Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

OneForecast: A Universal Framework for Global and Regional Weather Forecasting

Yuan Gao, Hao Wu, Ruiqi Shu, Huanshuo Dong, Fan Xu, Rui Ray Chen, Yibo Yan, Qingsong Wen, Xuming Hu, Kun Wang, Jiahao Wu, Qing Li, Hui Xiong, Xiaomeng Huang

TL;DR

OneForecast addresses the challenge of accurate weather forecasting across global and regional scales by unifying a multi-scale graph neural network with region refinement and a neural nested grid. It introduces an adaptive Multi-stream Messaging mechanism with dynamic gating to preserve high-frequency signals and extreme-event features, and a neural nesting grid to inject global forecasts into regional high-resolution predictions. Evaluations on WeatherBench2 ERA5 data show improved RMSE, ACC, CSI, and SEDI, with strong performance in long-term rollouts and cyclone tracking, and robust ensemble forecasts via $N=50$ perturbed initial conditions. The approach reduces boundary-information loss and demonstrates a scalable, end-to-end trainable framework that better captures atmospheric dynamics across spatial and temporal scales.

Abstract

Accurate weather forecasts are important for disaster prevention, agricultural planning, etc. Traditional numerical weather prediction (NWP) methods offer physically interpretable high-accuracy predictions but are computationally expensive and fail to fully leverage rapidly growing historical data. In recent years, deep learning models have made significant progress in weather forecasting, but challenges remain, such as balancing global and regional high-resolution forecasts, excessive smoothing in extreme event predictions, and insufficient dynamic system modeling. To address these issues, this paper proposes a global-regional nested weather forecasting framework (OneForecast) based on graph neural networks. By combining a dynamic system perspective with multi-grid theory, we construct a multi-scale graph structure and densify the target region to capture local high-frequency features. We introduce an adaptive messaging mechanism, using dynamic gating units to deeply integrate node and edge features for more accurate extreme event forecasting. For high-resolution regional forecasts, we propose a neural nested grid method to mitigate boundary information loss. Experimental results show that OneForecast performs excellently across global to regional scales and short-term to long-term forecasts, especially in extreme event predictions. Codes link https://github.com/YuanGao-YG/OneForecast.

OneForecast: A Universal Framework for Global and Regional Weather Forecasting

TL;DR

OneForecast addresses the challenge of accurate weather forecasting across global and regional scales by unifying a multi-scale graph neural network with region refinement and a neural nested grid. It introduces an adaptive Multi-stream Messaging mechanism with dynamic gating to preserve high-frequency signals and extreme-event features, and a neural nesting grid to inject global forecasts into regional high-resolution predictions. Evaluations on WeatherBench2 ERA5 data show improved RMSE, ACC, CSI, and SEDI, with strong performance in long-term rollouts and cyclone tracking, and robust ensemble forecasts via perturbed initial conditions. The approach reduces boundary-information loss and demonstrates a scalable, end-to-end trainable framework that better captures atmospheric dynamics across spatial and temporal scales.

Abstract

Accurate weather forecasts are important for disaster prevention, agricultural planning, etc. Traditional numerical weather prediction (NWP) methods offer physically interpretable high-accuracy predictions but are computationally expensive and fail to fully leverage rapidly growing historical data. In recent years, deep learning models have made significant progress in weather forecasting, but challenges remain, such as balancing global and regional high-resolution forecasts, excessive smoothing in extreme event predictions, and insufficient dynamic system modeling. To address these issues, this paper proposes a global-regional nested weather forecasting framework (OneForecast) based on graph neural networks. By combining a dynamic system perspective with multi-grid theory, we construct a multi-scale graph structure and densify the target region to capture local high-frequency features. We introduce an adaptive messaging mechanism, using dynamic gating units to deeply integrate node and edge features for more accurate extreme event forecasting. For high-resolution regional forecasts, we propose a neural nested grid method to mitigate boundary information loss. Experimental results show that OneForecast performs excellently across global to regional scales and short-term to long-term forecasts, especially in extreme event predictions. Codes link https://github.com/YuanGao-YG/OneForecast.

Paper Structure

This paper contains 37 sections, 2 theorems, 45 equations, 30 figures, 7 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Method
  3. Earth-specific Region Refined Graph Encoder
  4. Multi-stream Messaging
  5. Decoding and Optimization
  6. Downstream Tasks
  7. Experiments
  8. Benchmarks and Baselines
  9. Comparison with state-of-the-art methods
  10. Regional High Resolution Forecast
  11. Extreme Events Assessment
  12. Long-term and Ensemble Forecasts
  13. Ablation Studies
  14. Conclusion
  15. Proofs of Theorems
  16. ...and 22 more sections

Key Result

Theorem 2.1

High-pass Filtering Property of Multi-stream Messaging. Considering the improved multi-stream message passing mechanism, suppose the graph signal $\bm{f} \in \mathbb{R}^N$ has a spectrum $\hat{\bm{f}} = \bm{U}^\top \bm{f}$ under the graph Fourier basis $\bm{U} = [\bm{u}_1, ..., \bm{u}_N]$, where $\b then there exist constants $\alpha > 0$ and $\kappa > 0$ such that the frequency response of the op

Figures (30)

  • Figure 1: Forecast results of extreme typhoons. (a) OneForecast's predicted wind speed for Typhoon Molva (2020) at 850 hPa pressure level with a 60-hour lead time. (b)–(c) the predicted cyclone tracks of Typhoon Yagi (2018) and Typhoon Molva (2020) using different models
  • Figure 2: Overview of Our OneForecast. (a) The overall architecture includes input variables, an encoder, a message passing module, a decoder, and visualization of forecast variables; (b) The global forecasts module uses rollout technology to generate future forecasts; (c) The neural nested grid method specializes in regional high-resolution weather forecasts tasks; and (d) The ensemble forecasting module generates long-term forecast results.
  • Figure 3: 10-day forecast results of different models.
  • Figure 4: We select the latitude-weighted RMSE (lower is better) and ACC (higher is better) of several variables.
  • Figure 5: Comparison results of 100-day forecasts between the two best models and our OneForecast.
  • ...and 25 more figures

Theorems & Definitions (3)

  • Theorem 2.1
  • Theorem 1.1: High-pass Filtering Property of Multi-stream Messaging
  • proof