AI-Based Weather Forecasting Has Blind Spots

Traditional weather forecasting models are physics-based and rely on supercomputers. Practically speaking, this means that they start from the basic governing equations (like the Navier-Stokes equations) and use approximations to model aspects of the problem in order to make the physics solvable, given constraints on time, computational power, spatial resolution, and so on.

So-called AI models approach the problem differently, training a model on past weather conditions in order to predict future weather. In some respects, this approach is very successful; AI-based models require less computational infrastructure to run and, in recent years, have greatly improved their predictions of everyday weather.

However, these AI models do poorly when predicting extreme weather events, because their training data contain relatively few examples of these events. They show limited ability to extrapolate their predictions to more extreme events. But these events–like the unprecedented 2021 heatwave in the Pacific Northwest or many of the Category 5 hurricanes we’ve seen in the last decade–are happening increasingly often due to climate change. Those events will keep happening, more frequently, as warming continues. Physics-based models can predict and forecast these events in ways that AI-based models fail to because they are limited by their trained experiences.

Researchers are working to find ways to better equip AI-based models with more physical sense, but, as these models proliferate, it’s important for their users (and those of us using their forecasts) to know what their current weaknesses are. (Image credit: B. McGowan; research credit: Y. Sun et al.; see also S. Nath and T. Palmer; via Gizmodo)

Understanding Pollen Dispersal

When the wind blows, trees shift and sway, reconfiguring their shape and their leaves in response. For parts of the year, that flow can also pluck pollen grains off the tree, carrying them on the winds. A new computational simulation models this pollen dispersal from a tree, with the aim of eventually integrating into a tool for urban planners.

Trees are an important component to fighting climate change, especially in cities, because they cool their surroundings in addition to providing fresh oxygen. But urban planners recognize the downsides to trees, too–allergies, anyone?–and, with the right tools, they could maximize the trees’ advantages while minimizing pollen spread for allergy-sufferers. (Image credit: M. Köles; research credit: T. Dbouk et al.; via Physics World)

📣 Registration is open for the Faculty Development Program on CFD using OpenFOAM by FOSSEE, IIT Bombay.

This free online program is specially designed for faculty members using CFD in research and teaching.

📅 2–5 June 2026
💻 Online | Free of cost

🔗 Register: shorturl.at/CYUzi

📲 Scan QR code in poster for registration.

#CFD #OpenFOAM #ComputationalFluidDynamics #EngineeringFaculty #FDP #FOSSEE #IITBombay #OpenSource #Research #EngineeringEducation #Simulation #OpenSource

37x Speedup in Lattice Boltzmann Cylinder Flow

https://github.com/alikamp/Parks-KPBM-Scaling

#HackerNews #LatticeBoltzmann #Speedup #ComputationalFluidDynamics #Research #Innovation

📣 Faculty members working in CFD and simulation are invited to join the Faculty Development Program on CFD using OpenFOAM by FOSSEE, IIT Bombay.

🗓 2–5 June 2026
💻 Online Mode

Learn OpenFOAM from basic to intermediate level with interactive sessions and receive a certificate upon fulfilling attendance criteria.

🔗 Register: https://shorturl.at/CYUzi

#CFD #OpenFOAM #FDP #FOSSEE #IITBombay #OpenSource #EngineeringEducation #Simulation #ComputationalFluidDynamics

Petite vidéo de l'énergétique d'une expérience de Kelvin-Helmholtz que je trouvais sympa

#CFD #ComputationalFluidDynamics #fluidDynamics #physics

This month, CTCS (IIT Madras) & @PIK_climate present a webinar:
📢: The asymptotic state of decaying turbulence
🎙️: Prof. K. R. Sreenivasan, New York University
📅: March 30 |⏰ 19:30 IST | 16:00 CEST | 10:00 EDT
🔗: https://us06web.zoom.us/webinar/register/WN_0BicRp5MTFqu8Yf3I2jCcQ

#Complexsystems #phd #computationalfluiddynamics #Turbulence #NonlinearDynamics #Bifurcations #NYU #ITCP #PIK #webinarinvite #Zoomcodes #ComplexNetworks #MachineLearning #fluidmechanics

Richtmyer-Meshkov Instability

If you send a shock wave through a magnetized plasma–something that happens in both supernova explosions and inertial confinement fusion–it can trigger an instability known as the Richtmyer-Meshkov instability. The image above shows a form of this, taken from a simulation. Rather than treating the plasma as a single idealized fluid, the researchers represented it as two fluids: an ion fluid and an electron fluid. This allowed them to better capture what happens when certain components of the plasma react to changes faster than others do.

The image itself shows the electron number density across the fluid, where darker colors represent higher electron number density. The interface between high and low-densities shows a roll-up instability that resembles the Kelvin-Helmholtz instability, but there are also regions of mushroom-like plumes that more closely resemble Rayleigh-Taylor instabilities.

The authors note that these structures don’t appear in simulations that represent a plasma as a single fluid; you need the two-fluid representation to see them. (Image and research credit: O. Thompson et al.)

Funded PhD - multiscale electrochemical modeling
University College Dublin

See the full job description on jobRxiv: https://jobrxiv.org/job/university-college-dublin-27778-funded-phd-multiscale-electrochemical-modeling/

#battery #computationalfluiddynamics #electrochemistry #moleculardynamics #quantumchemistry #ScienceJobs #hiring #research
https://jobrxiv.org/job/university-college-dublin-27778-funded-phd-multiscale-electrochemical-modeling/?fsp_sid=10273

Improving Turbulence Models

Calculating turbulent flows like those found in the ocean and atmosphere is extremely expensive computationally. That’s why forecasting models use techniques like Large Eddy Simulation (LES), where large physical scales are calculated according to the governing physical equations while smaller scales are approximated with mathematical models. Researchers are always looking for ways to improve these models–making them more physically accurate, easier to compute, and more computationally stable.

In a new study, researchers used an equation-discovery tool to find new improvements to these models for the smaller turbulent scales. They started by doing a full, computationally expensive calculation of the turbulent flow. The equation-discovery tool then analyzed these results, looking to match them to a library of over 900 possible equations. When it found a form that fit the data, the researchers were then able to show analytically how to derive that equation from the underlying physics. The result is a new equation that models these smaller scales in a way that’s physically accurate and computationally stable, offering possibilities for better LES. (Image credit: CasSa Paintings; research credit: K. Jakhar et al.; via APS)