📣 New preprint "Probabilistic measures afford fair comparisons of AIWP and NWP model output" with Tilmann Gneiting, Tobias Biegert, Kristof Kraus, Eva Walz and Alexander Jordan available at arxiv.org/abs/2506.03744. Some details below 🧵

Probabilistic measures afford fair comparisons of AIWP and NWP model output

We introduce a new measure for fair and meaningful comparisons of single-valued output from artificial intelligence based weather prediction (AIWP) and numerical weather prediction (NWP) models, calle...

arxiv.org

Useful law for estimating the overall speedup from ML/GPU optimizing subcomponents of scientific code en.wikipedia.org/wiki/Amdahl%...

Amdahl's law - Wikipedia

en.wikipedia.org

AI NWP models will solve this, right? RIGHT?

Dennis Mersereau 🛰️@wxdam.com · last yr.

NWS just announced a major reduction in upper-air (weather balloon) observations due to staffing shortages. Omaha and Rapid City losing balloon launches altogether, with launches cut in half at 6 other sites. Losing this many launches could affect severe weather forecasting and computer models.

Ever been told to ""sleep on it""? 😴 @alanajaskir.bsky.social’s research shows how snoozing helps your brain learn! #sleepbenefits #replay #hippocampus #academicsky #standupforscience

After losing 2 flight directors (my old job), the NOAA Hurricane Hunters are down to minimum staffing for continuous hurricane flights. And if I still had my old job, I'd be looking for new employment before the next onslaught of cuts. yaleclimateconnections.org/2025/03/noaa...

Potentially interesting job for the NOAA crowd - Verisk, one of the biggest modeling firms in the insurance industry, is looking for an experienced atmospheric scientist. Charles Jackson is hiring, who was on my PhD committee and is just a delight to work with, aside from being a kickass scientist.

Pleased to share my views in a new TEDx talk on how AI offers compelling new ways to simulate the Earth with unprecedented resolution and interactivity: youtu.be/I0nxFLh-iBg?...

AI is making whole earth simulation possible--why that matters | Mike Pritchard | TEDxBoston

YouTube video by TEDx Talks

youtu.be

Excited to share some great collaboration on a new data assimilation framework with transformer encoding arxiv.org/pdf/2502.02884. Exciting times for data assimilation with ML

☀️ APPS ARE OPEN for our Summer'25 MOMENTUM FELLOWSHIP ☀️ - For data sci PhD students @ a U.S. institution - Mentorship of REU students - June 9 - August 1 - 5 research projects based in NYC and/or Boulder, CO 📆 INFO + APPLY by 3/15/25: leap.columbia.edu/education-2/...

Kyleen Liao did stellar work with @karadlamb.bsky.social, Pierre Gentine, and me a while ago on simulating PM2.5 emissions from California wildfires using graph neural networks (GNNs). Her paper with cool new model validation results is out now: www.cambridge.org/core/journal...

Words of wisdom from @naomioreskes.bsky.social's Science on a Mission.

1/n Please see our new paper in Nature Communications! We used 1,851 tree-ring fire-scar sites and contemporary fire perimeters to quantify the prevalence of wildfire from 1600-1880 compared to 1984-2022. 🧪🌍🔥 Our key findings are as follows ...

A fire deficit persists across diverse North American forests despite recent increases in area burned - Nature Communications

Across many North American forests, recent years with exceptional area burned are not unprecedented when considering the multi-century perspective offered by fire-scarred trees. Nevertheless, abundant...

nature.com

Also consistent with my experience that PNAS is actually a terrible journal. I know of several PNAS papers that were either straight wrong or you could not reproduce their results.

Scott Kominers@skominers.bsky.social · 2y ago

Utterly astounding to me that it's possible to publish a paper in PNAS that shows a correlation between having a "hit paper" as a postdoc and later academic success... www.pnas.org/doi/10.1073/...

I should have led off this thread with the statement: this is how to think about wildlife risk and what we can do about it. No safety intervention makes a community perfectly safe. Many reduce risk. It is a layered and cumulative effort. Think about auto safety to understand what I mean:

Michael Wara@michaelwara.bsky.social · 2y ago

People die in car crashes. A lot of people. In 2022 in the US, about 46,000. But if we had not made the safety improvements in cars that we have over the past half century - and had the same death rate as a result, 153,000 people would have died.

Is there a link between #ClimateChange & increasing risk/severity of #wildfire in California--including the still-unfolding disaster? Yes. Is climate change the only factor at play? No, of course not. So what's really going on? [Thread] #CAfire #CAwx #LAfires iopscience.iop.org/a...

Notice power generation - high wind events are bad news for transmission lines. Suspect these statistics are worse for structures/fatalities - high wind fires make up 12% of burn area in western US, but 52% of lives/60% of structures lost, via @climate-guy.bsky.social @pyrogeog.bsky.social research

Downslope Wind‐Driven Fires in the Western United States

Downslope wind-driven fires accounted for 13.4% of fire occurrence and 11.9% of total burned area in the western United States during 1992–2020 Most structure losses and fatalities in fires 1999–...

agupubs.onlinelibrary.wiley.com

Andrea Thompson@andreatweather.bsky.social · 2y ago

We don't know the causes of the Palisades, Eaton or Hursts fires yet, but humans cause almost all of the ones in California: www.scientificamerican.com/article/the-... This graphic shows the breakdown from agency statistics 🧪

Skew-T Log-P xkcd.com/3032

“With the rising costs of and vulnerability to extreme weather in a changing climate for the United States, dismantling or defunding NOAA would be a catastrophic error” Good op-ed from Ryan Maue in todays @nytimes: www.nytimes.com/2024/12/01/o...

Opinion | Republicans Would Regret Letting Elon Musk Ax Weather Forecasting

Everyone benefits from government weather forecasting. Republicans have the chance to make it better.

nytimes.com

Arvind, me, and Jonah released a new pre-print on some pen and paper analysis of fundamental failure modes and old school stability analysis for neural PDEs typically used in AI for Science application. arxiv.org/abs/2411.15101. 1/n

What You See is Not What You Get: Neural Partial Differential Equations and The Illusion of Learning

Differentiable Programming for scientific machine learning (SciML) has recently seen considerable interest and success, as it directly embeds neural networks inside PDEs, often called as NeuralPDEs, d...

arxiv.org

Is there a similar conversation we can have among climate scientists as the new administration freezes or cuts federal funding levels for climate related research? Should we be more private sector focused in our collaborations while also doing more public outreach in non-university spaces?

Alexandria Ocasio-Cortez@aoc.bsky.social · 2y ago

There have been a lot of times where I feel like we need to have honest discussions on the left about what has worked and what hasn’t, share observations without judgement, but it feels like there isn’t much space for it. I think it’s holding us back. I wonder how and where we can accomplish that.

We have an opinion piece at the NYT on the need to operationalize some aspects of climate science. 🧪 Thoughts welcome! (Gift link) www.nytimes.com/2024/11/13/o...

Opinion | Climate Science Can’t Keep Up With the Warming Planet

We need more timely updates in response to the rapid changes to the climate.

nytimes.com

+1 Williams (2017): "While extreme drought has been rare in the SE U.S. over the past half century, the intensity and rapid onset of the SE U.S. drought in 2016, and its destructive impacts on wildfire...should motivate preparedness for reoccurrences of droughts of similar or stronger magnitude."

The 2016 Southeastern U.S. Drought: An Extreme Departure From Centennial Wetting and Cooling

New gridded monthly soil-moisture estimates indicate that the southeast U.S. drought in fall 2016 was the second most severe since at least 1895 The driver was low precipitation, but record-high ev...

agupubs.onlinelibrary.wiley.com

Kelly Hereid@kellyhereid.bsky.social · 2y ago

Paleoclimate agrees with this too - despite the decades-long wet conditions in the Northeast, drought is common in the paleo record, as in this example from Pearl @thirstygecko.bsky.social et al

Thrilled to make LEAP's 1st jump into #Bluesky by hi-lighting our next-gen #climate #data scientists, who spearheaded today's ClimateML Methods Workshop + brought together early-career scientists in the #NYC area to share progress @ the intersection of #climate science + #ML! #community #diversity

#HarveyLab @UW_SEFS is recruiting 1-2 funded PhD students in 2025. Research is part of @NSF project testing mechanisms of forest resilience to fire + changing fire regimes in PNW www.nsf.gov/awardsearch/... + opportunities to collab w/ WFFRC westernfireforest.org Apply to SEFS by Nov 15

With the (belated) September data now in, the updated prediction is that 2024 is almost certain to be a new annual surface temperature record, and possibly by more than 0.1ºC. 50% change of exceeding 1.5ºC above the late 19th C.

For several years, I have been mostly focused on ML models and their climate applications. So it felt good to have contributed something more physically motivated; here's a new algorithm we developed to simulate aerosol perturbations (or seeding) in PySDM: github.com/open-atmos/P...

Release PySDM v2.75 · open-atmos/PySDM

What's Changed Seeding dynamic and associated initialisation logic (super-droplet injection during simulation) in #1367 Kudos @jtbuch & @claresinger! Full Changelog: v2.74...v2.75

github.com

Bidenomics architect Deese set out a cheaper, green-ish “Marshall Plan” this week. @70sbachchan.bsky.social and I looked at key developments that led to this (a lot has happened!) and what’s wrong / missing (some big things). buttondown.com/polycrisisdi...

As the fire season intensifies over parts of the western United States, I wanted to share some results from ongoing work to improve seasonal and subseasonal-to-seasonal scale forecasts of wildfire frequency and burned area with machine learning. The model forecast for July 2024:

My colleagues at LEAP STC are organizing a ML for Earth System Modeling workshop at ICML 2024 this July. Please consider submitting and share the call for papers among interested folks! leap-stc.github.io/ml4esm-works...

An outstanding report from ProPublica - an indictment of how the agencies are mistreating their most valuable resource and it's not just the woefully inadequate level of compensation but also the failure to attend to cancer and emotional trauma and stress. www.propublica.org/article/wild...

“It Feels Impossible to Stay”: The U.S. Needs Wildland Firefighters More Than Ever, but the Federal Government Is Losing Them

Highly skilled firefighters are the last line of defense against wildfires, but that line is fraying because the government decided long ago that they’re not worth very much.

propublica.org