An approach using elastic graph dynamic link model for automating the satellite interpretation of tropical cyclone patterns

摘要

During the past decades, satellite interpretation has become a vital method for the determination of weather patterns especially for the identification of severe weather patterns such as tropical cyclones (TC). The recognition of TC is based on the Dvorak Technique which relies on subjective human interpretation. In this paper, an Elastic Graph Dynamic Link Model (EGDLM) is proposed to automate the satellite interpretation process and provide an objective analysis of tropical cyclone patterns. The method integrates a Dynamic Link Architecture (DLA) for neural dynamics and an Active Contour Model (ACM) for contour extraction of TC patterns. 145 tropical cyclones in the period between 1990 to 1998 were analysed by the system with an overall correct classification rate above 95%. For hurricanes with distinct eye formation, the model reported a deviation within 3 km from the actual eye location, which was obtained from the reconnaissance aircraft measurements of minimum surface pressure.