University of Natural Resources and Life Sciences, Vienna (BOKU) - Research portal
Gewählte Master / Diploma Thesis:
Bernhard Gill
(2022):
Bestandesvorratsermittlung mittels Airborne Laserscanning Daten.
Master / Diploma Thesis - Institut für Geomatik,
BOKU-Universität für Bodenkultur,
pp 51.
UB BOKU
obvsg
FullText
- Abstract:
- The aim of this thesis is the estimation of growing stock. On the one hand, only height metrics are calculated from a normalized digital surface model (nDSM) on different radii of circles. The other way was to derive height metrics, intensity metrics and the stand density from the point cloud. The Random Forest regression algorithm was used. Challenging for the modelling was the temporal discrepancy between survey of the inventory (2014) to the time of the aerial survey (2018). The inventory was collected using an angle count sample (ACS). An ACS has the disadvantage of not having an area reference. However, this is essential for the application of an area-based approach. The ACS points were visually corrected in their position. Circles with radii from 5 to 27.5 m in 2.5 m increments were placed on the ACS points and metrics were calculated on these areas. To improve model quality, the inventory was divided into strata (coniferous, deciduous, mixed). When comparing the models from the nDSM and the point cloud, it is clearly evident that the additional information obtained from the point cloud has a strong influence on the model quality. The use of the circle combinations was also superior to the single circle radii. The calculated stem density on the plots as well as the information from the Sentinel- 2 scene (4 bands and NDVI) was not chosen by the feature selection. The best models from the point cloud for softwood achieved an R² of 0.67 and an RMSErel of 32.26 %. The best model for calculating the growing stock for hardwoods was at the 15 m single circle with metrics derived from the point cloud (R² = 0.64 with RMSErel = 34.49 %). In comparison, the radius of the single circle of the nDSM (best model radius = 20 m) performed worse with an R² of 0.48 and RMSErel of 41.76 %. This clearly shows that the additional information from the point cloud causes an improvement of the models.
- Beurteilende*r: Atzberger Clement
- 1.Mitwirkender: Immitzer Markus