README.md

metadata

title: AFDBench
emoji: 🌦
colorFrom: blue
colorTo: green
sdk: gradio
sdk_version: 4.19.2
app_file: app.py
pinned: true
license: apache-2.0
python_version: 3.11
short_description: The Weather Forecast Discussion Alignment Benchmark

AFDBench: Area Forecast Discussion Benchmark

AFDBench evaluates how well AI models generate professional meteorological text compared to Human NWS Forecasters.

Core Metrics:

1. Met-Align: Physical accuracy vs. Human numerical choices.
2. Style-Align: Linguistic alignment with NWS AFD professional prose.

Initial results on 7,734 human samples reveal a massive Meteorological Hallucination Gap in zero-shot open models.