Weather prediction is well known to be a difficult task. Regardless, as a consumer of the weather prediction websites, it is very useful to get some measure of which websites are better at consistently getting their prediction right. There are a lot of anecdotal views in circulation, claiming the superiority of one website over another or of local website over foreign websites. The basic concept is simple: Sample the weather prediction for a future date and compare it against the measured value when that date arrives.
Some of us have cooked up some python scripts in the ForecastCompare project on GitHub to evaluate the weather prediction accuracy within a very limited set of parameters. Sampling and querying data for all cities is beyond our resource capabilities, so we've set it up to gather the data for Trondheim, Norway. Although this is a somewhat arbitrary choice, the hope is that others will clone the repository and create their own mini-databases for their own cities.
The scripts currently sample the weather predictions from the following:
These predictions are compared against a local weather observation tower. The data and results can be seen at omer.pythonanywhere.com
- Data from Yr.no picks 5 day forecast and parses it for extracting min and max temps. The granularity for extracting min/max temps in this way is too big. Should parse the 24hour forecast (http://www.yr.no/place/Norway/S%C3%B8r-Tr%C3%B8ndelag/Trondheim/Trondheim/forecast_hour_by_hour.xml) in addition. Hour by hour only gives forecast for next 24 hours, but that is more detailed. - Add more data comparison tables to the html page. This should include Cloud cover, precipitation etc. - Move raw data display into its own html page - Why does Voll tower give so many -99999 values - Get key free access to OWM. Currently it must use an appid for a free user, but OWM website states that open-source applications can get key-free access - automate table generation in the html. Currently column names are hardwired... instead get names from tables - curl wttr.in/Trondheim