ML models are increasingly used in weather forecasting, offering accurate predictions and reduced computational costs compared to traditional numerical weather prediction (NWP) models. However, current ML models often have limitations such as coarse temporal resolution (usually 6 hours) and a narrow range of meteorological variables, which can limit their practical use. Accurate forecasting is crucial for renewable energy, aviation, and marine shipping sectors. Despite advancements, ML models still struggle with prediction continuity and temporal resolution. While some models have made strides in accuracy and efficiency, improving their temporal granularity and including a broader set of meteorological variables remains challenging.
Researchers from Fudan University and the Shanghai Academy of Artificial Intelligence have introduced FuXi-2.0, an advanced ML model for global weather forecasting that provides 1-hourly predictions and covers a broad range of meteorological variables. FuXi-2.0 outperforms the European Centre for Medium-Range Weather Forecasts (ECMWF) high-resolution forecasts (HRES) in key areas such as wind power forecasting and tropical cyclone intensity. The model integrates atmospheric and oceanic components, offering improved accuracy over its predecessor, FuXi-1.0, and other models like Pangu-Weather. FuXi-2.0’s enhanced temporal resolution and comprehensive variable set significantly advance practical weather forecasting applications.
The study employs the ERA5 reanalysis dataset from ECMWF, which provides hourly meteorological data with a spatial resolution of approximately 31 km starting from January 1950. For this research, two subsets of ERA5 data were used: one spanning 2012-2017 for training a 6-hourly forecast model and another from 2015-2017 for a 1-hourly forecast model. FuXi-2.0 forecasts 88 meteorological variables, including upper-air and surface variables, with additional static and temporal encodings of geographical information. The model’s training involved resetting accumulated variables to zero to match operational conditions and setting oceanic variables to NaN where applicable. Data from wind farms in the UK and South Korea were also used for wind power forecasting, incorporating quality control measures to ensure accuracy.
FuXi-2.0 introduces a dual-model system to deliver continuous 1-hourly forecasts, integrating a primary model for 6-hourly forecasts and a secondary model for hourly interpolation. This architecture improves reliability and efficiency compared to previous models. The 6-hourly model processes data through convolution layers and Swin Transformer blocks, while the 1-hourly model generates hourly forecasts within a 6-hour window. Training used the robust Charbonnier loss function and involved extensive GPU cluster iteration. Wind power forecasting was conducted using an MLP model focusing on day-ahead forecasts. Evaluation metrics included RMSE, ACC, and forecast/observation activity, with normalized differences used to compare model performance.
The study evaluates FuXi-2.0’s 1-hourly forecasts using 2018 data, comparing its performance with ECMWF HRES and Pangu-Weather. FuXi-2.0 shows superior accuracy in variables important for weather prediction, such as temperature and wind speed, outperforming ECMWF HRES in root mean squared error (RMSE) and anomaly correlation coefficient (ACC) across most forecast lead times. Its forecasts are more detailed than those of Pangu-Weather, and it has better activity measures. Additionally, FuXi-2.0’s wind power forecasts for wind farms and tropical cyclone intensity predictions are more accurate than those from ECMWF HRES, showcasing its improved forecasting capabilities.
In conclusion, Recent advancements in ML for weather forecasting have led to models outperforming the ECMWF HRES in global prediction accuracy. These ML models typically offer 6-hour temporal resolution and 0.25° spatial resolution but are limited by their focus on basic meteorological variables. The FuXi-2.0 model addresses these limitations by providing 1-hourly forecasts and including a wider range of variables crucial for sectors like wind and solar energy, aviation, and maritime shipping. FuXi-2.0 outperforms ECMWF HRES and integrates atmospheric and oceanic data for improved tropical cyclone forecasts. Future improvements include higher spatial resolutions, additional variables, and enhanced precipitation accuracy.
Check out the Paper. All credit for this research goes to the researchers of this project. Also, don’t forget to follow us on Twitter and join our Telegram Channel and LinkedIn Group. If you like our work, you will love our newsletter..
Don’t Forget to join our 50k+ ML SubReddit
Sana Hassan, a consulting intern at Marktechpost and dual-degree student at IIT Madras, is passionate about applying technology and AI to address real-world challenges. With a keen interest in solving practical problems, he brings a fresh perspective to the intersection of AI and real-life solutions.
Leave a comment