Generative 3D Gaussian Splatting for Arbitrary-ResolutionAtmospheric Downscaling and Forecasting

2026-04-09

AI Summary

This study addresses the limitations of existing AI-based weather forecasting methods, which struggle to efficiently generate atmospheric fields at arbitrary high resolutions due to insufficient multi-scale adaptability and inefficient data representation. To overcome these challenges, this work proposes a novel framework that models latitude–longitude grid points as 3D Gaussian centers and employs a generative mechanism to predict their covariance, attributes, and opacity. The approach integrates a scale-aware Vision Transformer to enable cross-scale information fusion and continuous resolution adaptation. By uniquely combining generative 3D Gaussian modeling with scale-aware attention, the method supports unified multi-scale atmospheric prediction and flexible downscaling. Evaluated on ERA5 and CMIP6 datasets, the model accurately forecasts 87 atmospheric variables and achieves significantly superior downscaling performance compared to current state-of-the-art methods.

Abstract

While AI-based numerical weather prediction (NWP) enables rapid forecasting, generating high-resolution outputs remains computationally demanding due to limited multi-scale adaptability and inefficient data representations. We propose the 3D Gaussian splatting-based scale-aware vision transformer (GSSA-ViT), a novel framework for arbitrary-resolution forecasting and flexible downscaling of high-dimensional atmospheric fields. Specifically, latitude-longitude grid points are treated as centers of 3D Gaussians. A generative 3D Gaussian prediction scheme is introduced to estimate key parameters, including covariance, attributes, and opacity, for unseen samples, improving generalization and mitigating overfitting. In addition, a scale-aware attention module is designed to capture cross-scale dependencies, enabling the model to effectively integrate information across varying downscaling ratios and support continuous resolution adaptation. To our knowledge, this is the first NWP approach that combines generative 3D Gaussian modeling with scale-aware attention for unified multi-scale prediction. Experiments on ERA5 show that the proposed method accurately forecasts 87 atmospheric variables at arbitrary resolutions, while evaluations on ERA5 and CMIP6 demonstrate its superior performance in downscaling tasks. The proposed framework provides an efficient and scalable solution for high-resolution, multi-scale atmospheric prediction and downscaling. Code is available at: https://github.com/binbin2xs/weather-GS.

Problem

• numerical weather prediction
• high-resolution forecasting
• atmospheric downscaling
• multi-scale adaptability
• data representation

Innovation

• 3D Gaussian Splatting
• Scale-Aware Attention
• Atmospheric Downscaling
• Generative Modeling
• Arbitrary-Resolution Forecasting