Estimation of Ground Wind Speed in Complex Terrain Using Deep Learning

• Published in: Journal of Japan Society for Atmospheric Environment / Taiki Kankyo Gakkaishi Vol. 59; no. 6; pp. 73 - 82
• Main Authors: Michioka, Takenobu, Kosaka, Keita, Takimoto, Hiroshi, Sato, Ayumu
• Format: Journal Article
• Language: Japanese
• Published: Japan Society for Atmospheric Environment 08.10.2024; 公益社団法人 大気環境学会
• Subjects: complex terrain; convolutional neural network; deep learning; meso-scale meteoritical model; wind speed
• ISSN: 1341-4178; 2185-4335
• DOI: 10.11298/taiki.59.73

A method for predicting ground-level wind speeds over complex terrain was developed using deep learning. A mesoscale meteorological model with a horizontal resolution of 300 m was implemented to obtain horizontal wind speed components at 10 m above ground level at four mountainous areas. By utilizing these horizontal wind speed components as inputs and field observation data as training data, a deep learning model was developed. The Convolutional Neural Network (CNN) model uses wind speed data from six months of odd-numbered months as training data and from six months of even-numbered months as test data, revealing that the errors in the wind speeds are smaller than those obtained by the WRF at all four locations. Additionally, the CNN model improves the wind direction estimation. Using one month of observation data, the CNN model also predicts wind speeds and wind directions for the remaining eleven months, irrespective of the seasonal variations.