Kaja Kandare successfully defended her doctoral thesis, “Fusion of airborne laser scanning and hyperspectral data for predicting forest characteristics at different spatial scales”, on 25 August 2017.
The topic for the trial lecture was “New techniques for obtaining forest resource data – prospects and areas of application”. We congratulate!
Forests can be characterized by many attributes such as mean height, volume, diameter at breast height (DBH), age, tree species distribution, and different indices describing productivity and diversity. All these characteristics can be estimated using a wide range of remote sensing data from aerial photography and airborne laser scanning (ALS) to spaceborne or airborne multispectral or hyperspectral sensors, etc. Remote sensing is a science to obtain information of objects without making any physical contact with it, typically from aircraft or satellites. In particular, this thesis focused on two remotely sensed data sources that at the moment seem to be the most promising for abovementioned purposes: ALS and airborne hyperspectral data. Their combined use or fusion can be beneficial as they provide a complementary information for characterizing forest attributes. ALS and hyperspectral technologies provide very high spatial resolution allowing us to map the forest attributes at a very high spatial detail. This can be useful for certain applications but increasing the spatial detail does not always improve the accuracy of the predictions. Indeed, many predicted forest characteristics can be explored at many spatial scales, e.g. from tree to stand. Thus, the major objective of this thesis was to evaluate the potential of fusing ALS and hyperspectral data for the prediction of forest characteristics and to evaluate the benefits of different spatial details in the prediction of such characteristics. The fusion of ALS and hyperspectral data and the spatial scale exploration were carried out simultaneously in this thesis, and in particular it started with a focus on the spatial scale (development of a new ITC delineation algorithm) and it finished with a focus only on data fusion (prediction of forest structural diversity measures).
The ALS and hyperspectral data were fused at two different levels, product and variable-level fusion. The product-level fusion was used for the prediction of the site index and species-specific volume, while the variable-level fusion was used for total and species-specific volume, as well as structural diversity measures. For the evaluation of different spatial details in the prediction of forest characteristics we considered three remotely sensed-based inventory approaches, namely the individual tree crown (ITC) approach, the semi-ITC approach, and the area-based approach (ABA). In order to apply the ITC and semi-ITC approaches, the individual tree delineation algorithm was needed and developed based on the ALS point cloud. The forest characteristics evaluated in this thesis were: individual tree attributes (such as tree height, DBH, stem volume, age, and species), forest attributes (such as site index, total and species-specific volume), and forest structural diversity measures.
The ITC approach allowed an accurate determination of the height, species, DBH, and stem volume, while the age was subject to a greater error. The ITC approach for site index determination in combination with ALS and hyperspectral data was found to be an efficient and a stable procedure and it reached similar accuracy as in the existing site index maps based on field surveys. For species-specific volume, the ITC approach reached high accuracies but there were also large systematic errors for minority species. For majority species, the semi-ITC approach resulted in slightly higher accuracies and smaller systematic errors compared to ABA. In all three approaches, ALS and hyperspectral data were important to provide higher accuracies. The fusion of ALS and hyperspectral data for forest structural diversity measures did not improve their accuracy but produced accuracy levels comparable to the models built on ALS data alone, except for one measure. In these experiments, ALS data showed the best predictions for the majority of the structural diversity measures taken into account.
To conclude, the ITC and semi-ITC approaches can provide higher spatial detail of the predicted forest characteristics. This information can also be aggregated to coarser scales, e.g. stands. The use of ITC and semi-ITC approaches has a potential in different forestry and ecology applications, where the accuracy of the semi-ITC also showed the capacity in operational forest applications. The fusion of ALS and hyperspectral data improves the predictions of forest characteristics, such as volumes and site index, while for some forest structural diversity measures the fusion did not improve the accuracy of results. Fusion of such data, especially for structural diversity measures has to be further explored.
Main supervisor: Researcher, Dr. Hans Ole Ørka, MINA-NMBU
Co-supervisors: Dr. Michele Dalponte, Fondazione Edmund Mach, Italy
Professor, Dr. Erik Næsset, MINA-NMBU
First opponent: Associate Professor, Dr. Valerie A. Thomas, Virginia Polytechnic Institute and State University, USA
Second opponent: Associate Professor, Dr. Petteri Packalén, University of Eastern Finland, Finland
Committee coordinator: Professor Hans Fredrik Hoen, MINA-NMBU
[bibtex file=SkogRoverAll.bib key=Kandare2017]