BROWSE

ITEM VIEW & DOWNLOAD

Improvement of Tropical Cyclone Track Forecast over the Western North Pacific Using a Machine Learning Method

Cited 0 times inthomson ciCited 0 times inthomson ci
Title
Improvement of Tropical Cyclone Track Forecast over the Western North Pacific Using a Machine Learning Method
Author
Kim, Kyoungmin
Advisor
Cha, Dong-Hyun
Keywords
Tropical cyclone; Western North Pacific; Regional climate model; Machine learning; Artificial neural network; K-means clustering
Issue Date
2020-02
Publisher
Graduate School of UNIST
Abstract
The accurate tropical cyclone (TC) track forecast is necessary to mitigate and prepare significant damage by a tropical cyclone. TC has been predicted by the numerical model, statistical model, and machine learning in previous researches. However, those models are separately used to predict the track of TC, and historical data with satellite image were used as input variables for machine learning without predicted data about the tropical cyclone in previous researches. In this study, we corrected the predicted track of TC by the regional climate model to ANN. TCs that occurred during the period from 2006 to 2015 over the western North Pacific were simulated by WRF, and TCs in this study include all categories of TCs except tropical depression (i.e., tropical storm, severe tropical storm, and typhoon) from June to November. We evaluated the performance of predicting TC track based on length, speed, and direction of forecast compared with observation. The simulated positions of TCs with historical data were used as variables for training and testing ANN targeted to TC position after 24-hour, 48-hour, and 72-hour. For optimizing the number of neurons in ANN, simulated TCs were divided into two parts, which are the TCs in 2006-2014 for ANN optimization and the TCs in 2015 for a blind test. Also, the output selection method, which has range based on the mean absolute error of WRF, was applied to exclude outlier of ANN results. By the output selection, the prediction error of ANN was more reduced than the prediction error of WRF. As a result, ANN can improve more the performance of WRF when the error of WRF was higher, and the error of ANN result, which wasn’t excluded by the output selection, increased less than ANN without applying output selection in the lower error of WRF. Also, cluster analysis was done in this study to investigate the effect of ANN depending on the location of predicted TC. This study used k-means clustering to divide the simulated TCs, and the TCs were divided into four parts, considering the silhouette coefficient value. The ANN with the output selection had better performance than WRF in cluster 1 (western Pacific) and cluster 2 (south of Korea) for 24-hour and 48-hour forecast. The ANN without the output selection had better performance than WRF in cluster 3 (Southeast Asia and China) and cluster 4 (south of Japan) for 72-hour forecast.
Description
Department of Urban and Environmental Engineering (Disaster Management Engineering)
URI
Go to Link
Appears in Collections:
UEE_Theses_Master
Files in This Item:
Improvement of Tropical Cyclone Track Forecast over the Western North Pacific Using a Machine Learning Method.pdf Download

find_unist can give you direct access to the published full text of this article. (UNISTARs only)

Show full item record

qrcode

  • mendeley

    citeulike

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.

MENU