Super-resolution (SR) is crucial for enhancing the spatial resolution of Earth System Model (ESM) data, thereby enabling more precise analysis of environmental processes. This paper introduces ViFOR, a novel SR algorithm integrating Vision Transformers (ViTs) with Fourier-based Implicit Neural Representation Networks (INRs). ViFOR effectively captures global context and high-frequency details essential for accurate SR reconstruction by embedding Fourier-based activation functions within the transformer architecture. Extensive experiments demonstrate that ViFOR consistently outperforms state-of-the-art methods, including ViT, SIREN, and SRGANs, in terms of Peak Signal-to-Noise Ratio (PSNR) and Mean Squared Error (MSE) for both global and local imagery. ViFOR achieves PSNR improvements of up to 4.18 dB, 1.56 dB, and 1.73 dB over ViT on full-image Source Temperature, Shortwave, and Longwave Flux datasets. These results highlight ViFOR's effectiveness and potential for advancing high-resolution climate data analysis.
View on arXiv@article{zeraatkar2025_2502.12427,
title={ ViFOR: A Fourier-Enhanced Vision Transformer for Multi-Image Super-Resolution in Earth System },
author={ Ehsan Zeraatkar and Salah A Faroughi and Jelena Tešić },
journal={arXiv preprint arXiv:2502.12427},
year={ 2025 }
}