A physics-driven hybrid transformer model for hydrologic simulation under nonstationary environmental conditions
文献类型：期刊论文
| 作者 | Zhang, Hongwei1,2; Tao, Zexing1; Wang, Kaiwen1; Zhao, Gang1; Chen, Jiewei1; Xu, Duanyang1; Liu, Ronggao1; Wang, Longhao1; Wang, Lei1; Ge, Quansheng1,2 |
| 刊名 | JOURNAL OF HYDROLOGY
 |
| 出版日期 | 2026-04-01 |
| 卷号 | 669页码:135133 |
| 关键词 | Time-varying parameters Nonstationary hydrologic processes Physics-driven deep learning |
| ISSN号 | 0022-1694 |
| DOI | 10.1016/j.jhydrol.2026.135133 |
| 产权排序 | 1 |
| 文献子类 | Article |
| 英文摘要 | Conventional hydrological models with static parameters often struggle to represent the nonstationary behavior of hydrological processes driven by climate change and human activities. To address this challenge, we propose a Physics-Driven Hybrid Transformer Hydrologic Model (PD-HTHM), which integrates an LSTM-Transformer hybrid encoder with the conceptual SIMHYD model within an end-to-end differentiable framework. The model is physics-driven, using the hybrid encoder to generate time-varying parameters that dynamically adjust key hydrological processes. To systematically evaluate model performance, eight process-oriented dynamic parameterization strategies were designed, covering surface, soil, and groundwater processes, and tested across 71 catchments in the River Severn Basin (UK). Results show that all dynamic parameterization strategies outperform the static SIMHYD model on common evaluation metrics, with the configuration jointly adjusting the soil moisture storage capacity (SMSC) and the baseflow recession coefficient (K) achieving the best overall performance. However, the optimal parameter combinations vary among catchments, indicating pronounced spatial heterogeneity. Further analysis reveals that, rather than simply increasing the number of variable parameters, dynamically adjusting a few key parameters associated with dominant hydrological processes is more effective, improving model robustness and predictive accuracy under nonstationary conditions. Overall, PD-HTHM demonstrates the potential of integrating deep learning with process-based modeling to improve the representation of nonstationary hydrological dynamics. |
| URL标识 | 查看原文 |
| WOS关键词 | PARAMETERS ; ASSIMILATION ; CATCHMENTS ; MEMORY ; QUANTIFICATION ; PREDICTIONS ; PERFORMANCE ; STREAMFLOW ; ENSEMBLE ; ERROR |
| WOS研究方向 | Engineering ; Geology ; Water Resources |
| 语种 | 英语 |
| WOS记录号 | WOS:001694733400001 |
| 出版者 | ELSEVIER |
| 源URL | [http://ir.igsnrr.ac.cn/handle/311030/220950]  |
| 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 | Tao, Zexing; Ge, Quansheng |
| 作者单位 | 1.Chinese Acad Sci, Inst Geog Sci & Nat Resources Res, Beijing 100101, Peoples R China; 2.Univ Chinese Acad Sci, Beijing 100049, Peoples R China |
| 推荐引用方式 GB/T 7714 | Zhang, Hongwei,Tao, Zexing,Wang, Kaiwen,et al. A physics-driven hybrid transformer model for hydrologic simulation under nonstationary environmental conditions[J]. JOURNAL OF HYDROLOGY,2026,669:135133. |
| APA | Zhang, Hongwei.,Tao, Zexing.,Wang, Kaiwen.,Zhao, Gang.,Chen, Jiewei.,...&Ge, Quansheng.(2026).A physics-driven hybrid transformer model for hydrologic simulation under nonstationary environmental conditions.JOURNAL OF HYDROLOGY,669,135133. |
| MLA | Zhang, Hongwei,et al."A physics-driven hybrid transformer model for hydrologic simulation under nonstationary environmental conditions".JOURNAL OF HYDROLOGY 669(2026):135133. |
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。