Tropical Cyclone Detection and Tracking Using Spatial–Temporal Characteristics of Regional Wind Speed From CYGNSS: Proof of Concept and Assessment

Reliable data for global rapid coverage and short-period detection of tropical cyclones (TCs) are scarce. CYGNSS captures GNSS signals reflected from the Earth’s surface, delivering high-revisit global observations. This study presents a proof-of-concept evaluation of CYGNSS’s capability for global TC detection and tracking. A machine learning-based algorithm is introduced, which extracts wind speed features through regional segmentation. Using comparative analysis, statistics that effectively distinguish TC from non-TC samples were selected, while the spatiotemporal correlation of TC motion was incorporated for detection and tracking. This method addresses challenges including the irregular spatial sampling of CYGNSS data and the class imbalance between TC and non-TC samples. For daily time-slots, the method achieves a probability of detection (POD) of 82% with a false alarm rate (FAR) of 12% for tropical storm (TS) level TC detection. For hurricane-level TCs, the POD increases to 83% while the FAR decreases to 9%. The tracked TS-level tracks show a cosine similarity of 92.0% with IBTrACS reference paths. For semi-daily time-slots, the method attains a POD of 75% with a FAR of 8% for TS-level detection, while for hurricane-level TCs, the POD reaches 79% with a FAR reduced to 3%. The cosine similarity between tracked TS-level tracks and IBTrACS references reaches 93.1%. However, the limited quality of irregular point observations and the challenges in observing the fine-scale structure of TCs with CYGNSS data continue to constrain its capability for global TC detection with short revisit periods. Further refinement of the constellation’s design and antenna performance, integrated with complementary remote sensing systems, is essential to enhance the TC monitoring capability.