FuXi-ENS: A machine learning model for medium-range ensemble weather forecasting

FuXi-ENS: A machine learning model for medium-range ensemble weather forecasting

9 Aug 2024 | Xiaohui Zhong1†, Lei Chen1,2†, Hao Li1,2*†, Jun Liu1, Xu Fan2, Jie Feng3, Kan Dai4, Jing-Jia Luo5, Jie Wu6 and Bo Lu6*
**FuXi-ENS: A Machine Learning Model for Medium-Range Ensemble Weather Forecasting** This paper introduces FuXi-ENS, an advanced machine learning (ML) model designed to deliver 6-hourly global ensemble weather forecasts up to 15 days. The model operates at a spatial resolution of 0.25°, incorporating 5 atmospheric variables at 13 pressure levels and 13 surface variables. By leveraging the probabilistic nature of Variational AutoEncoder (VAE), FuXi-ENS optimizes a loss function that combines the continuous ranked probability score (CRPS) and the Kullback–Leibler (KL) divergence, enhancing the incorporation of flow-dependent perturbations in both initial conditions and forecasts. **Key Contributions:** 1. **High Spatial Resolution:** FuXi-ENS operates at a high spatial resolution of 0.25°, which is crucial for many applications. 2. **Probabilistic Forecasting:** The model incorporates CRPS and KL loss, outperforming traditional L1 loss combined with KL loss in VAE models. 3. **Flow-Dependent Perturbations:** The model introduces perturbations at both initial conditions and each forecast step, mirroring the ensemble generation process in traditional numerical weather prediction (NWP) models. **Performance Evaluation:** - **Compared to ECMWF Ensemble:** FuXi-ENS outperforms the ECMWF ensemble in 98.1% of 360 variable and forecast lead time combinations, as measured by CRPS. - **Extreme Weather Events:** The model demonstrates superior performance in predicting extreme weather events, including tropical cyclones (TCs) and the 2018 Northeast Asia heatwave. - **Computational Efficiency:** FuXi-ENS completes a 15-day forecast with a 6-hourly temporal resolution in approximately 10 seconds per member on an Nvidia A100 GPU, significantly faster than conventional NWP models. **Conclusion:** FuXi-ENS represents a significant advancement in medium-range ensemble weather forecasting, offering superior performance and computational efficiency. The model's ability to handle high-resolution forecasts and its innovative loss function design make it a promising tool for enhancing ensemble weather predictions, particularly for extreme events.**FuXi-ENS: A Machine Learning Model for Medium-Range Ensemble Weather Forecasting** This paper introduces FuXi-ENS, an advanced machine learning (ML) model designed to deliver 6-hourly global ensemble weather forecasts up to 15 days. The model operates at a spatial resolution of 0.25°, incorporating 5 atmospheric variables at 13 pressure levels and 13 surface variables. By leveraging the probabilistic nature of Variational AutoEncoder (VAE), FuXi-ENS optimizes a loss function that combines the continuous ranked probability score (CRPS) and the Kullback–Leibler (KL) divergence, enhancing the incorporation of flow-dependent perturbations in both initial conditions and forecasts. **Key Contributions:** 1. **High Spatial Resolution:** FuXi-ENS operates at a high spatial resolution of 0.25°, which is crucial for many applications. 2. **Probabilistic Forecasting:** The model incorporates CRPS and KL loss, outperforming traditional L1 loss combined with KL loss in VAE models. 3. **Flow-Dependent Perturbations:** The model introduces perturbations at both initial conditions and each forecast step, mirroring the ensemble generation process in traditional numerical weather prediction (NWP) models. **Performance Evaluation:** - **Compared to ECMWF Ensemble:** FuXi-ENS outperforms the ECMWF ensemble in 98.1% of 360 variable and forecast lead time combinations, as measured by CRPS. - **Extreme Weather Events:** The model demonstrates superior performance in predicting extreme weather events, including tropical cyclones (TCs) and the 2018 Northeast Asia heatwave. - **Computational Efficiency:** FuXi-ENS completes a 15-day forecast with a 6-hourly temporal resolution in approximately 10 seconds per member on an Nvidia A100 GPU, significantly faster than conventional NWP models. **Conclusion:** FuXi-ENS represents a significant advancement in medium-range ensemble weather forecasting, offering superior performance and computational efficiency. The model's ability to handle high-resolution forecasts and its innovative loss function design make it a promising tool for enhancing ensemble weather predictions, particularly for extreme events.
Reach us at info@study.space
Understanding FuXi-ENS%3A A machine learning model for medium-range ensemble weather forecasting