Tree Crown Delineation using Watershed Techniques and Forest Metrics from NEON LiDAR Data

Abstract:

LiDAR is a powerful remote sensing tool allowing for forest metrics to be taken on varying scales, which ultimately provide important forestry variables used to calculate factors such as total biomass or leaf area index. These variables are most useful when calculated for individual trees throughout a stand, but in very dense forests, identifying single trees becomes more difficult by traditional means. Full forests can be quantified uniquely for the best understanding of ecological contributions as opposed to purely in situ tree inventories which are time consuming and extremely localized. Canopy height models (CHM) can be used to understand the forest as a whole. By inverting the CHM, the tree data becomes sinks in the ground, mimicking ponds; by applying watershed-related spatial analyst tools in ArcGIS and GrassGIS, the trees are delineated by makeshift "flooding." Within this algorithm, the crown peaks are also extracted as an intermediate step to delineation, but this is a reliable means to obtain an accurate number of trees, as well as their individual heights with high reliability (R² = 0.87). Delineated tree polygons can be directly overlaid onto different rasters to get many forest variables. In tightly clustered and very sparse stands, this method of delineation has a high level of accuracy. Following the workflow studies conducted on NEON LiDAR data on the Soaproot Saddle site, a ground-truth comparison was made with the Teakettle Experimental Forest site due to the availability of tree inventory data.