matching the forecasts with the ground truth

[randomgambit]

Hello there and thanks for this amazing package! It is extremely useful.

One question though: I am using it to retrieve the daily forecasts from the NOAA. However, I would like to match them with the ground truth (the actual conditions that the forecasts try to forecast). Can I use your API to get this data?

Thanks!

[jamaa]

Which NOAA forecasts are you you retrieving exactly? In the docs I only see GHCN data available from NOAA: https://wetterdienst.readthedocs.io/en/latest/data/coverage/noaa.html

[randomgambit]

yes, I saw this source. I guess my question was more conceptual: how would you match these forecasts with their ground truth? would you know where the data can be found? Thanks!

[jamaa]

GHCN is not a forecast but contains observed data, i.e. the "ground truth" you are looking for. I believe the wetterdienst package generally only provides observed data, so it is the forecasts that you have to find elsewehere.

[gutzbenj]

Dear @randomgambit ,

you could potentially compare NOAA GHCN (global hourly historical data) and DWD Mosmix (global hourly forecast data). I will provide you with an example in the upcoming days.

[randomgambit]

@gutzbenj that would be great! Thanks!!

[randomgambit]

I i magine the difficulty is not matching the day of the forecast, but it is matching the different geolocations (and ensuring forecasts h-days ahead and actual observations are aligned). Any help greatly appreciated!

[gutzbenj]

I believe from what I saw both NOAA GHCN and DWD Mosmix take the existing station identifier and only extend it by some prefix, so it should be be possible to get the union of existing ids, at least for Germany because in this case both data would come from the same maintainer (DWD), who also had an established identifying system for decades.

[gutzbenj]

Dear @randomgambit ,

I added an example at https://github.com/earthobservations/wetterdienst/blob/main/examples/forecast_validation.py that shows how to compare DWD Observation and Mosmix data. The problem here is that old Mosmix forecasts are not being kept that long on the servers so that we can basically only compare about 1 day of data at most.

[randomgambit]

hello @gutzbenj thanks but I cannot find the notebook. did you change its name?

[gutzbenj]

I'm so sorry, must have mixed up something. Now it is there!