Tropical Cyclone intensity prediction based on hybrid learning techniques

Coastal regions in India are very frequently hit by Tropical Cyclones (TCs), which result in tremendous loss. Its intensity prediction has been a challenging task because of drastic climatic changes over the past few years in the world. Intensity of Tropical Cyclone is highly influenced by ocean, at...

Full description

Saved in:
Bibliographic Details
Published in:Journal of Earth System Science 2023-02, Vol.132 (1), p.28, Article 28
Main Authors: Varalakshmi, P, Vasumathi, N, Venkatesan, R
Format: Article
Language:English
Subjects:
Analysis
Artificial neural networks
Atmospheric models
Blended learning
Climate change
Coastal zone
Cyclones
Deep learning
Earth and Environmental Science
Earth Sciences
Feature extraction
Hurricanes
Learning algorithms
Literature reviews
Machine learning
Meteorological parameters
Neural networks
Parameters
Predictions
Rain
Regression analysis
Regression models
Sea level
Space Exploration and Astronautics
Space Sciences (including Extraterrestrial Physics
Spatial analysis
Stochasticity
Storm damage
Tidal waves
Tropical cyclone intensities
Tropical cyclones
Wind
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Coastal regions in India are very frequently hit by Tropical Cyclones (TCs), which result in tremendous loss. Its intensity prediction has been a challenging task because of drastic climatic changes over the past few years in the world. Intensity of Tropical Cyclone is highly influenced by ocean, atmospheric and meteorological parameters which makes the task difficult to define the mechanism of Tropical Cyclone intensity prediction. Here, a hybrid deep learning model is built using historical observations collected from various sources to perform a data-driven prediction of Tropical Cyclone’s intensity using regression model. This hybrid model utilizes convolutional neural network (CNN) architectures for feature extraction and machine learning models for regression. The hyper-parameters are optimized to fine-tune the model. The weights and biases are optimized using stochastic gradient descent (SGD). The proposed system is also compared with other regression models in machine learning and deep learning. The spatial analysis is accomplished using Modern-Era Retrospective analysis for Research and Applications, Version 2 Project Overview (MERRA-2) data for various coastal regions that are hit by cyclones. The temporal analysis is carried out using intensity skill forecast for time-series observation. The proposed system is also tested for Cyclone Amphan that hit eastern coastal regions of India.
ISSN:0973-774X
0253-4126
0973-774X
DOI:10.1007/s12040-022-02042-5