UEPA-GAN: An Unsupervised Framework for Daily Precipitation Downscaling
文献类型：期刊论文
| 作者 | Mo, Jinlin1; Zhang, Xun1; Wang, Fang2; Liu, Yu1 |
| 刊名 | IEEE GEOSCIENCE AND REMOTE SENSING LETTERS
 |
| 出版日期 | 2026 |
| 卷号 | 23页码:1501605 |
| 关键词 | Precipitation Generators Image reconstruction Training Spatial resolution Climate Computer architecture Generative adversarial networks Data models Artificial intelligence Extreme precipitation attention (EPA) GAN physical constraint precipitation downscaling unsupervised |
| ISSN号 | 1545-598X |
| DOI | 10.1109/LGRS.2026.3657199 |
| 产权排序 | 2 |
| 文献子类 | Article |
| 英文摘要 | Precipitation downscaling aims to recover high-resolution (HR) precipitation fields from low-resolution (LR) inputs and is critical for representing precipitation intensity, spatial patterns, and regional extreme events. Although deep learning (DL)-based methods exhibit strong nonlinear modeling capability, supervised approaches are constrained by scarce paired HR data, while existing unsupervised methods often struggle to reconstruct extreme precipitation patterns and lack physical consistency. To address these challenges, we propose UEPA-GAN, an unsupervised cycle-consistent generative adversarial network for precipitation downscaling that combines a Uformer-based generator with an extreme precipitation attention (EPA) mechanism, enabling effective modeling of fine-scale spatial details and extreme precipitation features from unpaired data, while physical constraints are incorporated to preserve physical consistency. Experiments on the CLDAS dataset show that UEPA-GAN improves precipitation downscaling quality, achieving higher PSNR and SSIM and lower heavy rainfall reconstruction error (HRRE) compared with other unsupervised methods, demonstrating superior accuracy and stability in reconstructing extreme precipitation events. |
| URL标识 | 查看原文 |
| WOS研究方向 | Geochemistry & Geophysics ; Engineering ; Remote Sensing ; Imaging Science & Photographic Technology |
| 语种 | 英语 |
| WOS记录号 | WOS:001694322200004 |
| 出版者 | IEEE-INST ELECTRICAL ELECTRONICS ENGINEERS INC |
| 源URL | [http://ir.igsnrr.ac.cn/handle/311030/220926]  |
| 专题 | 陆地表层格局与模拟院重点实验室_外文论文 |
| 通讯作者 | Zhang, Xun |
| 作者单位 | 1.Beijing Technol & Business Univ, Sch Comp & Artificial Intelligence, Beijing 102401, Peoples R China; 2.Chinese Acad Sci, Inst Geog Sci & Nat Resources Res, Beijing 100101, Peoples R China |
| 推荐引用方式 GB/T 7714 | Mo, Jinlin,Zhang, Xun,Wang, Fang,et al. UEPA-GAN: An Unsupervised Framework for Daily Precipitation Downscaling[J]. IEEE GEOSCIENCE AND REMOTE SENSING LETTERS,2026,23:1501605. |
| APA | Mo, Jinlin,Zhang, Xun,Wang, Fang,&Liu, Yu.(2026).UEPA-GAN: An Unsupervised Framework for Daily Precipitation Downscaling.IEEE GEOSCIENCE AND REMOTE SENSING LETTERS,23,1501605. |
| MLA | Mo, Jinlin,et al."UEPA-GAN: An Unsupervised Framework for Daily Precipitation Downscaling".IEEE GEOSCIENCE AND REMOTE SENSING LETTERS 23(2026):1501605. |
入库方式： OAI收割
来源：地理科学与资源研究所
浏览0
下载0
收藏0
其他版本
除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。