Vegetation Mapping Using Remote Sensing and Deep Learning Learning from Sparse and Incomplete Labels at Scale

PhD defence by Hui Zhang

Assessment Committee

Associate Professor Laura Vang Rasmussen, Geosciences and Natural Resource Management, University of Copenhagen (Chairperson)
Adjunct Professor Compton J. Tucker, University of Maryland, Baltimore, CO, USA
Associate Professor Andrea Nascetti, KTH Royal Institute of Technology in Stockholm, Sweden

Supervisors

Professor Christian Igel
Postdoc Nico Lang
Assistant Professor Stefan Oehmcke

Department

Department of Computer Science

Place

The defence is conducted as a hybrid defence.

To attend the defence in person:
Building: Pioneer Centre for AI, Room: Seminar Room (OEV3-Seminar),
Øster Voldgade 3, 1350 København

To attend the defence online:
Please follow the link to attend the defence online: https://ucph-ku.zoom.us/j/65119104961
MeetingID, if relevant: 651 1910 4961
Password, if relevant: 456294

Email address to gain access to the thesis: huzh@di.ku.dk.
You will either receive a copy of the thesis or be informed where you can read a physical copy.
Recipients of copies of the thesis are not allowed to share or distribute it due to copyright compliance.

Short description of the thesis

How much can we learn about vegetation from sparse observations?
This thesis explores how deep learning can turn sparse and incomplete reference data into dense, detailed vegetation maps. Across three studies, it progresses from mapping urban tree cover with incomplete point labels to reconstructing three-dimensional vegetation structure across Europe and globally using satellite imagery and sparse LiDAR. The resulting maps capture ecologically relevant information beyond canopy height and, at global scale, provide calibrated prediction uncertainty.