Consideration of terrain features from satellite imagery in machine learning of basic wind speed. (1st April 2022)

• Record Type: Journal Article
• Authors: Lee, Donghyeok; Jeong, Seung Yong; Kang, Thomas H.-K.
• Is Part Of: Building and environment. Volume 213(2022)
• Publication Date: 2022-04-01
• Subjects: Basic wind speed -- Satellite imagery -- Terrain effect -- Machine learning -- K-NN -- SimCLR; Buildings -- Environmental engineering -- Periodicals; Building -- Research -- Periodicals; Constructions -- Technique de l'environnement -- Périodiques; Electronic journals; 696
• DOI: 10.1016/j.buildenv.2022.108866
• Languages: English
• ISSNs: 0360-1323

Abstract

Abstract: Basic wind speed is a basis for calculating design wind loads (including wind environment evaluation) on structures at a specific site. Because structural design of high-rise buildings is typically governed by wind loads, accurate estimation of basic wind speed, which has been done by converting observed data for a region to that imposed at a height of 10 m on flat open terrain, is important. Although equations within codes attempt to take into account terrain features by considering effects such as surface roughness and topography, it is often difficult to apply them to real conditions due to terrain complexity. To overcome the limitation of engineering judgment, consideration of the terrain features from satellite imageries using machine learning algorithm is proposed. The number of selected weather stations, terrain similarity, distance from station, and machine learning method of multilayer perceptron (MLP) are also investigated as parameters or methodology. The estimation accuracy is shown to be high in the order of the MLP method and methods of considering both terrain similarity and distance, terrain similarity only, and distance only (traditional engineering judgment). Highlights: Selecting methods of neighboring weather station(s) were compared to improve accuracies of basic wind speed estimation. Satellite image preprocessing methods for machine learning were compared. Effects of terrain similarity and/or distance from station on basic wind speed were compared. This kind of AI-based approaches to determine the basic wind speed was attempted to be researched for the first time.