← Back to all news
arXiv Machine Learning By Jiapei Liu, Ruidong Li, Ting Sun, Guangheng Ni

Tropical Cyclone Attributes in Sparse Autoencoder Features of an AI Weather Model

Read the original on arXiv Machine Learning →

The paper investigates how tropical cyclone attributes are represented within the sparse autoencoder (SAE) features of the AI weather model GraphCast. By analyzing 2,115 six‑hourly timesteps from 205 cyclones, the authors map SAE activations to cyclone position, intensity, and pressure structure across multiple layers and decompositions, finding that each attribute is encoded in multiple features and that deeper layers better predict future position and pressure structure. The study demonstrates that the organization of physical‑attribute relations in SAEs is reproducible and can aid in interpreting cyclone‑related information in GraphCast.

Machine-generated by The Flow from the publisher's headline and feed description — not written or checked by a human. The full article lives at arXiv Machine Learning.

Related stories

arXiv Machine Learning
Oct 16

Can AI Forecast the Extremes? Evaluating Probabilistic Weather Models on Tropical Cyclones

The study evaluates probabilistic AI weather models—GenCast, FourCastNet 3, and FGN—against the ECMWF IFS-ENS ensemble for tropical cyclone forecasting. Using new probabilistic metrics, the authors find that AI models, particularly FGN, outperform IFS-ENS in cyclone tracking, achieving better location scores for 90% of storms in 2025. The analysis also identifies intensity and spread limitations, model‑specific failure modes, and the influence of pretraining data, highlighting opportunities for targeted improvement.

By Nathan Chalumeau, Xavier Renard, Remi Meynadier, Hugo Rakotoarimanga, Marcin Detyniecki, Emmanuel de Bezenac
benchmarks
More like this →
Hugging Face Trending Papers
Sep 2

TC-Next: Zero-Shot Multimodal Cyclone Forecasting

TC-Next is a multimodal deep learning model that forecasts tropical cyclone track and intensity 6–24 hours ahead by combining a foundation model’s atmospheric forecast fields with GridSat infrared satellite imagery. Trained solely on GraphCast forecasts for the Western Pacific, it reduces track error by 15–44 % and intensity error by a factor of 3–6 compared to the rule‑based tracker TempestExtremes, and maintains superior performance when applied zero‑shot to other forecast systems such as Pangu‑Weather, IFS HRES, and WeatherNext Cyclones. Ablation studies confirm that incorporating the additional satellite modality consistently improves tracking accuracy at all lead times and enhances intensity predictions, especially at longer horizons.

multimodalbenchmarks
More like this →