A dataset for machine learning model to convective initiation detection and nowcasting over southeastern China

This dataset identifies severe convective weather events using surface meteorological observation data, determines convective initiation (CI) labels based on ground-based weather radar data, and extracts feature data from weather radar and satellite observations. All surface meteorological data, weather radar data, and satellite data are sourced from the operational databases of the National Meteorological Information Center, China Meteorological Administration. Radar features: The dataset provides ten radar features at 10-minute intervals, including: Composite Reflectivity (CR), Hybrid Scan Reflectivity (HBR), Constant Altitude Plan Position Indicator Reflectivity at 2-7 km (CAPPI 02-07), Echo Top (ET), and Vertical Integrated Liquid (VIL). The spatial resolution is 0.01°, and the temporal resolution is 10 minutes. These products are radar mosaic products, generated by merging quality-controlled single-radar products from all radars within and around the study area using specific algorithms. The characteristic value of radar features are -32768 and -1280. -32768 represents no data or is not within the observation range. -1280 represents minimum clear sky echo. To facilitate use in AI model training, all feature data are saved in NumPy '.npy' format, which is then compressed into '.bin' format during storage to reduce storage space and improve data download efficiency. Reflectivity data have undergone specific conversion processing。 Satellite features: Satellite features are derived from spectral channel observation data obtained by the Advanced Geosynchronous Radiation Imager (AGRI) onboard the Fengyun-4A (FY-4A) geostationary meteorological satellite of the China Meteorological Administration. Nine channels with wavelengths of 0.65μm, 1.61μm, 3.75μm, 6.25μm, 7.1μm, 8.5μm, 10.8μm, 12μm, and 13.5μm are selected. The channel data are temporally aligned with the 10-minute interval radar data, possessing the same 10-minute temporal resolution. Spatially, they cover the same area as the radar observations. The satellite channel data have undergone parallax correction. The spatial resolution of the visible channel (0.65 μm) is resampled to 0.005°; the two near-infrared channels (1.61, 3.75 μm) are resampled to 0.02°; and the other channel data are resampled to 0.04 °. During storage, 'NaN' values are replaced by -9, and all non-NaN values are scaled by a factor of 10,000. To facilitate use in AI model training, all feature data are saved in NumPy '.npy' format, which is then compressed into '.bin' format during storage to reduce storage space and improve data download efficiency. CI Labels ：Convective Initiation (CI) data include the coverage area (outer contours of all CI cells identified at that time), CI type (0 for declining, 1 for developing), and basic attributes of each CI cell. These attributes include area, aspect ratio of the cell contour, center point location, contour orientation angle, and the mean, maximum, and minimum radar reflectivity values within the CI cell. CI label data are stored in ASCII text format.CI Labels Details:For each severe convective weather event and at each 10-minute interval, the Convective Initiation areas extracted via the cell identification algorithm are labeled. Each map contains several CI areas. For each area, the following are recorded: the contour 'Contours', the contour 'Area', the aspect 'Ratio' of the contour (the ratio of the major axis to the minor axis of the ellipse fitted to the strong echo region), the center point location ('Cx', 'Cy'), the orientation 'Angle' of the ellipse contour (the angle between the major axis and the y-direction), and the mean 'Rmean', maximum 'RMax', and minimum 'RMin' reflectivity values within the region.The labels are represented by two types of files:a) Configuration Profile and Attribute File (CI)Each file records the positions and attributes of multiple CI entities identified on a radar image for a single sample. The attributes of each CI entity are described in three lines, with data separated by commas within each line.b) Convective Initiation Region Mask Data File (CImask)The CI range is represented by mask data:Value 1 indicates pixels with convective initiation that are developing.Value 2 indicates pixels with convective initiation that are declining.Value 0 indicates pixels without convective initiation.This label data records the extent and type of convective initiation observed by radar.Spatial resolution: 0.01° × 0.01°Temporal resolution: 10 minutes