This repository contains an early-stage prototype of an interactive application designed to explore and analyze extreme weather events, with a particular focus on hurricanes.
The core objective of the project is to build the foundations for systematic benchmarking of weather forecast models, by comparing:
- observed atmospheric fields derived from reanalysis data (ERA5),
- and forecasts produced by different modeling approaches, including traditional numerical weather prediction models and more recent AI-based models.
At this stage, the work is exploratory. Visual analysis is used as a first step to understand model behavior, identify meaningful variables, and highlight typical failure modes (intensity errors, structural biases, timing issues, etc.) before moving toward more formal quantitative benchmarks.
The long-term goal is to provide a robust framework in which multiple extreme events (hurricanes, heatwaves, atmospheric rivers, etc.) can be analyzed in a consistent way, eventually enabling reproducible and interpretable comparisons between forecasting systems.
The current implementation focuses on a first case study: Hurricane Dorian (2019). It includes visualizations of wind and pressure fields, simple intensity indicators such as the evolution of minimum surface pressure, and the initial building blocks required for future trajectory and skill-score analyses.
An early proof-of-concept version of the application is available here:
👉 https://weather-model-comparison.streamlit.app/
This demo represents only the first exploratory step of the project. Future iterations will progressively introduce additional events, forecast datasets, and quantitative evaluation metrics.