Tropical Cyclone Intensity Estimation Using Deep Convolutional Neural Networks

Estimating tropical cyclone intensity using only satellite imagery is challenging. For over 30 years, the Dvorak technique has been successfully applied, with modifications and improvements, to solve this problem, and experts worldwide continue to use it for tropical cyclone intensity estimation. A number of related techniques have been derived using the original Dvorak technique, some of which are automated. However, a common problem is that independent Dvorak analysts and/or automated implementations of the Dvorak technique can arrive at very different intensity estimates even when using the same satellite imagery. As a recent example, the 15 UTC 10 October 2017 National Hurricane Center discussion for Tropical Storm Ophelia noted that the Dvorak intensity estimates ranged from T2.3/33 kt (by UW-CIMSS) to T3.0/45 kt (by TAFB) to T4.0/65 kt (by SAB). In this particular case, human experts at TAFB and SAB differed by 20 kts in their Dvorak analyses, and the automated version at the University of Wisconsin was 12 kt lower than either of them. The National Hurricane Center (NHC) estimates about 10-20% uncertainty in its post analysis when only satellite-based estimates are available.

The success of the Dvorak-based techniques over time proves that spatial patterns in infrared (IR) imagery strongly relate to tropical cyclone intensity, but subjectivity in its application adds uncertainty to the estimates. This study aims to utilize deep learning, the current state of the art in pattern recognition and image recognition, to address the need for an automated and objective tropical cyclone intensity estimation. Deep learning describes a multi-layer neural network of simple computational units that learns discriminative features without relying on a human expert to identify which features are important. Our study mainly focuses on convolutional neural network (CNN), a deep learning algorithm, to develop an objective tropical cyclone intensity estimation. CNN is a supervised learning algorithm that requires a large number of training data. Since the archives of intensity data and tropical cyclone-centric satellite images are openly available for use, the training data are easily created by combining the two. Results, case studies, prototypes, and advantages of this approach will be discussed.