Estimating Grassland Biomass - Potentials and Limitations of Point Cloud Analysis
Quantifying above ground biomass of grasslands is important information for grassland management and the understanding of ecological processes in grassland habitats. Often, allometric relationships between grassland height and biomass are used for biomass estimation. While these methods may be used in intensively used grassland with a homogenous canopy surface, in heterogenous grasslands it is not possible to repeat these measurements on larger areas. Recent technological advances in active and passive remote sensing offering new opportunities for estimations of grassland biomass. Many studies using remote sensing data for biomass estimation are based on the analysis of optical remote sensing sensors and are situated in forests and agricultural crops. Small temporal and spatially heterogenous grasslands were often neglected due to their complex vegetation structure. Just recently, point cloud data based either on terrestrial laser measurements (TLS) or on photogrammetric image analysis (SfM) approaches were investigated for their potential of biomass estimation in grasslands. The focus of this talk will be on evaluating the potential of TLS and SfM derived point clouds in deriving biomass estimation of grasslands with very different land use intensities. Both approaches show promising results for predicting grassland biomass (R2 ranging from 0.48 to 0.79 and from 0.35 to 0.81 for TLS and SfM respectively). TLS always performs better, which could be explained by the higher point densities and thus higher information content about the vegetation structure. However, under consideration of price and expert knowledge UAV based point clouds also produce satisfying results. Another aspect of the talk will be the comparison of performance aspects (e.g. computing time) of different point cloud analysis strategies. It can be shown that two scans of the same location from different aspects already provide detailed information about biomass and additional scans only lead to an unnecessary increase in data volume while maintaining consistent prediction quality. Various analysis methods will be test for extracting information from the point clouds. Here methods based on canopy surface height show the best prediction performance for biomass. Concluding, it is possible to say that both TLS and SfM-based point clouds have a good potential for deriving biomass information of grasslands, independent of land use intensity. However, to derive final conclusions the stability of the statistical relationships needs to be test over several growing periods. For the future, also the fusion of point cloud information with spectral information should be tested, as better biomass prediction models can be expected from this.