Relate forecast errors to accuracy of weather forecasts

[FrankKr]

DoR:

• Compare ICON weather forecasts to reference #24

What:/Why:
We expect our power forecasts for OS Middenmeer (mainly wind generation) to be good when the weather forecasts are accurate and to be worse when the weather forecasts are off. The goal of this issue is to validate this hypothesis.

• If the hypothesis turns out to be correct, then it strengthens our case to investigate the use of ensemble weather forecasts and to investigate the use of weather forecast from different sources.
• If the hypothesis turns out to be incorrect; in other words, our forecast are sometimes bad even though the weather forecasts were good. In that case, we have a reason to suspect that we could improve the way our models use the weather forecasts to make predictions.

Comments:

• Outcomes of this story can lead to recommendations for the full project.

Acceptance criteria:

• The fair dataset that results from Compare ICON weather forecasts to reference #24 is used for this story.
• Realized weather Reanalysis weather data has been gathered for the same weather station location that is used for the weather forecasts.
• The relationship between the accuracy of weather forecasts and the accuracy of OpenSTEF's power forecasts has been analysed.
  • Extra/nice to have idea: Check how much better OpenSTEF's forecasts get when actual weather measurements are used instead of weather forecasts.
  • Are there specific conditions where the relationship between weather forecast errors and power forecast errors are stronger or weaker?
• A HTML of the notebook including visualizations, metrics, and conclusions, has been saved in the AIFES SharePoint.
• The notebook containing the code used for the analysis, has been committed to GitHub. (Notebook results/data have been cleared beforehand.)
• Follow-up idea's are documented on the backlog and/or in a document on the SharePoint that can be used as input for the full project proposal.

[MartijnCa]

@MartijnCa, prepare this one for the next refinement.

[FrankKr]

Simon:

Ik heb jullie input data (KNMI met korte leadtime) vergeleken met zowel de D-2 voorspellingen van de DWD als met ERA5.
Conclusies zijn als volgt:

• Error D-2 forecasts is ongeveer 25% groter dan van de voorspellingen die jullie gebruiken.
• De rest van de error kan niet echt verklaart worden aan verschillen tussen voorspelling en reanalysis aangezien era5 niet beter is dan korte termijn KNMI voorspellingen.