TAASRAD19

TAASRAD19 provides a high-resolution radar reflectivity dataset for precipitation nowcasting and benchmarking of analog ensemble and machine learning models.

Key Features:

• Spatial resolution: Horizontal resolution of 500 meters covering an area with a diameter of approximately 240 kilometers.
• Temporal sampling: 5-minute sampling interval across the dataset.
• Temporal span and size: Dataset spans over nine years and comprises 894,916 timesteps of precipitation data.
• Data format: Reflectivity provided as 2D images representing the maximum reflectivity on vertical sections.
• Geographic source: Collected by the Civil Protection weather radar of the Trentino South Tyrol Region (Italian Alps).
• Curated subset: A curated subset of 1,732 sequences totaling 362,233 radar images is included.
• Annotations: Images annotated with precipitation type tags assigned by expert meteorologists.

Scientific Applications:

• Nowcasting model development: Training and testing of machine learning and deep learning models for short-term precipitation forecasting.
• Analog ensemble research: Development and evaluation of analog ensemble methods for precipitation prediction.
• Benchmarking: Standardized benchmarking of novel spatiotemporal forecasting algorithms and nowcasting methods.
• Supervised classification: Supervised learning for precipitation-type classification using expert meteorologist annotations.

Methodology:

The dataset includes 2D maximum-reflectivity images, a curated subset of 1,732 sequences (362,233 images), expert-assigned precipitation type tags, validation using a TrajGRU forecasting model, interactive exploration and dimensionality reduction with UMAP, and accompanying software tools for data pre-processing, model training, and inference.