U-Cast: A Surprisingly Simple and Efficient Frontier Probabilistic AI Weather Forecaster

2026/04/10 by Salva Rühling Cachay, Duncan Watson-Parris, Rose Yu · 1 voice
Computer Science · Mathematics · Physics and Astronomy · #cs.AI #cs.LG #physics.ao-ph #stat.ML

paper · pdf · doi:10.48550/arxiv.2604.09041

arxiv published 2026/04/10 · arxiv updated 2026/05/29

Abstract

AI-based weather forecasting now rivals traditional physics-based ensembles, but state-of-the-art (SOTA) models rely on specialized architectures and massive computational budgets, creating a high barrier to entry. We demonstrate that such complexity is unnecessary for frontier performance. We introduce \ours, a probabilistic forecaster built on a standard U-Net backbone trained with a simple recipe: deterministic pre-training on Mean Absolute Error followed by short probabilistic fine-tuning on the Continuous Ranked Probability Score (CRPS) using Monte Carlo Dropout for stochasticity. As a result, our model matches or exceeds the probabilistic skill of GenCast and IFS ENS at 1.5^∘ resolution while reducing training compute by over 10× compared to leading CRPS-based models and inference latency by over 10× compared to diffusion-based models. U-Cast trains in under 12 H200 GPU-days and generates a 15-day ensemble forecast in 3 seconds. These results suggest that scalable, general-purpose architectures paired with efficient training curricula can match complex domain-specific designs at a fraction of the cost, opening the training of frontier probabilistic weather models to the broader community.

Citations

• Controllable Probabilistic Forecasting with Stochastic Decomposition Layers
• OmniCast: A Masked Latent Diffusion Model for Weather Forecasting Across Time Scales
• Swift: An Autoregressive Consistency Model for Efficient Weather Forecasting
• AERIS: Argonne Earth Systems Model for Reliable and Skillful Predictions
• FourCastNet 3: A geometric approach to probabilistic machine-learning weather forecasting at scale
• Elucidated Rolling Diffusion Models for Probabilistic Forecasting of Complex Dynamics
• Skillful joint probabilistic weather forecasting from marginals
• AIFS-CRPS: Ensemble forecasting using a model trained with a loss function based on the Continuous Ranked Probability Score
• ArchesWeather & ArchesWeatherGen: a deterministic and generative model for efficient ML weather forecasting
• Probabilistic weather forecasting with machine learning
• The Importance of Being Scalable: Improving the Speed and Accuracy of Neural Network Interatomic Potentials Across Chemical Domains
• Continuous Ensemble Weather Forecasting with Diffusion models
• A Deep Learning Earth System Model for Efficient Simulation of the Observed Climate
• What if? Numerical weather prediction at the crossroads
• Probabilistic Emulation of a Global Climate Model with Spherical DYffusion
• Probabilistic Weather Forecasting with Hierarchical Graph Neural Networks
• Scaling transformer neural networks for skillful and reliable medium-range weather forecasting
• Learning skillful medium-range global weather forecasting
• WeatherBench 2: A benchmark for the next generation of data-driven global weather models
• Accurate medium-range global weather forecasting with 3D neural networks
• FuXi: A cascade machine learning forecasting system for 15-day global weather forecast
• Scaling Spherical CNNs
• Idealized Baroclinic Wave Test Results
• Uncertainty Quantification over Graph with Conformalized Graph Neural Networks
• FourCastNet: Accelerating Global High-Resolution Weather Forecasting using Adaptive Fourier Neural Operators
• Elucidating the Design Space of Diffusion-Based Generative Models
• WeatherBench Probability: A benchmark dataset for probabilistic medium-range weather forecasting along with deep learning baseline models
• Forecasting Global Weather with Graph Neural Networks
• Diffusion Models Beat GANs on Image Synthesis
• The ERA5 global reanalysis
• Simple and Scalable Predictive Uncertainty Estimation using Deep Ensembles
• Dropout as a Bayesian Approximation: Representing Model Uncertainty in Deep Learning

Cited by

• Kastor: An efficient fine-tuning strategy for generative emulation of PDE simulations

Discussions

• *Fine-tuning takes ~1 day (1.2 H200-GPU-days) on a pre-trained deterministic backbone; add ~7 H200-days to train that backbone from scratch (~8 total). 📄 arxiv.org/abs/2604.09041 💻 github.com/Rose-S [bsky, 2 points, 1 comments]

Related

• FourCastNet 3: A geometric approach to probabilistic machine-learning weather forecasting at scale
• GenCast: Diffusion-based ensemble forecasting for medium-range weather
• ArchesWeather & ArchesWeatherGen: a deterministic and generative model for efficient ML weather forecasting
• SEEDS: Emulation of Weather Forecast Ensembles with Diffusion Models
• Probabilistic weather forecasting with machine learning