Rapid Extraction of Convective Clouds from High Spatiotemporal Resolution Advanced Geostationary Imager Measurements

Automatic recognition of convective clouds from geostationary satellite imagers is essential for enhancing situational awareness and severe storms nowcasting. A dynamic classification framework is developed for rapid extraction of convective clouds from high temporal and spatial resolution Advanced Geostationary Radiation Imager (AGRI) infrared (IR) band measurements. While existing cloud property retrievals (including cloud phase and cloud type discrimination) mostly rely on radiative transfer modeling with empirical thresholds, such static approaches demonstrate limited operational reliability across diverse meteorological regimes and synoptic conditions, and do not output convection areas directly. To address this gap, built on the cloud type products and AGRI IR radiance measurements, a regionally adaptive K-means dynamic clustering algorithm is implemented for seven regions (Northeast, North, East, Central, South, Northwest, and Southwest China) over the Chinese mainland, counting for spatial heterogeneity in cloud regime characteristics. While this methodology overcomes the limitation of using only thresholds in cloud type production through adaptation of K-means technique, it also avoids the lack of prior information in K-means technique by using cloud type product. Validation against ground-based radar data during mid-June and mid-July of 2023 reveals good agreement between convective clouds and high dBZ values, and further comparisons with rain-gauge measurements show high correlation between convective clouds and heavy rainfall. Implementation of this dynamic classification approach enables near real-time convective monitoring with 15-min refresh cycles for Fengyun-4, significantly enhancing nowcasting capabilities for severe weather events.