Hi all,
To improve the evaluation of forecasts, I’ve made a small update to the way Ranked Probability Scores (RPS) are calculated for precipitation.
Specifically, grid points where all quintile climatology values are zero (i.e., completely dry land regions) are now masked out. This change avoids ambiguity when all quintile boundaries are equal and allows for a fairer comparison between models.
I also want to acknowledge that updates to the AI Weather Quest Python package are currently frequent. To improve transparency and keep everyone informed, I’ve created a dedicated update webpage that will be maintained with important changes to the package.
To upgrade to the latest version, simply run:
python3 -m pip install --upgrade AI-WQ-package
I will now begin re-evaluating the dynamical models and will update the forum post accordingly.
Thanks everyone,
Josh