← News

Models & toolsSources: arXiv · 18 Nov 2026

AEGIS brings differentiable modeling to Mars climate simulation

Researchers introduce AEGIS, a Mars general circulation model rebuilt to expose gradients, pairing the planet's atmospheric physics with a modern dynamical core and hooks for neural network components.

In plain words

Climate models simulate a planet's weather and atmosphere, but Mars versions are hard to check against observations because data from the planet is sparse. AEGIS is a new Mars climate model built so its calculations can be differentiated, meaning the model's outputs can be mathematically traced back to its inputs. That allows automated calibration and lets machine learning components be trained inside the model.

What was announced

The authors, presenting the work on arXiv, argue that existing Mars models such as the LMD PCM, the NASA Ames Mars GCM, and PlanetWRF are mature and physically detailed but written in legacy Fortran with finite-difference or finite-volume solvers, so they do not expose gradients for calibration or machine learning. AEGIS instead couples Mars's atmospheric physics to the Dinosaur dynamical core and includes interfaces for neural closures, the learned components that can stand in for or augment physical parameterizations.

In the demonstrations reported, AEGIS ran stable simulations spanning ten Mars years. These reproduced the seasonal carbon dioxide cycle while conserving the total CO2 inventory, captured large-scale surface-temperature structure, and produced surface pressure consistent with Mars Orbiter Laser Altimeter topography. Gradients computed through coupled trajectories matched finite-difference calculations, which the authors say supports physical calibration and neural training, and they highlight the framework's computational efficiency and differentiability compared with conventional models.

What it does not show

  • The source is an abstract only; detailed methods, validation results and comparisons are not available in the supplied passages.
  • Claims of efficiency, gradient accuracy and simulation quality come from the paper's authors and have not been independently confirmed.

For security and IT teams

Scope
The paper targets Mars atmosphere simulation, including weather forecasting for entry, descent and landing operations, not Earth climate modeling.

Sources

Source-support review completed 19 Nov 2026. This story was drafted with AI assistance and checked against retained source passages. Automated checking is not independent verification.

Coverage: arXiv: abstract.

Updates and corrections

No corrections so far.

First published 19 Nov 2026

New to the subject? AI from the Ground Up starts from the beginning.