA CMAQ Model Development of a WRF-Chem model based on EPA s Community Multiscale/Multipollutant Air Quality (CMAQ) model to meet both on-line and off-line modeling needs (Institute for Multidimensional Air Quality Studies, U. Houston). Intended use of coupled air-quality model - forecasting chemical-weather, - testing air pollution abatement strategies, - planning and forecasting for field campaigns, - analyzing measurements from field campaigns - assimilation of satellite and in-situ chemical measurements (Daewon W. Byun and Seung-Bum Kim, University of Houston)
Simulated Surface O 3 and Horizontal Wind 2130 UTC August 27, 2000 Ozone Surface Winds High ozone plumes are located in the downwind side of high emission sources in the urban and industrial complexes due to steady southeasterly sea breeze winds (Daewon W. Byun and Seung-Bum Kim, University of Houston)
Comparison with NOAA Aircraft Obs. ALTITUDE (km) 6 5 4 3 2 1 Aug. 27, 2000 O3_obs O3_m_geos O3_m_prof alt_obs 120 100 80 60 40 20 O3 (ppbv) 0 18 19 20 21 22 23 0 0 TIME (UTC) Model shows higher background ozone; plume locations are well matched with observations. (Daewon W. Byun and Seung-Bum Kim, University of Houston)
Online WRF-Chem Implementation (FSL) Consistent: all transport done by meteorology Same vertical and horizontal coordinates (no horizontal and vertical interpolation) Same physics parameterization for subgrid scale transport No interpolation in time, or flow/mass adjustments Chemistry Weather interactions / feedbacks Radiation, microphysics, etc Easy handling (Data management) Meteorology and chemistry data in same history file Often more efficient (CPU costs)
Model Forecasts Surface O 3 forecast Similar results! Wind direction Front location Peak O 3 Other AQ models are similar Figure: Stu McKeen (NOAA/AL)