Towards Prediction and Mapping of Grassland Aboveground Biomass Using Handheld LiDAR

Abstract

The Oostvaardersplassen nature reserve in the Netherlands is grazed by large herbivores. Due to their increasing numbers, the area became dominated by short grazed grasslands and biodiversity decreased. From 2018, the numbers are controlled to create a diverse landscape. Fine-scale mapping and monitoring of the aboveground biomass is a tool to evaluate management efforts to restore a heterogeneous and biodiverse area. We developed a random forest model that describes the correlation between field-based samples of aboveground biomass and fifteen height-related vegetation metrics that were calculated from high-density point clouds collected with a handheld LiDAR. We found that two height-related metrics (maximum and 75th percentile of all height points) produced the best correlation with an R2 of 0.79 and a root-mean-square error of 0.073 kg/m2. Grassland segments were mapped by applying a segmentation routine on the normalized grassland’s digital surface model. For each grassland segment, the aboveground biomass was mapped using the point cloud and the random forest AGB model. Visual inspection of video recordings of the scanned trajectories and field observations of grassland patterns suggest that drift and stretch effects of the point cloud influence the map. We recommend optimizing data collection using looped trajectories during scanning to avoid point cloud drift and stretch, test horizontal vegetation metrics in the model development and include seasonal influence of the vegetation status. We conclude that handheld LiDAR is a promising technique to retrieve detailed height-related metrics in grasslands that can be used as input for semi-automated spatio-temporal modelling of grassland aboveground biomass for supporting management decisions in nature reserves.