Reproduce “Investigating Inductive Biases for Machine Learning Emulation of Sudden Stratospheric Warmings in Idealised Isca Simulations” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Oskar Bohn Lassen, Simon Driscoll, Stephen I. Thomson, Sebastian Schemm, Francisco C. Pereira
This study demonstrates that machine learning emulators with explicitly three-dimensional architectural inductive biases are superior for capturing stratospheric dynamics, especially Sudden Stratospheric Warmings (SSWs), in idealized atmospheric simulations. It reveals that low pointwise prediction error does not guarantee physically faithful wave-mean-flow interactions, highlighting the need for mechanism-aware evaluation through Eliassen–Palm flux diagnostics to identify specific emulator failure modes.
Key insights
- Explicitly three-dimensional ML architectures, which inherently encode vertical coupling, consistently outperform two-dimensional and graph-based models in emulating stratospheric dynamics, particularly under active SSW-like variability.
- Low pointwise forecast error in individual variables (temperature, wind) does not guarantee physically faithful wave-mean-flow interaction; mechanism-aware diagnostics like EP-flux are critical for identifying dynamically important inconsistencies.
- Emulator errors in capturing stratospheric wave activity were disproportionately large for the wave-1 component, suggesting difficulty in reproducing nonlinear stratospheric processes that generate this signal, even with dominant wave-2 forcing.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.