Detection of Bow Echoes in French Kilometer-Scale Weather Models

Western Europe weather data for detecting bow echoes using convolutional neural networks. The dataset contains 9,174 simulated reflectivity fields and manually labeled ground truth contours from Météo-France's AROME-EPS and AROME models. It was created by Arnaud Mounier of the Centre National de la Recherche Scientifique for training and validating a U-Net model.

Use Cases

• Training convolutional neural networks for severe weather detection based on simulated reflectivity fields.
• Validating machine learning models for meteorological feature identification based on manually labeled ground truth.
• Applying trained U-Net architectures to deterministic weather model output for operational forecasting.
• Comparing model performance between ensemble (AROME-EPS) and deterministic (AROME) prediction systems.

Strengths

• Includes 6,206 training and 2,620 validation samples from the AROME-EPS model.
• Provides 348 samples from the deterministic AROME model for application testing.
• Contains pre-trained model weights (optimal_UNet_weights.h5) and architecture (optimal_UNet_architecture.json).
• Data covers a specific geographic grid of Western Europe (717x1121 points, 12°W-16°E and 37.5°N-55.4°N).

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Last update date is unknown; freshness unverified.

Provenance

Source

Météo-France operational weather models (AROME-EPS & AROME).

Collection Method

Simulated reflectivity fields from models with manually labeled ground truth contours.

Geography

Western Europe (12°W-16°E and 37.5°N-55.4°N).