Agentic reinforcement learning improves AI weather report accuracy
Zephyrus-R1: Agentic Reinforcement Learning for Weather Reporting
Machine Learning
Summary
It’s hard to train AI to write weather reports because there are many valid ways to describe the weather, making it tricky to check if the AI’s answers are right. The authors developed a system, Zephyrus-R1, that helps AI learn by rewarding it for matching expert weather reports and using evidence correctly. They trained their models to use tools for gathering weather data and to write clear, accurate reports. Their trained models scored much better at producing reliable weather forecasts, even beating some leading large models.
What this means in practice
- •For meteorological software developers: Build AI systems that generate reliable, evidence-based weather reports by training models with reward functions aligned to expert assessments.
- •For weather data platform teams: Integrate AI agents that use multiple data sources autonomously to produce clear weather forecasts with quantified uncertainty for decision support.
Authors
Fengfei Yu, Marshall Fisher, Isadora White, Sumanth Varambally, Yasaman Jafari, Shivam Yadav, Samuel M. Bartlett, Shawn Roj, Benjamin J. Moore, Michael D. Warner, Luyu Han, Zhirui Xia, Duncan Watson-Parris, Taylor Berg-Kirkpatrick, Yian Ma, Rose Yu
Abstract
Reinforcement learning (RL) has emerged as a key method for training AI models in scientific applications, but RL remains difficult when outputs cannot easily be verified against a known ground-truth answer. Natural-language weather forecast reports are one such domain: they translate atmospheric data into assessments of weather evolution, hazards, and uncertainty for both the general public and key decision-makers. Unlike numerical weather-analysis questions, they admit many valid outputs, which limits deterministic verification. To address this, we develop Zephyrus-R1, a practical framework for RL on scientific weather forecast reporting. We focus on three key components of the agentic RL pipeline: reward design, curriculum learning, and an analysis of human-expert alignment. To encourage both report legibility and factual accuracy, we design a reward function that combines a meteorological rubric with bidirectional claim entailment, both evaluated against National Weather Service reference reports. Our framework optimizes this reward in a simulated tool-calling environment, enabling vision-language agents to gather and analyze forecast evidence across regions and lead times before producing a report. We use annotations from three meteorologists to audit the claim labels produced by our LLM judges. Finally, a curriculum helps small models first develop tool use and scientific reasoning through RL on verifiable weather-analysis questions, then optimize the open-ended report reward. After training Qwen3.5-4B with Zephyrus-R1, we improve the mean across six judge-metric scores by 26% over the base model and achieve results competitive with frontier open-weight models. For Qwen3.8-27B, we improve the same mean by 13% over the base model, by 6.7% over Kimi-K3, and by 2.7% over GPT-5.6-Luna.