Stefano Puliti successfully defended his doctoral thesis, “Use of photogrammetric 3D data for forest inventory”, on 31 March 2017.
The topic for the trial lecture was “Using multi-temporal remotely sensed data to monitor changes in forest structure”. We congratulate!
Aerial imagery have long been used as auxiliary information to reduce the costs of forest inventories. Due to the high correlation between tree height and forest biophysical properties, manual photogrammetric techniques have been applied to aerial imagery for the measurement of vertical canopy structure, an important variable in forest inventories. In the past decade years, major advances have resulted in the development of photogrammetric software for the automatic generation of photogrammetric data from digital imagery. As a result, user-friendly advanced photogrammetric software are now available on the market, allowing for an increasing number of users to produce dense three-dimensional (3D) photogrammetric point clouds. The increased accessibility to advanced software in addition to the large availability of aerial imagery has led to a renaissance in the use of photogrammetry for forest inventory. The smaller costs of acquiring photogrammetric data compared to alternative 3D remote sensing data (i.e., airborne laser scanning; ALS), make their use appealing. The four studies included in this thesis addressed the use of photogrammetric data for the two main categories of forest inventories, namely: forest management inventories (FMI) and large-scale forest surveys (LSFS). For both categories, this thesis illustrated potential applications for which photogrammetric data may be advantageous over alternative 3D remote sensing data.
Wall-to-wall photogrammetric data produced from imagery collected using different
platforms, i.e. a manned aircraft in paper I and an unmanned aerial vehicle (UAV) in paper II, were used to model forest biophysical properties of interest in FMI. Both structural and spectral variables from photogrammetric data were used as predictor variables. Furthermore, when available, accuracy figures from ALS based inventory were used as a benchmark. The accuracy assessment revealed that photogrammetric data were able to predict forest biophysical properties with similar accuracy to ALS data. Furthermore, the first two papers highlighted some advantages related to photogrammetry, namely: 1) the possibility to use spectral information for species-specific FMI, and 2) the versatility of acquiring photogrammetric data using UAVs.
Moreover, the possibility to use UAVs in forest inventories was further addressed by illustrating LSFS applications for which UAVs could be cost-efficient. As a means of reducing the costs for RS auxiliary data acquisition, UAV data were acquired as a sample (i.e. partial-coverage) over a large area. In paper III the sample of UAV data, together with a subsample of field data were used in a hybrid inferential framework to estimate growingvistock volume (GSV) and assess its uncertainty. Such an approach enabled an increase in precision compared to design-based estimates using only field data. In paper IV, these data sources were augmented by a third wall-to-wall layer of Sentinel-2 multispectral data as a means of further increasing the precision of the estimator. In the latter case, the recently developed hierarchical model-based inference was adopted to enable a statistically rigorous estimation of the GSV and its uncertainty. This approach resulted in a slight increase in the precision of the hybrid estimator. Nevertheless, it allowed for a reduction of the UAV sampling intensity, hence reducing the UAV acquisition costs substantially.
Overall, the thesis concluded that photogrammetric data will have an increasingly important role in forest inventories. Not only are comparable levels of accuracy achievable, but their use can be more cost-efficient than alternative 3D remotely sensed data. Even though further research in different forestry settings should confirm our findings, the applications described in this thesis were found to have potential for operational use.
Main supervisor: Professor, Dr. Terje Gobakken, MINA, NMBU
Co-supervisors: Professor, Dr. Erik Næsset, MINA, NMBU
Researcher, Dr. Hans-Ole Ørka, MINA, NMBU
Professor Dr. Hailemariam Temesgen, Oregon State University, USA
Dr. Lars Torsten Waser, Swiss National Forest Inventory and Department of Landscape Dynamics, Swiss Federal Institute for Forest, Snow and Land-scape Research WSL, Switzerland
Professor Dr. Tron Eid, MINA, NMBU
[bibtex file=SkogRoverAll.bib key=Puliti2017]