The Rise of Data-Driven Weather Forecasting: A First Statistical Assessment of Machine Learning–Based Weather Forecasts in an Operational-Like Context

• DOI: 10.1175/bams-d-23-0162.1
• Publication Date: 2024-02-29T16:04:53Z
• Authors: Zied Ben Bouallègue; Mariana C. A. Clare; Linus Magnusson; Estibaliz Gascón; Michael Maier-Gerber; Martin Janoušek; Mark Rodwell; Florian Pinault; Jesper S. Dramsch; Simon T. K. Lang; Baudouin Raoult; Florence Rabier; Matthieu Chevallier; Irina Sandu; Peter Dueben; Matthew Chantry; Florian Pappenberger
• Keywords: Weather prediction; Model output statistics

Abstract Data-driven modeling based on machine learning (ML) is showing enormous potential for weather forecasting. Rapid progress has been made with impressive results some applications. The uptake of ML methods could be a game changer the incremental in traditional numerical prediction (NWP) known as “quiet revolution” computational cost running forecast standard NWP systems greatly hinders improvements that can by increasing model resolution and ensemble sizes. An emerging new generation models, developed using high-quality reanalysis datasets like ERA5 training, allows forecasts require much lower costs are highly competitive terms accuracy. Here, we compare first time ML-generated NWP-based an operational-like context, initialized from same initial conditions. Focusing deterministic forecasts, apply common verification tools to assess what extent data-driven produced one recently models (PanguWeather) matches quality attributes leading global (the ECMWF IFS). very promising, comparable accuracy both metrics extreme events, when verified against operational IFS analysis synoptic observations. Overly smooth bias lead time, poor performance predicting tropical cyclone intensity identified current drawbacks ML-based forecasts. A paradigm relying inference state-of-the-art initialization training.