ERA-‐20C: the ECMWF reanalysis of the 20th century

ERA-­‐20C: the ECMWF reanalysis of the 20th century (1899-­‐2010) using surface observaCons only Paul Poli, Hans Hersbach, Dick Dee, Adrian Simmons, Paul Berrisford, David Tan, and Carole Peubey Slide 1
© ECMWF MoCvaCons ●  Can a century-­‐long dataset bridge weather and climate scales? –  Inspired by Gil Compo’s 20th Century Reanalysis –  Mandated by call for proposals from EU FP7  ERA-­‐CLIM project ●  Challenges –  ScienCfic ●  How to handle a changing observing system (great increase in the global number of observaTons over the course of the 20th century) ●  How to detect automaTcally data issues (breaks in staTon Tme-­‐series, staTonary …) ●  How to handle greater unknowns in the Earth System components as we go back in Tme ●  Can we really produce meaningful maps (by today’s standards) for 1900? –  Technical ●  ProducTon speed and throughput ●  Assembling large datasets prior to and aXer producTon Slide 2
© ECMWF SimplisCc view of the reanalysis system ●  Input –  Invariant forcings –  Time-­‐varying forcings –  ObservaTons ●  AssimilaTon and model –  Atmosphere –  Land-­‐surface –  Ocean waves ●  Output – 
– 
– 
– 
Monitoring DocumentaTon Budgets Products Slide 3
Integration works forward in time
Dataset is built from the past into the present
Started on 1 January 1899
Production used several parallel ‘streams’
© ECMWF At the core of it: EsCmaCng and using uncertainCes in an ensemble of 24-­‐hour 4DVAR data assimilaCons 10
members
One can see this as a “Monte-Carlo” equivalent to the reanalysis
estimation problem: The ensemble is supposed to contain all sources
of uncertainties (nay, it doesn’t…)
so that product spread is a measure of product uncertainty
Slide 4
© ECMWF Great Blizzard of February 1899 (i.e., only 1 month aQer spin-­‐up) Kocin et al., Weather and Forecasting, 1988
doi: http://dx.doi.org/10.1175/1520-0434(1988)003<0305:TGAOAE>2.0.CO;2
Slide 5
© ECMWF Surface pressure (top) and wind (boRom) data counts/year Slide 6
© ECMWF Total uncertainty verificaCon for observaCons assimilated: error budget closure Showing only observaTons assimilated in the first 90 minutes of the 24-­‐hour 4DVAR window in ERA-­‐20C ensemble – hence not an independent verifica1on Assumed
Slide 7
© ECMWF Actual
Temperature analysis mean global increments (Jan 1979-­‐Dec 2010) [K]
ERA-Interim
ERA-20C ensemble
ERA-20C deterministic
Slide 8
© ECMWF Longwave downwelling radiaCon at NDBC Buoy#44008 (hourly) Observations (in red) come from a NDBC buoy augmented by sensors funded by project "New England
Shelf Fluxes“ sponsored by JAII: Massachusetts Technology Collaborative's John Adams Innovation
Institute. Data retrieved from Woods Hole Oceanographic Institution website on 26 April 2014
Slide 9
© ECMWF Total column water vapor over oceans, laCtudes 20oS-­‐20oN Slide 10
© ECMWF SSM/I CM-­‐SAF FCDR 22 GHz channel (comparison using NWP-­‐SAF RTTOV-­‐11), over oceans, laCtudes 20oS-­‐20oN, non-­‐rainy pixels Average brightness temperatures at obs. date/time/location/sensor/satellite
Slide 11
© ECMWF Top: Total column water vapor BoRom: SSM/I CM-­‐SAF FCDR 22 GHz brightness temperatures over oceans, laCtudes 20oS-­‐20oN (boRom non-­‐rainy pixels only) Obs4MIPs-type of evaluation: “bringing observations to the modellers”
Reverse: “bringing model outputs to the observers” … MIPs4Obs?
Slide 12
© ECMWF Comparison with surface temperatures measured by Russian ice staCons over the ArcCc (intersec1on of observa1ons available at ECMWF and in ICOADS 2.5.1 to guarantee data source/origin) ERA-Interim
Submitted to QJRMS
Slide 13
© ECMWF ERA-20C
Unpublished work
Conclusions & outlook ●  Our first century ensemble reanalysis producTon taught us many lessons –  Applied in a determinisTc run (couldn’t afford a full ensemble re-­‐run) ●  Stepping stone to adding now upper-­‐air observaTons –  And later satellite data ●  Next century-­‐long reanalysis will use the CERA system (Coupled atmosphere/
ocean), developed by Laloyaux et al. –  Surely many more issues will appear then –  But we’ll learn a great deal about discrepancies between atmosphere & ocean models & observaTons ●  Reanalysis is an iteraTve work –  Serving users in the mean Tme ●  We hope to have ERA-­‐20C data out on the public data server ‘soon’ –  IniTal evaluaTon was essenTal to avoid issuing raw incorrect ensemble data –  Data copying proves more Tme-­‐consuming than producTon itself! ●  Fair assessment of ERA-­‐20C with a few observaTonal, independent datasets Slide 14
© ECMWF ERA-­‐20C fine print ●  Horizontal resoluTon T159 (approx. 125 km, as in ERA-­‐40), 91 model levels up to 0.01 hPa ●  Analysis increments at T95 (approx. 210 km) ●  Model version IFS CY38R1, with added Tme-­‐varying forcings: HadISST2.1.0.0 (sea-­‐surface temperature and ice ● 
● 
● 
● 
● 
● 
● 
● 
● 
● 
fracTon), greenhouse gases (O3, CO2, CH4, N2O, CFC-­‐11, CFC-­‐12, CFC-­‐22, CCL4), solar cycle, aerosols opTcal depth, as described in model-­‐only integraTon documentaTon, ERA Report Series 16, Hersbach et al., 2013 www.ecmwf.int ObservaTons source: ISPD v3.2.6 and ICOADS v2.5.1 Two producTons so far: a 10-­‐member ensemble (~200 days in 2013) and a determinisTc re-­‐run (~50 days in 2014) Data assimilaTon method different from ECMWF NWP operaTons in 2013/14: –  24-­‐hour 4DVAR, 3-­‐hourly output for analyses –  Ensemble updates of background error correlaTons and global variances every 10 days –  DeterminisTc uses constant (current) background error correlaTons, and global variances scaled to the ensemble –  ModulaTon of background errors in vorTcity by daily maps of ensemble spread to reduce linear model instabiliTes –  Digital filtering of increments in vorTcity and temperature acTvated VariaTonal bias correcTon of surface pressure observaTons –  Using prior detecTon of Tme-­‐series breaks based on SNHT homogeneity test using NOAA-­‐CIRES 20CR departures –  DeterminisTc also includes detecTon and rejecTon of staTonary observed Tme-­‐series (quite a few in the 1990s…) AssimilaTon of marine surface wind observaTons from ICOADS2.5.1 Data produced in 6x 20-­‐year streams for the ensemble, and 22x 5-­‐year streams for the determinisTc Data volume generated about 700 Tb for ensemble, 75 Tb for determinisTc Model Tme-­‐step of 60 minutes for ensemble, 30 minutes for determinisTc ( beoer atmospheric Tdes) Only ensemble documentaTon available so far: ERA Report Series 14, Poli et al., 2013 www.ecmwf.int –  Documents several issues found with ensemble producTon, all fixed in determinisTc producTon Data currently being copied to a public data server for release Slide 15
© ECMWF