@bgroenke
Scientist, (Bayesian) data nerd, software engineer, and researcher at the Potsdam Institute for Climate Impact Research @pik-potsdam.bsky.social working at the intersection of machine learning, statistics, and computational geoscience. Opinions my own.
Weeks after the US Congress rejected unprecedented cuts to science budgets that the administration of US President Donald Trump had sought for 2026, funding to several agencies that award research grants is still not freely flowing go.nature.com/3OvYRSZ
@michaelemann.bsky.social In case you'd like another example of climate doom for your next book or blog post, see above...
If you'd like a more nuanced and well studied analysis of where we currently stand, I can highly recommend this video from Simon Clark @simonoxfphys.com
youtu.be/fsah3Z2pC6Q?...
I was deeply disappointed to see @hossenfelder.bsky.social embrace full climate doomerism in this video. Despite the limited but nevertheless noteworthy progress made in recent years, Sabine cherry picks a few headlines and concludes that humanity has "given up" on climate change. This is ludicrous.
Future #EU contribution to #GlobalHeating could be halved without placing extra burden on the economy.
An economic PIK study in the top journal JEEM calculates the effectiveness of “clean-up certificates” (emission right + removal obligation).
#Climate #Edenhofer
www.pik-potsdam.de/en/news/late...
Why is losing ice from land a major problem for people living near the ocean? Using a miniature beach to demonstrate one of the many impacts of going past #planetaryboundaries, the Planetary Boundaries Science lab asks "Is sea level rise worse than you think?"👇
www.youtube.com/watch?v=G4ao...
"LLMs did not create what economists, with their penchant for meaning-destroying dimensionality reduction, would call science’s incentive 'structure.'"
Damn, shots fired!
I don't remember if I saw this posted by you or someone else, but there's also this study
metr.org/blog/2025-07...
which suggests that we maybe shouldn't be trusting self-reported productivity gains. Turns out people are actually pretty bad at perceiving time, shocker, I know.
The examples provided by the python `sbi` package may also be worth looking at?
@sbi-devs.bsky.social
sbi.readthedocs.io/en/latest/
The software engineering on this project is admittedly not up to my usual standards, but maybe our hopefully soon-to-be published work on stochastic weather generators would be useful here? It also makes use of probabilistic programming.
Might be too much though.
github.com/bgroenks96/w...
I do know what those words mean! Do you need just data or also a suitable simulator?
"There aren't supposed to be wildfires in winter"
Checkmate, climate deniers! 😂
But this seems more like a problem of predictability limits rather than outliers?
And surely there are other causes of apparent outliers in social science data than just selecting on the dependent variable? Measurement error? Unobserved group structure in the population?
It's wrong to assume that you can't predict outliers. This is classic mind projection fallacy. Probability distributions are models, not exact representations of reality. If we have access to the right information and causal mechanisms, we absolutely can predict or retroactively diagnose outliers.
I think they just mean citing others' work as a way of "signalling" that you are "well read" (even if not) and to avoid being accused of failing to cite relevant work, while not bothering to actively and critically engage with said work.
@lilybellesweet.bsky.social
In geospatial data analysis, a common mistake is performing significance tests on each spatial point and then displaying this on a map with stippling or similar. This very often results in a massively inflated FPR due to spatial correlation.
journals.ametsoc.org/view/journal...
Based on what I have been hearing about the NCAR situation, no one knew anything until the USA Today article came out. That means there isn’t much actual organization inside the administration to pull this off and a chance to fight back. Call your congresspeople today! Especially republicans.
Attacking NCAR is enormously self-destructive for America. NCAR is the leading centre in weather and climate research in the world. We Europeans often travel to NCAR to collaborate, discuss and learn. Hurting the centre will cause damage that is irreversible for years to decades with global impact.
This news is nothing short of a tragedy and an outrage. I urge my American friends and colleagues to call your congressional representatives and resist this blatant attack on science and academic freedom!
www.rmpbs.org/blogs/scienc...
PIK at #IPCC: Nine PIK scientists joined the first Lead Authors’ Meeting for the 7th Assessment Report on the state of climate science in Paris last week. The meeting marked the start of drafting the 3 Working Group reports: www.pik-potsdam.de/en/news/late... @ipcc.bsky.social @wflamb.bsky.social
The prior is only half the story though, and usually the least interesting part. I'm usually working with physical models where the likelihood also incorporates a lot of prior knowledge.
Anyway, now that we're done trolling, I am actually a fan of your work on equivalence testing :)
Ah, no, you misunderstand. In the natural sciences, we're simply not interested in throwing away valuable background information in exchange for fantasies about error control.
Quite the contrary. In my domain, frequentist theory is pretty much useless, so I don't bother thinking about it much, until I see your posts in my feed!
That's just because the frequentists are too busy whining about priors to identify as anything ;)
The irony here is that it is, and always has been, the frequentist view that is dogmatic in its gatekeeping of what can and cannot be assigned a probability. A good Bayesian can and will adopt frequentist probabilities when the problem calls for it, as the framework quite naturally allows for it.
Back by popular demand: At #EGU26 we'll organize another BUGS session: Blunders, Unexpected Glitches, and Surprises!
Submit abstracts on ideas that seemed great but didn't work, errors and bugs that led to new insights (or funny stories), or any other unexpected results.
www.egu26.eu/session/56997
Amen!
CO making me proud ♥️⛰️
