R. Saravanan
@sarava.net
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
"Runctitiononal features"? "Medical fymblal"? "1 Tol Line storee"? This gets worse the longer you look at it. But it's got to be good, because it was published in Nature Scientific Reports last week: www.nature.com/articles/s41... h/t @asa.tsbalans.se
November 27, 2025 at 7:13 PM
CESM atmospheric model is working with Fortran for GPUfication in the near-term (unlike DOE, which is headed towards all C++/Kokkos, I believe).
NCAR recently received a grant to build C++ version of the CESM ocean model (MOM) using AMReX framework
Not aware of major other language efforts for CESM.
NCAR recently received a grant to build C++ version of the CESM ocean model (MOM) using AMReX framework
Not aware of major other language efforts for CESM.
November 27, 2025 at 7:07 PM
CESM atmospheric model is working with Fortran for GPUfication in the near-term (unlike DOE, which is headed towards all C++/Kokkos, I believe).
NCAR recently received a grant to build C++ version of the CESM ocean model (MOM) using AMReX framework
Not aware of major other language efforts for CESM.
NCAR recently received a grant to build C++ version of the CESM ocean model (MOM) using AMReX framework
Not aware of major other language efforts for CESM.
Yes, DOE is ahead of the game in GPU friendly modeling, although there's still physics (parameterizations) that need to be ported. Other US modeling efforts (e.g., CESM) are quite behind. Therefore, the most popular models currently used by the community will not be able to use GPUs well.
November 27, 2025 at 3:57 PM
Yes, DOE is ahead of the game in GPU friendly modeling, although there's still physics (parameterizations) that need to be ported. Other US modeling efforts (e.g., CESM) are quite behind. Therefore, the most popular models currently used by the community will not be able to use GPUs well.
Also, I believe DOE no longer supports research for decadal and longer timescales; the focus has shifted to shorter term.
The available computing is likely to be all GPU. Most legacy climate code runs very poorly on GPUs. There is new ML code (emulators etc.) for GPUs but it's still early days there
The available computing is likely to be all GPU. Most legacy climate code runs very poorly on GPUs. There is new ML code (emulators etc.) for GPUs but it's still early days there
November 26, 2025 at 9:57 PM
Also, I believe DOE no longer supports research for decadal and longer timescales; the focus has shifted to shorter term.
The available computing is likely to be all GPU. Most legacy climate code runs very poorly on GPUs. There is new ML code (emulators etc.) for GPUs but it's still early days there
The available computing is likely to be all GPU. Most legacy climate code runs very poorly on GPUs. There is new ML code (emulators etc.) for GPUs but it's still early days there
Tipping point framing is often used to motivate SRM, as in this article: "Earth’s climate and nature are already passing tipping points"
I know gloom and doom sells better in reporting but I wish climate reporters would also read this paper: www.nature.com/articles/s41...
I know gloom and doom sells better in reporting but I wish climate reporters would also read this paper: www.nature.com/articles/s41...
‘Tipping points’ confuse and can distract from urgent climate action - Nature Climate Change
The tipping points framing is widely used in climate discussions but receives mixed feedback. This Perspective critiques it for oversimplifying the complexities of natural and social systems and faili...
www.nature.com
November 25, 2025 at 10:30 PM
Tipping point framing is often used to motivate SRM, as in this article: "Earth’s climate and nature are already passing tipping points"
I know gloom and doom sells better in reporting but I wish climate reporters would also read this paper: www.nature.com/articles/s41...
I know gloom and doom sells better in reporting but I wish climate reporters would also read this paper: www.nature.com/articles/s41...
Tehran appears to have survived extreme low rainfall in 1950 and 1961, but the population now is ~10 million (compared to 2-3 million 50 years ago)!
November 24, 2025 at 7:25 PM
Tehran appears to have survived extreme low rainfall in 1950 and 1961, but the population now is ~10 million (compared to 2-3 million 50 years ago)!
There is no strong case to invest massive resources in SRM or other geoengineering research (and certainly not to prepare for deployment).
I don't buy the tipping point or overshoot arguments for large investments in SRM research (while insisting that no deployment is intended). 4/4
I don't buy the tipping point or overshoot arguments for large investments in SRM research (while insisting that no deployment is intended). 4/4
November 24, 2025 at 4:49 PM
There is no strong case to invest massive resources in SRM or other geoengineering research (and certainly not to prepare for deployment).
I don't buy the tipping point or overshoot arguments for large investments in SRM research (while insisting that no deployment is intended). 4/4
I don't buy the tipping point or overshoot arguments for large investments in SRM research (while insisting that no deployment is intended). 4/4
Do we want to trust uncertain SRM to counter uncertain warming impacts? Not really. We can be much more certain that reducing carbon emissions will reduce the warming.
I do support SRM research, but in a low-key fashion, like any other fundamental scientific research with a modest research budget 3/
I do support SRM research, but in a low-key fashion, like any other fundamental scientific research with a modest research budget 3/
November 24, 2025 at 4:49 PM
Do we want to trust uncertain SRM to counter uncertain warming impacts? Not really. We can be much more certain that reducing carbon emissions will reduce the warming.
I do support SRM research, but in a low-key fashion, like any other fundamental scientific research with a modest research budget 3/
I do support SRM research, but in a low-key fashion, like any other fundamental scientific research with a modest research budget 3/
This means that we cannot completely trust models we use to project global warming, but they are the best we have to plan for the future.
One could argue since we trust GCMs to project warming, we should also trust them for SRM studies. Well ... two wrongs do not make a right! 2/
One could argue since we trust GCMs to project warming, we should also trust them for SRM studies. Well ... two wrongs do not make a right! 2/
November 24, 2025 at 4:49 PM
This means that we cannot completely trust models we use to project global warming, but they are the best we have to plan for the future.
One could argue since we trust GCMs to project warming, we should also trust them for SRM studies. Well ... two wrongs do not make a right! 2/
One could argue since we trust GCMs to project warming, we should also trust them for SRM studies. Well ... two wrongs do not make a right! 2/
Some thoughts on this topic:
All models can be wrong for aspects of the climate system. Model convergence does not imply that models are "correct", especially if models that converge represent key processes poorly due to "herd mentality" in parameterizations.
E.g.
www.pnas.org/doi/10.1073/... 1/
All models can be wrong for aspects of the climate system. Model convergence does not imply that models are "correct", especially if models that converge represent key processes poorly due to "herd mentality" in parameterizations.
E.g.
www.pnas.org/doi/10.1073/... 1/
Observed humidity trends in dry regions contradict climate models | PNAS
Arid and semi-arid regions of the world are particularly vulnerable to greenhouse
gas–driven hydroclimate change. Climate models are our primary to...
www.pnas.org
November 24, 2025 at 4:49 PM
Some thoughts on this topic:
All models can be wrong for aspects of the climate system. Model convergence does not imply that models are "correct", especially if models that converge represent key processes poorly due to "herd mentality" in parameterizations.
E.g.
www.pnas.org/doi/10.1073/... 1/
All models can be wrong for aspects of the climate system. Model convergence does not imply that models are "correct", especially if models that converge represent key processes poorly due to "herd mentality" in parameterizations.
E.g.
www.pnas.org/doi/10.1073/... 1/
Impressive...but no Wallace & Gromit reference?
November 20, 2025 at 7:47 PM
Impressive...but no Wallace & Gromit reference?