Doctor's Theses (authored and supervised):
"Improving Forest Mensurations with High Resolution Point Clouds";
Supervisor, Reviewer: N. Pfeifer, M. Hollaus;
Department of Geodesy and Geoinformation, TU Wien,
oral examination: 2021-04-27.
Forests are of great economic and ecological benefit because they function as timber source and provide a variety of vital ecosystem services. To maintain the functioning of forest ecosystems and the provision of these diverse services, regular acquisitions of the forest conditions, including timber stock estimations, are required. Forest in- ventories (FI) which have evolved from economic demands provide such information and allow to optimize management strategies. FIs also form the basis to monitor abundance, health and changes of forest ecosystems. FIs traditionally rely on field measurements that include a number of forest parameters within plots. With the increasing availability of airborne laser scanning (ALS) systems, such local plot level inventories can be upscaled to larger areas in a convenient, robust and reproducible way.However, current FI acquisition methods suffer from a number of deficiencies. Firstly, FIs still rely on measurements in the field which are labor-intensive to acquire and partially subjective (e.g. the determination of crown closure). Secondly, allometric functions, which are deduced from field inventories and applied to ALS or auxiliary remote sensing data to predict FI parameters (e.g. growing stock) over larger areas, do not necessarily reflect site-specific characteristics. This leads to errors in the predicted parameters. Locally adjusted allometric functions, however, are cumbersome and expensive to achieve through field inventories. Therefore, such functions often do not exist. Thirdly, today´s forest stands, to which the respective allometric functions are applied, are derived from manual delineations based on ortho-images. Stand delineation, however, is a complex task, and resulting stands do not necessarily represent homogeneous compartments. Available forest stand boundaries are often originating from historical forest management units that have changed over time. The aim of this thesis is to investigate possibilities to improve current FI acquisition strategies through the integration of the various laser scanning systems we have at hand today.The miniaturization of laser scanning sensors and positioning systems allows the systems to be operated on light weight unmanned aerial vehicle-platforms (UAV- borne laser scanning, ULS) or placed on the ground (terrestrial laser scanning, TLS). There is a gap, however, between ALS and these close-range systems. Firstly, the systems deliver completely different point cloud qualities in terms of the resolution of the resulting point clouds. Secondly, the viewing direction and geometry differ fundamentally, particularly between TLS, which records the canopy from the bottom, and ALS, which can hardly acquire forest areas close to the ground. Yet, FIs could be improved through the combination of the different laser scanning systems. Close- range systems with the high level of detail allow for a fast, accurate and cost- efficient acquisition of field reference data, from which site-specific allometries can be deduced. Improved allometries can subsequently be applied to ALS data acquired from large areas with low point densities.Thus, the combination of the different systems allows to accomplish the scale transition from plot to landscape scales. Locally detailed single tree information could be brought to wide areas, allowing for landscape-wide FI parameter estimations. Yet, open aspects remain for such a proceeding, which are addressed in this thesis. In a first study, the point cloud quality requirements are investigated in order to allow a complete and accurate scene reconstruction from ULS data. The results show that the point density is crucial for the stem detection, whereas the accuracy of the stem reconstruction mainly depends on the accuracy of the sensor. A second study analyses differences in the ways ALS and ULS systems capture the forest structure. ULS systems acquire the forest structure more completey. In combination with the high level of detail, this allows to directly measure single tree components within the point cloud, apart from the computation of classical ALS metrics. In contrast, derived structure metrics differ between ALS and ULS, depending on the way the point cloud information is used for the metrics computation. In a third study, an approach is proposed to delineate homogeneous forest compartments with a similar forest structure from ALS point clouds. Homogeneous forest compartments can be used to upscale locally measured FI parameters from close-range laser scanning to larger areas by applying locally adjusted statistical models to ALS point cloud metrics. A final study deals with the temporal aspect of the inventories and investigates the potential to update ALS forest structure information with Sentinel-1 (S-1) C-band time series. Since the general height and stand density structures, respectively, are well reproduced, S-1 data facilitates to fill gaps between ALS acquisitions.The insights gained in the four studies will help making optimal use of the information content each system provides. Thus, the integration of the different laser scanning devices into the scale transition contributes to improving FI acquisition strategies. With such a procedure, landscape-wide accurate spatial forest information could be acquired in a robust, fast and cost-efficient way. This opens up new possibilities to integrate remote sensing in operational forest management activities, for instance for the selection of tree species, for shading studies, or for the modelling of forest fires, the sun irradiance or the water cycle. Ultimately, the integration of data from different systems will improve our ability to monitor the condition and development of forests.
"Official" electronic version of the publication (accessed through its Digital Object Identifier - DOI)
Electronic version of the publication:
Created from the Publication Database of the Vienna University of Technology.