Skip to content

20CR (v3)

NOAA-CIRES-DOE Twentieth Century Reanalysis Version 3

Discontinued

No longer updated

A fixed-period dataset produced once for 1836–2015 (with an experimental extension to 1806–1835); it will not be extended beyond 2015. As of 2026-10-11 the PSL 20CR pages mention no further extension or a version 4. Because it is almost the only ensemble reanalysis of the 19th- and early 20th-century atmosphere, it remains the standard dataset for that period.

Successor: None (ERA5 recommended from 1940) — No new version succeeding 20CR v3 has been released. ERA-20C and CERA-20C (1900–2010, both ended) are other long reanalyses that likewise assimilate only surface observations and are useful for comparison. If you only need the period from 1940, ERA5, which assimilates the full observing system, is more accurate.

At a glance

Item Details
Section Atmosphere › Reanalyses without satellite data
Type Reanalysis
Domains Atmosphere, Land
Provider NOAA Physical Sciences Laboratory (PSL), University of Colorado CIRES and U.S. Department of Energy (DOE) (distributed by NOAA PSL, NSF NCAR GDEX and NERSC)
Period 1836-01 ~ 2015-12
Coverage Global
Horizontal resolution Model T254 (0.703° × 0.702°, 512 × 256 Gaussian; described by PSL as 75 km grids) / PSL pressure-level output on a 1° × 1° grid (360 × 181)
Vertical resolution 64 model levels / 28 PSL pressure levels (1000–1 hPa; specific humidity up to 100 hPa)
Temporal resolution 3-hourly (8 times a day), daily and monthly means; ensemble mean and spread of 80 members
Update frequency Ended (fixed-period dataset)
Licence PSL data are freely available without restrictions (acknowledgement of PSL and copies of publications requested) / NCAR GDEX copy under CC BY 4.0
Account required No

Note on period: The released period is January 1836 to December 2015, with 1806–1835 as an experimental extension (released in 2021). In the NCAR metadata, the full range is 1805-12-31 18 UTC to 2016-01-01 00 UTC. Uncertainty is very large in the early 19th century, when observations are sparse, so always look at the ensemble spread.

How to read the resolution
  • Horizontal resolution lists both the native model grid and the grid you download, as given in the official documentation. The same dataset can come on different grids from different providers or products, so check the grid of your downloaded file.
  • Grid spacing is not the effective resolution. A provided grid that is finer than the native one carries no extra information.
  • Intervals and whether a value is instantaneous, mean or accumulated differ between variables — see 'Available variables' below.
  • The 'provided resolution' in lists and the comparison table is a simplified km/degree value for quick comparison. Use this page for exact values.

Overview

The Twentieth Century Reanalysis assimilates only surface observations of synoptic pressure into NOAA's Global Forecast System and prescribes sea surface temperature and sea-ice distribution, producing 3-hourly estimates of the global atmosphere from the early 19th century to 2015. Using a single observation type avoids artificial changes from the introduction of satellite and upper-air data, and the 80-member ensemble filter provides an uncertainty estimate for every analysis. It is mainly used for historical extreme events, long-term circulation variability, and the pre-1940 period that other reanalyses do not cover.

What changes with the successor

20CR v2c → 20CR v3

Transition: 2019

Version 3 moves to the autumn 2017 operational GFS, raising the resolution from T62 to T254, increases the ensemble from 56 to 80 members, and adds many more surface pressure observations from ISPD version 4.7. Adaptive covariance inflation and localization give more realistic uncertainty (ensemble spread) estimates, and the mid-19th-century sea-level pressure bias and precipitation spin-up problems are reduced. The record now starts in 1836 (1806 experimentally).

Aspect 20CR v2c 20CR v3
Period 1851–2014 (2013–2014 as an experimental extension) 1836–2015 (1806–1835 as an experimental extension, released in 2021)
Forecast model NCEP GFS circa 2009 (T62, 28 levels) Autumn 2017 operational NCEP GFS v14.0.1 + Noah land surface model + 2.5-layer thermodynamic sea-ice model, with dry-air mass conserved
Horizontal resolution T62 (about 1.9°), analysis output on a 2° × 2° grid T254 (about 0.7°, 512 × 256 Gaussian); described by PSL as 75 km grids
Vertical levels 28 levels 64 levels (PSL pressure-level output: 28 levels, 1000–1 hPa)
Ensemble size 56 members 80 members
Data assimilation Ensemble Kalman filter with fixed covariance localization (4000 km) and fixed inflation; digital filter initialization Ensemble square-root filter with 4D incremental analysis update (4D-IAU), adaptive localization, adaptive inflation via relaxation to prior spread (RTPS), adaptive quality control
Observations Surface and marine pressure only, ISPD version 3.2.9 Surface and marine pressure only, ISPD version 4.7 (including ICOADS 3.0, IBTrACS and newly digitized Old Weather and Weather Detective data). Marine pressures before 1870 bias-corrected against climatology
SST and sea-ice boundary conditions SST: SODAsi.2 pentads (to 2012), daily high-resolution blended analysis from 2013 / sea ice: COBE-SST2 SST: 8 members of SODAsi.3 (to 1980) and 8 members of HadISST2.2 (1981–2015) / sea ice: HadISST2.3
Output interval and format 6-hourly analyses, GRIB1 (NCAR) / netCDF (PSL) 3-hourly (8 times a day), netCDF4. Variable abbreviations differ from versions 2 and 2c (as noted by NERSC)
Known issues addressed Mid-19th-century low bias in global sea-level pressure (influence of Maury marine observations, 1851–1865), precipitation spin-up, unreliable confidence (spread) estimates Reduced sea-level pressure bias and artificial trend, no precipitation spin-up, more reliable confidence estimates, better storm and hurricane intensities
Remaining issues — Systematic biases in tropical precipitation and time-varying biases in Southern Hemisphere high-latitude pressure remain in version 3 (Slivinski et al. 2019)

Things to watch when switching

  • Versions 2c and 3 differ in boundary conditions, observations and model, so values differ even for the same year. Do not mix versions; when reproducing v2c results with v3, recompute climatologies and trends from scratch.
  • In v3 the SST input switches from SODAsi.3 to HadISST2.2 between 1980 and 1981, and files are split into SI (1806–1980) and MO (1981–2015) folders. Check for a discontinuity in any trend analysis that crosses this boundary.
  • Variable abbreviations and folder layout changed (e.g. sea-level pressure prmsl, 3-hourly accumulated precipitation apcp), so v2c scripts will not run unchanged.
  • The v3 ensemble spread is more reliable, so when interpreting 19th-century results where observations are sparse, report uncertainty from the spread files rather than relying on the ensemble mean alone.

Sources: Slivinski et al. (2019), Towards a more reliable historical reanalysis — 20CRv3, QJRMS, 145, 2876–2908 · NSF NCAR GDEX — 20CR Version 3 (d131003) · NSF NCAR GDEX — 20CR Version 2c (d131002) · reanalyses.org — 20CRv3 system details (compiled by the 20CR team) · reanalyses.org — New 20CR DA system presentation (comparison with the previous system) · NCAR Climate Data Guide — 20CR Version 2 and 2c · NOAA PSL news — Version 3 of the 20CR dataset will provide a 180-year history of the weather (2019-08-16)

Available variables

Main variables from the NOAA PSL netCDF files (OPeNDAP attributes). Short names are the in-file names; the PSL folder name is given in each group heading. Folder names ending in SI hold 1806–1980 files and MO hold 1981–2015 files (e.g. prsSI, prsMO). Spread files follow the same layout under spreads/.

Time-stamp convention — instantaneous, mean or accumulated? — click to show

The 20CR v3 files at PSL are 3-hourly (8 times a day) means of the 80-member ensemble, with time defined as the start of the interval (coordinate_defines 'start'). 00, 06, 12 and 18 UTC are analyses; 03, 09, 15 and 21 UTC are filled with 3-hour forecasts from those analyses (original pgrbensmeananl / +3 files). Variables available only from the forecast, such as precipitation, heat fluxes and radiation, are 3-hour averages or accumulations.

  • Original GRIB descriptors: analyses are 'anl', intermediate times '3 hour fcst', fluxes '0-3 hour ave fcst', precipitation and runoff '0-3 hour acc fcst', maximum and minimum temperature '0-3 hour max/min'
  • apcp is a 3-hour accumulation (kg/m^2 = mm); fluxes such as lhtfl and dswrf are 3-hour mean rates (W/m^2)
  • The PSL file attributes carry no averaging information (avg_period, cell_methods), and every statistic is 'Ensemble Mean'. Whether a field is a mean or an accumulation was inferred from the long_name ('3-hourly accumulated …') and the original GRIB descriptors
  • The PSL files do not state whether accumulation or averaging intervals start or end at the time stamp. The original GRIB labels them as 3-hour windows from the forecast start (0–3 h, 3–6 h)
  • Values in the spreads/ folder are the ensemble standard deviation at the same time; Dailies and Monthlies are daily and monthly averages of the 3-hourly values

Source: 20CRv3 output file lists from the 20CR team (reanalyses.org: pgrbensmeananl, +3, sflxgrbensmeanfg) + PSL netCDF file attributes

instantaneous mean accumulated maximum minimum static — instantaneous: value at the time stamp · mean/accumulated/maximum/minimum: over the indicated interval · static: does not change with time

Pressure levels (prsSI, prsMO; 3-hourly, 1000–1 hPa)

Variable Short name Units Time convention Notes
Air temperature air degK instantaneous
Geopotential height hgt m instantaneous
U- and V-components of wind uwnd, vwnd m/s instantaneous
Specific humidity (1000–100 hPa) shum kg/kg instantaneous

Near surface (2mSI, 2mMO, 10mSI, 10mMO)

Variable Short name Units Time convention Notes
2 m air temperature air degK instantaneous file air.2m
10 m U-component of wind uwnd m/s instantaneous file uwnd.10m

Single level and column (miscSI, miscMO)

Variable Short name Units Time convention Notes
Pressure reduced to MSL prmsl Pa instantaneous
Total cloud cover (entire atmosphere) tcdc % 3-hour mean file tcdc.eatm · Original GRIB descriptor for entire-atmosphere total cloud cover: '0-3 hour ave fcst'

Accumulations and snow (accumsSI, accumsMO; 3-hourly accumulation)

Variable Short name Units Time convention Notes
Total precipitation apcp kg/m^2 3-hour accumulated
Water equivalent of accumulated snow depth weasd kg/m^2 instantaneous (3-hour) Stored in the accums folder but a state variable; original GRIB descriptor '3 hour fcst' (instantaneous)

Surface fluxes (sfcFlxSI, sfcFlxMO)

Variable Short name Units Time convention Notes
Latent heat net flux lhtfl W/m^2 3-hour mean
Downward short-wave radiation flux at the surface dswrf W/m^2 3-hour mean file dswrf.sfc

The full variable list is in the official documentation.

How to get the data

Example code

Data from 1981 onward are in the MO folders; data up to 1980 are in the SI folders.

import xarray as xr

base = "https://psl.noaa.gov/thredds/dodsC/Datasets/20thC_ReanV3"
slp = xr.open_dataset(f"{base}/miscMO/prmsl.2014.nc")["prmsl"]   # Pa, 3-hourly ensemble mean
print(slp.sel(lat=37.5, lon=127.0, method="nearest").resample(time="1D").mean())

Practical tips

  • Reliability varies widely with observation density, even at the same time. Do not use the ensemble mean alone; check the spread in the spreads/ folder and avoid interpreting regions where the spread approaches the size of climate variability.
  • The SST boundary condition is SODAsi.3 through 1980 and HadISST2.2 from 1981, which is why the PSL folders are split into SI and MO. Analyses spanning 1980/1981 must read both folders, and should check for a discontinuity at that point.
  • Version 3 uses different variable abbreviations from versions 2 and 2c, as the NERSC portal also points out. Check variable names first when reusing old scripts.
  • The developers' evaluation shows systematic biases in tropical precipitation and time-varying biases in Southern Hemisphere high-latitude pressure. For these topics, cross-check results against other long reanalyses such as ERA-20C.

Notices and change history

2021-01-07 · Info · 20th Century Reanalysis v3 now extends back to 1806

PSL announced an experimental extension of version 3 back to 1806, making the full record 210 years long. Observations in the extension are very sparse, so uncertainty is large.

Read the original

2019-08-16 · Info · Version 3 of the 20CR dataset will provide a 180-year history of the weather

PSL announced that version 3, extending back to 1836, would be hosted at PSL starting in late September 2019. Compared with version 2c it offers more consistent uncertainty estimates and smaller errors, and it reduces an artificial trend in global sea-level pressure.

Read the original

References for your paper

Double-check before using in a paper

The citations below were compiled by the maintainer and may contain errors. Always double-check DOIs, authors, years and access-date formats against the provider's official dataset page and your journal's guidelines.

Dataset citation

NCAR GDEX 20CR v3 (d131003) — DOI 10.5065/H93G-WS83

NOAA-CIRES-DOE Twentieth Century Reanalysis Version 3. NSF NCAR Geoscience Data Exchange (GDEX), dataset d131003. https://doi.org/10.5065/H93G-WS83 (Accessed DD MMM YYYY) — copy the exact format, including authors and year, from the GDEX Citation tab.

Reference papers

20CR v3 system paper — DOI 10.1002/qj.3598

Slivinski, L. C., Compo, G. P., Whitaker, J. S., Sardeshmukh, P. D., Giese, B. S., McColl, C., Allan, R., Yin, X., Vose, R., Titchner, H., et al. (2019): Towards a more reliable historical reanalysis: Improvements for version 3 of the Twentieth Century Reanalysis system. Quarterly Journal of the Royal Meteorological Society, 145(724), 2876–2908.

Show BibTeX
@article{slivinski2019_20crv3,
  author  = {Slivinski, Laura C. and Compo, Gilbert P. and Whitaker, Jeffrey S. and Sardeshmukh, Prashant D. and Giese, Benjamin S. and McColl, Chesley and Allan, Rob and Yin, Xungang and Vose, Russell and Titchner, Holly and others},
  title   = {Towards a more reliable historical reanalysis: Improvements for version 3 of the Twentieth Century Reanalysis system},
  journal = {Quarterly Journal of the Royal Meteorological Society},
  volume  = {145},
  number  = {724},
  pages   = {2876--2908},
  year    = {2019},
  doi     = {10.1002/qj.3598}
}

20CR v3 performance evaluation (optional) — DOI 10.1175/JCLI-D-20-0505.1

Slivinski, L. C., et al. (2021): An Evaluation of the Performance of the Twentieth Century Reanalysis Version 3. Journal of Climate, 34(4), 1417–1438.

Original 20CR project paper (optional) — DOI 10.1002/qj.776

Compo, G. P., et al. (2011): The Twentieth Century Reanalysis Project. Quarterly Journal of the Royal Meteorological Society.

Acknowledgements

PSL asks papers using version 3 to include: "Support for the NOAA-CIRES-DOE Twentieth Century Reanalysis Project version 3 dataset is provided by the U.S. Department of Energy, Office of Science Biological and Environmental Research (BER), by the National Oceanic and Atmospheric Administration Climate Program Office, and by the NOAA Physical Sciences Laboratory." PSL also asks you to acknowledge PSL as the data provider and to send a copy of the relevant publications.

If this page helped, please acknowledge the Korean Climate Data Navigator

Please add the sentence below to the Acknowledgements of your paper.

Acknowledgements

Information on the datasets used in this study was compiled with the help of the Korean Climate Data Navigator (KCDN) (Sim, 2026; https://doi.org/10.5281/zenodo.23300923).