PS 33-214 - Estimation of understory biomass with vertical metrics and voxel-based volume obtained from backpack-LiDAR

Abstract: Temperate forests have multiple layers: canopy, subcanopy, shrubs, and herbs. Understory vegetation, which indicates vegetation in shrubs and herbs layers, plays a significant role in forest functioning, including carbon sequestration. However, traditional field-based methods of estimating forest biomass often exclude understory vegetation. In this study, we propose a method for estimating understory biomass using Backpack Laser Scanning (BPLS) in temperate forests. We selected four plots and installed five 1m × 1m subplots in each of the afforestation sites located in Gyeonggi Province, South Korea. We obtained laser scanned point cloud data of the four plots using BPLS and harvested all understory vegetation in the subplots to measure its biomass. For estimating understory biomass, we extracted twenty-six vertical metrics and volumes (of voxel size 17.3, 8.6, 4.3, and 2.1cm) from the point cloud data of each subplot. Then we selected the best subsets of variables, which were used to develop a multiple regression analysis model. Then, we used the Leave-One-Out-Cross-Validation (LOOCV) method to evaluate the performance of the model. As a result, the subset consists of the mean height of the points, point density of the 4^th floor when we divided point cloud data into 10 floors with equal intervals, and volume of voxel size 4.3cm were chosen as the best subset of predictors. The understory biomass estimation equation was developed using these three variables: (understory biomass) = -26.2 + (mean height) × 308.5 + (density of 4^th floor) × 189 + (volume of 4.3cm voxel) × 70.6. The LOOCV results showed an R² value of 0.70 and an RMSE of 30.6g. The variable with the most important impact on the dependent variable was mean height, with a p-value of less than 0.001. The volume of 4.3cm voxel also had a significant impact, with a p-value of less than 0.05. The density of the 4^th floor had the least impact on the dependent variable. This equation can be applied to estimate understory biomass of the study area. However, some points corresponding to leaves, branches, or stems of trees, not the understory, could affect the results to overestimate biomass, so removing unwanted points will be a key challenge. Overall, our study provides a method for estimating understory biomass using BPLS, and our results can improve estimates of carbon sequestration in forests and provide useful information for forest management.