Maria Åsnes Moan defended her PhD thesis «Advancing site index determination using point cloud data» November 29th, 2024.
The topic of the trial lecture was «Measuring Changes in Forests Using
Remote Sensing».
Thesis abstract
Site index (SI) is the top height at a given reference age and is used to describe the forest site’s potential to produce wood volume. SI determination has evolved from purely field-based assessments based on age and height measurements to approaches using point cloud data. Point cloud data like airborne laser scanning (ALS) data and digital aerial photogrammetry (DAP) data have been used to determine SI using the direct and height differential approach. These approaches use the top height development over a known period to determine SI using point cloud data from at least two points in time, i.e., multitemporal point cloud data. The direct approach determines SI using prediction models of field-measured SI with predictor variables calculated from the point cloud data. The height differential approach determines the SI where the expected top height development is closest to the predicted top height development. This thesis aimed to advance SI determination using point cloud data by investigating some current challenges and opportunities. To this end, four separate studies were carried out.
Disturbances might make an area unsuitable for SI determination with point cloud data. This has previously been addressed by using non-decreasing top height and biomass as proxies for suitability, but this does not necessarily ensure undisturbed top height development. The first study of this thesis classified suitability using variables from multitemporal ALS data. The suitability was defined from field registered disturbances to the dominant trees. The results showed that suitability could be classified using multitemporal ALS data even though the definitions of suitability in that study were conservative as only one dominant tree being dead resulted in the plot being classified as unsuitable.
Using a time series of ALS data might increase the accuracy of SI determination compared to using data from two consecutive ALS acquisitions. This is because a longer period of the top height development will be represented. The second study used a time series of ALS data from three points in time to determine SI with the direct and height differential approach. The prediction errors were not statistically significantly different when using the full length of the time series of ALS data compared to using ALS data from two consecutive points in time, i.e., either first and second or second and third point in time. However, the area suitable for SI determination increased when any subset of consecutive points in time from the time series could be used for SI determination due to increased flexibility to avoid using periods where disturbances had occurred.
The value of improved information may be used to assess the multi-objective utility of different SI determination approaches for forest management. The third study used stochastic programming to assess the value of improved information of the direct and height differential approach using either multitemporal ALS data or ALS and subsequent DAP data. The value of improved information was closest to zero and thus best for the height differential approach in this case study.
The height differential approach could potentially be used for SI determination in young forests. The fourth study detected the positions of branch whorls from very dense point cloud data using a deep learning model and used detected branch whorls to determine SI with the height differential approach. The root mean square error of SI determined from detected branch whorls ranged between 19.85 and 20.87%. One of the challenges of SI determination in young forests is that the SI curves are steeper for younger ages than for older ages which means that the effect of branch whorl detection errors on the determined SI would be larger in young forests.
This thesis has addressed some current challenges and opportunities in order to advance SI determination with point cloud data. Still, the first and fourth study revealed that more research is needed on how to best define suitability and determine SI in young forests.
Supervisors:
Lennart Noordermeer, Norwegian University of Life Sciences (NMBU)
Ole Martin Bollandsås, Norwegian University of Life Sciences (NMBU)
Evaluation committee:
Mikko Vastaranta, University of Eastern Finland (UEF)
Henrik Jan Persson, Swedish University of Agricultural Sciences (SLU)
Meley Mekonen Rannestad, Norwegian University of Life Sciences (NMBU)