NAISS
SUPR
NAISS Projects
SUPR
Deep learning segmentation of historical urban tree canopy in Swedish cities
Dnr:

NAISS 2026/4-1108

Type:

NAISS Small

Principal Investigator:

Blaz Klobucar

Affiliation:

Sveriges lantbruksuniversitet

Start Date:

2026-06-08

End Date:

2027-07-01

Primary Classification:

40505: Landscape Architecture

Allocation

Abstract

This project develops a deep learning pipeline for mapping urban tree canopy cover change from the 1960s to present using historical panchromatic aerial imagery of Malmö, Gothenburg and Stockholm. We fine-tune BWTreeNet, a U-Net based segmentation architecture, on Swedish aerial imagery paired with manually annotated canopy labels across multiple decades. GPU resources are required for iterative model training experiments and city-scale inference across multiple time periods. The project is funded by the Swedish University of Agricultural Sciences (SLU) until end of 2027.