Climate Physics Dynamic Matching
Abstract
Deep generative models such as flow matching and diffusion models have shown potential for learning complex dynamical systems, but typically act as black boxes that neglect underlying physical structure, while physics-based models governed by partial differential equations are often incomplete due to missing source terms, or uncertain parametrisations. We present Climate Physics Dynamic Matching (ClimPhyDM), a variational simulation-free dynamics informed framework for weather forecasting that combines an advection-type physics prior with data-driven components in a variational framework. % to capture the stochasticity and multi-modality of unresolved atmospheric dynamics. On the ERA5 benchmark at hourly (42-hour) and monthly (5-month) resolutions, ClimPhyDM outperforms ClimODE, and GB-DM, keeping the lower error at extended horizon, indicating improved temporal stability and resistance to error accumulation, while its simulation-free paradigm also enables training on a single modest 12 GB consumer GPU.
Related
- Physics-Guided Conditional Flow Matching with Energy Regularization for Robust PDE Inverse Problems
- Estimating dynamic models by matching random features
- Physics-Guided Flow Matching: A Generative AI Framework for Global Total Electron Content Forecasting
- Longest-Queue Policies in Dynamic Matching with Abandonments
- Dynamic Matching with Abandonment: Optimal Guarantees via Static Priority Policies
BibTeX
@misc{singh2026climate,
title = {{Climate Physics Dynamic Matching}},
author = {Gurjeet Sangra Singh and Frantzeska Lavda and Alexandros Kalousis},
howpublished = {arXiv},
year = {2026},
month = aug,
doi = {10.48550/arxiv.2608.26907},
eprint = {2608.26907},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2608.26907},
}
RIS
TY - UNPB
TI - Climate Physics Dynamic Matching
AU - Gurjeet Sangra Singh
AU - Frantzeska Lavda
AU - Alexandros Kalousis
PY - 2026
DA - 2026/08/27
PB - arXiv
DO - 10.48550/arxiv.2608.26907
UR - https://arxiv.org/abs/2608.26907
AB - Deep generative models such as flow matching and diffusion models have shown potential for learning complex dynamical systems, but typically act as black boxes that neglect underlying physical structure, while physics-based models governed by partial differential equations are often incomplete due to missing source terms, or uncertain parametrisations. We present Climate Physics Dynamic Matching (ClimPhyDM), a variational simulation-free dynamics informed framework for weather forecasting that combines an advection-type physics prior with data-driven components in a variational framework. % to capture the stochasticity and multi-modality of unresolved atmospheric dynamics. On the ERA5 benchmark at hourly (42-hour) and monthly (5-month) resolutions, ClimPhyDM outperforms ClimODE, and GB-DM, keeping the lower error at extended horizon, indicating improved temporal stability and resistance to error accumulation, while its simulation-free paradigm also enables training on a single modest 12 GB consumer GPU.
ER -