The Latest In Utilizing
Machine Learning to
Complete Tree Inventories
Josh Behounek

“Solutions Through Innovation and Expertise”

1931
Pen & Paper Computer
1992
GIS-based
2002
Machine learning
2022

Tree Inventory vs Assessment
Assessments
Google Streets
DRG Standard
LiDAR

Current Process

Buffalo, NY Inventory Update
2001 2014 Difference
Sites 124,445 127,080 2,635
Total DBH 871,173” 817,627” -53,546”
Average
DBH
7” 6” -1”
# Species 281 247 -34
# Removals 668 2,707 2,039
# Planting
Sites
48,761 44,619 -4,142

Top 5 Species Still Poor Recommended for Removal
45, 43%
24, 23%
20, 19%
9, 9%
6, 6%
MAPLE, NORWAY
MAPLE, SILVER
LINDEN, LITTLELEAF
HORSECHESTNUT, COMMON
CHERRY, KWANZAN
162 total trees

Condition Change Assessment
2001
Inventory
2014
Inventory
5,698
Poor
3 Dead
145 Poor
11 Fair
0 Good
293 Plant
38,199
Fair
298 Dead
3,445 Poor
26,830 Fair
962 Good
1,365 Plant
25,632
Good
259 Dead
717 Poor
8,952 Fair
9,783 Good
1,878 Plant

3 Types of
Machine Learning &
Tree Inventories
✔ Nothing
✔ Inherited
❏ Current

Machine Learning Process with Google Streets
Geo-localization of tree
canopy (Step 1)
● Aerial imagery is
used to identify
where trees are.
● Canopy pixels are
extracted and
vectorized to define
the boundary called
the tree canopy
zone.
Estimating tree
count (Step 2)
● Within the tree
canopy zone, street
view imagery is used
to find the trees
under street view
Estimating distance
from observer (Step 3)
● A heat map is generated
that defines the distance
of each pixel from the
observer.
● Using this, the average
distance of tree pixels is
calculated within the
bounding box extracted
in step 2
Identifying location of
individual trees (Step 4)
● Observer location and
field of view is projected
in aerial view (the right
angle in blue above)
● Using the distance
calculated in step 3,
individual trees are
placed on aerial image
map (yellow points).
Photo credit - SiteRecon

1. No Idea of Number of Trees to be Inventoried

Scenario 1: No Idea # of Trees – Tucson, AZ

2. Inherited

Inherited - San Diego Results
26% Vacant
Sites
~ 436,770 Total Sites

Updating Missing Trees

Updating Removals

Utilizing Point Data
1. Number of Trees
2. Location of Trees
3. TreeKeeper Software
4. Tree Equity
5. Pruning Cycles
6. Planting Locations
7. Updating

Implementing an
Urban Forestry Mapping &
Monitoring Program

Advantages
Photo credit - greehill

Implementing Tree Monitoring Program
Year 1
Initiate tree monitoring
program
Perform advanced
assessments
Install TreeKeeper 9
Year 2
Implement information
via TreeKeeper 9
Year 3
Perform tree
monitoring data
collection
Perform advanced
assessments of
flagged trees
Perform change
analysis
Update TreeKeeper 9
Year 5
Perform tree
monitoring data
collection
Perform advanced
assessments of
flagged trees
Perform change
analysis
Update TreeKeeper 9
Year 4
Implement information
via TreeKeeper 9
Photo credit - greehill

Initial assessment
greehill drives streets &
parks per contract specs
Data Delivery
Data is delivered into
TreeKeeper 9 with API to
greehill software
Data extraction
Data is processed via
machine learning to provide
information per data specs.
Advanced Assessments
Davey provides Level 2
assessments to flagged trees.
Outlier Trees
Based on results of data,
client goals, & budget a
certain # of trees are
identified for advanced
assessments
Tree Monitoring
Program
Operation workflow

Corridor clearance
Human Thermal Comfort

Machine Learning
Advantages
● Objective
● Repeatable
● Efficient
● Precise

“You do not rise to the level of your
goals, you fall to the level of your
systems.”
Atomic Habits by James Clear

The Latest In Utilizing
Machine Learning to
Complete Tree Inventories
Josh Behounek
573-673-7530
Josh.Behounek@davey.com