Detecting Multi-layered Forest Stands Using High Density Airborne LiDAR Data

Abstract

Since two decades, the use of terrestrial laser scanning (TLS) and Airborne Light Detection and Ranging (LIDAR) has become very prominent in analysing 3D forest structures (AKAY et al. 2009). The potential of full waveform analysis of high density Airborne LiDAR data (ALS) for the detection and structural analysis of multi-layered forest stands is not yet well investigated (JASKIERNIAK et al. 2011), although ALS data provide exact information on tree heights of multi-layered forest stands using particular laser pulse characteristics (GAULTON & MALTHUS 2010). Since the mid-19th century, managed forests in Brandenburg have been dominated by Scots pine monocultures. In the last fifteen to twenty years many forest stands were converted into multi-layered mixed forests by silvicultural conversion of forests and natural succession (MLUR 2004). Today, the majority of forest stands in the federal state of Brandenburg remain dominated by Scots pine (Pinus sylvestris) in the canopy layer, while European beech (Fagus sylvatica) or Sessile oak (Quercus petraea) are predominant in the understorey. In this study, we investigate and discuss the potential of full waveform high density airborne LiDAR data (ALS) for detecting, classifying, and stratifying discrete vegetation layers at forest stand level, based on 0.1ha investigation plots. Full waveform high density ALS data on each 5th percentile level was used in combination with binary logistic regressions to discover the structural layer type of multi-layered forest stands from normalized discrete ALS pulses. The results of the descriptive statistics of ALS point clouds and binary logistic regression models produce particular forest layer profile indices of understorey vegetation and canopy layer. Such parameters can further be used as variables for forest structure analysis algorithms, and can be empirically tested against stand characteristics. The validation of ALS data and model results is tested against empirical forest mensuration data of the “Datenspeicher Wald 2 (DSW 2-Forest inventory data)” and field survey reference points using error matrices. We demonstrate that binary logistic regression analyses are functional for establishing a prediction model. The model was applied successfully on larger forest stands and forest areas, and can become useful for identifying and separating single from multi-layered forest stands using percentiles of total amounts of ALS return pulses on a 10x10m raster size with a high overall accuracy of 90%. The established model has the potential for a broad range of forest management applications, such as timber inventory evaluation, forest growth modelling, monitoring of vegetation dynamic and succession, as well as ecological classifications and the detection of deadwood in forest stands (KIM et al. 2009).