Identification of Riparian Vegetation Types with Machine Learning Based on LiDAR Point-Cloud Made Along the Lower Tisza’s Floodplain

• Main Authors: Fehérváry István, Kiss Tímea
• Format: Article
• Language: English
• Published: Sciendo 2020-04-01
• Series: Journal of Environmental Geography
• Subjects: airborne lidar; scikit-learn; gini impurity; decision tree; riparian forest
• Online Access: https://doi.org/10.2478/jengeo-2020-0006

The very dense floodplain vegetation on the artificially confined floodplains results in decreased flood conveyance, thus increase in flood levels and flood hazard. Therefore, proper floodplain management is needed, which must be supported by vegetation assessment studies. The aims of the paper are to introduce the method and the results of riparian vegetation classification of a floodplain area along the Lower Tisza (Hungary) based on automatized acquisition of airborne LiDAR survey. In the study area 15x15 m large training plots (voxels) were selected, and the statistical parameters of their LiDAR point clouds were determined. Applying an automatized parameter selection and 10-fold cross-validation he most suitable decision tree was selected, and following a series of classification steps the training plots were classified. Based on the decision tree all the pixels of the entire study area were analysed and their vegetation types were determined. The classification was validated by field survey. On the studied floodplain area the accuracy of the classification was 83%.