Search

Data-driven modeling based on machine learning (ML) is showing enormous potential for weather forecasting. Rapid progress has been made with impressive results for some applications. The uptake of ML methods could be a game changer for the incremental progress in traditional numerical weather prediction (NWP) known as the "quiet revolution" of weather forecasting. The computational cost of running a forecast with standard NWP systems greatly hinders the improvements that can be made by increasing model resolution and ensemble sizes. An emerging new generation of ML models, developed using high-quality reanalysis datasets like ERA5 for training, allows forecasts that require much lower computational costs and that are highly competitive in terms of accuracy. Here, we compare for the first time ML-generated forecasts with standard NWP-based forecasts in an operational-like context, initialized from the same initial conditions. Focusing on deterministic forecasts, we apply common forecast verification tools to assess to what extent a data-driven forecast produced with one of the recently developed ML models (PanguWeather) matches the quality and attributes of a forecast from one of the leading global NWP systems (the ECMWF IFS). The results are very promising, with comparable accuracy for both global metrics and extreme events, when verified against both the operational IFS analysis and synoptic observations. Overly smooth forecasts, increasing bias with forecast lead time, and poor performance in predicting tropical cyclone intensity are identified as current drawbacks of ML-based forecasts. A new NWP paradigm is emerging relying on inference from ML models and state-of-the-art analysis and reanalysis datasets for forecast initialization and model training. SIGNIFICANCE STATEMENT: We compare the traditional approach of generating weather forecasts with a new approach based on machine learning (ML). The traditional approach is based on resolving physical equations with numerical methods and can be quite expensive to run on supercomputers. ML models are trained on a very large dataset that combines observations and physically based models to derive our best estimate of the state of the atmosphere hour after hour, over the past decades. In this work, we assess the forecast performance with statistical tools and a focus on extreme events. Our results suggest that ML models, which are much less expensive to run than standard methods, could have a promising future in numerical weather prediction. Our study also points out weaknesses and limitations that would require further investigations and future ML model improvements. KEYWORDS: Forecast verification/skill; Operational forecasting; Machine learning, 1. Introduction Numerical weather prediction (NWP) is the dominant approach for weather forecasting. A weather forecast is the result of the numerical integration of partial differential equations starting from the [...]