Use of Artificial Intelligence and Machine Learning for Determination of Basic Wind Speed and Design Wind Speed


Wind load is a crucial structural design consideration, especially for tall buildings. The wind load and its potential impact is primarily dependent on the speed of the wind at the site. Currently, design codes representing the respective wind speeds of a particular region, commonly known as basic wind speed, are widely used to determine the wind load. Basic wind speed is based on decades of recorded wind speed data. In most design codes, including ISO 4354 and ASCE 7 – 16, it is the speed set at ten meters above the ground in flat and open terrain and the definition of flat terrain is similar for these design codes. For practical application purposes, the basic wind speed is often generalized by the specific terrain conditions.

It is recommendable to consider the terrain effects such as topography, surface roughness and height for accurate calculation of basic wind speed from the available wind data. Although equations within most design codes attempt to consider the effects of topography and surfaces roughness, it is difficult to apply them in real conditions due to the complexity of the terrain. Besides, different terrain categories are determined based on engineering judgment that may lead to errors. These limitations make an accurate quantitative evaluation of wind speeds a big challenge.

Due to the rapid changes in terrains and global climates, updating the available basic wind speeds is necessary. Recently, machine learning has emerged as a promising technique for addressing various issues within the wind engineering field, including predicting wind speeds. However, this approach only uses numerical data and does not consider the effects of the terrain. In addition, the machine learning approach is only applicable to regions where data exists. Addressing these limitations will make machine learning a valuable tool in wind engineering.

On this account, Mr. Donghyeok Lee (PhD Student) and Mr. Seung Yong Jeong (PhD Candidate) led by Professor Thomas Kang from Seoul National University proposed a new machine learning procedure for accurately predicting the basic wind speeds based on the terrain features from satellite imageries. Their main objective was to improve the accuracy of existing basic wind speeds by overcoming the limitation of engineering judgment. Baseline experiments, based on three methods: Select-One, K-NN and Each-One, involving wind speed data observed in 318 stations in Korea, were performed for comparison. The effects of the three main parameters: terrain similarity, number of selected weather stations and distance from the stations, were evaluated. Their work is currently published in the research journal, Building and Environment.

The research team showed that using the machine learning approach allowed quantitative terrain evaluation and subsequent prediction of the basic wind speeds. K-NN based on both terrain similarity and distance showed better performance than either K-NN basic wind speeds based only on terrain similarity or based only on distance. Compared with baseline experiments, where the effects of topography and surface roughness were based on engineering judgment, the presented machine learning approach provides the best performance than all the K-NN methods. This was attributed to its ability to accurately capture terrain features that are difficult to recognize by human judgment or conventional methods for determining basic winds speeds.

In a nutshell, Professor Thomas Kang and his research group developed a machine learning-based method for determining the basic wind speed considering the effects of terrain features from satellite imagery. By overcoming the shortcomings of the conventional methods, it proved to be an appropriate method for accounting for the different terrain effects. In a statement to Advances in engineering, Professor Thomas Kang stated that the machine learning approach is more accurate and would lead to accurate determination of basic wind speeds for effective structural design.

Use of Artificial Intelligence and Machine Learning for Determination of Basic Wind Speed and Design Wind Speed - Advances in Engineering

About the author

Dr. Thomas Kang is a Professor in the Department of Architecture & Architectural Engineering and Director for Engineering Education Innovation Center & GECE at Seoul National University, Korea. Prior to that, he was an Assistant Professor in the School of Civil Engineering and Environmental Science at the University of Oklahoma, Norman, OK, USA. He has held various affiliated positions in the U.S. and Japan, including Adjunct Professor at the University of Oklahoma, Adjunct Professor at the University of Illinois at Urbana-Champaign, and Lecturer at UCLA, the University of Hawaii at Manoa and the University of Tokyo. Prof. Kang received his PhD from UCLA, his MS from Michigan State University, and his BS from Seoul National University.

Prof. Kang is a Fellow of Post-Tensioning Institute (PTI) and a Fellow of American Concrete Institute (ACI). Prof. Kang received the Kenneth B. Bondy Award for Most Meritorious Technical Paper as Lead Author from PTI in 2012, and the Wason Medal for Most Meritorious Paper as Lead Author from ACI in 2009 with the subject of post-tensioned concrete. He regularly teaches the course of Post-Tensioned Concrete Structures at the University of Illinois at Urbana-Champaign every other summer (both on campus and online) and at the University of Hawaii at Manoa every fall (live online lectures). Prof. Kang has served an Editor-in-Chief for four journals: Wind and Structures, International Journal of Concrete Structures and Materials, Journal of Structural Integrity and Maintenance, and Advances in Computational Design; and Associate Editor for PTI Journal of Post-Tensioning Institute. He is one of the founding and voting members of PTI DC-20 Committee, Building Design, and has been a voting member for ACI Committee 369, Seismic Repair and Rehabilitation; Joint ACI-ASCE Committees 335, Composite and Hybrid Structures, 352 Joints and Connections in Monolithic Concrete Structures, and Joint ACI-ASCE Committee 423, Prestressed Concrete; and Joint ACI-ASME Committee 359, Concrete Containments for Nuclear Reactors.

Prof. Kang published more than 150 international journal papers and more than 150 international conference proceedings, including 50 in ACI Structural Journal and 10 in PTI Journal. He has chaired many sessions/symposiums of structural engineering; delivered many keynote/invited speeches; and organized international conferences/workshops as a Chair. Additionally, Dr. Kang has done a lot of practice as a consulting engineer in Korea and the U.S. Prior to joining the academia, he had a working experience in California, USA (e.g., John A. Martin & Associates), and was a licensed Professor Engineer (PE) in California.

He is currently working with Oncrets, a start-up company in the field of post-tensioning, as a creator of new-to-market “Smart Jack” and its patent holder.

About the author

Donghyeok Lee is a PhD student in the Department of Artificial Intelligence at Seoul National University. He received his BS from Chung Ang University and MS from Seoul National University. His research interests include machine learning for computer vision, graph neural network and applying machine learning to the building structure.


About the author

Seung Yong Jeong is a PhD Candidate in the Department of Architecture and Architectural Engineering at Seoul National University. He received his BS and MS from Konkuk University and Seoul National University, respectively. His research interests include performance-based seismic and wind design, nonlinear analysis, and high-rise concrete buildings.



Lee, D., Jeong, S., Y. & Kang, T. H.-K (2022). Consideration of terrain features from satellite imagery in machine learning of basic wind speedBuilding and Environment, 213, 108866.

Go To Building and Environment

Check Also

Size effect study of the fracture properties of steel fiber reinforced concrete using a novel 3D mesoscale modelling approach - Advances in Engineering

Size effect study of the fracture properties of steel fiber reinforced concrete using a novel 3D mesoscale modelling approach