What question did this study set out to answer?

The study aims to create a dataset for machine learning models to effectively identify and forecast convective initiation (CI).

March 2, 2026Open Access

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

YLYujia Liu AXAnyuan Xiong NLNa Liu

Key Points

The study aims to create a dataset for machine learning models to effectively identify and forecast convective initiation (CI).
Developed a dataset named CIDS for AI CI detection and forecasting.
Included feature data from 10 radar mosaic products and FY-4A satellite across multiple spectral bands.
Analyzed intense convective weather events from March to September between 2018 and 2023.
Applied an algorithm on radar data to identify CI, storm cell locations, and assign CI category labels.
The dataset contains 136,728 samples of convective events.
Identified a total of 4,159,491 instances of CI, including 1,789,208 developing CIs.
CIs are tagged with enhancements in area and average echo intensity over a 30-minute period.

Abstract

Effective identification and forecasting of Convective Initiation (CI) are crucial for the early warning of severe convective weather. Artificial intelligence (AI) has significant potential in CI forecasting and warning. This study presents a dataset, named CIDS, designed for AI models to identify and forecast CI. The dataset covers southeastern China and samples intense convective weather events. These events include short-duration heavy precipitation, thunderstorm winds, and hail occurring from March to September between 2018 and 2023. It provides feature data and labels at 10-minute intervals. Feature data includes 10 radar mosaic products (spatial resolution: 0.01° × 0.01°) and FY-4A satellite radiance (or reflectance) across nine spectral bands (the spatial resolution of the visible channel is 0.005° × 0.005°, the shortwave infrared and mid-wave infrared channels are 0.02° × 0.02°, and the other channels are 0.04° × 0.04°). An algorithm based on radar composite reflectance factors identifies CI, determines if a storm cell is incipient, and marks its location. By analyzing the spatial extent and intensity evolution of CI over the next 30 minutes, the algorithm assigns CI category labels. The CIDS comprises 136,728 samples, identifying 4,159,491 CIs, including 1,789,208 developing CIs, which refer to CIs with enhancements of area and average echo intensity over the next thirty minutes.

Ask AI

Helpful

Bookmark

View Full Paper