Supervised Segmentation of Ultra-High-Density Drone Lidar for Large-Area Mapping of Individual Trees

Martin Krůček,Kc Cushman,Kamil Král,Azim Missarov,James R Kellner

doi:10.3390/rs12193260

Abstract

We applied a supervised individual-tree segmentation algorithm to ultra-high-density drone lidar in a temperate mountain forest in the southern Czech Republic. We compared the number of trees correctly segmented, stem diameter at breast height (DBH), and tree height from drone-lidar segmentations to field-inventory measurements and segmentations from terrestrial laser scanning (TLS) data acquired within two days of the drone-lidar acquisition. Our analysis detected 51% of the stems >15 cm DBH, and 87% of stems >50 cm DBH. Errors of omission were much more common for smaller trees than for larger ones, and were caused by removal of points prior to segmentation using a low-intensity and morphological filter. Analysis of segmented trees indicates a strong linear relationship between DBH from drone-lidar segmentations and TLS data. The slope of this relationship is 0.93, the intercept is 4.28 cm, and the r2 is 0.98. However, drone lidar and TLS segmentations overestimated DBH for the smallest trees and underestimated DBH for the largest trees in comparison to field data. We evaluate the impact of random error in point locations and variation in footprint size, and demonstrate that random error in point locations is likely to cause an overestimation bias for small-DBH trees. A Random Forest classifier correctly identified broadleaf and needleleaf trees using stem and crown geometric properties with overall accuracy of 85.9%. We used these classifications and DBH estimates from drone-lidar segmentations to apply allometric scaling equations to segmented individual trees. The stand-level aboveground biomass (AGB) estimate using these data is 76% of the value obtained using a traditional field inventory. We demonstrate that 71% of the omitted AGB is due to segmentation errors of omission, and the remaining 29% is due to DBH estimation errors. Our analysis indicates that high-density measurements from low-altitude drone flight can produce DBH estimates for individual trees that are comparable to TLS. These data can be collected rapidly throughout areas large enough to produce landscape-scale estimates. With additional refinement, these estimates could augment or replace manual field inventories, and could support the calibration and validation of current and forthcoming space missions.

Highlights

Numerous studies have demonstrated the utility of terrestrial laser scanning (TLS) for automated segmentation of individual trees in high-density point clouds [1,2,3]
Diameter at breast height (DBH) estimates using these data are strongly correlated with diameter at breast height (DBH) of the same trees quantified using TLS [7], and a study in eucalypt forest in Australia showed that tree DBH from TLS data are unbiased with respect to field measurements [6]
There is a linear relationship between the estimated DBH from segmentations applied to drone lidar and TLS data for the same individual trees (Figure 5)

Summary

Introduction

Numerous studies have demonstrated the utility of terrestrial laser scanning (TLS) for automated segmentation of individual trees in high-density point clouds [1,2,3] Because these data can be acquired and processed rapidly in comparison to the time required for traditional field inventories, they have the potential to increase the accuracy and frequency of stand-level forest assessment. Scaling these methods to landscapes larger than a few hectares has been challenging [1,2]. Because high-density lidar data can model wood volume by segmenting individual trees [3,9], lidar data provide better estimates of tree-level aboveground biomass (AGB) and wood volume than manual methods using diameter at breast height (DBH) and allometric scaling equations [6,10]

Methods

Results

Conclusion