1. Introduction * Typhoons are meteorological phenomena that cause harmful urban damage. Accurate typhoon prediction is essential for disaster preparedness in reducing the greater severe risk. The wind speed at multiple heights is one of the main atmospheric variables contributing to the typhoon which can be predicted using Numerical Weather Prediction (NWP). However, there are still inaccuracy issues in NWP caused by an imperfect interpretation of the model in depicting several complex conditions of an atmospheric physical process (Chen et al., 2020). Therefore, an accuracy improvement needs to perform in NWP output to better identify the typhoon track. Efforts to improve the accuracy of NWP have been conducted. Rasp and Lerch (2018) utilized a statistical approach for bias correction of 2-m temperature prediction from the ensemble weather numeric method. The correction of NWP was also carried out for the * 동아대 ICT융합해양스마트시티공학과 박사과정 ** 동아대 해양도시건설방재연구소 연구교수, 공학박사 *** 동아대 ICT융합해양스마트시티공학과 부교수, 공학박사 **** 동아대 ICT융합해양스마트시티공학과 교수, 공학박사 (Corresponding author: Department of ICT Integrated Safe Ocean Smart Cities Engineering, Dong-A University, jjyee@dau.ac.kr) 이 연구는 2022년도 한국연구재단 연구비 지원에 의한 결과의 일 부임. 과제번호:2016R1A6A1A03012812 temperature variable using the support vector machine (SVM) algorithm (Zeng et al., 2020). However, to the best of the authors’ knowledge, there are no studies on determining the algorithms for improving the accuracy of wind speed prediction at multiple atmospheric height levels. This study aims to investigate the appropriate machine learning arrangement among various algorithms, training dataset scenarios, multiple atmospheric predictors, and different NWP resolutions to improve the accuracy of NWP output for wind speed at 850 mb and 200 mb height levels. 2. Methodology 2.1 Data The present study uses atmospheric data at two levels (850 mb and 200 mb) from 33 soundings spreading over the western North Pacific region. These data are used as the target in the supervised ML dataset training. Data of the Global Forecasting System (GFS) is also retrieved from the web server of the National Oceanic and Atmospheric Administration (NOAA) for the input of the Weather Research and Forecasting (WRF) model. 2.2 Method The NWP used in this study is the WRF model for periods of 15 typhoon occurrences with a double-nested 2022년 추계학술발표대회 : 일반부문 태풍 경로 식별을 위한 기계 학습 알고리즘을 활용한 서로 다른 높이에서의 향상된 풍속 예측법 Improved Wind Speed Prediction at Two Height Levels Using Different Machine Learning Algorithms for Typhoon Track Identification 타마마딘 마마드 * 이창계 ** 기성훈 *** 이정재 **** Tamamadin, Mamad Lee, Chang-kye Kee, Seong-Hoon Yee, Jurng-Jae Abstract Typhoons are meteorological disasters that can have even more severe consequences. To reduce the higher impact, a Numerical Weather Prediction (NWP) is used to predict wind speed as one of the variables for the typhoon indicators. However, there are still many issues, especially accuracy, that must be continually improved. This paper aims to find the appropriate machine learning algorithm and its predictors to improve the accuracy of NWP for wind speed prediction at lower level (850 mb) and upper level (200 mb) tested in the resolution of 27 km and 9 km. This study found that the finest accuracy at 850 mb and 200 mb is achieved using Deep Neural Multilayer Perceptron (DNMLP) and Naive Bayes Gaussian, respectively. The best accuracy of wind speed at 850 mb and 200 mb is 0.75 and 0.78, respectively. 키워드 : 수치기상예측, 기계학습, 풍속예측, DNMLP, Naive Bayes Gaussian Keywords : Numerical Weather Prediction, Machine Learning, Wind Speed Prediction, DNMLP, Naive Bayes Gaussian - 430 - 2022년 대한건축학회 추계학술발표대회논문집 제42권 제2호(통권 제78집) 2022. 10. 26 ~ 10. 28