Geophysical Research Letters, Journal Year: 2025, Volume and Issue: 52(8)
Published: April 23, 2025
Abstract A new approach called cross‐attractor transforms (Agarwal et al., 2025, https://doi.org/10.1029/2024gl110472 ) aims to improve weather forecasts by using neural networks learn optimal maps between nature and imperfect numerical prediction (NWP) models. Unlike the latest generation of machine learning (MLWP) models, this leverages prior knowledge via known governing equations learns only what is needed map that model target system being forecasted (e.g., real‐world weather). This draws from same underlying principles dynamical systems theory chaos have been foundation operational NWP for last half century, extend upon based post‐processing efforts. The results show an enhanced potential outperform both MLWP models post‐processed with ML, highlighting value in merging data‐driven ML methods.
Language: Английский