Search papers, labs, and topics across Lattice.
This study investigates how different machine learning architectures鈥攃onvolutional, transformer, and graph-based鈥攁ffect the emulation of sudden stratospheric warming (SSW) dynamics in idealized Isca simulations. The findings reveal that while model performance is similar under quiet stratospheric conditions, significant differences emerge during active SSW events, highlighting the importance of three-dimensional vertical coupling as a critical inductive bias. Notably, low forecast errors do not necessarily correlate with accurate representations of wave-mean-flow interactions, indicating potential limitations in current emulation approaches.
Significant architectural differences in machine learning models only emerge during active sudden stratospheric warming events, revealing the crucial role of vertical coupling in accurate predictions.
Machine-learning emulators are increasingly used for weather prediction and have the potential to extend skill on subseasonal-to-seasonal timescales by learning dynamically important sources of predictability. A key challenge is whether the models can exploit predictability anchors, such as stratospheric variability, that influence tropospheric circulation beyond short lead times. We test how architectural inductive bias affects emulation of sudden stratospheric warming (SSW) dynamics using paired idealised Isca simulations that differ only in an imposed wave-2 heating perturbation. Across convolutional, transformer, and graph-based architectures trained for one-step prediction, model differences are modest when the stratosphere is dynamically quiet but widen substantially when SSW-like variability is active. Our results identify explicit three-dimensional vertical coupling as a key inductive bias for machine-learning emulation of stratospheric dynamics. However, Eliassen-Palm flux diagnostics show that low forecast error does not guarantee physically faithful wave-mean-flow interaction, with coherent errors remaining in stratospheric wave-driving structure.
