FastNet: Improving the physical consistency of machine-learning weather prediction models through loss function design
Published in Artificial Intelligence for the Earth Systems, 2026
This work explores how changes to the training loss, including a spherical harmonic loss, horizontal gradient terms and the representation of wind, improve the physical consistency of FastNet’s machine-learning weather forecasts.
Recommended citation: T. Dunstan, O. Strickson, T. Bennett, ..., N. Eftekhari, et al., "FastNet: Improving the physical consistency of machine-learning weather prediction models through loss function design," Artificial Intelligence for the Earth Systems, vol. 5, no. 3, e250090, 2026.
Download Paper