The China Meteorological Administration (CMA) recently unveiled its AI meteorological service system, Fenghe (literally "Wind Harmony"), aiming to revolutionize meteorological service models and achieve intelligent transformation in weather forecasting and public services.

Currently, Fenghe comprises five core modules: a meteorological knowledge center, a model square, a meteorological AI toolbox, an agent factory and an evaluation lab.

Its key capabilities include understanding meteorological service demands, generating tailored content, performing weather-related reasoning and decision-making, and calling specialized meteorological tools.

By accessing Fenghe via a mini-program or web platform, users can input weather-related queries and receive automated responses. Covering scenarios from science popularization Q&A and weather inquiries to disaster warnings and industry-specific advice, Fenghe provides personalized recommendations based on intelligent analysis.

For instance, for a family driving from Shanghai to Baiyangdian Lake during a thunderstorm, it can suggest alternative indoor attractions while avoiding waterlogged highway sections during peak storm hours.

Fenghe will undergo regular evaluations by an expert team to ensure its competitiveness as a vertical domain model.

Behind Fenghe's launch are breakthroughs in core technologies, such as a trillion-scale meteorological corpus for training, LoRA-based fine-tuning to enhance domain-specific knowledge, reinforcement learning from human feedback for advanced reasoning, and multi-agent collaboration tailored for meteorological service scenarios, ensuring stable performance and user experience.