1. Preparation of a tool for automatic detection of dead trees
in selected forests in Luxembourg using remote sensing
(RS4DeadTrees 1.0)
Technical and legal aspects of flying a drone (UAV)
Karolina Zięba-Kulawik1, Konrad Skoczylas1, Piotr Wężyk2
Luxembourg Institute of Socio-Economic Research (LISER)
1Luxembourg Institute of Socio-Economic Research, 2Faculty of Forestry University of Agriculture in Krakow
3. Introduction - Dead trees as a burning topic for decision-makers,
politicians and local communities
Based on European monitoring plots, an estimated 27%
of forest trees were moderately, severely defoliated,
or dead (State of Europe’s Forests, 2020).
Photo: K. Zięba-Kulawik
6/15/2023 Seed Grant - ccEXPAR 3
Causes: extreme droughts and heat waves, widespread
bark beetle outbreaks and a wider occurrence of forest
fires.
Forest map of Europe based on remote sensing and forest
inventory statistics (European Forest Institute; EFI).
227 mln ha
Photo: M. Ratajczak
Photo: K. Zięba-Kulawik
4. Introduction - Situation of trees in Luxembourg's forests
The results of the 2022 phytosanitary inventory Luxembourg's forests
Source: Administration de la nature et des forêts: „Les résultats de l’inventaire
phytosanitaire 2021 des forêts du Luxembourg”.
Photo: K. Skoczylas
According to studies carried out in Luxembourg by The
Nature and Forest Agency on the state of the forests, it
was estimated that around 61% of trees in 2022 were
significantly damaged, heavily damaged, or dead. The
inventory was conducted on only 1200 trees by five
forest experts using traditional methods.
All species combined, during the summer of 2022:
15.40% of the trees show no damage (class: 0),
22.90% of the trees are slightly damaged (class: 1),
61.70% of the trees are seriously damaged
(generally dying) or dead trees (class: 2, 3 and 4).
6/15/2023 Seed Grant - ccEXPAR 4
Traditional forest inventory
Mobile Reporter/ www.lessentiel.lu
A falling tree injures a child's head in Diekirch (Luxembourg):
5. Introduction - Remote sensing technologies in forests monitoring
Unmanned Aerial Vehicle (UAV)
Satellite imagery Aerial images and LiDAR
6/15/2023 Seed Grant - ccEXPAR
PLANETSCOPE (3 m)
LANDSAT 8 (30 m)
Whirlwind path in the Tuchola Forest.
Fot. Kacper Kowalski / FORUM.
Satellite Systems for Environmental Monitoring
LiDAR point cloud
Orthophoto RGB, CIR
6. 6/15/2023 Seed Grant - ccEXPAR 6
The main goals of the project
development of an algorithm for the semi-automatic detection of
standing dead trees (SDT) based on aerial remote sensing data;
assessment of the risk posed by dead trees standing near of routes
for a selected municipality in Luxembourg;
creation of maps with the location of dead trees to help manage
these areas.
Additional objectives:
testing of point clouds to extract selected tree parameters, which can
help estimate the volume of lost trees;
Investigating to what extent dead stands influence land surface
temperature changes and the formation of local forest heat islands.
RS4DeadTrees 1.0
7. Source: Zieba-Kulawik /LandCover, CBK (PL), ESA; 2017.
Study area - municipality of Bourscheid
6/15/2023 Seed Grant - ccEXPAR 7
Forests (50%)
Roads/trails (280 km)
Photo: K. Zięba-Kulawik
8. Data - RGB and CIR aerial images (ortophotomaps 2021; 10cm)
10.07.2023 Seed Grant - ccEXPAR 8
RGB CIR
Dead
trees
9. 10.07.2023 Seed Grant - ccEXPAR 9
Airborne Laser Scanning (ALS) point cloud height models
LiDAR 2019 - nationwide data coverage
LiDAR 2019 - Color point by RGB
Cross section - color point by class
LiDAR 2019
LiDAR 2017
DSM
DTM
DSM - Digital
Surface Model
DTM - Digital
Terrain Model
H = 28,14 m
point cloud converted into voxels
ALS - color points by class
10. Sensor Type
Hasselblad 4/3" CMOS Sensor;
3-Axis Gimbal with Dual Cameras
20MP 5.1K Wide-Angle
12MP Telephoto with 28x Hybrid Zoom
Sensor Resolution Effective: 20 Megapixel (5280 x 3956)
Min/Max Aperture f/2.8 - f/11
ISO Sensitivity
Video: 100 to 6400
Photo: 100 to 6400
6/15/2023 Seed Grant - ccEXPAR 10
UAV - DJI MAVIC-3 CINE
Maximum Takeoff Weight 899 g
Maximum Horizontal Speed
11.2 mph / 5 m/s (C Mode)
33.6 mph / 15 m/s (N Mode)
47.0 mph / 21 m/s (S Mode)
Maximum Ascent Speed 17.9 mph / 8 m/s
Maximum Descent Speed 13.4 mph / 6 m/s
Flight Ceiling 3.7 Miles / 6000 m
Maximum Flight Time 46 Minutes
Maximum Hover Time 40 Minutes
Platform
Camera
11. Methods – Geographic Object-Based Image Analysis (GEOBIA)
eCognition algorithms using machine learning
CIR orthophoto
GEOBIA
segmentation
and classification
Layer with dead trees
Export of results
Dead trees layer
in GIS environment
Dead trees GEOBIA workflow
12. examples of segmentation
GEOBIA Step 1 - Segmentation
eCognition software - Rule-set
6/15/2023 Seed Grant - ccEXPAR 12
Segmentation:
- input layers:
three orthophoto bands:
infrared, red and green,
- algorithm: multiresolution
Segmentation with parameters:
Scale=7, Shape=0.2, Compactness=0.7,
- merging of adjacent segments
with spectral difference DN<10.
13. GEOBIA Step 2 - Classification
6/15/2023 Seed Grant - ccEXPAR 13
Classification:
- mean value in the infrared
channel: SP>=40 and SP<=150,
- average value in red channel:
SC>=80 and SC<=175,
- NDVI vegetation index<0.
Export of results
to GIS environment
NIR
red
green
14. Drone and camera images to assess classification accuracy
6/15/2023 Seed Grant - ccEXPAR 14
Verification of results in the field –
„Ground truth”
150 photos (drone) + 60 photos (camera)
15. Accuracy Assesment - stratified random points
260 points randomly distributed within each class (proportional to its relative area)
6/15/2023 Seed Grant - ccEXPAR 15
Accuracy
Assesment
points:
50 points:
photos from
drone
50 points:
photos from
camera
160 points:
randomly
distributed
=
260 points
16. 6/15/2023 Seed Grant - ccEXPAR 16
Results - Classification of dead-tree regions
The confusion matrix for the classification results (1 – dead trees,
0 – other) for the verification dataset (no. of pixels).
79 4 83
10 167 177
89 171 260
94,61%
0,88
88,7% 95,1%
dead trees
detected
dead trees
not
detected
Example where several dead trees were not detected
by the algorithm:
Overall accuracy - what proportion of all reference locations have been correctly mapped.
User's accuracy - how often the class on the map will actually be present on the ground.
17. Results - maps with locations of dead trees in Bourscheid
6/15/2023 Seed Grant - ccEXPAR 17
Sections of municipalities Dead trees [ha]
Michelau 7,6
Lipperscheid 7,5
Bourscheid 6,8
Kehmen et Scheidel 3,3
Welscheid 5,1
Schlindermanderscheid 6,3
SUM 36,6 ha
Dead trees area =
36.6 ha
(10% of coniferous
trees in the
municipality)
18. 6/15/2023 Seed Grant - ccEXPAR 18
Results - trails with the highest risk
Dangerous paths where dead trees may fall (marked in red).
In the municipality of Bourscheid, 56 kilometres of
trails have been designated at risk due to the
possible fall of dead trees (20% of all trails: roads,
cycle routes and footpaths in the municipality).
19. 6/15/2023 Seed Grant - ccEXPAR 19
Results - trails with the risk (tree height + buffer 10m)
85 km of trails at risk of falling trees (30% of the trails in the municipality)
20. Future of the project?
"Let'z Forest" mobile application + Citizen Science
- Free mobile application with map of dead trees in Luxembourg and danger areas on the trail;
- Topographic and satellite map and trail layer for online and offline use;
- Info about the current position (GPS coordinates), which can be quickly sent by SMS, e-mail.
6/15/2023 Seed Grant - ccEXPAR 20
21. Volume estimation of dead trees - preliminary results
3D point cloud
6/15/2023 Seed Grant - ccEXPAR 21
point cloud Voxel 50cm =
150m3
Same place in 2019 and 2021 –
significant progression of tree dieback noted on orthophotomaps
2019 2021
2022
Volume of voxels – tree no. 1
Example tree
No. 1
Based on voxels
created on the 2019
ALS point cloud, the
volume of the group
of dead trees was
estimated at
11 253 m3.
Group of dead trees on MLS 2022 point cloud
22. 6/15/2023 Seed Grant - ccEXPAR 22
Land Surface Temperature (LANDSAT-8; 30m)
Do areas with dead trees affect the growth of heat islands?
23. Highlights
Currently there are no solutions available for the generation of maps with
dead trees in Luxembourg
• We developed and evaluated tree inventory methods of detecting dead
trees.
• We evaluate the potential dead tree risk assessment to communication
routes movement.
• We evaluated remote sensing contribution to the potential risk
assessment.
• We provide needed and sophisticated data for forest and tourism
management.
6/15/2023 Seed Grant - ccEXPAR 23
RS4DeadTrees 1.0