WIND GUST ESTIMATION APRROCAHES USING MACHINE LEARNING AT INDONESIAN AIRPORT

• Main Author: Nanda Pramono, Triadinda
• Format: Final Project
• Language: Indonesia
• Subjects: Geologi, hidrologi & meteorologi
• Online Access: https://digilib.itb.ac.id/gdl/view/76773
• Institution: Institut Teknologi Bandung

The risk of airplane accidents in Indonesia is still high, with one of the factors causing the accident being the meteorological variable, that is wind gust. A wind gust is a brief change of wind with significant changes that will cause trouble in flight, especially when it lands. Wind gusts which is occurs in brief would be important to predict for aircraft safety. This research estimates the probability of occurrence which classified as biner classification and magnitude of wind gusts at Indonesian airports. The predictor data uses ERA-5 data with hourly resolution and the predictant data is METAR wind gust data with 10-minute resolution converted into hourly averages. The data period used is five years (2017 to 2022). The method used is Machine Learning (ML) through Artificial Neural Networks (ANNs) and compared with the regression method. The predictor variables used are wind speed, specific humidity, vertical wind speed, relative vortices, temperature gradient, and the auto-regressive component (n-1) of the wind gust data. 70% training data is used and 30% testing data. Prediction models are made separately for the DJF, MAM, JJA, and SON seasons due to the different wind gust characteristics in each season. The results of METAR data analysis showed that three airport stations in Indonesia had the most wind gusts, tehre are Soekarno-Hatta International Airport, El Tari International Airport, and Sam Ratulangi International Airport . The wind gust prediction model using the ANNs method is able to predict the wind gust event better than the regression method. For the wind gust prediction model, the regression method represents the wind gust magnitude much better than the ANN method because it produces a larger correlation value. Stepwise regression for optimizing predictor showed that autoregresif (n-1), wind speed, and specific humidity on 950 hPa level pressure have a dominant affecting to wind gust near surface around the airport.