The major research question of the project is: How should the future multitemporal, multispectral mobile laser scanning (MLS) data be computationally processed to provide timely information for environmental sustainability and especially for mapping of the forest health, tree species classification, mapping of dead trees, and forest fire risk. The sub-objectives include: 1) We conduct pioneering multispectral MLS measurements and case studies for next-generation forest data, 2) We apply novel computational methods utilizing both spectral and geometric features in MLS point cloud analysis to improve estimation and prediction for tree species, forest health and forest risk management, 3) We take the international collaboration into account and accomplish a global benchmarking study of new computational methods, especially for tree species and dead wood, inside the project.

Climate change is causing great threat to the boreal forests. We propose a methodology that integrates the latest innovations in drones, hyperspectral (HS) imaging, and machine learning to implement an efficient and precise framework for forest health monitoring. To solve the problem of generating extensive labeled training datasets for deep learning, we propose a novel approach producing simulated HS drone image datasets of forests with selected stress factors and using those to train machine learning models for vegetation analysis. We will use the method to optimize the drone procedures in forest health analysis, use simulated data in transfer learning, and validate the results using the existing and new in-situ datasets collected using drone systems flying above and inside of forests. We believe that the proposed approach will result in a breakthrough in usability of machine learning methods in drone and HS imaging based forest health and disturbance analysis.