Physics-Assisted and Topology-Informed Deep Learning for Weather Prediction

Authors: Jiaqi Zheng, Qing Ling, Yerong Feng

IJCAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We conduct experiments on the 5.625degree resolution ERA5 data set, demonstrating the competitive performance of PASSAT compared to the state-of-the-art deep learning models and the NWP model IFS T42. We conduct ablation studies to evaluate the effectiveness of the physics and topology information used in PASSAT.
Researcher Affiliation Academia Jiaqi Zheng1 , Qing Ling1 and Yerong Feng2 1Sun Yat-Sen University 2Shenzhen Institute of Meteorological Innovation EMAIL, EMAIL, yerong EMAIL. Sun Yat-Sen University is an academic institution. Shenzhen Institute of Meteorological Innovation is a public research institution. All authors are from academic or public research institutions.
Pseudocode Yes Algorithm 1: PASSAT: Predicting any weather variable u for ̄ = t + 0.2, t + 0.4, , t + t at time t
Open Source Code Yes We release an open-source Pytorch implementation of PASSAT online1. 1https://github.com/Yumenomae/PASSAT5p625
Open Datasets Yes We conduct the experiments on the European Centre for Medium-Range Weather Forecasts Reanalysis V5 (ERA5) 5.625 -resolution data set, spanning from 1979 to 2018 and provided by Weather Bench [Hersbach et al., 2020; Rasp et al., 2020].
Dataset Splits Yes The data samples from 1979 to 2015 are used in the training set, 2016 in the validation set, as well as 2017 and 2018 in the test set.
Hardware Specification No The paper mentions "NVIDIA s Modulus" in relation to training baseline models but does not specify the hardware (e.g., GPU models, CPU types) used by the authors for their own experiments.
Software Dependencies No The paper mentions a "Pytorch implementation" and "NVIDIA s Modulus" but does not specify version numbers for PyTorch or any other software dependencies.
Experiment Setup Yes In PASSAT, we set the integration step size as 0.2 hours. We use PASSAT and the baseline models to predict these weather variables, at a temporal resolution of 6 hours (6am, 12am, 6pm, and 12pm of each day) and lasting for 24 steps (144 hours).