NOAA ESRL GSD Assimilation and Modeling Branch WoF / Hi-Impact Wx Workshop 1 April 2014 – Norman, OK From RAPv3/HRRR-2014 to the NARRE/HRRRE era Stan Benjamin Curtis Alexander David Dowell Ming Hu Steve Weygandt John Brown Tanya Smirnova Haidao Lin Georg Grell Eric James Georg Grell Joe Olson Steven Peckham Jaymes Kenyon Tracy Smith Brian Jamison NOAA NCEP Geoff Manikin Geoff DiMego EMC colleagues HRRR 1 Apr z +15h composite reflectivity (valid 02z) Norman colleagues NSSL, SPC, CAPS, U.Oklahoma

NCEP implementations of RAP version 2 and HRRR RAP HRRR RAP “Version 2” real-time at GSD, final testing NCEP “Version 2” NCEP implementation occurred 25 Feb 2014 HRRR Real-time experimental at GSD, testing at NCEP Planned NCEP implementation August 2014

Pre-radar assimilation ~ w/ radar refl assim (RUC/HRRR) km radar DA - HRRR Crossover in forecast skill between nowcasting/extrapolation vs. NWP 2014 HRRR

Rapid Refresh Hourly Update Cycle 1-hr fcst 1-hr fcst 1-hr fcst Time (UTC) Analysis Fields 3DVAR Obs 3DVAR Obs Back- ground Fields Partial cycle atmospheric fields – introduce GFS information 2x/day Cycle hydrometeors Fully cycle all land-sfc fields (soil temp, moisture, snow) Hourly ObservationsRAP 2014 N. Amer Rawinsonde (T,V,RH)120 Profiler – NOAA Network (V)Down to 9 (!) Profiler – 915 MHz (V, Tv)20-30 Radar – VAD (V)125 Radar reflectivity - CONUS1km Lightning (proxy reflectivity)NLDN ( GLD360 later) Aircraft (V,T)2-15K Aircraft - WVSS (RH)0-800 Surface/METAR (T,Td,V,ps,cloud, vis, wx) Buoys/ships (V, ps) MesonetFlagged for now GOES AMVs (V) AMSU/HIRS/MHS radiancesUsed GOES cloud-top press/temp low-level (<1200m AGL) bldg added GPS – Precipitable water260 WindSat scatterometer2-10K Observations Used 4

Data Assimilation Type PBL Pseudo Innovations Soil Adjustment Low-Cloud Building Snow Cover Building Lightning Obs (radar refl proxy) RAPv1GSI 3D-VARNo RAPv2 GSI EnKF-3DVAR Hybrid + temp-dep radar-hydrometeor Yes Yes + revised trimming Yes Model Version Horiz/Vert Advection Scalar Advection Upper-Level Damping SW Radiation Update Land Use Land-sfc model PBL Micro- physics Radiation LW/SW RAPv1 WRF-ARW V th /3 rd Monotonic w-Rayleigh minUSGS RUC 6-lev MYJ Thompson V3.2 RRTM/ Goddard RAPv2 WRF-ARW v th /5 th Positive- Definite w-Rayleigh min MODIS Fractional RUC 9-lev MYN N Thompson v3.4.1 RRTM/ Goddard w/ snow fix Major changes for RAPv2 Assimilation – hybrid ensemble assimilation, soil/PBL/surface/cloud assimilation Model – -land-surface, PBL, cloud physics RAPv2 enhancements over RAPv1 5

RAPv2 PBL moisture pseudo-obs Improved PBL moisture structure, crucial for RAP / HRRR MEM 21z 21 Dec 2013 Analyzed sounding RAPv1 RAPv2 RAPv1RAPv2 Model-based sfcOA MLCAPE/CIN & EFF. SHEAR Images courtesy Israel Jirak, Steven Weiss, Phillip Bothwell, Andy Dean, Storm Prediction Center

Data AssimilationModel GFS EnKF-3DVAR hybrid data assimilation PBL-based pseudo- innovations Surface-obs-based soil moisture/temp adjustment Low cloud building / other cloud enhancements Temp-dependent radar hydrometeor building Lightning data assimilation GSI trunk update Physics changes MYJ  MYNN v3.5.1-Joe Olson PBL scheme 9-layer RUC LSM (from 6-layer, Tanya Smirnova) MODIS land-use (fractional) Modified roughness length Revised Thompson cloud m-phys. Higher-order numerics introduced Positive definite scalar advection 5 th -order vertical advection Update to WRFv3.4.1 RAP version 2 upgrades 7

RMSE Vertical Profiles Nov – 22 Dec 2012 RAPv2 Hybrid RAPv1 No Hybrid (3D-VAR ) Wind Temp RH Wind 03-hr Forecast 12-hr Forecast RHTemp RAPv2 Hybrid Data Assimilation Improved upper-air environment 8

Soil Temperature Soil Moisture Valid 20 UTC 03 June 2013 Cooling Warming RAPv2 Soil Adjustment Moistening Drying Based upon surface temperature and dewpoint analysis increments – new option within GSI. Critical also for HRRR 9 Improved soil, near-surface temperature, moisture

RAPv2 WITH lightning assim RAPV1 No lightning assim Rapid Refresh – 0h valid 02z 26 Jan 2012 RAP Lightning Assimilation Improved convective coverage off the coast with lightning assimilation Observed Reflectivity NSSL Seminar Series04 June 2013High-Resolution Rapid Refresh 10 NCEP Production Suite Review4-5 December 2012Rapid Refresh / HRRR 10 NOAA/ESRL/GSD18 Feb RAPv2 for NWS Regions

Radar observed HRRR 12-h forecast Composite Reflectivity (dBZ) NOAA Next-Generation RAP / HRRR System Forecast of Mid-Atlantic Derecho – 29 June 2012 Valid 11PM EDT Real-time HRRR forecast HRRR Experiment with 2011 RAP DA Key 2012 DA improvements PBL-pseudo-obs for moisture Soil moisture/temp adjustment

3-km Interp 2013 HRRR Initialization from RAP GSI Hybrid GSI cld Anx Digital Filter 1 hr fcst 18 hr fcst GSI Hybrid GSI cld Anx Digital Filter 1 hr fcst 18 hr fcst GSI Hybrid GSI cld Anx Digital Filter 18 hr fcst 3 km HRRR 13z 14z 15z 13 km RAP Refl Obs 1 hr pre-fcst Obs GSI cld Anx GSI 3D-VAR Cld Obs 15 hr fcst HRRR 2h fcst ~ +1:10 hr 1 hr fcst

Radar Obs 05:00z 18 May hr fcst 3-km radar DA 0-hr fcst NO 3-km radar DA Improved 0-2 hr Convective Fcsts 05z + 0 min 18th IOAS-AOLS / 4th R2O4 February 2014HRRR Implementation 13

Radar Obs 05:15z 18 May 2013 Improved 0-2 hr Convective Fcsts 05z + 15 min 15-min fcst 3-km radar DA 15-min fcst NO 3-km radar DA 18th IOAS-AOLS / 4th R2O4 February 2014HRRR Implementation 14

Radar Obs 05:30z 18 May 2013 Improved 0-2 hr Convective Fcsts 05z + 30 min 30-min fcst 3-km radar DA 30-min fcst NO 3-km radar DA 18th IOAS-AOLS / 4th R2O4 February 2014HRRR Implementation 15

Radar Obs 05:45z 18 May 2013 Improved 0-2 hr Convective Fcsts 05z + 45 min 45-min fcst 3-km radar DA 45-min fcst NO 3-km radar DA 18th IOAS-AOLS / 4th R2O4 February 2014HRRR Implementation 16

Radar Obs 06:00z 18 May 2013 Improved 0-2 hr Convective Fcsts 05z + 1 hour 1-hr fcst 3-km radar DA 1-hr fcst NO 3-km radar DA 18th IOAS-AOLS / 4th R2O4 February 2014HRRR Implementation 17

18th IOAS-AOLS / 4th R2O4 February 2014HRRR Implementation HRRR improvements over 2012 HRRR 2012 HRRR 2013 HRRR CSI vs. lead-time 25 dBZ reflectivity 40-km verif Eastern US Real-time 1 June – 31 Aug respective years NOT event matched, but long-term and season match Improvements from RAP DA (moisture pseudo-obs, soil-adjust, hybrid) and from HRRR 3-km radar DA

ModelData Assimilation RAP- ESRL (13 km) WRFv incl. physics changes Physics changes: Grell-Freitas convective scheme MYNN PBL scheme - Olson version RUC LSM update Thompson microphysics – v3.5.1 RRTMG radiation scheme Shallow cumulus parm w/ rad feed MODIS veg fraction/leaf area index Merge with GSI trunk Increase ensemble weight in hybrid DA 8m  2m bkg for sfc Td assim Radiance bias correction New sat assimilation (NOAA-19, METOP-B, GOES, direct readout – RARS) HRRR (3 km) WRFv incl. physics changes Physics changes: MYNN PBL scheme - Olson version RUC LSM update Thompson microphysics – v3.5.1 RRTMG radiation scheme Numerics changes: 6 th order diffusion in flat terrain smooth BC 3-km hybrid ens/var assimilation (was var-only in 2013) 8m  2m bkg for sfc Td assim Radar LH – 4x less intense than 2013 (2x less intense than RAP but more local) Changes with high/medium importance for overall forecast skill Candidate RAPv3/HRRRv2 Changes

1200 UTC 19 May dBZ CSI 3 km 25 dBZ CSI 20 km 25 dBZ CSI 40 km 25 dBZ CSI 80 km HRRR using 2014 RAPv3 IC with ens=0.75 DA, new physics (not yet 2m qv DA) HRRR in 2013 real-time w/ RAPv2 IC HRRR tests – May

 Included new sensors/data  GOES sounding data from GOES-15  amsua/mhs from noaa-19 and metop-b ；  Included the RARS data (Just on Zeus now)  Removed some high peaking channels to fit the model top of RAP and removed the ozone channels  Implemented the enhanced bias correction scheme with cycling Summary of radiance updates for RAP V3

Radiance channels used for RAP V3 AMSU-A (remove high-peaking channels) noaa_n15: channels 1-10, 15; noaa_n18: channels 1-8, 10,15; noaa_n19: channels 1-7, 9-10, 15; metop-a: channels 1-6, 8-10,15; metop-b: channels 1-10, 15; HIRS4 (remove high-peaking and ozone channels) metop-a: channels: 4-8, 10-15; MHS noaa_n18, noaa-19, metop-a, and metop-b: channels 1-5; GOES (remove high-peaking and ozone channels) GOES-15 (sndD1,sndrD2,sndrD3,sndrD4): channels 3-8,

Use of 2m q v background (instead of k=1 (~8m w/ σ=0.997)) New in RAPv3 – GSI mod 2m dewpoint measurement is at…. 2m Moist bias introduced with use of k=1 background for q v 8m q v usually drier than 2m q v Moisture bias much reduced in hPa in daytime – 00z verification Also evident at 12z => improvement in convective environment RH (%) bias vs. raobs, May h forecasts - RAP experiments RAPv2-fall ver RAPv3 RAPv3 w/ 2m qv DA Verif at 00z 6h fcsts Verif at 12z 6h fcsts

ESRL – experimental version RAPv1 – used in 2011 – Initialized 2011 HRRR – effective but too many storms RAPv2 – used in – Initialized HRRR – Better surface DA, hybrid DA radar HRRR – 2012 – Major improvement over 2011 (storm bias, coverage/accuracy) HRRR – 2013 – 3km/15min radar assimilation – Initialized from RAPv – Available 45 min earlier, much more accurate 0-15h storm forecasts, more reliable 2-computer NWS-NCEP - operational Implemented 1 May 2012 RAPv2 - implemented scheduled on 25 Feb 2014 running in NCEP/NCO testing now HRRR – Implementation scheduled for Aug 2014 HRRR testing on WCOSS with real-time end-to-end runs underway (Curtis, Ming heroics) HRRR (and RAP) Future MilestonesHRRR Milestones RAP/HRRR Implementation Map

ESRL – experimental version RAPv3 – to be used in 2014 – Will initialize 2014 ESRL-HRRR(v2) – Improved PBL, convection, sfc assim RAPv4 – to be used in 2015 – Will initialize 2015 ESRL-HRRR(v3) – Target: hourly RAP ensemble for ensemble data assimilation HRRRv2 – 2014 – Improved radar assimilation, surface assimilation, PBL/cloud physics HRRRv3 – 2015 – To be initialized from RAP ensemble data assimilation + improved 3km physics (incl. aerosol-aware microphysics from NCAR/Thompson) NWS-NCEP - operational Implement early 2015 Implement 2016 Implement 2015 Implement 2016 HRRR (and RAP) Future MilestonesHRRR Milestones RAP/HRRR Implementation Map NOTE: ~42h HRRR forecast run 1-2x daily to start by May 2014 for energy and severe wx forecasts – on DOE computer. (

NARRE/HRRRE Roadmap - Highlights – (for FAA Model Development & Enhancement Product Development Team) 2014 Implement HRRR (v1) at NCEP (convection, icing, terminal). Develop RAPv3 and HRRRv2 toward 2015 implementations at NCEP. Extension of HCPF and NARRE/TL into improved aviation probabilistic fields Start on 3km-ensemble-DA on small domain retro test 2015Implement RAPv3 and HRRRv2 at NCEP. Implement regional ensemble data assimilation into parallel experimental RAP at ESRL into test version as part of continued NARRE development. Development of RAP and HRRR continues Implement NAMRR at NCEP, using same single regional ensemble assimilation. Implement regional ensemble assimilation to initialize RAP/NAMRR Implementation for 6-member NARRE with ARW and NMMB members at NCEP for improved aviation probabilistic forecasts. 2019Implementation of multi-member HRRRE with ARW and NMMB 3km-supercell-capable members at NCEP 26

RUC / RAP / HRRR growth in number of model gridpoints * compared to Moore’s law* * NOTE: other aspects of the model system (# variables, etc.) also increased, making this a conservative estimate of model memory growth * NOTE: Moore’s law baselined to 1994 RUC1 RUC1 RUC2 RUC20 RUC13 RAP13 HRRR Year RUC / RAP / HRRR model growth rate similar to Moore’s Law Moore’s Law (2x per 2 yrs) (60) (40) (3) growth [dx (km)]

HRRR – 1800 x 1060

HRRR – 1800 x 1060 (=1.9M) Initial HRRR-EnKF-DA test 670x360 = 0.24M 7.5x smaller than CONUS HRRR (aim for 10x smaller)

HRRR-ens-DA demo – some numbers – HPC requirements in 2013 cores 1,000 – current HRRR (CONUS fcst in min) By 2017 (research), at NCEP? 10,000 - HRRR-full-domain hourly EnKF – 80-mem 10,000 – HRRRE with 10 members HRRR EnKF DA demo + HRRRE – mini-10% NE domain 1,000 – EnKF DA 500 – 10-member HRRRE, run to 6h only

Directions for MDE for to improve AHP forecasting & NAS efficiency Further improvement in data assimilation – Ensemble-based DA, later toward 4d-ens-var Improve cloud/hydrometeor assimilation use of radar, PBL, satellite observations Probabilistic AHP fcsts thru hourly updated ensembles – NARRE, HRRRE, improved global ensemble Update frequency – Needs to go to every 30 min by 2020, then to 15-min in early 2020s and every 5-min by late 2020s – Use full NWP with high-frequency radar and satellite data – Coordination with NOAA Warn-On-Forecast plan (improved 1-2h forecasts updated on 5-min basis with storm-scale ensemble data assimilation) 31

MDE directions – , continued (2) Resolution – 3km horizontal resolution may be sufficient (with higher-order numerics as used in the current WRF-ARW dynamic core) until computer power allows resolution of m or less – Vertical resolution must be increase to at least levels to improve better forecasts of terminal conditions, convection (via improved boundary-layer development), icing (via improved vertical resolution of clouds). – There will be a trade-off between horizontal/vertical resolution and ensemble size. If there is fairly little improvement in depiction of convective storms from 3km down to 0.5km, then additional computer power may first be applied to ensemble forecasting and ensemble data assimilation. – 1km fixed nests around hubs may be appropriate. Cloud microphysics – Further improvements in cloud microphysics are essential, including aerosol- aware microphysics and higher-order moment or even bin microphysics. Fog – Again, increased vertical resolution in the lowest m and aerosol-aware microphysics will lead to an improvement for these forecasts 32

MDE directions, continued (3) Inline chemistry – Should be used in all NCEP models – RAP-chem already running experimentally over N.America every 6h – Critical for improved visibility – near-terminal and slant visibility – Critical for cloud/hydrometeor effects Volcanic ash – Real-time forecasts of volcanic ash with an hourly updated model (like RAP) are now feasible. Inline chemistry is available and is already running in an experimental real- time version of the RAP over North America. Real-time global eruption data is now available from the USAF and is being copied in real-time to NOAA/ESRL. Boundary-layer physics interaction with land-surface, radiation, cloud physics – An integrated unified approach to boundary-layer, sub-grid-scale mixing is needed and already an area of initial research. Output frequency – Down to 15min for selected 2-d fields from HRRR, expect full-resolution 3-d output at 5- 15min output frequency New emphasis on terminal weather – E.g., lake breeze at ORD – Wake mitigation 33

MDE directions, continued (4) Contrail forecasting – Will be improved significantly by aerosol-aware microphysics and inline chemistry Global Rapid Refresh – FIM model now produces improved wind forecasts over GFS at 12h and longer durations – WRF/RAP physics can be included, 15-km in real-time Computing advances expected – GPU or MIC or other, 2-5x jump in CPU/$ New observations expected – Radar – MPAR, integration of CASA, TDWR – Satellites – GOES-R and NPOESS – Regional aircraft observations? – Vast increase in obs from m – proprietary – tower, nacelle, sodar 34

Evolution of hourly updated NOAA modeling Feb 2014 Rapid Refresh v2 – NCEP oper PBL/soil/radar assimilation enhancements  Improved surface forecasts, convective environment fields Hybrid ensemble-variational GSI assimilation Model – improved cloud / PBL / LSM, numerics improvements, updated WRF July 2014 – HRRR (3km) - planned NCEP oper with 3km/15min radar refl. assimilation 2015 – RAPv3 / HRRRv2 North American Rapid Refresh Ensemble (NARRE) ~2017 NMM, ARW cores Hourly updating with GSI-hybrid EnKF Initially 6 members, 3 each core, physics diversity (stochastic only or with RAP/NAM/NCAR suites) Hourly forecasts to 24-h NMMB (+ARW?) members to 84-h 4x/day Rapid Refresh  NARRE 2017 – Ensemble Rapid Refresh/NAM – NARRE (w/ hybrid 4d-ens/var DA) 2019? – Ensemble HRRR – HRRRE – ( ultimately with hourly ~3km ensemble DA) HRRR NARRE / HRRRE at NCEP