Deep learning approach for detecting tropical cyclones and their precursors in the simulation by a cloud-resolving global nonhydrostatic atmospheric model

Daisuke Matsuoka,Seiichi Uchida,Daisuke Sugiyama,Masuo Nakano

doi:10.1186/s40645-018-0245-y

Abstract

We propose a deep learning approach for identifying tropical cyclones (TCs) and their precursors. Twenty year simulated outgoing longwave radiation (OLR) calculated using a cloud-resolving global atmospheric simulation is used for training two-dimensional deep convolutional neural networks (CNNs). The CNNs are trained with 50,000 TCs and their precursors and 500,000 non-TC data for binary classification. Ensemble CNN classifiers are applied to 10 year independent global OLR data for detecting precursors and TCs. The performance of the CNNs is investigated for various basins, seasons, and lead times. The CNN model successfully detects TCs and their precursors in the western North Pacific in the period from July to November with a probability of detection (POD) of 79.9–89.1% and a false alarm ratio (FAR) of 32.8–53.4%. Detection results include 91.2%, 77.8%, and 74.8% of precursors 2, 5, and 7 days before their formation, respectively, in the western North Pacific. Furthermore, although the detection performance is correlated with the amount of training data and TC lifetimes, it is possible to achieve high detectability with a POD exceeding 70% and a FAR below 50% during TC season for several ocean basins, such as the North Atlantic, with a limited sample size and short lifetime.

Highlights

Tropical cyclones (TCs), referred to as typhoons, cyclones, and hurricanes, cause significant damage to human life, agriculture, forestry, fisheries, and infrastructure
Detection results This section first introduces one of the best cases of detection results under the condition that the number of TCs and precursors is larger than eight and probability of detection (POD) is larger than 80.0%
In this work, the detectability of TCs and their precursors for each basin, season, and lead time was investigated based on a deep neural network approach using 20 year simulated outgoing longwave radiation (OLR) by the Nonhydrostatic Icosahedral Atmospheric Model (NICAM)

Summary

Introduction

Tropical cyclones (TCs), referred to as typhoons, cyclones, and hurricanes, cause significant damage to human life, agriculture, forestry, fisheries, and infrastructure. We performed a CNN-based binary classification that categorizes 2D cloud data (OLR) into “developing TCs and their precursors” or “non-developing depressions.”. At first glance, it appears natural to categorize the data into the following three classes: TCs, precursors, and non-developing depressions Since they are defined by the threshold value of maximum wind speed, it is difficult to identify them from cloud images. In order to reduce the number of candidate regions, we set a limit to the cloud cover in the range of 30.0 to 95.0% and 50% or more over sea areas In this manner, 97.0% of precursors of TCs in the simulation data could be covered. We adopt binary classification to facilitate the evaluation of prediction results, we can output detection results as probabilistic information using p

Evaluation metrics of prediction results

Results and discussion

Conclusions