- 期刊
人工智慧深度學習演算法在颱風強度估算之應用研究
Application of Artificial Intelligence Deep Learning Algorithms in Typhoon Intensity Estimation Research
摘要
臺灣位於全球颱風發生頻率最高的西北太平洋海域，其帶來的強風與豪雨對臺灣的影響不容忽視，因此有效且及時對颱風強度的估算為目前重要的天氣議題之一。近年來，隨著電腦硬體的進步，機械深度學習透過機器對大量樣本的分析和學習，能夠隱式提取圖像中深層抽象的複雜特徵，因此廣泛地被應用於氣象領域中。本研究以2017－2020年，西北太平洋區域的強烈颱風個案之生命週期的向日葵8號紅外線衛星影像為樣本，以影像邊緣技術進行衛星影像的預處理，所得之結果以卷積型類神經網路（CNN）進行深度學習，所建構的颱風強度偵測模型對2021年獨立強烈颱風影像樣本進行估測，其與實際風速於輕度、中度以及強烈階段的均方根誤差分別為7.3kts、13kts與16.7kts，而均方根誤差百分比於各強度階段均小於20%，結果顯示本研究所建構的颱風模型能適用於颱風強度的估算。
並列摘要
Taiwan is located in the Northwest Pacific Ocean where typhoons occur most frequently in the world. The immediately estimation of typhoon intensity is a very important issue, due to the strong winds and heavy rainfalls accompanying typhoons cause the disasters. Deep learning can implicitly extract the deep complex features in the images through the learning of a large amount of samples, and this technique has been increasing applied to the meteorological studies recently. In this study, we collected infrared band images from Himawari 8 captured during severe typhoon events in the Northwestern Pacific Ocean spanning 2017 to 2021. We employed image edge detection techniques to preprocess the collected images, and Convolutional Neural Networks (CNNs) were utilized to construct a typhoon intensity estimation model. Model training involved data from 2017 to 2020, while the model's performance was validated against observed intensities from the Joint Typhoon Warning Center (JTWC) for the year 2021. The root mean square error (RMSE) for independent samples in 2021, at various typhoon development stages (mild, moderate, and severe), were 7.3 kts, 13 kts, and 16.7 kts, respectively. Furthermore, the root mean square error percentage (RMSPE) for validation results was found to be less than 20%. These results underscore the viability of this technique for accurate typhoon intensity estimation.