DeepTornado: A Tornado Radar Dataset over China

DeepTornado is a tornado radar dataset constructed from operational S-band Doppler weather radar observations of the China New Generation Weather Radar (CINRAD) network, with reference to confirmed tornado cases over China during the period 2017–2024. The dataset will be updated as more confirmed tornado cases become available.The dataset consists of radar data patches extracted from low-elevation scans and includes three radar variables: reflectivity factor, radial velocity, and spectrum width, each provided at the lowest three elevation angles. Each data patch is centered on its labeled position (if available). All radar data are interpolated to a Cartesian grid with a spatial resolution of 250 m, and each sample is stored as an individual compressed NumPy (.npz) file.An empirical labeling scheme is adopted that jointly considers radar signatures and confirmed tornado reports. Samples are categorized into four classes: confirmed supercell tornadoes (TOR), warning tornado-like mesocyclones without confirmed tornado reports (WRN), confirmed tornadoes with weak or ambiguous radar signatures (WEK), and non-tornadic background scenes (NUL). The dataset contains 2,064 labeled samples derived from 143 confirmed tornado cases, along with 5,314 NUL samples representing diverse non-tornadic conditions.DeepTornado is intended to support statistical analyses of tornado-related radar features, data-driven and machine learning–based methodological studies, and related research on severe convective storms under realistic observational constraints. Users are encouraged to consider the inherent limitations of weather radar observations and to construct additional background samples as needed for specific applications.