Wind Speed Forecasting Using Long Short-Term Memory (LSTM) and Meteorological Variables at Tanjung Perak, Surabaya, Indonesia

Yesiska M Siallagan (Physics Study Program, Faculty of Science and Technology, University of Jambi, Jambi 36361, Indonesia)
Tugiyo Aminoto (Physics Study Program, Faculty of Science and Technology, University of Jambi, Jambi 36361, Indonesia)
Husnul Hamdi (Physics Study Program, Faculty of Science and Technology, University of Jambi, Jambi 36361, Indonesia)

Journal of The Physical Society of Indonesia
Vol. 2 No. 2 (2026): October 2026

This study evaluated Long Short-Term Memory (LSTM) models for wind-speed prediction using meteorological observations from Tanjung Perak, Surabaya, for 2015–2024. Air temperature, relative humidity, atmospheric pressure, and rainfall were used as predictors. Data preprocessing, alternative training–test partitions (70:30, 80:20, and 90:10), and model configurations were evaluated using root mean square error (RMSE) and mean absolute error (MAE). Among the configurations reported for the 80:20 partition, 50 LSTM units and 50 epochs yielded the lowest test errors (RMSE = 1.2965 knots; MAE = 1.0175 knots). For January 2025, reported forecast errors were RMSE = 1.4253 and MAE = 1.2261 knots for 7 days; RMSE = 1.3178 and MAE = 1.0943 knots for 14 days; and RMSE = 1.2117 and MAE = 0.8886 knots for 31 days. These results refer to different verification periods and do not establish superior performance at longer forecast horizons. Multiple linear regression identified significant associations with temperature, humidity, and pressure (p < 0.001), while rainfall was not significant (p = 0.164). Further chronological and operational validation is needed.