ICLR 2026 | 时空数据(Spatial-Temporal)论文总结[下]（物理时空和气象时空：气象预测，时空点过程等）

时空探索之旅

发布于 2026-03-10 16:07:24

1210

ICLR 2026将在2025年4月24日到28日于巴西里约热内卢（Rio de Janeiro, Brazil）举行。ICLR 2025共有19,000多篇投稿，录用5,359篇，录取率28.18%。本文总结了2026 ICLR上有关时间序列（time series）相关论文。如有疏漏，欢迎大家补充。

本文总结了ICLR 2026时空数据（Spatial-Temporal）的论文，总计36篇，本文涉及13篇，如有疏漏，欢迎补充。

注：由于论文数目较多，分为上下篇，基于数据生成机制与应用场景的本质差异对论文进行分类：

上篇主要涵盖交通时空数据以及城市科学等内容，包括交通预测，人群移动，轨迹挖掘，交通模拟，自动驾驶，信号控制等内容。
下篇主要涵盖：气象时空和物理时空，如气象预测，时空点过程，时空动力系统等

观察：下篇文章统计值

最大均分	均值	最小均分
7	5.31	4

其中均分≥6的有3篇。

1. OmniField: Conditioned Neural Fields for Robust Multimodal Spatiotemporal Learning2. DGNet: Learning Spatiotemporal PDEs with Discrete Green Networks3. Enabling arbitrary inference in spatio-temporal dynamic systems: A physics-inspired perspective4. TEN-DM: Topology-Enhanced Diffusion Model for Spatio-Temporal Event Prediction5. Zephyrus: An Agentic Framework for Weather Science6. ARROW: An Adaptive Rollout and Routing Method for Global Weather Forecasting7. Omni-Weather: Unified Multimodal Foundation Model for Weather Generation and Understanding8. TianQuan-S2S: A Subseasonal-to-Seasonal Global Weather Model via Incorporate Climatology State9. Extreme Weather Nowcasting via Local Precipitation Pattern Prediction10. DeepPrim: a Physics-Driven 3D Short-term Weather Forecaster via Primitive Equation Learning11. Task-Adaptive Parameter-Efficient Fine-Tuning for Weather Foundation Models12. STORM: Synergistic Cross-Scale Spatio-Temporal Modeling for Weather Forecasting13. GeoFAR: Geography-Informed Frequency-Aware Super-Resolution for Climate Data

1 OmniField: Conditioned Neural Fields for Robust Multimodal Spatiotemporal Learning

链接：https://openreview.net/forum?id=VpWDZ3yTBn

关键词：Conditioned Neural Fields, Multimodal Learning, Spatiotemporal Learning, Scientific Data, Neural Fields

作者：Kevin Valencia, Thilina Balasooriya, Xihaier Luo, Shinjae Yoo, David Park

分数：6, 6, 4

信心：2, 3, 3

均分：5.3333333

TL; DR：Spatiotemporal scientific data are inherently multimodal yet sparse, noisy, and irregular; we introduce OmniField, a multimodal conditioned neural field for unified robust spatiotemporal representation learning.

2 DGNet: Learning Spatiotemporal PDEs with Discrete Green Networks

链接：https://openreview.net/forum?id=EJ8HnNTEAv

关键词：Partial Differential Equations, Graph Neural Networks, Physics-Informed Machine Learning, Generalization

作者：Yingjie Tan, Quanming Yao, Yaqing Wang

分数：6, 4, 6, 4

信心：4, 4, 4, 4

均分：5.0

3 Enabling arbitrary inference in spatio-temporal dynamic systems: A physics-inspired perspective

链接：https://openreview.net/forum?id=b6Py2zy0fK

关键词：Neural operators, Spatio-temporal systems, Graph neural networks, Data mining

作者：Yan Ge, Zhengyang Zhou, Qihe Huang, Yuxuan Liang, Yang Wang

分数：6, 4, 6

信心：3, 3, 4

均分：5.333333333333333

4 TEN-DM: Topology-Enhanced Diffusion Model for Spatio-Temporal Event Prediction

链接：https://openreview.net/forum?id=BZ1vutP53o

关键词：Spatio-temporal point process, Diffusion model, Topological data analysis

作者：Yuxin Liu, Kaiming Wang, Chenguang Yang, Yulia Gel, Yuzhou Chen

分数：6, 2, 4

信心：3, 4, 4

均分：4.0

5 Zephyrus: An Agentic Framework for Weather Science

链接：https://openreview.net/forum?id=aVeaNahsID

关键词：Agents, Large Language Models, Weather Science, Code Generation

作者：Sumanth Varambally, Marshall Fisher, Jas Thakker, Yiwei Chen, Zhirui Xia, Yasaman Jafari, Ruijia Niu, Manas Jain, Veeramakali Vignesh Manivannan, Zachary Novack, Luyu Han, Srikar Eranky, Salva Rühling Cachay, Taylor Berg-Kirkpatrick, Duncan Watson-Parris, Yian Ma, Rose Yu

分数：6, 8, 6, 8

信心：3, 3, 3, 4

均分：7.0

TL; DR：We built an AI weather assistant that lets scientists explore meteorological data through natural conversation, and created a benchmark to evaluate LLMs for weather science.

6 ARROW: An Adaptive Rollout and Routing Method for Global Weather Forecasting

链接：https://openreview.net/forum?id=Qs0BieWYEN

关键词：Deep Learning; Spatiotemporal Analysis; Weather Forecasting

作者：Jindong Tian, Yifei Ding, Ronghui Xu, Hao Miao, Chenjuan Guo, Bin Yang

分数：8, 4, 6

信心：4, 4, 3

均分：6.0

7 Omni-Weather: Unified Multimodal Foundation Model for Weather Generation and Understanding

链接：https://openreview.net/forum?id=3WnXsp72v6

关键词：AI for Science, Unified foundation model, Interpretable reasoning

作者：Zhiwang Zhou, Yuandong Pu, Xuming He, Yidi Liu, Yixin Chen, Junchao Gong, Xiang Zhuang, Wanghan Xu, Qinglong Cao, SHIXIANG TANG, Yihao Liu, Wenlong Zhang, LEI BAI

分数：8, 4, 6, 6

信心：3, 4, 3, 3

均分：6.0

8 TianQuan-S2S: A Subseasonal-to-Seasonal Global Weather Model via Incorporate Climatology State

链接：https://openreview.net/forum?id=7Dvmq7MhwU

关键词：Subseasonal Weather Forecasting

作者：Guowen Li, Xintong Liu, Yang Liu, Mengxuan Chen, Shilei Cao, Xuehe Wang, Juepeng Zheng, Jinxiao Zhang, Haoyuan Liang, Lixian Zhang, Jiuke Wang, Meng Jin, Hong Cheng, Haohuan Fu

分数：6, 4, 6

信心：5, 3, 4

均分：5.333333333333333

TL; DR：This paper introduces TianQuan-S2S, a novel machine learning model that provides accurate global daily mean forecasts up to 45 days by integrating climatology state information.

9 Extreme Weather Nowcasting via Local Precipitation Pattern Prediction

链接：https://openreview.net/forum?id=fDknsQhSgm

关键词：Spatiotemporal Forecasting, Extreme Heavy Rain, Video Transformer

作者：Chang hoon Song, Teng Yuan Chang, Youngjoon Hong

分数：4, 4, 6, 6

信心：4, 4, 4, 3

均分：5.0

10 DeepPrim: a Physics-Driven 3D Short-term Weather Forecaster via Primitive Equation Learning

链接：https://openreview.net/forum?id=EyyWd0hH0q

关键词：Weather forecasting, Physics-informed neural networks, Primitive equations, Earth atmospheric dynamics, Deep learning.

作者：Jiawei Chen, Weiqi Chen, Rong Hu, Peiyuan Liu, Haifan Zhang, Liang Sun

分数：6, 6, 2, 6

信心：4, 4, 4, 3

均分：5.0

TL; DR：We propose DeepPrim, a physics-informed 3D deep weather forecaster designed to learn primitive equations of the Earth’s atmospheric dynamics.

11 Task-Adaptive Parameter-Efficient Fine-Tuning for Weather Foundation Models

链接：https://openreview.net/forum?id=eFExhM3tKr

关键词：Weather Foundation Model, Parameter-Efficient Fine-Tuning, Earth Science

作者：Shilei Cao, Hehai Lin, Jiashun Cheng, Yang Liu, Guowen Li, Xuehe Wang, Juepeng Zheng, Haoyuan Liang, Meng Jin, Chengwei Qin, Hong Cheng, Haohuan Fu

分数：0, 4, 8, 6

信心：5, 4, 3, 4

均分：4.5

TL; DR：This paper introduces WeatherPEFT, a new, more efficient fine-tuning method that performs as well as full training but with fewer resources by using task-specific adjustments and focusing on the most critical parameters.

12 STORM: Synergistic Cross-Scale Spatio-Temporal Modeling for Weather Forecasting

链接：https://openreview.net/forum?id=JLF6XDnscF

关键词：spatial-temporal forecasting

作者：Qihe Huang, Zhengyang Zhou, Yangze Li, Jiaming Ma, Kuo Yang, Binwu Wang, Xu Wang, Yang Wang

分数：4, 4, 8, 4

信心：5, 4, 5, 4

均分：5.0

13 GeoFAR: Geography-Informed Frequency-Aware Super-Resolution for Climate Data

链接：https://openreview.net/forum?id=0WHpOekph0

关键词：climate downscaling, image super-resolution, implicit neural representation, earth observation, environmental science

作者：Chang Xu, Gencer Sumbul, Li Mi, Robin Zbinden, Devis Tuia

分数：4, 4, 8, 6

信心：4, 4, 3, 3

均分：5.5