Other minor updates: - Where available, we added 2022 as an eval year in the interactive graphics. - We added forecast activity as a metric for deterministic models, a simple measure of blurring. - More regions. Don't hesitate to file bugs or suggestions as GitHub issues. end/

RT @DopplerChase: 2025 is here tomorrow, so let's reflect on 2024. Even without the final counts and the new AMS and AGU ML journals, 2024…

@crisbodnar Actually for Z500, resolution doesn't make a big difference to the absolute values, see: But yeah, of course resolution matters. It's simply a best attempt at getting the different models in a single graph.

@crisbodnar Well, there is a disclaimer on the spreadsheet but I get the point. But what super low res model are you talking about?

@crisbodnar So definitely not perfect and the point is not really to compare models exactly but rather to see a trend. For that, I hope it's sufficient. WeatherBench is an attempt to do everything exactly but it doesn't have nearly as many models.

What is the conclusion from GraphDOP being so far away from SotA? Is the setup still suboptimal in some way or is pure obs-based forecasting harder than some might have thought.

@danrothenberg @RyanKeisler Yup, fully agree. I didn't mean to criticize the authors. Just wondering what the implications of this are.

Well, here we go --> 🦋

Interestingly, the first observation-based models, Aardvark and FuXi Weather, (even though both still use ERA5 as a training crutch) are roughly where ERA5 initialized models were 4 years ago. Similar progress incoming?

@turbulentjet Nice set of experiments but this is using FourCastNet v1 which is not a SotA AI-weather model. Given it's relative lack of skill I don't think you can generalize the results to all AI-weather models.

RT @NMcGreivy: Our new paper in @NatMachIntell tells a story about how, and why, ML methods for solving PDEs do not work as well as adverti…

@PierreGentine Open-access link? 😉