International Journal of Applied Earth Observations and Geoinformation (Apr 2023)
Reduced model complexity for efficient characterisation of savanna woodland structure using terrestrial laser scanning
Abstract
Advances in terrestrial laser scanning (TLS) enable the extraction of ecologically meaningful data from detailed 3D representations of individual trees. Computer models deliver a comprehensive suite of tree structural metrics that are difficult, if not impossible, to obtain using traditional field methods. However, best practice high-end TLS equipment and computer modelling are expensive and complex, and ground-based data acquisition is spatially limited, thus presenting significant hurdles for the implementation of this technology in land management. We investigated the utility of lower-cost TLS data acquisition and processing for efficient, large-scale assessment of tree volume as an ecologically meaningful parameter. A 1 ha plot in a tropical savanna woodland was scanned twice over consecutive years using an entry-level TLS scanner (Leica BLK360), with the second survey conducted immediately after a high-intensity fire event. The performance of low-complexity voxel models for calculating individual tree volume was tested and calibrated against more established and more complex Quantitative Structure Models (QSM) estimates of a 100-tree subset. Of the models tested, a filled voxel model with a voxel size of 0.04 m achieved 96% accuracy when compared to QSM estimates. Processing time for individual trees was over 100 times faster. To further explore the utility of lower-cost, lower-complexity data in large-scale monitoring, the best-performing optimised volume model was then applied to the hectare-scale data set and used to establish an allometric model based on metrics that can be obtained from aerial surveys. The best-performing allometric model used tree height and crown area as a compound variable in a logarithmic linear regression and was able to explain 99% of variance in the total tree volume. Furthermore, as the training data contained trees from recently burnt vegetation, the model was able to account for fire damage, important for carbon accounting in fire prone ecosystems such as savannas. With the utility of LiDAR scanning for vegetation mapping and monitoring firmly established in the literature, development of methods for non-specialist practitioners is now essential for greater utilisation of this technology by land managers. We provide a case study highlighting the utility of lower-cost data acquisition and efficient processing for locally adapted vegetation mapping and monitoring.
