Detecting tree mortality using waveform features of airborne LiDAR

Tree mortality impacts biodiversity, carbon dynamics and the management of forests. Climate change is expected to increase tree mortality, but understanding of tree mortality rates and the underlying processes is limited; thus, more accurate and efficient tree mortality mapping methods are required. In this study, we investigated the feasibility of using airborne light detection and ranging (LiDAR) waveform (WF) features in detecting dead trees to monitor tree mortality and studied how the WF features of dead trees change over time.

We used three consecutive LiDAR campaigns using fixed sensor and flight parameters in a boreal forest in Southern Finland (61.5°N, 24.2°E). The campaigns spanned four years and were carried out in 2011, 2013 and 2015. A Riegl LMS-Q680i LiDAR sensor, which operates at 1550 nm wavelength, provided return WF data to study the geometric-optical properties of living and dead trees and monitor mortality of Norway spruce (Picea abies H. Karst.).

Our findings highlight the differences in radiometric and geometric WF features between living and dead trees. The return WFs from dead trees were consistently elongated and contained more backscattering energy. We also found that as a tree died, the canopy and branch structures became less dense and more irregular, leading to more complex return WFs. The WF features were used for binary classification of living and dead trees, resulting in classification accuracies between 94.7 and 98.5 %, depending on the campaign. Distinguishing between living and dead trees is challenging for trees that have died recently when there are only minor defects in the crown and discoloration of foliage. Tree decay after death improved the discernability between living and dead trees as the geometric-optical properties of the crown change. There is, however, a limit after which further stages of decay might impair the discernability of living and dead trees.

The radiometric and geometric WF features and canopy mortality effects on the WF features are consistent across datasets implying intrinsic quality of information in the WF features. The within-class variance of WF features in dead trees is greater than that in living trees, indicating significant variations in the geometric and radiometric properties of trees between stages of decaying and dying. Our results imply that LiDAR WFs can be used for the accurate detection of dead trees to map tree mortality.