Detecting Tropical Cyclone Secondary Eyewalls with a Microwave-Based Scheme
Open Access
- Author:
- Cheung, Alex
- Area of Honors:
- Meteorology
- Degree:
- Bachelor of Science
- Document Type:
- Thesis
- Thesis Supervisors:
- Steven J Greybush, Thesis Supervisor
Raymond Gabriel Najjar, Jr., Thesis Honors Advisor
Jenni Evans, Thesis Supervisor - Keywords:
- Tropical Cyclones
Secondary Eyewall Formation
Eyewall Replacement Cycle
Meteorology
Tropical Meteorology
Machine Learning
Detection Algorithm
Hurricanes
Satellite
Remote Sensing - Abstract:
- Intense tropical cyclones (TCs) often form secondary eyewalls, triggering a process known as an eyewall replacement cycle (ERC). This can lead to short-term fluctuations in intensity and an increase in the size of the TC wind field. When occurring near landfall, the short-term variations can dramatically alter coastal watch, warning, and storm surge forecasts, potentially altering pre-storm preparation plans, including evacuations. However, documenting these events can be a time-consuming, subjective, and sometimes difficult task. Here, we use 89 –92 GHz microwave imagery from the NOAA Cooperative Institute for Research in the Atmosphere’s Tropical Cyclone PRecipitation, Infrared, Microwave, and Environmental Dataset (TC PRIMED) to develop image-based variables to identify concentric structures related to deep convection. The image-based variables are combined with various environmental and storm variables (e.g, deep-layer shear magnitude, current maximum wind speed, 24-h difference in radius of 5 kt (1 kt = 0.514 m s–1) winds, and 24-h difference in infrared brightness temperature), to create a probabilistic secondary eyewall classification scheme using a machine learning classifier (linear discriminant analysis). This classification scheme is trained and tested using subjectively created secondary eyewall labels (2016–2019) of storms from the North Atlantic, East Pacific, West Pacific, and Southern Hemisphere basins. We trained the classifier using 36 storms and retained 16 storms for testing. From the classifier output, we calculate the probability of detection, false alarm ratios, skill scores, and bias ratio for various probability thresholds. Using the best default probability threshold (50%), the model produced a secondary eyewall probability of detection of about 64% with a false alarm ratio of 34% and a Peirce’s Skill Score of 0.52, indicating fair skill in the model.