cs.LGNov 11, 2026

PhysXAI: A Theoretical Framework for Diagnosing Physical Hallucinations in Weather Foundation Models

Authors: Anjan Mahapatra, Nikhil Raj

Organizations: Department of Mechanical Engineering, IIITDM Kancheepuram, Chennai, Tamil Nadu, India · Department of Mechanical Engineering, Indian Institute of Technology Kanpur, Kanpur, Uttar Pradesh, India

Abstract

Modern machine learning models for global weather forecasting achieve remarkable predictive accuracy, yet they frequently produce states that violate fundamental physical laws, including conservation of mass, hydrostatic balance, or spectral energy scaling. We term these outputs physical hallucinations and propose PhysXAI, a rigorous diagnostic framework that detects, localises, and attributes such inconsistencies. Unlike existing post-hoc metric suites, PhysXAI is a learnable system that explicitly decouples forecast error from physical plausibility. We formalise the problem through a novel joint condition on numerical and physical residuals, and introduce a multi-branch encoder that extracts representations from conservation, dynamical, spectral, cross-variable, and spatiotemporal residuals. A key innovation is a physics-consistency contrastive learning objective, which we theoretically show the latent representation to be approximately isometric to the physical residual space, ensuring that diagnostic decisions remain grounded in physically meaningful deviations. We further derive an adaptive multi-objective loss with gradient-based weighting to balance competing diagnostic objectives. The framework produces spatially resolved hallucination maps, violation-type classifications, and calibrated Bayesian uncertainty, enabling both detection and interpretable attribution of physical inconsistencies. We conduct a comprehensive comparative evaluation of PhysXAI across ERA5, WeatherBench 2, and MERRA-2, using forecasts from GraphCast, Pangu-Weather, and NeuralGCM, with Python-written code and a fully reproducible experimental protocol. Our theoretical analysis and empirical evaluation establish PhysXAI as a principled framework for diagnosing physical hallucinations and advancing trustworthy AI for weather forecasting.

Figures & tables

Appendix figures & tables4 assets

Supplementary material from the paper’s appendix.

Appendix

Explore similar work

Jun 9, 2026

PhysMetrics.Weather: An Evaluation Framework for Physical Consistency in ML Weather Models

Machine learning weather prediction (MLWP) models have achieved impressive forecasting performance at a small fraction of the computational costs required for traditional physics-based methods. However, they are primarily (1) data-driven and (2) evaluated using pixel-wide error metrics (e.g., RMSE), so there are no guarantees that their forecasts are consistent with known physical laws. We introduce PhysMetrics..Weather, an evaluation framework that assesses the physical realism of MLWP models across three types of metrics: conservation, spectral, and dynamical. By quantifying physical realism, this tool guides the development of physics-informed architectures and helps evaluate whether MLWP models are reliable for operational use. Our framework is available on Github at https://github.com/Emmakast/PhysMetrics.Weather.
May 22, 2026

The physics of AI weather models

Could it be that AI weather models are solving physical equations, although they may not be the equations used by conventional NWP models? We compute correlations of forecast skill and Centered Kernel Alignment, providing evidence that different AI weather models represent the atmosphere in similar ways, despite differences in architecture and capacity. We argue that the architecture and training of the AI models constrains the form of the physical laws that they might simulate. In particular, we propose that the models implement a particle description of the atmosphere, where the latent variables at each mesh point correspond to the position of a particle in the high dimensional latent space. We hypothesize that the movement of the particles follows a gradient flow in the latent space towards a minimum of a learned free energy functional. Analysis of the GraphCast and Aurora models show that they make changes on large spatial scales in the early processor layers and move to smaller scale with increasing layer depth, consistent with the gradient flow hypothesis.
Aug 26, 2026

Missing the Butterfly and Predicting the Past: Features or Bugs of Accurate AI Weather Models?

AI weather prediction (AIWP) models rival physics-based models, yet the sources of their unexpected forecast accuracy and the degree of their physical fidelity remain unclear. Here, across a hierarchy spanning observation-based reanalysis, a general circulation model, and the multi-scale Lorenz system, we show that AI models can be trained to skillfully predict the past (backcast), though backcasts are systematically less accurate than forecasts. However, skillful backcasting appears to violate the second law of thermodynamics, and all these forecasting and backcasting models miss the butterfly effect. We trace the surprising forecast accuracy, missing butterfly, and skillful backcasting to a single cause: inevitable coarse-graining of training data, which removes fast, small scales and/or some variables. From the Lorenz system to official Pangu-Weather models, reducing coarse-graining makes AI predictions more physics-like (arrow of time and butterfly-like effects emerge), but forecast accuracy declines. Results offer an explanation for AIWP models' forecast skill: unlike physics-based models, they implicitly learn how fast, small scales affect large scales without inheriting their rapid error growth. Broader implications are that AI models' proliferation calls for revisiting predictability theories and long-term climate emulation strategies, and backcasting offers a useful, new lens for such analyses.