Python-based Prediction of Rapid Intensification from MIMIC-TC Ensemble (PRIME)

Abstract

Rapid intensification (RI), as defined by the National Hurricane Center (NHC), is an increase in the maximum sustained winds of a tropical cyclone (TC) of at least 30 knots (~34-35 mph) within a 24-hour period.  Intensity forecasting is one of the most difficult aspects of TC analysis forecasting, with RI prediction being one of the most challenging issues.  Predicting intensity and RI is critical for emergency responses, including evacuation and disaster prevention.  Deep learning (DL) and its application in TC analysis holds much potential.  Morphed Integrated Microwave Imagery at the Cooperative Institute for Meteorological Satellite Studies (MIMIC) is a product that synthesizes “morphed” images of TCs.  MIMIC-TC is a product that uses 85-92 GHz microwave imagery to create the images.  Using the Python programming language, a DL convolutional neural network (CNN) ensemble was developed as a proof-of-concept for prediction of RI, known as the Prediction of Rapid Intensification from MIMIC-TC Ensemble (PRIME).  Six members comprise PRIME, split into three 10 and 20 epoch models.  Each model has either 2, 3, or 4 convolutional layers.  A MIMIC-TC dataset was created using available North Atlantic Basin (NATL) storms from 2019 and 2020, and a total of 1508 images were used for training the models.  After running the Ensemble on all available storms from 2019 and 2020, it appeared all models were overfit, and subsequently gave inaccurate classifications.  The average percentage of correct classifications of “No RI” (nRI) was 30%, and the average percentage of correct classifications of “Possible RI” (pRI) was 27%.