Otter Weather: Skillful and Computationally Efficient Medium-Range Weather Forecasting

2026-06-24

AI Summary

This work addresses the high computational cost of state-of-the-art mesoscale AI weather forecasting models, which hinders their adoption and iteration by resource-constrained teams. The authors propose Otter Weather, a lightweight and efficient spatiotemporal forecasting architecture capable of both deterministic and probabilistic predictions under the ERA5 dataset and WeatherBench evaluation protocol. Otter substantially advances the skill–compute Pareto frontier: its base variant improves 24-hour forecast accuracy by 9.6% over the best numerical weather prediction (NWP) baseline while requiring fewer than 3.5 A100 GPU-days for training. The larger Otter-XL model surpasses the IFS ENS baseline by 9.7% in Continuous Ranked Probability Score (CRPS), achieving this with an order-of-magnitude lower computational cost than current leading models, and demonstrates strong generalization to other scientific computing tasks.

Abstract

State-of-the-art medium-range AI weather models can outperform traditional Numerical Weather Prediction (NWP) but require massive training budgets. This restricts usage for under-resourced groups and severely limits fast model iteration. Here we develop Otter Weather, a highly efficient spatiotemporal forecasting model designed to democratise high-performance weather prediction with AI. Evaluated on ERA5 reanalysis data at 1.5° resolution using standard WeatherBench protocols, the Otter family significantly advances the skill-compute Pareto frontier. The deterministic version outperforms the best NWP baseline by 9.6% at a 24-hour lead time while requiring fewer than 3.5 A100-days for training. It provides a 2x efficiency gain over lightweight AI models and a 100-fold reduction in compute compared to resource-intensive frontier architectures. We extend these efficiency gains into probabilistic forecasting by training via the Continuous Ranked Probability Score (CRPS). Scaling to a larger architecture, Otter-XL achieves a 9.7% CRPS improvement over the IFS ENS baseline. This yields an almost two-fold increase in predictive skill over comparable lightweight models at similar compute budgets. Otter-XL also outperforms frontier architectures like GenCast by over 2%, while using an order of magnitude less compute. Finally, Otter is applied out-of-the-box to a complex acoustic scattering PDE task where it outperforms a state-of-the-art foundation modelling approach, suggesting that the advances made here might apply across a range of scientific domains.

Problem

• medium-range weather forecasting
• computational efficiency
• AI weather models
• training budget
• democratization

Innovation

• efficient AI weather forecasting
• spatiotemporal modeling
• CRPS-based probabilistic forecasting
• compute-efficient deep learning
• scientific foundation models