Convective Cloud Detection and Tracking Using the New-Generation Geostationary Satellite Over South China

Abstract

As a core element of weather and climate change research, convective clouds have complex physical structures and dynamic processes, and knowledge of their evolution is limited by remote sensing data along with detection and tracking algorithms. In this study, we construct convective cloud detection and tracking algorithms for the Himawari-8 AHI data by combining machine learning model, area overlapping, and Kalman filter algorithms. First, we establish complicated and strict spatiotemporal data-matching conditions between the AHI and CloudSat CPR data, and a parallax correction model is developed to minimize the impact of parallax. To expand the sample volume of convective clouds, region growing algorithm is further used. Then, convective clouds during the day and night are detected based on the machine learning model which is trained through feature selection and hyper-parameter optimization. Finally, the automatic tracking of convective clouds, including those with relatively small scales or fast-moving speeds, is achieved by comprehensively using the area overlapping and Kalman filter algorithms. Validation results indicate that the algorithm developed can detect convective clouds of different scales with high accuracy, and the results show good continuity during day to night transitions. Moreover, the convective cloud detection algorithm can capture the convective clouds in advance compared with the traditional threshold algorithm. A case study of convective cloud demonstrates that the tracking algorithms we constructed can track convection at different scales. Overall, our study provides a new approach for convective cloud detection and tracking, which can contribute to better understanding of weather and climate change.

Topics & Concepts