NextGen FAB Progress and Plans Steve Albers, Isidora Jankov, Zoltan Toth, Scott Gregory, Kirk Holub, Yuanfu Xie, Paula McCaslin NOAA/ESRL/GSD Forecast Application Branch Updated July , 0000UTC
Presentation Outline LAPS Overview Recent Progress (Year 1 – AIV Validation) Future Plans (Years 2,3 – Model Bias Correction) Role of LAPS in RUA?
What is Local Analysis and Prediction System (LAPS) -- Variational LAPS? LAPS Observation oriented analysis Efficient and fine resolution analysis, short latency Portability and ease of use Multiscale analysis Hot-start analysis Cloud analysis Good performance in verification of real time forecast Moving LAPS toward variational LAPS Gradually merging LAPS processes into a unified variational system Possessing the above traditional LAPS features Providing spatial consistent analysis Using CRTM for assimilation satellite data (AMSU under testing) Terrain-following coordinate variational analysis is being tested
LAPS Motivation High Resolution (500m – 20km), rapid update (10-60min), local to global Highly portable system Collaboration with user community - about 150 world wide Federal Gov’t – NWS, RSA, PADS, FAA, DHS, SOS State Gov’t – California Dept of Water Resoures International – Finnish Met. Inst., China Heavy Rain Inst. Private Sector – Toyota, WDT Wide variety of data sources: OAR/ESRL/GSD/Forecast Applications Branch*
Presentation Outline LAPS Overview Recent Progress (Year 1 – AIV Validation) Future Plans (Years 2,3 – Model Bias Correction) Role of LAPS in RUA?
AIV Validation Progress Real-time statistics comparing LAPS with observations available Analyses compared with mostly dependent observations Forecasts compared with independent observations State variables (wind, temperature, humidity, precipitation) Surface and aggregated 3-D variables Available on-line at laps.noaa.gov/verif/
Cloud / Reflectivity / Precip Type (1km 15-min analysis) DIA Obstructions to visibility along approach paths *
AIV Validation Progress Statistics of analyzed and forecast AIVs being investigated One approach is using IR (11 micron) satellite to help verify clouds Compare gridded forecast and observed/analyzed brightness temp Verifying both forecasts and analyses Compare forecast (or analyzed) cloud ceiling with METARs Presently done qualitatively (with overlays of data) Consider doing quantitatively, possibly collaborating with verification group in ACE
Observed & Forecast IR Satellite Brightness Temp HWT 3km Domain 25 Jun Z Simulated VIS also available (derived from cloud amount) Forecasters are naturally familiar with satellite images Used for objective cloud forecast verification OBSForecast
Observed and Forecast IR Satellite Brightness Temp 23 Apr Z
Observed and Analyzed Cloud Base Height 24 Apr Z
AIV Validation Progress Precipitation related AIVs Threat Score (ETS, Bias) calculated for radar reflectivity thresholds Threat Score (ETS, Bias) calculated for precipitation amount
HWT 1km V-LAPS 0-3 h Composite Reflectivity Verification Higher ETS (best at short lead time) Compare WRF initialization schemes, work with DTC? Var. LAPS Initialization
Cloud Analysis Independent Validation All-sky Imager Compare LAPS simulated all-sky analyses (or forecasts) to actual all-sky imagery Validates quality of analyses (or forecasts) of clouds / visibility obstructions Courtesy: Longmont Astronomical Society All-Sky Camera Sun Glare
Cloud Analysis Independent Validation All-sky Imager This example has more clouds with high opacity Validation leads to improvements (e.g. parallax correction, thin cirrus) Can be extended to airplane point of view Courtesy: Longmont Astronomical Society Sun Glare
Presentation Outline LAPS Overview Recent Progress (Year 1 – AIV Validation) Future Plans (Years 2,3 – Model Bias Correction) Role of LAPS in RUA?
Statistical Post-processing of Ensemble Forecasts for Aviation Applications Premise: Statistically corrected ensemble forecasts will provide ultimate 6D datacube from which all forecast information, including covariability across variables, space, and time will be derivable Current State NAEFS - North American Ensemble Forecast System global ensemble data, 1x1 degree resolution LAMP- Processed at obs sites, spread to grid No systematic processing of AIVs yet Objective Develop methods and test them in collaboration with EMC & MDL
Statistical Post-processing of Ensemble Forecasts for Aviation Applications Produce ensemble of statistically bias corrected and calibrated 3-D AIV and other variables Why GSD/FAB? Combination of expertise in these areas Statistical post-processing Data assimilation Numerical Weather Prediction Proven record of collaboration Involvement in DTC Collaboration planned with EMC/NCEP & MDL (K. Gilbert et al)
Statistical Post-processing of Ensemble Forecasts for Aviation Applications Produce ensemble of statistically bias corrected and calibrated 3-D AIV and other variables Gridded NWP analyses checked with observations used as "truth" Assess systematic errors in ensemble mean and spread
Data Analysis Use variational version of 3-D LAPS analysis Installed in AWIPS-II and used operationally by the WFOs Ensemble ExREF (Experimental Regional Ensemble Forecast System) 9-km experimental ensemble developed among GSD, HMT, EMC Used experimentally by NWS/WR, WPC Goal is to transfer new methods to EMC for operational SREF use
Choice of Variables / Methods Model Prognostic Variables and Derived Variables All will be bias corrected AIVs derived from bias corrected prognostic variables Will test if these AIVs are well calibrated Bias correction represents new capability for NCEP 3-D Cloud Liquid, Cloud Ice, Precipitating Hydrometeors Prognostic variable to be calibrated Derived variables include cloud base, visibility Determine ratio of ensemble spread and mean error This spread correction method considered by EMC for NAEFS use 3-D Winds
Bayesian Methods (in FY`15) Bayesian Processor of Ensemble (BPE) Developed by R. Krzystofowicz et al for statistical AIV correction Advantages Proper treatment of non-Gaussian variables More advanced methods to correct 2nd and higher moments of forecast distribution Uses analyzed climatological distribution in correction process Fuses predictive information from latest obs and/or analysis into correction process BPE method will be implemented and tested with EMC Transferrable to NCEP operations
Presentation Outline LAPS Overview Recent Progress (Year 1) Future Plans (Years 2,3) Role of LAPS in RUA?
Transition to Rapid Updating Analysis * WHAT IS RUA? Courtesy Jason Levit ? 7 8
ROLE OF LAPS IN RUA - PLUSES 1)Very frequent update (10-15 mins, can be 5 mins) 2)Run at 1 km resolution (see eg HWT real time experiments) 3)Can be run either 2D or 3D 4)Uses multitude of observations 5)Uses multi-radar mosaicing, reflectivity, cloud liquid/ice, lightning, etc 6)LAPS executes operationally on AWIPS & AWIPS2 - can be ported to NCEP? What are criteria? 7)Variational LAPS - state of the art DA, with following innovations: multiscale, control variables, obs preconditioning, etc. 8)Used both as real time analysis for situational awareness & for initializing NWP WOF models (see, e.g., HWT)
ROLE OF LAPS IN RUA - NEGATIVE Variational LAPS meets most if not all criteria by Jason except: Not "GSI-based", not in "GSI framework" o GSI is not flexible or modular, unyieldy for development E.g., LAPS multiscale and control variable choices very difficult to implement in GSI What does this criterion cover? o What warrants this? GSI has been used at NCEP for 20+ yrs? What criteria we think should be considered primarily? o Performance E.g., Reflectivity ETS - LAPS competitive with persistence in 0-3 hrs o Speed LAPS 18 times faster than GSI on same grid etc o Modularity Both GSI and LAPS has work to do o Other considerations? Please share
OUR VISION - NOAA DA REPOSITORY NOAA's DA scheme 5-10 yrs from now will not be like current GSI or LAPS o Will have components from both and other systems Create NOAA DA repository o Bring GSI, LAPS, and selected other NOAA DA systems onto common platform (eg, DA systems at NSSL, AOML) Modularize each Test exchanging components to find optimal configuration for each application o Engage DTC - difficult undertaking Define goals and rules of engagement Accelerating NOAA's DA development that will o Set the foundation for development of NOAA's next generation DA system(s) o Be configurable from common repository
PROPOSED LAPS WORK FOR RUA Compare 2D RTMA with 2D variational LAPS o Subjectively o Objectively against dependent / independent observations What additional, not listed features are desired of RUA? LAPS can focus on and add those If LAPS is deemed "not implementable" at NCEP o Fix shortcomings Less costly than adding special LAPS features into GSI? Add other desired features into LAPS such as o Visual / quantitative products for Visibility Particles
Thanks much ! Questions? More info at
Backup slides for additional information.
NextGen FAB Team Members Steve Albers - FAB contact, DA, Verification Scott Gregory - Ensemble Statistics Isidora Jankov - Ensemble Statistics Kirk Holub - AIV Verification Paula McCaslin - AIV Verification, Visualization Zoltan Toth - Project Guidance Yuanfu Xie - Data Assimilation
LAPS System Overview Data Ingest Intermediate data files GSI ENSEMBLE FORECAST MODEL Verification Analysis Scheme Downscaling can work as a stand alone module from background → GSI or other applications such as Fire wx. Downscaling is also an integral part of variational LAPS (aka. STMAS). Data Background (or cycled forecast) Observations Standalone downscaling module Traditional LAPS Variational LAPS (with downscaling) Model prep
Transition from Traditional to Fully Variational LAPS state vars, wind (u,v) clouds / precip balance and constraints in multi-scale variational analysis Wind analysis Temp/Ht analysis Humidity analysis Cloud analysis balance Traditional LAPS analysis: Wind, Temp, Humidity, Cloud, Balance Ultimately Temporary hybrid system : Traditional LAPS cloud analysis and balance Numerical Forecast model Large Scale Model First Guess Cycling Option var. LAPS
Example of Surface analysis Temp, wind 10 Apr Z
Three-Dimensional Cloud Analysis METAR LAPS HOT-START INITIALIZATION FHFLFHFL + FIRST GUESS *
3-hr Diabatically (hot-start) initialized WRF-ARW forecast Analysis
Cloud Analysis Flow Chart Cloud Fraction 3-D Isosurface * (From radars and model first guess)