Jonathan Wider

someone should create a swimming ring from this!

Figure 1 shows this nicely visually. I also only learned about this recently

Quantifying changes in climate variability and extremes: Pitfalls and their overcoming Conventional normalization of spatiotemporal data sets with respect to a reference period induces artifacts Normalization-induced artifacts are most severe if variability or extremes are under scru...

Not to say it would alter any conclusions, but it's good to keep in mind that this methodology of computing σ on a calibration set and then defining extremes on a test set based on whether they exceed n * σ can lead to overestimating the extremeness of the event. doi.org/10.1002/2015...

On my way to @ellisunitjena.bsky.social for the "AI for Earth and Climate Sciences" summer school 🚲🏞️🙌

Greetings from @jonathanwider.bsky.social at the Machine Learning Summer School, which is taking place at the @Universidad Católica San Pablo in Peru at an altitude of 2,300 meters. The summer school runs from 2 to 13 August and includes lectures by numerous renowned ML/AI researchers.

I checked a few ones and could see the prompt in the arxiv HTML version

Heute gelernt: Die physikalischen Einheiten-Vorsätze kommen zum Teil aus dem Dänischen. Femten als dänisches Wort für 15 wurde zur Vorsilbe "Femto", wobei z.B. ein Femtometer 10 hoch minus 15 Meter sind. Analog bei atten, dänisches Wort für 18, ein Attometer sind 10 hoch minus 18 Meter.

Eine Straße mit Häusern aus der Gründerzeit. Erker und Türmchen.

Gestern war dieses Foto von André Kempner in der LVZ.

In dem ersten Haus mit Erker rechts wohnten meine Großeltern und eine zeitlang auch ich.

Wenn ich als Kind zu Besuch kam, stand meine Großmutter am Erkerfenster und winkte. Das war schön.

Alle Häuser waren grau damals.

Microsoft also announced its first AI weather forecasting model (ClimaX) in early 2023 - microsoft.github.io/ClimaX/

(This is not to say that Aurora is any less interesting as a result)

Are "ConvNets" missing an N?

Interesting read despite the annoyingly clickbaity title

My guess is that the person joined BlueSky after clicking the link to your starter pack. Someone on Reddit thought the same.

jetzt sind alle Kreise ausgezählt - wenn es dabei bleibt waren es etwa 10,000 Stimmen zu wenig für die 5% Hürde.

und es fehlt noch 3x BaWü, 3x NRW, 1x MeckPomm & 1x Schleswig Holstein...

jetzt auf 284 - und sie stehen bei 5,001% wenn ich mich nicht verrechnet habe beim (BSW / Gültige)-Stimmen rechnen.

Studie zu den Pariser Klimazielen: Hat die 1,5-Grad-Ära begonnen, Herr Bevacqua? Das 1,5-Grad-Ziel ist gefährdet, wie aktuelle Berechnungen zeigen. Trotzdem rät Klimaforscher Emanuele Bevacqua, daran festzuhalten: Noch sei es möglich, die Erderwärmung zu verlangsamen.

Das 1,5-Grad-Ziel ist gefährdet, wie aktuelle Berechnungen zeigen. Trotzdem rät Klimaforscher Emanuele Bevacqua, daran festzuhalten: Noch sei es möglich, die Erderwärmung zu verlangsamen.

2022-2025 X Twitter mention data

Here's Twitter's descent since 2022

From my first impression, the largest change might be that you don't need to run the model yourself anymore, which is nice for people who don't have fancy GPUs available. I guess it also streamlines things for people who actually use the forecasts in decision making.

I'm not sure if this is a big change from a scientist's perspective. I thought the code and trained models for GenCast and GraphCast were already on GitHub and some of the forecasts were available on WeatherBench 2? (which is both great of course!) Maybe someone else can clarify :)

Tausende bei Kundgebung auf dem Leipziger Markt vor der CDU Zentrale. Menschen bauen #Brandmauer aus Kartons auf, rufen „alle zusammen gegen den Faschismus“. Reden von Einzelpersonen und Vertreter:innen von Flüchtlingsrat, Gewerkschaften, Kirche. #le3001

The value of failure: Learning from what didn’t work “What failure prepared you for later success?” Tim Ferriss often asks this question to the high-performing guests on his podcast, giving them the opportunity to reflect on how setbacks have shaped the...

My esteemed former colleague @danklotz.bsky.social co-chairs a similar session at EGU this year: blogs.egu.eu/divisions/hs...

Isn't the first panel of this figure what you are looking for? bsky.app/profile/cope...

It's the same for me with "downscaling" and "upscaling" - it's used in the opposite way as "downsampling" and "upsampling" in signal processing & machine learning.

... and I saw it in a presentation by @mdebrito.bsky.social

Do you mean the 38th Chaos Communication Congress (CCC =C^3 -> C3) or did I miss a joke?

Or also 2025 = 1^3 + 2^3 + 3^3 + 4^3 + ... + 9^3

Example of realistic rainfall simulations generated by a machine learning model over England and Wales (Addison et al., 2024)

PhD opportunity at Bristol: "Machine Learning for Understanding Extreme Weather in a Changing Climate". Please share with potential applicants. Deadline is Jan 31 12:00.

www.bristol.ac.uk/geography/co...

www.bristol.ac.uk/media-librar...

Which makes training hard, I guess?

Assimilation windows also differ between ERA5 and HRES forecasts (at some time steps), which complicates the evaluation - this is e.g. described in chapter 5.2.2 of the GraphCast supplementary materials

I think the 12h timestep is mostly a consequence of the data ECMWF releases: the assimilation used for ERA5 has differences between the 00z/12z and 06z/18z initialization times (unequal "lookahead", i.e., up to how far in the future are observations used?).

That's an interesting setup - I wasn't aware of it. They seem to use FourCastNet, which from my experience isn't the strongest ML weather prediction model, I would guess it's possible that repeating the analysis with GraphCast or Pangu would yield better results.