Notes on reforecasting and the computational capacity needed for future SREF systems Tom Hamill NOAA Earth System Research Lab presentation for 2009 National.

Slides:



Advertisements
Similar presentations
Numerical Weather Prediction Models
Advertisements

Weather-Climate Linkage Dave Novak (WPC) & Jon Gottschalck (CPC) January 21,
SPC Potential Products and Services and Attributes of Operational Supporting NWP Probabilistic Outlooks of Tornado, Severe Hail, and Severe Wind.
Statistical post-processing using reforecasts to improve medium- range renewable energy forecasts Tom Hamill and Jeff Whitaker NOAA Earth System Research.
“Where America’s Climate, Weather and Ocean Services Begin” NCEP CONDUIT UPDATE Brent A Gordon NCEP Central Operations January 31, 2006.
CHOICES FOR HURRICANE REGIONAL ENSEMBLE FORECAST (HREF) SYSTEM Zoltan Toth and Isidora Jankov.
A short term rainfall prediction algorithm Nazario D. Ramirez and Joan Manuel Castro University of Puerto Rico NOAA Collaborator: Robert J. Kuligowski.
2012: Hurricane Sandy 125 dead, 60+ billion dollars damage.
Instituting Reforecasting at NCEP/EMC Tom Hamill (ESRL) Yuejian Zhu (EMC) Tom Workoff (WPC) Kathryn Gilbert (MDL) Mike Charles (CPC) Hank Herr (OHD) Trevor.
The NCEP operational Climate Forecast System : configuration, products, and plan for the future Hua-Lu Pan Environmental Modeling Center NCEP.
Transitioning unique NASA data and research technologies to the NWS 1 Evaluation of WRF Using High-Resolution Soil Initial Conditions from the NASA Land.
A statistical downscaling procedure for improving multi-model ensemble probabilistic precipitation forecasts Tom Hamill ESRL Physical Sciences Division.
MOS Performance MOS significantly improves on the skill of model output. National Weather Service verification statistics have shown a narrowing gap between.
Introduction to Numerical Weather Prediction and Ensemble Weather Forecasting Tom Hamill NOAA-CIRES Climate Diagnostics Center Boulder, Colorado USA.
Exploitation of Ensemble Output (and other operationally cool stuff) at NCEP HPC Peter C. Manousos NCEP HPC Science & Operations Officer
Ensemble Post-Processing and it’s Potential Benefits for the Operational Forecaster Michael Erickson and Brian A. Colle School of Marine and Atmospheric.
Chapter 13 – Weather Analysis and Forecasting. The National Weather Service The National Weather Service (NWS) is responsible for forecasts several times.
1 Ensemble Reforecasts Presented By: Yuejian Zhu (NWS/NCEP) Contributors:
1 Global Ensemble Strategy Presented By: Yuejian Zhu (NWS/NCEP) Contributors: EMC staff.
Forecasting and Numerical Weather Prediction (NWP) NOWcasting Description of atmospheric models Specific Models Types of variables and how to determine.
WWRP Toward a Seamless Process for the Prediction of Weather and Climate: On the Advancement of Sub-seasonal to Seasonal Prediction TTISS, September 2009,
Experimental seasonal hydrologic forecasting for the Western U.S. Dennis P. Lettenmaier Andrew W. Wood, Alan F. Hamlet Climate Impacts Group University.
Exploring sample size issues for 6-10 day forecasts using ECMWF’s reforecast data set Model: 2005 version of ECMWF model; T255 resolution. Initial Conditions:
Weather and Climate Prediction Cliff Mass University of Washington.
Alan Robock Department of Environmental Sciences Rutgers University, New Brunswick, New Jersey USA
Preliminary Results of Global Climate Simulations With a High- Resolution Atmospheric Model P. B. Duffy, B. Govindasamy, J. Milovich, K. Taylor, S. Thompson,
A Comparison of the Northern American Regional Reanalysis (NARR) to an Ensemble of Analyses Including CFSR Wesley Ebisuzaki 1, Fedor Mesinger 2, Li Zhang.
1 Precipitation verification Precipitation verification is still in a testing stage due to the lack of station observation data in some regions
National Weather Service Application of CFS Forecasts in NWS Hydrologic Ensemble Prediction John Schaake Office of Hydrologic Development NOAA National.
Improving Ensemble QPF in NMC Dr. Dai Kan National Meteorological Center of China (NMC) International Training Course for Weather Forecasters 11/1, 2012,
Celeste Saulo and Juan Ruiz CIMA (CONICET/UBA) – DCAO (FCEN –UBA)
Notes on reforecasting and the computational capacity needed for future SREF systems Tom Hamill NOAA Earth System Research Lab presentation for 2009 National.
1 Agenda Topic: National Blend Presented By: Kathryn Gilbert (NWS/NCEP) Team Leads: Dave Myrick, David Ruth (NWS/OSTI/MDL), Dave Novak (NCEP/WPC), Jeff.
1 Using reforecasts for probabilistic forecast calibration Tom Hamill & Jeff Whitaker NOAA Earth System Research Lab, Boulder, CO NOAA.
1 Climate Test Bed Seminar Series 24 June 2009 Bias Correction & Forecast Skill of NCEP GFS Ensemble Week 1 & Week 2 Precipitation & Soil Moisture Forecasts.
1 An overview of the use of reforecasts for improving probabilistic weather forecasts Tom Hamill NOAA / ESRL, Physical Sciences Div.
Short-Range Ensemble Prediction System at INM José A. García-Moya SMNT – INM 27th EWGLAM & 12th SRNWP Meetings Ljubljana, October 2005.
Course Evaluation Closes June 8th.
Interoperability at INM Experience with the SREPS system J. A. García-Moya NWP – Spanish Met Service INM SRNWP Interoperability Workshop ECMWF –
Plans for Short-Range Ensemble Forecast at INM José A. García-Moya SMNT – INM Workshop on Short Range Ensemble Forecast Madrid, October,
APPLICATION OF NUMERICAL MODELS IN THE FORECAST PROCESS - FROM NATIONAL CENTERS TO THE LOCAL WFO David W. Reynolds National Weather Service WFO San Francisco.
Slides for NUOPC ESPC NAEFS ESMF. A NOAA, Navy, Air Force strategic partnership to improve the Nation’s weather forecast capability Vision – a national.
Insights from CMC BAMS, June Short Range The SPC Short-Range Ensemble Forecast (SREF) is constructed by post-processing all 21 members of the NCEP.
18 September 2009: On the value of reforecasts for the TIGGE database 1/27 On the value of reforecasts for the TIGGE database Renate Hagedorn European.
Working group III report Post-processing. Members participating Arturo Quintanar, John Pace, John Gotway, Huiling Yuan, Paul Schultz, Evan Kuchera, Barb.
Evapotranspiration Estimates over Canada based on Observed, GR2 and NARR forcings Korolevich, V., Fernandes, R., Wang, S., Simic, A., Gong, F. Natural.
Statistical Post Processing - Using Reforecast to Improve GEFS Forecast Yuejian Zhu Hong Guan and Bo Cui ECM/NCEP/NWS Dec. 3 rd 2013 Acknowledgements:
CTB computer resources / CFSRR project Hua-Lu Pan.
Federal Department of Home Affairs FDHA Federal Office of Meteorology and Climatology MeteoSwiss A more reliable COSMO-LEPS F. Fundel, A. Walser, M. A.
Reanalysis Needs for Climate Monitoring Craig S Long Arun Kumar and Wesley Ebisuzaki CPC NOAA Climate Test Bed (CTB) Meeting – November 9-10, NCWCP.
Exploring Multi-Model Ensemble Performance in Extratropical Cyclones over Eastern North America and the Western Atlantic Ocean Nathan Korfe and Brian A.
Ensemble prediction review for WGNE, 2010 Tom Hamill 1 and Pedro de Silva-Dias 2 1 NOAA/ESRL 2 Laboratório Nacional de Computação Científica
1 NCEP’s Climate Forecast System as a National Model “Where America’s Climate, Weather and Ocean Services Begin” 32 nd Climate Diagnostics and Prediction.
Fly - Fight - Win 2 d Weather Group Mr. Evan Kuchera HQ AFWA 2 WXG/WEA Template: 28 Feb 06 Approved for Public Release - Distribution Unlimited AFWA Ensemble.
DOWNSCALING GLOBAL MEDIUM RANGE METEOROLOGICAL PREDICTIONS FOR FLOOD PREDICTION Nathalie Voisin, Andy W. Wood, Dennis P. Lettenmaier University of Washington,
VERIFICATION OF A DOWNSCALING SEQUENCE APPLIED TO MEDIUM RANGE METEOROLOGICAL PREDICTIONS FOR GLOBAL FLOOD PREDICTION Nathalie Voisin, Andy W. Wood and.
Probabilistic Forecasts Based on “Reforecasts” Tom Hamill and Jeff Whitaker and
HYDROCARE Kick-Off Meeting 13/14 February, 2006, Potsdam, Germany HYDROCARE Actions 2.1Compilation of Meteorological Observations, 2.2Analysis of Variability.
NOAA Northeast Regional Climate Center Dr. Lee Tryhorn NOAA Climate Literacy Workshop April 2010 NOAA Northeast Regional Climate.
Regional Re-analyses of Observations, Ensembles and Uncertainties of Climate information Per Undén Coordinator UERRA SMHI.
© Crown copyright Met Office Review topic – Impact of High-Resolution Data Assimilation Bruce Macpherson, Christoph Schraff, Claude Fischer EWGLAM, 2009.
Current Issues and Challenges in Ensemble Forecasting Junichi Ishida (JMA) and Carolyn Reynolds (NRL) With contributions from WGNE members 31 th WGNE Pretoria,
Overview of Deterministic Computer Models
Jennifer Boehnert Emily Riddle Tom Hopson
Nathalie Voisin, Andy W. Wood and Dennis P. Lettenmaier
University of Washington Center for Science in the Earth System
Christoph Gebhardt, Zied Ben Bouallègue, Michael Buchhold
Environmental Statistics
MOGREPS developments and TIGGE
Presentation transcript:

Notes on reforecasting and the computational capacity needed for future SREF systems Tom Hamill NOAA Earth System Research Lab presentation for 2009 National Workshop on Mesoscale Probabilistic Prediction NOAA Earth System Research Laboratory

Part I: notes on reforecasting and its applicability to SREF

What are reforecasts? Forecasts for dates in the past using the same forecast model and data assimilation system as is used operationally. Also called “hindcasts” but we prefer “reforecast” to make the connection with reanalyses. The (relatively) consistent data set allows effective statistical post-processing algorithms to be developed, deterministic and probabilistic.

Reforecasting, lessons learned Common misapprehension: reforecasting is not a calibration technique, but a data set that facilitates a wide array of calibration techniques. Reforecasts provide statistical downscaling (as opposed to dynamic, i.e. hi-res models) Reforecasts improve calibration of long-lead forecasts, rare-event forecasts. In both cases, need large sample size. Less impact relative to short training data sets for short-term temperature forecasts. Reforecasts are important to the hydrologic community, who need to train and test on large samples to cover many flooding events. Can’t do post-processing without quality reanalysis and observation data sets – Up-to date reanalyses, to initialize previous year’s forecasts with the same data assimilation system used in real time, achieving consistency between training data set and real-time forecasts. – Long time series of observations (for post-processing model development) Reforecasts may be more difficult with multi-model SREF, may affect choice of SREF system. Reforecasts are no longer an exotic technology; operational at ECMWF, soon at NCEP with Climate Forecast System Reanalysis & Reforecast.

5 Statistical downscaling using T62 GFS reforecasts The relative merits of statistical vs. dynamical downscaling is an important research question to answer.

6 Example of imperfections in 2005 ECMWF, 1998 GFS 5-mm reliability diagrams, raw ensembles horizontal lines indicate distribution of climatology error bars from block bootstrap Raw forecasts have poor skill in this strict BSS

7 5-mm reliability diagrams, after reforecast calibration

8 Precipitation skill with 1x weekly (x 20 year), 30-day, and full training data sets Note the substantial benefit of weekly relative to 30-day training data sets, especially at high thresholds. Note diminishing returns of increasing sample size beyond 1x weekly.

Part II: sizing the computers we need.

Current ensemble forecasts at NCEP Current NCEP Global EPS configuration: 4 cycles daily, 20 members/cycle, to 16 days. Uses single-model T128L28 (~150 km). Near-term plans to increase to T190L28, which we will assume as the current baseline. Through North American Ensemble Forecast System (NAEFS), access to Canadian ensemble data (0.9-degree, L58, 20 members/cycle). Current NCEP SREF configuration: 4 cycles daily, to 87h lead, 21 members/cycle, horizontal grid spacings ~40 km, !~28 levels. Initial conditions and LBCs interpolated from NCEP global model. Soon resolution upgrade. Assume 32kmL28M20 as baseline for existing SREF system.

What are other weather prediction facilities running currently? Global models: current reference standard and world-best is ECMWF modeling system. – ECMWF implemented a 2x daily, T255 global ensemble (50 members) in Nov. 2000; a T399 ensemble in Nov. 2006; are planning for T639 ensemble in Their strategic plan calls for a 10-km deterministic and 20-km global ensemble (~ T1000) by 2015.strategic plan Regional ensemble systems. Future plans for regional ensembles are more vague, and computational requirements depend on domain size, which vary radically. Current regional ensemble prediction systems include: – UK Met Office: 24 members, 24-km grid spacing to 54 h. Plans for 18-km L70 regional and 1.5-km small regional ensemble around British Isles by – DWD (Germany): 2.8-km, 20 member regional ensemble over Germany, run currently. Plans beyond 2013 TBD. – CMC (Canada): Plans for regional ensemble of 33 km/L60, 20 members to 48-h lead by km L70 to 72h, 20 members in – Several consortia in Europe to share regional ensemble model output and distribute computations, e.g., COSMO consortium; talk. w. Jose Garcia-Moya for more details.

2-m temperature skill from TIGGE data set T-2m, DJF 2008/09 NH (20°N - 90°N) BC vs. ERA-interim ECMWF’s investment in better, higher-resolution models, data assimilation provides approximately 2 extra days of forecast lead over the nearest competitor. An investment to improve resolution thus can be expected to provide substantial benefit to NWS.

The impacts of improved data assimilation and higher-resolution models North America Z500 RMSE for the control experiments and latest upgrades of the MSC global analysis-forecast system (January and February 2007) Acknowledgements to S. Laroche, Environment Canada

A proposal for the sizing of future NCEP ensemble systems Straw(person) : NCEP should be aiming for the following operational capacity by 2016, in order to be competitive with the best in the world (100% capacity): Global ensembles: T1000 L100 M20 (~20-km global ensemble) 4x daily to day 3; T500L50M20 to 7 days; T250L50M20 to 15 days lead. Accompanying reforecasts in order to statistically post-process. 255x more CPU than current; 147x more storage than current. Regional ensemble: – (1) SREF on large regional domain (similar to current domain, covering N America) at 4 km L100 M20, 4x daily to 24h lead, thereafter 12kmL50 M20 to 87-h lead. Accompanying reforecasts (5 members every 3 rd day over 20 years) in order to statistically post-process. EnKF initialization. – (2) Limited-area, high-impact, on-demand ensemble prediction system run within SREF domain. 2 km, L100 M20 to 24-h lead. Assume 4 regional domain ensembles computed per day. Assume domain will cover 1/4 the size of the existing SREF domain in both E-W and N-S directions (1/16 th the area). No reforecasts. – 3906x more CPU than current; 1283x more mass storage than current. A laughably high cost estimate, with NOAA’s method of calculating CPU costs (same future CPU cost/cycle as in 2008.

A (more modest) proposal for the sizing of future NCEP ensemble systems Global ensembles: T1000 L100 M20 (~20-km global ensemble) 4x daily to day 3; T500L50M20 to 7 days; T250L50M20 to 15 days lead. Accompanying reforecasts in order to statistically post-process. 255x more CPU than current; 147x more storage than current.. Regional ensemble: – (1) 12-km L50M20 4x daily to 87h. Regional 4-km ensemble to 24-h, run ~4x daily. – (2) Reforecasts with 12-km ensemble, 10 members, every 3 rd day. – (3) Total: 265x more CPU than current; 61x more mass storage A more modest cost estimate, with NOAA’s method of calculating CPU costs (same future CPU cost/cycle as in 2008). This didn’t cause sticker shock at HQ NOAA

My perception Increased CPU for numerical weather prediction is perceived as important, but it’s not yet near the list of the most important upgrades NOAA wants to fund. And it darned well ought to be.

A signed statement to the NWS, NOAA, and Dept. of Commerce Leadership? A Recommendation from the National Workshop on Mesoscale Ensemble Prediction concerning the Procurement of High-Performance Computing by the National Weather Service. Whereas the United States lags badly in the skill of its numerical weather predictions relative to international competitors; Whereas the United States must constrain the resolution of its numerical weather prediction models to much less than those of its international peers so that they may run on far smaller computers; Whereas NOAA’s investment in high-performance computing for weather prediction is approximately two orders of magnitude smaller than its investment in weather satellites it is required to process; Whereas the lack of sufficient high-performance computing is a primary impediment to the development and testing of improved weather prediction models and methods; Whereas the transition to probabilistic weather prediction involves additional computational burdens, such as the computation of ensembles of forecasts and the statistical post-processing of these using additional past forecasts and observations; Whereas improving model resolution and implementing ensemble prediction techniques has consistently resulted in improved numerical weather predictions; Whereas improving the skill of these numerical weather predictions dramatically improves forecasts of high- impact events such as hurricanes, tornadoes, floods, heat waves, and cold spells; The members of the 2009 National Workshop on Mesoscale Ensemble Prediction unanimously recommend that NOAA procure greatly increased computational capacity for weather prediction to support higher- resolution models and improved ensemble forecast techniques. Signed, Lewis F. Richardson et al.

Forecasting-system improvement at ECMWF % % Historical trend Updated from Simmons & Hollingsworth (2002) Acknowledgements to A. Simmons Historical re-forecast project trend using re-analyses c/o Gilbert Brunet THORPEX presentation