Tropical Cyclone Intensity Change Prediction Based on Surrounding Environmental Conditions with Deep Learning

Abstract

Tropical cyclone (TC) motion has an important impact on both human lives and infrastructure. Predicting TC intensity is crucial, especially within the 24 h warning time. TC intensity change prediction can be regarded as a problem of both regression and classification. Statistical forecasting methods based on empirical relationships and traditional numerical prediction methods based on dynamical equations still have difficulty in accurately predicting TC intensity. In this study, a prediction algorithm for TC intensity changes based on deep learning is proposed by exploring the joint spatial features of three-dimensional (3D) environmental conditions that contain the basic variables of the atmosphere and ocean. These features can also be interpreted as fused characteristics of the distributions and interactions of these 3D environmental variables. We adopt a 3D convolutional neural network (3D-CNN) for learning the implicit correlations between the spatial distribution features and TC intensity changes. Image processing technology is also used to enhance the data from a small number of TC samples to generate the training set. Considering the instantaneous 3D status of a TC, we extract deep hybrid features from TC image patterns to predict 24 h intensity changes. Compared to previous studies, the experimental results show that the mean absolute error (MAE) of TC intensity change predictions and the accuracy of the classification as either intensifying or weakening are both significantly improved. The results of combining features of high and low spatial layers confirm that considering the distributions and interactions of 3D environmental variables is conducive to predicting TC intensity changes, thus providing insight into the process of TC evolution.

A Novel Tropical Cyclone Size Estimation Model Based on a Convolutional Neural Network Using Geostationary Satellite Imagery

“…A CNN is a hierarchical neural network system that can extract and analyze feature vectors from complex multidimensional data [32]. CNNs are mostly used in computer vision studies because they can efficiently extract local features [35]. To extract features related to the relationships between distant pixels in a CNN, many convolutional layers need to be stacked.…”

“…The CNN used in this study was composed of several layers that continuously extracted abstract features from the input data to perform regression or classification tasks by matching these features with the target of the study [33][34][35]. Each layer consisted of several neurons that computed weighted combinations of input data [35]. The model was trained to optimize parameters based on the nonlinear behavior of an activation function [35].…”

“…Recently, convolutional neural networks (CNNs), an artificial intelligence deep learning technique [28], have been used in TC studies in conjunction with satellite-based TC image analysis [29][30][31][32][33][34][35][36]. However, despite the importance of TC size, most previous studies utilized CNNs only to estimate the TC intensity and tracks [29][30][31][32][33][34][35][36].…”

“…Recently, convolutional neural networks (CNNs), an artificial intelligence deep learning technique [28], have been used in TC studies in conjunction with satellite-based TC image analysis [29][30][31][32][33][34][35][36]. However, despite the importance of TC size, most previous studies utilized CNNs only to estimate the TC intensity and tracks [29][30][31][32][33][34][35][36]. This study aimed to address this research gap by proposing an objective TC size estimation technique using a CNN model and geostationary satellite-based multispectral images from the western North Pacific (WNP).…”

Polar Vortex Multi-Day Intensity Prediction Relying on New Deep Learning Model: A Combined Convolution Neural Network with Long Short-Term Memory Based on Gaussian Smoothing Method

“…DL algorithms and models have the ability to learn from a large number of long-time signal data and extract image features, that is, providing a powerful nonlinear function fitting ability to forecast the polar vortex index. Compared with ordinary atmospheric models and traditional mathematical methods, many DL neural networks have strong advantages [44][45][46][47][48].…”