Reanalysis of the 20th Century Gilbert P. Compo, Jeffrey S. Whitaker, and Prashant D. Sardeshmukh Univ. of Colorado/CIRES Climate Diagnostics Center and NOAA Earth System Research Laboratory/PSD Special thanks to N. Matsui, R. Vose, R. Allan, S. Woodruff

The 20th Century Reanalysis Project Summary: An international collaborative project led by NOAA and CIRES to produce high-quality tropospheric reanalyses for the last 100 years using only surface observations. The reanalyses will provide: -First-ever estimates of near-surface and tropospheric 6-hourly fields extending back to the beginning of the 20 th century; -Estimates of biases and uncertainties in the basic reanalyses; -Estimates of biases and uncertainties in derived quantities (storm tracks, etc.) Initial product will have higher quality in the Northern Hemisphere than in the Southern Hemisphere. US Department of Energy INCITE computing award and NOAA Climate Program Office support to produce by early Initially produce

Ensemble Filter Algorithm x j b = b + x’ j b = first guess jth ensemble member ( j=1,…,56 ) y o = single observation with error variance R First guess interpolated to observation location : b = H b, y’ j b = H x’ j b Form analysis ensemble x j a = a + x’ j a from a = b + K ( y o - b ) x’ j a = x’ j b +  M (-y’ j b ) Note the different gain K =  j x’ j b y’ j b (  j y’ j b y’ j b + R) -1 Kalman Gain  M = (1 + {R/(  j y’ j b y’ j b + R)} –1/2 ) -1 K Modified Kalman Gain  shrinks the ensemble (1/(n-1)) is included in  j Analysis ensemble becomes first guess ensemble for next observation. Using 56 member Ensemble Filter T62 (about 2 degree), 28 level NCEP CFS03 model HadISST monthly boundary conditions (Rayner et al. 2003)

International Surface Pressure Databank All Union Research Institute of Hydrometeorological Information WDC Atmospheric Reconstructions over the Earth Australian Bureau of Meteorology British Antarctic Survey Danish Meteorological Institute Deutscher Wetterdienst EMULATE Environment Canada ETH-Zurich GCOS AOPC/OOPC Working Group on Surface Pressure Hong Kong Observatory ICOADS Instituto Geofisico da Universidade do Porto Japanese Meteorological Agency Jersey Met Dept. KNMI MeteoFrance Meteorological and Hydrological Service, Croatia National Center for Atmospheric Research NOAA Climate Database Modernization Program NOAA Earth System Research Laboratory NOAA National Climatic Data Center NOAA National Centers for Environmental Prediction NOAA Northeast Regional Climate Center at Cornell U. NOAA Midwest Regional Climate Center at UIUC Norwegian Meteorological Institute Ohio State U. – Byrd Polar Research Center Proudman Oceanographic Laboratory SIGN - Signatures of environmental change in the observations of the Geophysical Institutes South African Weather Service UK Met Office Hadley Centre U. of Colorado-CIRES/Climate Diagnostics Center U. of East Anglia-Climatic Research Unit U. of Lisbon-Instituto Geofisico do Infante D. Luiz U. of Lisbon- Instituto de Meteorologia U. of Milan-IFGA U. Rovira i Virgili-CCRG Subdaily observations assembled under GCOS AOPC/OOPC Working Group on Surface Pressure GCOS/WCRP Working Group on Observational Data Sets for Reanalysis NOAA NCDC, NOAA ESRL, and CU/CIRES: merging station data NOAA ESRL and NCAR (ICOADS): merging marine data Thank you to partners contributing observations:

Sea Level Pressure analyses for Tri-State Tornado Outbreak of 18 March 1925 (deadliest tornado in U.S. history) Manual Analysis, courtesy C. DoswellEnsemble mean from Ensemble Filter (4 hPa interval, 1010 hPa thick) NOTE!!! This analysis did not use ANY of the observations shown on the left.

Range of possibilities for Sea Level Pressure 18 March Z using 14 (of 56) members Ensemble of 56 possible realizations consistent with the observations

SLP 1 December hPa GPH SLP 1 December hPa GPH SLP 1 December hPa GPH SLP 1 December hPa GPH Analysis Ensemble Mean and Spread on selected dates in the reanalysis period Contours-ensemble mean Shading-blue: more uncertain, white: more certain Sea Level Pressure500 hPa Geopotential Height

Uncertainty estimate poleward of 20N blue actual difference red expected difference Nobs

Reanalysis skill is comparable to current Day-3 forecast skill ! Analyzed 500 mb Z compared with independent radiosonde obs ( ) * Assumes observation error of 20 m Median Absolute Error = 22.4 m Expected Error = 21.5 m* Analyzed Anomaly Observed Anomaly Correlation = 0.97 N = 7106 (courtesy S. Bronniman and A. Grant, ETH) Day 3 Pattern Correlation September-October Day-3, Day-5, Day mb Z anomaly correlation forecast skill from NCEP and ECMWF (Sept-Oct 2007)

300 hPa Geopotential Height Verification using independent radiosonde observations ( ) (courtesy S. Bronniman and A. Grant, ETH) N = 6619 Correlation = 0.94 * Assumes observation error of 35 m Median Abs Error = 39 m Expected Error = 37 m*

Summary Accuracy: Mid-tropospheric NH fields about as accurate as current 3-day forecasts. Uses: 1.Effectively doubling the reanalysis record length 2.Climate model validation dataset for large-scale synoptic anomalies during extreme periods, such as droughts (30’s, 50’s) and extreme tornado outbreaks. 3.Better understand events such as the ’s Arctic warming. Timeline: done, full 1892-present done by Spring Data Access: Will be freely available from NCAR, NOAA/ESRL and NOAA/NCDC in Fall in Spring , Partnering with GCOS Working Group on Surface Pressure, GCOS/WCRP Working Group on Observational Datasets for Reanalysis, Atmospheric Circulation Reconstructions over the Earth, RECLAIM, NOAA/CDMP, many National and University surface pressure data recovery efforts. For status updates, or

Future plans with ACRE : Early to mid-19th Century Reanalysis (with DoE and NOAA support) Version 3: mid-19th – 21st century would need all data by August 2010 improved version of NCEP model at higher resolution => hurricanes, high impact phenomena? potentially available in 2012

Obstacles to Improving Reanalyses of the 19th to 21st Centuries Model Error –Need improved representation of errors and unresolved processes (stochastic parameterization) Petascale Computing power –Need higher resolution for mesoscale extremes, e.g., hurricanes Additional observations needed –Millions of land and marine observations not yet recovered or exchanged –Working with partners in Met Services, Universities, and international initiatives such as ACRE, NOAA CDMP, GEOSS/GCOS, IEDRO, RECLAIM, WCRP, WMO DARE

ISPD only ISPD plus HURDAT-derived pressure Effect of HURDAT data on Sept 1938 “Long Island Express”

Any Skill Forecasting the Track? using 56 ensemble members T254L64 (about 0.5 degree) 36 hour forecast verifying 21 Sept Z

Summary Accuracy: Mid-tropospheric NH fields about as accurate as current 3-day forecasts. Uses: 1.Effectively doubling the reanalysis record length 2.Climate model validation dataset for large-scale synoptic anomalies during extreme periods, such as droughts (30’s, 50’s) and extreme tornado outbreaks. 3.Better understand events such as the ’s Arctic warming. Timeline: done, full 1892-present done by Spring Data Access: Will be freely available from NCAR, NOAA/ESRL and NOAA/NCDC in Spring in Spring , Partnering with GCOS Working Group on Surface Pressure, GCOS/WCRP Working Group on Observational Datasets for Reanalysis, Atmospheric Circulation Reconstructions over the Earth (Rob Allan at the UK Hadley Centre), RECLAIM, NOAA/CDMP, many National and University surface pressure data recovery efforts. For status updates, or

Analysis Ensemble Mean and Spread on selected dates in the reanalysis period SLP 1 December hPa GPH Contours-ensemble mean Shading-blue: more uncertain, white: more certain Sea Level Pressure500 hPa Geopotential Height

SLP 1 December hPa GPH Analysis Ensemble Mean and Spread on selected dates in the reanalysis period Contours-ensemble mean Shading-blue: more uncertain, white: more certain Sea Level Pressure500 hPa Geopotential Height

T700 Verification - Independent Obs (courtesy S. Bronniman and A. Grant, ETH) MAE = 1.59 K Expected Error = 1.73 K* MAE = 1.58 K Expected Error = 1.71 K* * Assumes observation error of 1.5 K Anom Corr = 0.87Anom Corr = 0.90

Uncertainty estimate poleward of 20N blue actual difference red expected difference Nobs

: 24-h forecast skill Nobs

Sea Level Pressure Analyses December 1947 US Weather BureauEnsemble Filter Intensity of East Coast Low is weaker in Ensemble Filter analyses, but major features are present. (Kocin and Uccellini, 2004) Contour interval is 4 hPa in all panels and 1012 hPa contour is thickened. L L L

Analyses of 500 hPa Geopotential Height 27 December 1947 Air Weather Service analysis valid 0400GMT Using only surface pressure observations, Ensemble Filter analysis compares well with analyses using upper-air observations m (18000 ft) contour is thickened Ensemble Filter T62 analysis valid 0600GMT NCEP T254 3 hr forecast valid 0600GMT Using only surface pressure obs Using all available surface and upper-air obs Severe snowstorm case in New York City, USA.

RMS error of 6-hourly 500 mb Z analyses for Jan-Feb 2005 using ECWMF 3DVAR and only Surface Pressure observations at 1930’s densities ECWMF System Retuned for only surface pressure ECWMF2005 T159L64 Courtesy J. Thepaut Northern Hemisphere Analysis Error

ECWMF2005 T159L64 Courtesy J. Thepaut Need system tailored for Climate Reanalyses Northern Hemisphere Analysis Error RMS error of 6-hourly 500 mb Z analyses for Jan-Feb 2005 using ECWMF 3DVAR, 4DVAR, NOAA/CIRES Ensemble Data Assimilation and only Surface Pressure observations at 1930’s densities ECWMF System Retuned for only surface pressure

Effect of HURDAT data at on Sept 1938 “Long Island Express” No HURDAT data

Effect of HURDAT data on Sept 1938 “Long Island Express” With HURDAT data