Airborne lidar data classification in complex urban area using random forest: a case study of Bergama, Turkey

Abstract

Airborne Light Detection and Ranging (LiDAR) data have been increasingly used for classification of urban areas in the last decades. Classification of urban areas is especially crucial to separate the area into classes for urban planning, mapping, and change detection monitoring purposes. In this study, an airborne LiDAR data of a complex urban area from Bergama District, İzmir, Turkey were classified into four classes; buildings, trees, asphalt road, and ground. Random Forest (RF) supervised classification method is selected as classification algorithm and pixel-wise classification was performed. Ground truth of the area was generated by digitizing classes into features to select training data and to validate the results. The selected study area from Bergama district is complex in urban planning of buildings, road, and ground. The buildings are very close to each other, and trees are also very close to buildings and sometimes cover the rooftops of buildings. The most challenging part of this study is to generate ground truth in such a complex area. According to the obtained classification results, the overall accuracy of the results is found as 70, 20%. The experimental results showed that the algorithm promises reliable results to classify airborne LiDAR data into classes in a complex urban area.