I am searching for a resource showing historical weather forecasts compared to actual weather data.

I'm mainly interested in seeing how accurate various forecast services have been in the past, when you compare their predictions to actual weather records.

I'll be interested in anything of the sort, however, some things that would be nice are:

  • Accuracy percentage for a certain forecast service

  • Accuracy percentage for a given time period

  • Accuracy percentage for a given geographical location

  • A combination of the three previous points

  • Detailed per-day comparisons

I realize these are quite demanding criteria, and it might not exist. However, I'm interested in anything that even remotely fits.

  • Are you willing to do the comparison yourself if you can get forecast data and weather data separately? I assume google searching for "how accurate are weather forecasts" (no quotes) was not helpful?
    – user3856
    Commented Sep 16, 2016 at 16:59
  • @BarryCarter Separate data sets technically fits the question, but I don't have the software or knowledge to process such a huge amount of data to get overall statistics. The google search you mention gives an interesting mix of things. None of what I've seen matches all the criteria from OP, but to be honest I haven't trawled through all the pages of results. I'd err on the side of "interesting". But obviously I'd prefer as many criteria from OP satisfied as possible. But as I said, it may not exist and my demands aren't necessarily that high.
    – Fiksdal
    Commented Sep 16, 2016 at 17:08
  • Aren't you the OP? This would make an interesting project, which suggests someone is probably working on it. The most machine readable forecasts I know of are TAFs (the forecast equivalent of METAR), but they don't include temperature forecasts. Historical copies of nws.noaa.gov/mdl/forecast/text/avnmav.txt ("guidance") may also exist.
    – user3856
    Commented Sep 16, 2016 at 17:11
  • 1
    I'm tempted to jump into it myself. Feel free to contact me (contact info in profile). I can't promise anything, but, if I do help, it'll be for free.
    – user3856
    Commented Sep 16, 2016 at 17:19
  • 2
    You may find this similar data request to be helpful.
    – ess
    Commented Nov 4, 2016 at 9:10

4 Answers 4


I found this on cousin site Stats.Stackexchange.com, from 2011:

Determining the reliability of weather forecast

One of the resources may be useful:

ESRL/PSD GEFS Reforecast Version 2

The ESRL/PSD 2nd-generation Reforecast Project has produced a dataset of historical weather forecasts generated with a fixed numerical model, using the 2012 version of NCEP's Global Ensemble Forecasting System (GEFS, Version 10).

This Reforecast V2 dataset consists of an 11-member ensemble of forecasts, produced every day from 00 UTC initial conditions from Dec 1984 to present. The horizontal resolution of GEFS is T254 (about 50 km) out to 8 days, and T190 (about 70 km) from 8-16 days. Real-time forecasts are ongoing. Among the advantages associated with this long reforecast dataset is that model forecast errors can be diagnosed from the past forecasts and corrected, thereby dramatically increasing the forecast skill. Past results have shown that the improvements may be particularly large for medium-range forecasts and for forecasts of relatively uncommon events such as heavy precipitation.

(and then an email address to contact with questions)

You'd still have to join this historical forecast data with historical measurement data, but it seems a good resource for historical forecasts - that doesn't involve having to collect daily on your own.


I came across this while doing some of my own "back of the envelope" research.

When I look at any given weather app, on the hourly or daily forecasts, I often wish I could go back in time to see what the prediction was, and how it then ended. In my current mindset, I'm thinking mainly of precipitation (total rainfall amounts).

Per what user3856 said about 8 years ago, I would benefit from having that forecast data, combined with the actual data. The forecast data would need to be multiple layers, as it would be a forecast at different intervals, eg "5days out" to "1day out", and then compared to the actual 0day values.

This is why I think daily rainfall/precipitation totals are a good metric to start with (and useful for my purposes).

So, is there something like an internet archive of weather forecast data? Where I could see what xyz weather service was predicting for a certain time/date, at a specific time/date?

I could keep going, but I'd rather open up the conversation again and see who else has thoughts on the topic, or perhaps any sources for these datasets.

Thanks y'all! (And yes I'm posting from a brand new account, it's been a while). -Rob


There may be links to data in the references of this paper: The quiet revolution of numerical weather prediction

  • appreciate the effort, but could you check and see if there are and report back please? thanks!
    – albert
    Commented Feb 21, 2020 at 1:32

If you don't mind waiting for crawling jobs, here are the API endpoints for Swiss weather forecasts and real-time measurements:

Real-time Measurement: https://gist.github.com/philshem/0812d610f11d26a042bbd3e617fa8d22

Forecast: https://www.meteoswiss.admin.ch/product/output/forecast-chart/version__20200220_0731/en/800100.json

(800100 indicates Zürich, you'd have to go through the browser's Dev Tools / Network Inspector to find the other endpoints. Also, because the forecast timestamp is in the URL, you may need to go to the source page and get the URL from there.)

enter image description here

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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