Exploring full-waveform LiDAR parameters for tree species classification

Precise tree species classification with high density full-waveform LiDAR data is a key research topic for automated forest inventory. Most approaches constrain to geometric features and only a few consider intensity values. Since full-waveform data offers a much larger amount of deducible information this study explores a high number of parameter and feature combinations. Those variables having the highest impact on species differentiation are determined. To handle the large amount of airborne full-waveform data and to extract a comprehensive number of variable combinations an improved algorithm was developed. The full-waveform point parameters amplitude, width, range corrected intensity and total number of targets within a beam are transferred into raster covering a test site of 10 km2. It was possible to isolate the three most important variables based on the intensity, the width and the total number of targets. Up to six tree species were classified with an overall accuracy of 57%, limiting to the four main species accuracy was improved to 78% and constraining just to conifers and broadleaved trees even 91% could be classified correctly.

Research highlights▶ Derivation of complex feature combinations from full-waveform LiDAR data. ▶ Classification of up to six tree species tested for mixed temperate forest. ▶ Main species classification accuracy of mixed temperate forest reaches 80%. ▶ Three most important feature combinations are extracted.