Comparing ALS- and Deep-Learning-Derived Urban Tree-Canopy Maps: An Uncertainty-Aware Framework for Green Infrastructure Planning

Urban tree canopy information is increasingly used to support climate adaptation and urban green infrastructure (UGI) planning. However, operational remote sensing approaches can produce substantially different canopy estimates. This study presents an uncertainty-aware framework comparing canopy extracted from an official airborne laser scanning (ALS)-derived normalised surface height product (nDMP) with canopy delineated from very-high-resolution (VHR) orthophotos using a pre-trained deep-learning (DL) model without local fine-tuning. ALS canopy candidates were identified using a minimum height threshold of 5 m, combined with local height standard deviation above 0.3 m or height above 12 m, followed by additional filtering. Spatial agreement and source-specific detections were analysed across designated land-use categories, district communities, and 1079 hexagonal units in Ljubljana, Slovenia. The ALS-derived product mapped 25.08% canopy cover, compared with 12.24% for the deep-learning product. Despite strong hexagon-level correlation (r = 0.872), numerical disagreement remained substantial (mean difference = 12.49 percentage points; MAE = 13.88 pp). Independent stratified validation of 400 locations yielded area-adjusted overall accuracies of 95.16% for ALS and 85.01% for deep learning, with canopy recall of 87.60% and 44.45%, respectively, against year-matched visual references. Municipal tree-inventory assessment further indicated higher positive-location correspondence for ALS (94.3%) than for deep learning (73.1%), without establishing individual-tree detection accuracy. The framework supports UGI planning by identifying comparatively low-canopy areas, revealing source-dependent uncertainty, and prioritising locations for further verification before planning interventions.

Authors

Institutions

Publication Details

Journal
Land
Published
2026-11-18
DOI
https://doi.org/10.3390/land15112208
Primary Topic
Remote Sensing and LiDAR Applications
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Comparing ALS- and Deep-Learning-Derived Urban Tree-Canopy Maps: An Uncertainty-Aware Framework for Green Infrastructure Planning

Krištof Oštir, Ana Potočnik Buhvald
4 links
Land
Remote Sensing and LiDAR Applications
article

Comparing ALS- and Deep-Learning-Derived Urban Tree-Canopy Maps: An Uncertainty-Aware Framework for Green Infrastructure Planning

Krištof Oštir, Ana Potočnik Buhvald
article en

Abstract

Urban tree canopy information is increasingly used to support climate adaptation and urban green infrastructure (UGI) planning. However, operational remote sensing approaches can produce substantially different canopy estimates. This study presents an uncertainty-aware framework comparing canopy extracted from an official airborne laser scanning (ALS)-derived normalised surface height product (nDMP) with canopy delineated from very-high-resolution (VHR) orthophotos using a pre-trained deep-learning (DL) model without local fine-tuning. ALS canopy candidates were identified using a minimum height threshold of 5 m, combined with local height standard deviation above 0.3 m or height above 12 m, followed by additional filtering. Spatial agreement and source-specific detections were analysed across designated land-use categories, district communities, and 1079 hexagonal units in Ljubljana, Slovenia. The ALS-derived product mapped 25.08% canopy cover, compared with 12.24% for the deep-learning product. Despite strong hexagon-level correlation (r = 0.872), numerical disagreement remained substantial (mean difference = 12.49 percentage points; MAE = 13.88 pp). Independent stratified validation of 400 locations yielded area-adjusted overall accuracies of 95.16% for ALS and 85.01% for deep learning, with canopy recall of 87.60% and 44.45%, respectively, against year-matched visual references. Municipal tree-inventory assessment further indicated higher positive-location correspondence for ALS (94.3%) than for deep learning (73.1%), without establishing individual-tree detection accuracy. The framework supports UGI planning by identifying comparatively low-canopy areas, revealing source-dependent uncertainty, and prioritising locations for further verification before planning interventions.

LandVol. 15(11)
University of Ljubljana (SI)
Javna Agencija za Raziskovalno Dejavnost RS
Sustainable cities and communities, Climate action
Remote Sensing and LiDAR Applications
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon