Lightnews — Scholar-powered news

R. Saravanan

@sarava.net

3.2K followers 580 following 690 posts

Climate scientist, Professor, and Dept Head at Texas A&M University. Author: ClimateDemon.com Now-and-then-blogger: https://Metamodel.blog Bio: https://r.saravanan.us/about

Metamodel.blog

Posts Media Videos Starter Packs

Pinned

R. Saravanan @sarava.net · May 28

I'll be speaking tomorrow, Thurs., May 29th, at 9:30AM Eastern Time in the Weather & Climate Livestream
www.youtube.com/@wclivestream

Topic: Using climate models to predict solar and wind power droughts, i.e., periods when it is Dark & Still

(agupubs.onlinelibrary.wiley.com/doi/full/10....)

Dark Stillness: Predicting Solar and Wind Power Droughts, R. Saravanan, https://r.saravanan.us

2 10

Reposted by R. Saravanan

Dr. Robin Tanamachi 🌪️ @tornatrix.bsky.social · 21h

While I’m at the UCAR Member’s meeting, I’m reminding everyone of the internships, workshops, and other student opportunities that they have available! www.ucar.edu/exhibit/stud...

Exhibit resources | University Corporation for Atmospheric Research

www.ucar.edu

7 14

Reposted by R. Saravanan

Andrew Dessler @andrewdessler.com · 7d

The Dept. of Atmospheric Sciences at Texas A&M has a YouTube channel. Check out our first video! Starring @knubez.bsky.social along with many faculty and students!

youtu.be/_e57SmX2Ijw?...

Texas A&M Atmospheric Sciences

YouTube video by TAMU Atmospheric Sciences

youtu.be

1 9 22

R. Saravanan @sarava.net · 8d

Any plot of future warming will depend on assumptions such as pledges being implemented or specific policies being continued.
Actual policies only affect the present. Mitigation only affects warming thus far and cannot "reduce future warming"
(if fossil fuel use for AI ramps up, the world will warm)

1 2

Reposted by R. Saravanan

Tomer Burg @burgwx.bsky.social · 9d

Adding my 2 cents to the discussion on Google DeepMind’s performance for Imelda… 🧵

You might see a map like this and be impressed that at no point did it predict landfall in the US. But there is a lot of context lost by simplifying it to just one image.

1 11 37

Reposted by R. Saravanan

Dr. Aaron Thierry @thierryaaron.bsky.social · 9d

"In science, to be ‘conservative’ is to understate your findings. In insurance, it means the opposite: erring on the side of overstatement of risks. For a clear assessment of the risks of #climate change, we need these two cultures to meet in the middle"

www.cambridge.org/core/books/a...

The Meaning of Conservative (Chapter 6) - Five Times Faster

Five Times Faster - November 2024

www.cambridge.org

3 8

R. Saravanan @sarava.net · 9d

As a prof who teaches coding, this line in Armin's post stood out:
"I still review every line, shape the architecture, and carry the responsibility for how it runs in production."
How do I teach the skills to be able to do this? It requires carefully controlled incremental exposure to LLM coding.

1 1 5

Reposted by R. Saravanan

Simon Willison @simonwillison.net · 9d

Most of the "90% of code written by AI" claims come from vendors selling AI tools and lack credibility as a result

Armin (creator of Flask, Jinja, Click) is different - when he says 90% of a new significant infrastructure project he's building was AI generated that's worth paying attention to

Armin Ronacher @mitsuhiko.at · 9d

“Is 90% of code going to be written by AI? I don’t know. What I do know is, that for me, on this project, the answer is already yes.” lucumr.pocoo.org/2025/9/29/90...

90%

AI is writing 90% of the code I was in charge of

lucumr.pocoo.org

4 20 110

Reposted by R. Saravanan

National Weather Service @nws.noaa.gov · 9d

Quite a satellite image this morning! 🛰️🌀

Clouds extend 1000 miles from #Imelda's center all the way into central Pennsylvania.

#Humberto continues to churn in the Atlantic.

Visible satellite image of the East Coast and Atlantic showing the two tropical system.

7 110 280

R. Saravanan @sarava.net · 10d

Nice piece. "Scientific" is a better adjective than "natural" when assessing efficacy and impacts. In the climate mitigation arena, most nature-based carbon offset schemes are either ineffective or just scams. (We may discover effective nature-based schemes, but only science can tell!)

1 5

Reposted by R. Saravanan

Michael Grunwald @mikegrunwald.bsky.social · 10d

My latest effort to be super-popular in @nytimes.com: “Spraying Roundup on Crops Is Fine. Really.” Yes, the world’s most hated herbicide is way less toxic than caffeine. Here’s a gift link where you can leave all your supportive comments: www.nytimes.com/2025/09/28/o...

Opinion | Spraying Roundup on Crops Is Fine. Really.

www.nytimes.com

13 7 34

Reposted by R. Saravanan

Kelly Hereid @kellyhereid.bsky.social · 13d

90 of 111 fatalities from Hurricane Rita happened due to the evacuation. 20 years later, would we do any better?

@sarava.net shouting out the value of much improved track forecasts

🎁 link:

20 years after Rita’s gridlock nightmare, is Houston ready to evacuate again?

Stalled and overheated cars, endless gridlock and nearly 100 deaths. The deadly Hurricane Rita evacuation is still etched in the memories of many Houstonians 20 years later.

www.houstonchronicle.com

4 19 46

Reposted by R. Saravanan

Nick Lutsko @nick-lutsko.bsky.social · 13d

I wrote a short piece on the debate about the future of climate modeling: Should we pool resources to do km-scale "masterpiece" runs or focus on using ML to build data-driven models that can still do large ensembles? 1/3

1 2 4

R. Saravanan @sarava.net · 13d

If we can't decide, how do we know if an ML model is "better" than a physical model?
For ML emulators, how do we know the extreme tail using a very large (~10,000) ensemble is correct if we can't afford a very large physical ensemble to verify it? (We don't have data to verify such extremes) 4/4

R. Saravanan @sarava.net · 13d

It is also important to agree on metrics. How you decide if predictions (of mean & spread) from Model A are better than predictions from Model B? For 1-day prediction, it is easy to verify forecast skill. For predicting the year 2100, how do we decide which prediction is better? 3/n

R. Saravanan @sarava.net · 13d

Do you want predictions at lead times of 1 day, 1 month, 1 year, 1 decade or 1 century? Do you want to predict emergent metrics like ECS/global-avg T, or future rainfall/T extremes at the county or zip code level?
it is important to define the goals. 2/n

R. Saravanan @sarava.net · 13d

It is indeed an emerging and interesting debate. At present, the physical models run on CPUs and the ML models run on GPUs. So they're not directly competing for computer resources. That will change though.
The debate is clouded because the prediction goals are poorly defined. 1/n

Reposted by R. Saravanan

Shel Winkley @shelwinkleywx.bsky.social · 13d

📱 Come make meaningful content with us!

This gives a creative, climate-motivated creator the chance to bring science & data to life through digital storytelling.

Grow your video & editing skills while communicating climate change impacts. (Probably work in a little fun along the way, too!)

Climate Central @climatecentral.org · 13d

What @ariellemargot.bsky.social said.

We're hiring a Social Media and Digital Content Creator!
www.climatecentral.org/open-positio...

1 2 3

R. Saravanan @sarava.net · 13d

If you compared the ACE2 emulator results to those obtained using a baseline multivariate linear regression model built using the same EAM training data (perhaps using a truncated EOF basis as predictors), how would the emulator fare?

R. Saravanan @sarava.net · 14d

What does "act now" mean?
Not saying 2C is achievable, but it is a more "realistic" goal, for a T target. I personally prefer emission targets, i.e., near net-zero. At present, there is no viable negative emissions tech (and there may never be). Geoengineering may create more problems than it solves

1 1

R. Saravanan @sarava.net · 14d

- Why replace the failed 1.5C target with the even more ambitious negative emissions target implied by the overshoot framing? Why not drop the overshoot framing and settle for a more achievable 2C target (in conjunction with net-zero)? 3/3

2 3

R. Saravanan @sarava.net · 14d

- Overshoot framing that wants to lower T back to 1.5C is even more ambitious (and expensive) than the 1.5C target because it requires net negative emissions. (This framing is obviously popular with carbon capture and geoengineering proponents.) 2/n

1 2

R. Saravanan @sarava.net · 14d

Now that it looks all but inevitable that we'll cross 1.5C soon, I have noticed talk of overshoot.
Some thoughts:
- 1.5C is not a physical tipping threshold; it was proposed as a more ambitious stabilization target than 2C (by island nations, to minimize SLR), in conjunction with net-zero. 1/n

1 1

R. Saravanan @sarava.net · 15d

At present, most AI models (emulators) are trained on just the atmospheric part of climate models, not the full oceanic part. So they're not true climate models, despite being called that. That may change with more research, but paucity of data makes it much harder to use AI well for ocean modeling.

1 1

R. Saravanan @sarava.net · 15d

As a climate modeler, I'm curious what you mean by "democratize climate models"? (AI has made weather forecasting models fast enough to run on your phone/laptop but you don't need to make centennial climate predictions everyday on your phone/laptop. Once a decade maybe enough.)

1 2

R. Saravanan @sarava.net · 15d

Here's another fantasy world that forms the basis for many geoengineering and tipping point discussions.

(I also have a longer comment responding to the original substack post on geongineering: www.theclimatebrink.com/p/revisiting...
)

1 3