[Paper Review] FuXi-2.0: Advancing machine learning weather forecasting model for practical applications

Xiaohui Zhong, Lei Chen+4 more|arXiv (Cornell University)|Sep 11, 2024

Hydrological Forecasting Using AI7 citations

TL;DR

FuXi-2.0 delivers 1-hourly global weather forecasts for the first 5 days and 6-hourly forecasts thereafter, incorporating a broad set of atmospheric and oceanic variables to outperform ECMWF HRES in key practical scenarios.

ABSTRACT

Machine learning (ML) models have become increasingly valuable in weather forecasting, providing forecasts that not only lower computational costs but often match or exceed the accuracy of traditional numerical weather prediction (NWP) models. Despite their potential, ML models typically suffer from limitations such as coarse temporal resolution, typically 6 hours, and a limited set of meteorological variables, limiting their practical applicability. To overcome these challenges, we introduce FuXi-2.0, an advanced ML model that delivers 1-hourly global weather forecasts and includes a comprehensive set of essential meteorological variables, thereby expanding its utility across various sectors like wind and solar energy, aviation, and marine shipping. Our study conducts comparative analyses between ML-based 1-hourly forecasts and those from the high-resolution forecast (HRES) of the European Centre for Medium-Range Weather Forecasts (ECMWF) for various practical scenarios. The results demonstrate that FuXi-2.0 consistently outperforms ECMWF HRES in forecasting key meteorological variables relevant to these sectors. In particular, FuXi-2.0 shows superior performance in wind power forecasting compared to ECMWF HRES, further validating its efficacy as a reliable tool for scenarios demanding precise weather forecasts. Additionally, FuXi-2.0 also integrates both atmospheric and oceanic components, representing a significant step forward in the development of coupled atmospheric-ocean models. Further comparative analyses reveal that FuXi-2.0 provides more accurate forecasts of tropical cyclone intensity than its predecessor, FuXi-1.0, suggesting that there are benefits of an atmosphere-ocean coupled model over atmosphere-only models.

Motivation & Objective

• Motivate the need for high-temporal-resolution ML weather forecasts across sectors like wind/solar, aviation, and marine shipping.
• Develop FuXi-2.0 to achieve 1-hourly global forecasts for the first 5 days and 6-hourly forecasts afterward.
• Expand the set of output variables to include both atmospheric and oceanic components for coupled modeling.
• Evaluate FuXi-2.0 against ECMWF HRES and Pangu-Weather across multiple practical variables and scenarios.

Proposed method

• Two-model framework: a 6-hourly forecast generator and a 1-hourly interpolator to ensure continuous 1-hourly forecasts.
• Transformer-based interpolation to reduce iterations and maintain forecast continuity.
• Inclusion of 5 upper-air variables across 13 pressure levels and 23 surface variables to support wind, solar, aviation, and marine applications.
• Incorporation of ocean variables to enable atmosphere-ocean coupling and improved tropical cyclone intensity forecasts.
• Evaluation against ERA5 as a reference and comparison with ECMWF HRES and Pangu-Weather for 1-hourly forecasts up to 90 hours.

Experimental results

Research questions

• RQ1 Can FuXi-2.0 deliver 1-hourly forecasts with continuous temporal coverage for the first 5 days and reliable 6-hourly forecasts thereafter?
• RQ2 Do FuXi-2.0 forecasts outperform ECMWF HRES and Pangu-Weather for key variables across wind, solar, aviation, and marine shipping applications?
• RQ3 Does atmosphere-ocean coupling in FuXi-2.0 enhance tropical cyclone intensity forecasts compared to atmosphere-only models?
• RQ4 What is the impact of the expanded variable set on forecast accuracy and activity across different lead times?

Key findings

• FuXi-2.0 outperforms ECMWF HRES in RMSE and ACC for commonly evaluated variables across up to 90-hour forecasts.
• FuXi-2.0 provides superior wind power forecasting using 1-hourly outputs compared to HRES.
• Coupled atmosphere-ocean outputs in FuXi-2.0 yield more accurate tropical cyclone intensity forecasts than FuXi-1.0 (atmosphere-only).
• Forecast activity analyses show FuXi-2.0 maintains realistic variability without over-smoothing, in contrast to Pangu-Weather’s smoother forecasts.
• FuXi-2.0 demonstrates advantages over both Pangu-Weather and HRES in key practical variables for wind/solar, aviation, and marine shipping.