Berkeley Earth¶

Berkeley Earth Land/Ocean Surface Temperature

Available

At a glance¶

Note on period: Global land + ocean TAVG starts in 1850, land-only TAVG in 1750, and high-resolution land TMAX and TMIN in 1850. When checked on 2026-10-12, the headers of the high-resolution global monthly files gave analysis run dates of 2026-09-11 (land) and 2026-08-13 (ocean); the last row (latest month) was not checked directly. The old 1° product was last run on 2025-01-04 and has not been updated since.

Overview¶

A global surface temperature record from Berkeley Earth, an independent U.S. non-profit, combining air temperatures from about 58,000 land stations with HadSST4 sea surface temperatures from 1850 onward. Its hallmarks are a statistical homogenization that splits station records at discontinuities ("scalpel") and Kriging interpolation; the land record extends back to 1750. Since 2025 the flagship product is the 0.25° high-resolution record built with regression Kriging on physically informed predictors, with anomalies relative to 1951–1980. The gridded files include a monthly climatology, so absolute temperatures can be reconstructed.

What changes with the successor¶

Berkeley Earth Land/Ocean → Berkeley Earth BEST-HR¶

Transition: 2025 Q2

From Q2 2025 the 0.25° high-resolution product became the operational record and the 1° time series and gridded products stopped updating. The description paper (in review at ESSD) presents it as version 2 of the Berkeley Earth surface temperature record.

Sources: Berkeley Earth data page · Rohde et al. (2026, in review at ESSD) BEST-HR paper

Available variables¶

Variable names from the high-resolution gridded NetCDF files (per the README). Variable names in the old 1° product files were not checked directly this time.

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

High-resolution grids (NetCDF, 0.25°, 1°, 5°)¶

Summary time series (text)¶

Old product daily land data (experimental, no longer updated)¶

The full variable list is in the official documentation.

How to get the data¶

• Berkeley Earth data page — HTTPS (web / wget / curl) No login needed
  • Lists the high-resolution global and land time series and 1° and 5° gridded files (storage.googleapis.com/berkeley-earth-temperature-hr/), the old 1° products (no longer updated) and station data. The full 0.25° grids and high-resolution country and city series are provided through a request form.
• High-resolution 1° grid (global TAVG) — HTTPS (about 400 MB) No login needed
  • The 5° grid is Global_TAVG_Gridded_5deg.nc (about 8 MB) in the same folder.
• Zenodo archive (April 2026 release, includes 0.25°) — HTTPS (about 55 GB, 21 files) No login needed
  • Contains monthly, annual and ensemble time series for global TAVG and land TAVG/TMAX/TMIN, 1° ensemble NetCDF files and 0.25° gridded NetCDF files. As a frozen release it suits reproducible publications; get the latest months from the data page files.
• Old 1° product (no longer updated, January 2025 run) — HTTPS (about 400 MB) No login needed
  • The Equal Area version (Land_and_Ocean_EqualArea.nc), the Alternate version (water temperatures under sea ice), land Complete_TAVG/TMAX/TMIN and the daily land files are in the same location.

Example code¶

Downloads the high-resolution 5° grid, reconstructs absolute temperature from anomaly plus climatology, and computes the area-weighted global mean anomaly.

import urllib.request
import numpy as np
import xarray as xr

url = ("https://storage.googleapis.com/berkeley-earth-temperature-hr/"
       "global/gridded/Global_TAVG_Gridded_5deg.nc")
urllib.request.urlretrieve(url, "be_5deg.nc")
ds = xr.open_dataset("be_5deg.nc")
print(ds)                                  # check dimension and coordinate names

anom = ds["temperature"]                   # °C, anomaly relative to 1951–1980
tdim = ds["time"].dims[0]
ydim, xdim = ds["latitude"].dims[0], ds["longitude"].dims[0]
t = ds["time"].values                      # decimal year, mid-month (e.g. 1981.125)
month = np.floor((t % 1) * 12).astype(int) # 0 = January … 11 = December

# Area-weighted global mean anomaly
w = np.cos(np.deg2rad(ds["latitude"]))
glob = anom.weighted(w).mean([ydim, xdim])
print(glob.isel({tdim: slice(-12, None)}).values.round(2))

# Absolute temperature = anomaly + climatology for the same month (last month shown)
clim = ds["climatology"]
mdim = [d for d in clim.dims if d not in (ydim, xdim)][0]
absolute_last = anom.isel({tdim: -1}) + clim.isel({mdim: month[-1]})

Practical tips¶

• The old 1° products (Land_and_Ocean_complete.txt, Land_and_Ocean_LatLong1.nc, etc.) have not been updated since Q2 2025. If your scripts still point to them, switch to the high-resolution files. The two products use different methods, so do not splice them into one series.
• The high-resolution product is labelled "PRELIMINARY DATA - SUBJECT TO CHANGE WITHOUT NOTICE - NOT YET PEER REVIEWED". In papers, record the download date or use the frozen Zenodo release (DOI).
• The CC BY-NC 4.0 licence means commercial use, including paid services and products, needs permission from Berkeley Earth.
• The ocean component is a reinterpolation of HadSST4, the same ocean input as HadCRUT5, so agreement between the two is not an independent confirmation.
• Anomalies are relative to 1951–1980 (as in GISTEMP), whereas HadCRUT5 uses 1961–1990. Align the baselines before comparing.
• The high-resolution gridded files carry no uncertainty information. Use the 95% ranges in the text time series or the 50-member ensemble files instead.
• A 0.25° grid can be finer than the station network. In data-sparse regions and early decades, small-scale structure may come from the statistical predictors rather than observations, so interpret it with care.

Notices and change history¶

References for your paper¶

Dataset citation¶

Zenodo dataset DOI (high-resolution, April 2026 release) — DOI 10.5281/zenodo.20428381

@misc{berkeleyearth_hr_2026apr,
  author    = {Rohde, Robert and Hausfather, Zeke and Calvert, Bruce},
  title     = {Berkeley Earth High Resolution Dataset - April 2026},
  year      = {2026},
  publisher = {Zenodo},
  doi       = {10.5281/zenodo.20428381}
}

Reference papers¶

Berkeley Earth Land/Ocean Temperature Record paper (old 1° product) — DOI 10.5194/essd-12-3469-2020

@article{rohde2020berkeley,
  author  = {Rohde, Robert A. and Hausfather, Zeke},
  title   = {The {Berkeley Earth} Land/Ocean Temperature Record},
  journal = {Earth System Science Data},
  volume  = {12},
  number  = {4},
  pages   = {3469--3479},
  year    = {2020},
  doi     = {10.5194/essd-12-3469-2020}
}

High-resolution product (BEST-HR) description paper — ESSD preprint in review — DOI 10.5194/essd-2026-412

@article{rohde2026besthr,
  author  = {Rohde, Robert A. and Calvert, Bruce and Hausfather, Zeke},
  title   = {{Berkeley Earth} Surface Temperature -- High-Resolution ({BEST-HR}): a 0.25{\textdegree} Global Gridded Temperature Data Set for Climate Monitoring},
  journal = {Earth System Science Data Discussions},
  year    = {2026},
  note    = {preprint, in review},
  doi     = {10.5194/essd-2026-412}
}

Acknowledgements¶

"Attribution under CC BY-NC terms should be given to Berkeley Earth", with a reference to www.berkeleyearth.org where possible (Berkeley Earth data page). The old 1° product files cite Rohde & Hausfather (2020). While the high-resolution description paper is in review, cite both the preprint and the Zenodo release DOI.

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).

Links¶

• Berkeley Earth data page
• High-Resolution Gridded README
• ESSD preprint — BEST-HR (essd-2026-412)
• Zenodo — Berkeley Earth High Resolution Dataset (April 2026)
• Rohde & Hausfather (2020), ESSD
• Global Temperature Report for 2025