Building Height Estimation from Open Optical Remote Sensing by Machine Learning Regression Technique: A Case Study of the Central of Bangkok

Abstract

Building height is crucial for understanding the urban environment and human activity. In Bangkok, there is open building footprint polygon layer, but it still lacks building height data. Recently, a digital surface model (DSM) from an unmanned aerial vehicle (UAV) was obtained for referencing. Additionally, open medium-resolution optical satellite images are freely provided almost a decade. Both data sources were used in research to estimate building heights. The study demonstrated that optical remote sensing, combined with reference height data, can effectively estimate building heights in urban areas. In this research we deploy two approaches namely, support vector machine regression (SVR) and random forest regression (RFR). The result produced similar root mean square error (RMSE) values: approximately 6.6 meters for buildings under 50 meters and around 12 meters for buildings under 100 meters. However, when evaluated with the 50-meter building height group in the second model testing, the SVR algorithm performed better than the RFR algorithm.