Multi-platform LiDAR approach for detecting coarse woody debris in a landscape with varied ground cover

Abstract

Coarse woody debris (CWD), or fallen logs, is known to be an essential habitat element for many organisms. CWD also supports ecosystem functioning through soil formation, nutrient cycling, and carbon storage. For these reasons, accurate assessments of CWD across landscapes are of interest to many ecologists and landscape managers, but traditional field-based measurements can be time-consuming and sampling strategies may not be representative of entire landscapes. Light detection and ranging (LiDAR) technologies may be able to provide a more rapid assessment of the number and volume of CWD across wide areas. However, most research using LiDAR for forest and woodland inventory assessment has focused on standing wood. Detection accuracy of CWD with LiDAR can be impacted by the point density of LiDAR data, ground layer vegetation, and sensor positioning relative to other vegetation or landscape structural features. We used a high-resolution terrestrial laser scanner (TLS), an unoccupied aerial vehicle (UAV) laser scanner (ULS), and a combination of data from both sensors (i.e. fused data, FLS) to estimate CWD in a grassy woodland ecosystem. The study area comprised plots with different types and amounts of vegetation cover and different types of CWD, both naturally occurring and introduced including dispersed, clumped, or a mixture of both types. This enabled a more detailed exploration of model performance across sensor types, vegetation types, and ground cover biomass. A random forest (RF) classification algorithm and noise removing operations on raster imagery were used to classify CWD. Completeness and correctness accuracy with the developed method were highly variable depending on the data and ground vegetation cover and ranged between 20% and 86%, and 12% and 96%, respectively, in comparison with field data. The LiDAR-derived digital surface model (DSM), surface roughness, and topographic position index were important variables for CWD detection. We found that the detection accuracy of CWD varied with the vegetation type, amount of ground vegetation cover, and LiDAR data. Ground cover density had a strong negative impact on accuracy, particularly for TLS and FLS data.