Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Storm-Centric 250 m NEXRAD Level-II Dataset for High-Resolution ML Nowcasting

Andy Shi

TL;DR

This work addresses the resolution bottleneck in ML-based radar nowcasting by introducing Storm250-L2, a storm-centric dataset that provides 250 m native-resolution reflectivity sequences derived from NEXRAD Level-II and GridRad-Severe. It crops around storm tracks to produce storm-centered, high-fidelity sequences in native polar coordinates, offering both native tilts and a derived pseudo-composite product, stored as ML-ready HDF5 tensors with comprehensive metadata and integrity manifests. The authors present a reproducible pipeline for storm tracking, radar matching, window optimization, and data serialization, along with a clear alpha-release scope and limitations. By enabling sub-kilometer radar inputs for nowcasting, Storm250-L2 aims to bridge high-resolution radar physics with data-driven forecasting, supporting more accurate predictions of high-impact convective events.

Abstract

Machine learning-based precipitation nowcasting relies on high-fidelity radar reflectivity sequences to model the short-term evolution of convective storms. However, the development of models capable of predicting extreme weather has been constrained by the coarse resolution (1-2 km) of existing public radar datasets, such as SEVIR, HKO-7, and GridRad-Severe, which smooth the fine-scale structures essential for accurate forecasting. To address this gap, we introduce Storm250-L2, a storm-centric radar dataset derived from NEXRAD Level-II and GridRad-Severe data. We algorithmically crop a fixed, high-resolution (250 m) window around GridRad-Severe storm tracks, preserve the native polar geometry, and provide temporally consistent sequences of both per-tilt sweeps and a pseudo-composite reflectivity product. The dataset comprises thousands of storm events across the continental United States, packaged in HDF5 tensors with rich context metadata and reproducible manifests.

A Storm-Centric 250 m NEXRAD Level-II Dataset for High-Resolution ML Nowcasting

TL;DR

This work addresses the resolution bottleneck in ML-based radar nowcasting by introducing Storm250-L2, a storm-centric dataset that provides 250 m native-resolution reflectivity sequences derived from NEXRAD Level-II and GridRad-Severe. It crops around storm tracks to produce storm-centered, high-fidelity sequences in native polar coordinates, offering both native tilts and a derived pseudo-composite product, stored as ML-ready HDF5 tensors with comprehensive metadata and integrity manifests. The authors present a reproducible pipeline for storm tracking, radar matching, window optimization, and data serialization, along with a clear alpha-release scope and limitations. By enabling sub-kilometer radar inputs for nowcasting, Storm250-L2 aims to bridge high-resolution radar physics with data-driven forecasting, supporting more accurate predictions of high-impact convective events.

Abstract

Machine learning-based precipitation nowcasting relies on high-fidelity radar reflectivity sequences to model the short-term evolution of convective storms. However, the development of models capable of predicting extreme weather has been constrained by the coarse resolution (1-2 km) of existing public radar datasets, such as SEVIR, HKO-7, and GridRad-Severe, which smooth the fine-scale structures essential for accurate forecasting. To address this gap, we introduce Storm250-L2, a storm-centric radar dataset derived from NEXRAD Level-II and GridRad-Severe data. We algorithmically crop a fixed, high-resolution (250 m) window around GridRad-Severe storm tracks, preserve the native polar geometry, and provide temporally consistent sequences of both per-tilt sweeps and a pseudo-composite reflectivity product. The dataset comprises thousands of storm events across the continental United States, packaged in HDF5 tensors with rich context metadata and reproducible manifests.
Paper Structure (14 sections, 4 figures)

This paper contains 14 sections, 4 figures.

Table of Contents

  1. Introduction
  2. Context and Motivation
  3. Current State of the Art in Radar Datasets
  4. Limitations and the Need for Higher Resolution
  5. This Work: The Storm250-L2 Dataset
  6. Contributions.
  7. Limitations (alpha release).
  8. Closing.
  9. Dataset Description
  10. Dataset Structure
  11. Event Selection & Composition
  12. Storm-Centric Processing & Cropping
  13. Data Representation & Integrity
  14. Known Limitations

Figures (4)

  • Figure 1: Comparison between coarse-resolution (2km) composite radar reflectivity and the 250m native-resolution data for the same convective storm.
  • Figure 2: Example of the storm-centric blob identification applied to a single convective event. The yellow X marks the storm center from the GR-S track, while gray outlines denote other candidate blobs that do not correspond to the GR-S location. The identified storm-centric blob is highlighted by the dashed-red outline.
  • Figure 3: Illustration of the window optimization procedure applied across all time steps of a storm event. Gray boxes represent the per-timestep storm bounds identified in the Storm Identification step. The dashed red rectangle denotes the final optimized window that minimally encloses all identified storm blobs across the storm’s life cycle.
  • Figure 4: Overview of the end-to-end process used to build the dataset.