Intraseasonal Equatorial Kelvin and Rossby Waves in Modern AI-ML Models

Abstract

We examine the structure of large-scale convectively coupled Kelvin and Rossby waves in a suite of modern AI-ML models. In particular, multiple runs of PanguWeather, GraphCast, FourCastNet and Aurora are performed to assess the structure of the aforementioned waves. Wavenumber-frequency diagrams of zonal winds from all models show a clear signature of Rossby and Kelvin waves with equivalent depths that are in accord with observations and reanalysis. Composites of Kelvin waves show correct lower and upper troposphere horizontal convergence patterns, vertical tilts in temperature, humidity and vertical velocity as well as the phase relation between temperature and vertical velocity anomalies. Though, differences between models are notable such as smaller vertical tilts and incorrect surface temperature anomalies in GraphCast and relatively weak convergent flows in PanguWeather. The models had much more difficulty with Rossby waves; while the horizontal gyres were captured, the vertical structure of temperature and divergence was incorrect. Apart from unexpected tilts in various fields, the temperature anomaly was inconsistent with the nature of the vertical velocity in all four models. Curiously, moisture and vertical velocity anomalies were much closer to observations. Further, only two models (GraphCast and FourCastNet) captured the simultaneous generation of deep vertical motion with moisture anomalies. In all, while the representation of these large-scale waves is encouraging, issues with the structure of Rossby waves and especially the inconsistency among fields require further investigation.