COBE-SST3¶

Centennial in situ Observation-Based Estimates, version 3 (COBE-SST3)

Available

At a glance¶

Note on period: According to the MRI server listing on 2026-10-12, yearly daily (daily/) and monthly (monthly/, 1x1/monthly/) files exist for 1850–2024 (the 2021–2024 files were uploaded on 2025-09-22). The paper (Ishii et al. 2025) and the NCAR Climate Data Guide still describe the period as 1850–2020, and we have not opened the 2024 file to confirm its last date. The perturbation ensemble has files only for 1890–2020 (0.25° daily) and 1870–2024 (1° monthly); COBE-SST3H covers 1982–2020.

Overview¶

A daily 0.25° sea surface temperature analysis developed by the JMA Meteorological Research Institute (MRI) as the successor to COBE-SST2, reconstructed from 1850 using in situ observations such as ICOADS Release 3.0. Biases are re-estimated for each observation type, and SST is analysed monthly together with land surface air temperature (COBE-LSAT3) for consistency between ocean and land; uncertainty is provided by a 300-member perturbation ensemble. It is a research dataset released with the 2025 paper; JMA's operational SST analysis remains COBE-SST2.

Available variables¶

According to the NCAR Climate Data Guide, the variables provided are sea surface temperature (SST) and sea ice concentration (SIC). Variable names and units inside the files are not given in official documents, so open a file (print(ds)) to check them. The file names below were confirmed from the MRI server listing.

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

Sea surface temperature¶

Sea ice¶

Uncertainty (perturbation ensemble)¶

Ancillary files¶

The full variable list is in the official documentation.

How to get the data¶

• MRI research data server — COBE-SST3 archive — Direct HTTPS directory download (wget / curl), NetCDF No login needed
  • Under cobe-sst3/ are daily/ (0.25° daily, one file per year), monthly/ (0.25° monthly), 1x1/ (1° monthly and perturbations) and perturb/ (300-member daily perturbations). The same parent folder holds COBE-SST3H (satellite-blended daily, 1982–2020), COBE-LSAT3 (land air temperature), COBE-STEMP3 (global surface temperature), mask/ and gm/. GrADS ctl files and a 00README are included. There is no OPeNDAP or THREDDS, and each yearly daily file is large, so download only the years you need.

Example code¶

There is no OPeNDAP, so download the file first and then open it. Variable and coordinate names are not documented officially — check them with print(ds) before use.

import urllib.request
import xarray as xr

base = "https://climate.mri-jma.go.jp/pub/archives/Ishii-et-al_COBE-SST3/cobe-sst3"
fn = "cobe-sst3.monthly.2020.nc"                 # 0.25° monthly file (daily: daily/cobe-sst3.2020.nc)
urllib.request.urlretrieve(f"{base}/monthly/{fn}", fn)

ds = xr.open_dataset(fn)
print(ds)                                         # check variable names, units, latitude order and time values
sst = ds[list(ds.data_vars)[0]]                   # replace with the SST variable name shown by print(ds)

Practical tips¶

• JMA's operational product is COBE-SST2. If you need recent months updated monthly, use COBE-SST2 (1°, monthly) or an operational daily product such as OISST, and keep COBE-SST3 for high-resolution and uncertainty analysis of past periods.
• Changes from COBE-SST2: resolution 1° → 0.25°; in situ input from ICOADS Release 3.0 (JMA operational data after 2015); biases re-estimated per observation type, with night marine air temperature (HadNMAT2) used in the bias correction; joint monthly analysis with land air temperature (COBE-LSAT3). The abrupt shift in the early 1940s is reduced, so mid-20th-century trends can differ from COBE-SST2.
• The climatology used in the analysis is 1985–2015 (31 years). We have not confirmed whether the distributed files contain an anomaly variable, so compute anomalies yourself against the base period you need (e.g. 1991–2020).
• If uncertainty matters, use the 300 members in perturb/. Note that the server has 0.25° daily perturbations only for 1890–2020 and 1° monthly perturbations for 1870–2024, which differ from the main analysis period (1850–2024).
• COBE-SST3H (1982–2020), which blends satellite SST directly, captures daily variability better. The paper finds COBE-SST3H statistics against buoys and Argo very close to OISST v2.1, while COBE-SST3, without direct satellite input, shows somewhat damped signals.
• The developer notes that reliability is low in periods and regions with sparse observations and that early 20th-century bias corrections are probably still insufficient. Pre-satellite sea ice concentration has low accuracy, and Antarctic sea ice is a climatology.
• The 2021–2024 extension appeared on the server in September 2025 but is not yet described in the paper or the NCAR CDG. If you use it, check the file contents and the server 00README, and record your download date.

Notices and change history¶

References for your paper¶

Reference papers¶

COBE-SST3 paper (primary citation) — DOI 10.2151/jmsj.2025-002

@article{ishii2025cobesst3,
  author  = {Ishii, Masayoshi and Nishimura, Akio and Yasui, Soichiro and Hirahara, Shoji},
  title   = {Historical High-Resolution Daily {SST} Analysis ({COBE-SST3}) with Consistency to Monthly Land Surface Air Temperature},
  journal = {Journal of the Meteorological Society of Japan. Ser. II},
  volume  = {103},
  number  = {1},
  pages   = {17--44},
  year    = {2025},
  doi     = {10.2151/jmsj.2025-002}
}

Acknowledgements¶

There is no separate dataset DOI, so cite Ishii et al. (2025) and state where you obtained the data (the MRI research data server URL) and when. The post-2020 extension is not described in the paper, so note separately that you used it and the date you downloaded it.

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¶

• Ishii et al. (2025) paper (J-STAGE, open access)
• JMSJ paper highlight
• NCAR Climate Data Guide — COBEv3 (with expert developer guidance)
• MRI research data server — COBE-SST3 archive
• JMA long-term reanalysis advisory meeting material — ocean boundary condition improvements (2026-02-13, Japanese)
• JMA website terms of use (also applied by MRI)