Machine learning algorithms for building height estimations using ICESat-2/ATLAS and Airborne LiDAR data

Abstract

Building height information is essential for determining urban morphology, urban planning studies, and manage sustainable growth. This study aims to use machine learning algorithms to estimate building heights from airborne LiDAR and spaceborne ICESat-2/ATLAS data. The performance of different machine learning algorithms was investigated when analyzing ICESat-2/ATLAS and airborne LiDAR data. The accuracy of building height information was compared with field measurements. Machine learning algorithms such as K-Nearest Neighbors (K-NN), Random Forest (RF), Support Vector Machines (SVM), Artificial Neural Networks (ANNs), and Random Sample and Consensus (RANSAC) were used to classify spaceborne and airborne LiDAR data. Among all the algorithms applied to ICESat-2/ATLAS, the RF algorithm provided the best results for the strong and weak beams with 0.9683 and 0.9614, respectively. The K-NN yielded the best result for the airborne LiDAR dataset with 0.9999. Statistical analyzes were applied to both LiDAR datasets. The results of statistical analyzes for the pair of field measurement and ICESat-2 were R² = 0.9894, RMSE = 0.4131, MSE = 0.1706, MAE = 0.3184, and ME = 0.0003; for the pair of field measurement and airborne LiDAR: R² = 0.8368, RMSE = 1.9646, MSE = 3.8597, MAE = 1.0586, and ME = -0.3450; and for the pair of airborne LiDAR and ICESat-2: R² = 0.8275, RMSE = 1.6664, MSE = 2.7770, MAE = 0.9040, and ME = 0.4598. As a result of the analysis, it was seen that the data obtained from the ICESat-2 system was successful in estimating building height and provided reliable data.