I am looking for a publicly available, easy to download, small data set of ensemble forecasts (or reforecasts), with corresponding observations or reanalysis. We are planning to illustrate some methods for a small software package, so ideally, the data set should be available for the foreseeable future, or even better, freely distributable. Further, ideally, the data set would be given as SV or text file (anything that can be read easily without requiring additional software packages, as it would be the case with GRIB for example). Any suggestions?

By easy to download I mean: Ideally, without creating any login, for example directly possible within Python or R; small in the sense of a couple of months and data for only one location are enough; there should be some verifying observation or re-analysis associated with the forecasts

  • 1
    $\begingroup$ Welcome to Earth Science SE! Easy to download means 'without creating a login' or 'download by a web interface'? Please define 'small'. Should it be a global or a regional data set or should it be data at one specific location (where also measurements are available)? $\endgroup$ Jul 6, 2016 at 7:59
  • $\begingroup$ @daniel.neumann I have added some details to the question $\endgroup$ Jul 6, 2016 at 8:04
  • $\begingroup$ I've never seen "small data set" and "ensemble forecast" used in the same sentence before. I would suggest atmos.washington.edu/~ens/uwme.cgi but their verification page is still under construction. $\endgroup$
    – f.thorpe
    Jul 6, 2016 at 21:20
  • $\begingroup$ @farrenthorpe I have also never used these terms in one sentence before, the reason here is that we are only looking for a small illustrative example for some software package. Thanks for the link to the UWME data, however, I do not see any option to download any data there $\endgroup$ Jul 7, 2016 at 7:30
  • 3
    $\begingroup$ NCEP's NOMADS server hosts the GFS ensemble forecasts output, you can get them here: nomads.ncdc.noaa.gov/data.php?name=access#EnsembProb. The data is in grib format, that is something you will have to deal with. Model output is almost never distributed as plain text for practical reasons. Observations from NOAA land stations and NDBC buoys are also publicly available. Google is your friend. $\endgroup$ Jul 7, 2016 at 16:07

3 Answers 3


For the historic forecasts check out the Global Ensemble Forecast System (GEFS) from NOAA. It has historic forecasts going back to 2015. For validation try Daymet (North American only) or PRISM (continental USA only) for gridded temperature and precip from historic observations.

The GEFS data is in GRIB format. But thats just something you can't get around. The European climate center (ECMWF) have their own suite of commands in a program called grib_api, which includes a handy GRIB to netCDF converter if that's a format your more familiar with.

All of these are free to download without signing up for anything. The GEFS data is also available through the R package rnoaa, and the PRISM has it's own R package prism.


You can try if this looks simple enough. Still requires registration, but it is free for reasonable use: Global Ensemble Forecast System (GEFS).


Synthesise your own dataset.

Get some real data, and take some averages and add some noise (or some other simple munging), such that it's not possible to get back to the original raw data.

Then you've got a realistic data set that you have all the rights on, that's exactly the size that you want, in exactly the format you want, and you can distribute it however you see fit.

  • $\begingroup$ That is our option to fall back on, though we would prefer a real dataset $\endgroup$ Jul 8, 2016 at 9:24
  • $\begingroup$ @user3825755 please can you edit your question to explain why? Doing so will help you get better answers $\endgroup$
    – 410 gone
    Jul 8, 2016 at 10:17

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.