Spherical multigrid neural operator for improving autoregressive global weather forecasting

Abstract Data-driven approaches for global weather forecasting have shown great potential. However, conventional architectures of these models struggle with spherical distortions, leading to unstable autoregressive forecasts. Although methods such as spherical Fourier neural operator (SFNO) based on...

Full description

Saved in:
Bibliographic Details
Main Authors: Yifan Hu, Fukang Yin, Weimin Zhang, Kaijun Ren, Junqiang Song, Kefeng Deng, Di Zhang
Format: Article
Language:English
Published: Nature Portfolio 2025-04-01
Series:Scientific Reports
Subjects:
Online Access:https://doi.org/10.1038/s41598-025-96208-y
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1849735003535048704
author Yifan Hu
Fukang Yin
Weimin Zhang
Kaijun Ren
Junqiang Song
Kefeng Deng
Di Zhang
author_facet Yifan Hu
Fukang Yin
Weimin Zhang
Kaijun Ren
Junqiang Song
Kefeng Deng
Di Zhang
author_sort Yifan Hu
collection DOAJ
description Abstract Data-driven approaches for global weather forecasting have shown great potential. However, conventional architectures of these models struggle with spherical distortions, leading to unstable autoregressive forecasts. Although methods such as spherical Fourier neural operator (SFNO) based on spherical harmonic convolution can alleviate these problems, they face the challenge of high computational cost. Here, we introduce a spherical multigrid neural operator (SMgNO) that integrates spherical harmonic convolution and low resolution SFNO in the multigrid framework, effectively alleviating data distortions while requiring few computational resources. Experiments for spherical shallow water equations and medium-range global weather forecasting demonstrate the effectiveness and robustness of SMgNO. For 500 hPa geopotential height with a 7 days lead time, SMgNO achieves a 9.31% and 6.83% improvement in anomaly correlation coefficient over IFS T42 and SFNO, respectively. Furthermore, SMgNO requires only 10% floating-point operations of SFNO for forward propagation and 30.90% less GPU memory than SFNO during training.
format Article
id doaj-art-1089decfe28a437cb8083cc862a1b51d
institution DOAJ
issn 2045-2322
language English
publishDate 2025-04-01
publisher Nature Portfolio
record_format Article
series Scientific Reports
spelling doaj-art-1089decfe28a437cb8083cc862a1b51d2025-08-20T03:07:40ZengNature PortfolioScientific Reports2045-23222025-04-0115111210.1038/s41598-025-96208-ySpherical multigrid neural operator for improving autoregressive global weather forecastingYifan Hu0Fukang Yin1Weimin Zhang2Kaijun Ren3Junqiang Song4Kefeng Deng5Di Zhang6College of Computer Science and Technology, National University of Defense TechnologyCollege of Meteorology and Oceanography, National University of Defense TechnologyCollege of Meteorology and Oceanography, National University of Defense TechnologyCollege of Meteorology and Oceanography, National University of Defense TechnologyCollege of Meteorology and Oceanography, National University of Defense TechnologyCollege of Meteorology and Oceanography, National University of Defense TechnologyCollege of Meteorology and Oceanography, National University of Defense TechnologyAbstract Data-driven approaches for global weather forecasting have shown great potential. However, conventional architectures of these models struggle with spherical distortions, leading to unstable autoregressive forecasts. Although methods such as spherical Fourier neural operator (SFNO) based on spherical harmonic convolution can alleviate these problems, they face the challenge of high computational cost. Here, we introduce a spherical multigrid neural operator (SMgNO) that integrates spherical harmonic convolution and low resolution SFNO in the multigrid framework, effectively alleviating data distortions while requiring few computational resources. Experiments for spherical shallow water equations and medium-range global weather forecasting demonstrate the effectiveness and robustness of SMgNO. For 500 hPa geopotential height with a 7 days lead time, SMgNO achieves a 9.31% and 6.83% improvement in anomaly correlation coefficient over IFS T42 and SFNO, respectively. Furthermore, SMgNO requires only 10% floating-point operations of SFNO for forward propagation and 30.90% less GPU memory than SFNO during training.https://doi.org/10.1038/s41598-025-96208-ySpherical convolutionAutoregressive forecastsMultigrid neural operatorSpherical shallow water equationsGlobal weather forecasting
spellingShingle Yifan Hu
Fukang Yin
Weimin Zhang
Kaijun Ren
Junqiang Song
Kefeng Deng
Di Zhang
Spherical multigrid neural operator for improving autoregressive global weather forecasting
Scientific Reports
Spherical convolution
Autoregressive forecasts
Multigrid neural operator
Spherical shallow water equations
Global weather forecasting
title Spherical multigrid neural operator for improving autoregressive global weather forecasting
title_full Spherical multigrid neural operator for improving autoregressive global weather forecasting
title_fullStr Spherical multigrid neural operator for improving autoregressive global weather forecasting
title_full_unstemmed Spherical multigrid neural operator for improving autoregressive global weather forecasting
title_short Spherical multigrid neural operator for improving autoregressive global weather forecasting
title_sort spherical multigrid neural operator for improving autoregressive global weather forecasting
topic Spherical convolution
Autoregressive forecasts
Multigrid neural operator
Spherical shallow water equations
Global weather forecasting
url https://doi.org/10.1038/s41598-025-96208-y
work_keys_str_mv AT yifanhu sphericalmultigridneuraloperatorforimprovingautoregressiveglobalweatherforecasting
AT fukangyin sphericalmultigridneuraloperatorforimprovingautoregressiveglobalweatherforecasting
AT weiminzhang sphericalmultigridneuraloperatorforimprovingautoregressiveglobalweatherforecasting
AT kaijunren sphericalmultigridneuraloperatorforimprovingautoregressiveglobalweatherforecasting
AT junqiangsong sphericalmultigridneuraloperatorforimprovingautoregressiveglobalweatherforecasting
AT kefengdeng sphericalmultigridneuraloperatorforimprovingautoregressiveglobalweatherforecasting
AT dizhang sphericalmultigridneuraloperatorforimprovingautoregressiveglobalweatherforecasting