Oakville, Ontario
Hyperspectral Imagery Analysis
for EAB Management
Ian Hanou
AMEC Earth & Environmental
[email protected]
(303) 503-4846
Wanted: Tree Killers – Using Hyperspectral
Technology for Beetles and Borers
Ian Hanou
AMEC Earth &
Environmental, Inc.
Ian Brown
City of Milwaukee
Image courtesy of
RFP Mapping, Inc.
EAB CONFIRMED IN WISCONSIN
Oakville Project Overview
Purpose/Scope: aide Oakville in EAB management by
collecting hyperspectral imagery, field spectra, and analyzing
the data to produce GIS locations of ash (Fraxinus) and
evaluate the detection of EAB stressed trees
Deliverables:
• Ash locations and potential EAB infestation in GIS format
• Imagery: hyperspectral and 4-band imagery
• 2010 Urban Tree Canopy (UTC) Assessment (compare w/2005)
• Report, metadata, accuracy assessment and PowerPoint
Components:
• Imagery Acquisition, Field Work, Iterations of Classification of ash and
EAB Infestation levels, Validations/Accuracy Assessment, Reporting
AMEC Earth & Environmental
Multispectral
Hyperspectral
HSI Technology Overview
• “Narrow band mapping”
• High Spectral Resolution: The
measure of a remote sensing
instrument‟s power to resolve
features in the electromagnetic
spectrum.
AMEC Earth & Environmental
Frequently Asked Questions
• Collection timeline and support required
• What affects cost
• What image resolution is required (spatial and spectral)
• Satellite or aerial-based
• Is field work needed, can we do it
• Can HSI replace traditional inventories
• Will HSI overlay with existing GIS data and aerials
• Can we assess tree stress, to what extent
• Other species and other uses of HSI
Field Campaign Overview
Dates: 19-22 July 2010
Area: 5 sites, 40 trees
Equipment: ASD Spectrometer
Purpose: gather spectral signatures on ash,
known EAB trees and background features.
AMEC Earth & Environmental
Field Spectra Signature
Collection
AMEC Earth & Environmental
IntTerra
Collection of spectral signatures using hand-held
spectrometer.
Reflectance
(amount of
light reflected
in each
wavelength)
BLUE – Construction Asphalt
SEA GREEN – Sage Brush
GREEN – Lawn Grass
ORANGE – Brown silty loamy soil
YELLOW – Pinyon Pine
Example Spectra from Spectral Libraries
Assessment of Spectra
AMEC Earth & Environmental
IntTerra
Ash and background spectra similar to spectra
observed in Milwaukee and Chicago.
• Same latitude and conditions
• Similar background species
WHITE – Oakville Spectra
GREEN – Milwaukee Spectra
AMEC Earth & Environmental
IntTerra
Assessment of Spectra - EAB
Direct Correlation of EAB Manifestation
RED – HIGH EAB
BLUE – MEDIUM EAB
YELLOW – LOW EAB
Note – ‘DOTTED’ spectra are additional spectra collected during offhours/rain. Included to demonstrate that overall trend still apparent
Field Collection Summary
Field campaign (handheld spectrometer):
• Confirms spectral response of Ash vs. Background as
similar to projects in Wisconsin and Illinois
• Supports potential for correlation between SWIR spectral
response and EAB infestation based on high vs. low
Next Step:
• Exploration of airborne spectral data to determine if
separation and correlation are evident
AMEC Earth & Environmental
IntTerra
Hyperspectral Collection Specs
Aerial Collection
• Collection days: 7/22, 7/26,
7/29, 8/1
• Flight Lines: 54
• Nominal Parameters: 1m
GSD, 30% overlap, 360 spectral
bands
Data Processing
• Radiance and Reflectance
• Georeferencing Information
AMEC Earth & Environmental
IntTerra
FEATURE EXPLOITATION
Hyperspectral Exploitation
HSI Grayscale
Ash Classification
(Note: Milwaukee data, not Oakville)
Initial Ash Analysis
2nd Interim Data Results
Flight Lines 06, 30, 31, 32, 41, 42, 43
AMEC Earth & Environmental
IntTerra
Ash Detection
31 Oct Methodology
Improved
Separation of Ash
Trees
5 Nov Methodology
28 Nov Methodology
Jan Methodology
GREEN – Ash
RED – Honey Locust
BLUE – Silver Maple
Small ash trees, park
background
AMEC Earth & Environmental
IntTerra
Ash Detection
Improved
Separation
of Honey
Locust
31 Oct Methodology
5 Nov Methodology
GREEN – Ash
RED – Honey Locust
BLUE – Silver Maple
Honey Locust,
most likely „false
alarm‟ species
AMEC Earth & Environmental
IntTerra
Ash Detection
31 Oct Methodology
5 Nov Methodology
28 Nov Methodology
Jan Methodology
GREEN – Ash
RED – Honey Locust
BLUE – Silver Maple
Improved
Separation of Ash
Trees
Additional Field Data: Training and Validation
AMEC Earth & Environmental
Field / Inventory
Validation
• Assessed several,
widespread locations
for accuracy in the
field
• Additionally, checked
classified results with
nearly 800 randomly
selected tree inventory
points
Accuracy Assessment
Process
- Ground Truth represented by Vector Layers
- Oakville Field Work
- Random Extract of Street Tree from buffered Crown Width
- Automated Comparison with Ash Detections
Observations
- Canopy Size & Background
- Understory (street vs tree/grass)
- Size and Detection (visual, database, overhead)
- Geospatial Offset (imagery registration, GPS and database
location)
Automated Accuracy Assessment
Overall User
Classified
Field
ASH
OTHER TOTAL
364
196
ASH
OTHER
Column total
560
116
948
1064
Un-weighted
77%
79%
Overall Producer
480
1144
Overall User
1624
Overall Producer
Sum of Correctly
Classified
Total Trees
1312.00
Overall Accuracy
81%
Weighted
82%
81%
1624.00
Species False Detects
Red Oak – 20/62
Silver Maple – 25/89
Norway Maple – 25/119
Honey Locust – 25/168
Accuracy Assessment Issues: Canopy Size and Background
Accuracy Assessment Issues: Geospatial Offset
EAB Hot-Spots
Heat map of Ash density
based on spatial
distribution of ash
detections from
hyperspectral imagery
AMEC Earth & Environmental
IntTerra
AMEC Earth & Environmental
EAB Infestation Spectra
RED – HIGH EAB
BLUE – MEDIUM EAB
YELLOW – LOW EAB
Gallery Count Field Data
Levels of Infestation
Canada's version
of field
From spectra name designation
Sample Number
(See Shape
Files and GPS
Field Designation
Logs)
(on-site)
M1
M2
1
Flower Gall
2 or 3 ?
Blank (2) or Treated (3)
4 or 5 ?
4 or 5 ?
4 or 5 ?
6
Treated
7
Blank
8
H
9
M
10
M
11
H
12
H
13
M
14
L
15 ?
M
EAB level of
infestation
N/A
N/A
N/A
N/A
N/A
N/A
N/A
N/A
N/A
N/A
M
H
H
H
M
L
M
NEW
ASSESSMENT
Infestation level based on
gallery count
NIL
NIL
M
H
H
M
H
NIL
H
H
H
H
L
M
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
Date
Address
19/07/2010
19/07/2010
19/07/2010
19/07/2010
19/07/2010
19/07/2010
19/07/2010
19/07/2010
19/07/2010
19/07/2010
20/07/2010
20/07/2010
20/07/2010
20/07/2010
20/07/2010
20/07/2010
20/07/2010
Golden Meadow Park
Golden Meadow Park
Golden Meadow Park
Golden Meadow Park
1372 Golden Meadow Trail
1371 Golden Meadow Trail
1376 Golden Meadow Trail
1373 Golden Meadow Trail
1385 Golden Meadow Trail
1404 White Oaks Blv.
1396 White Oaks Blv.
1392 White Oaks Blv.
1388 White Oaks Blv.
1384 White Oaks Blv.
1380 White Oaks Blv.
362 Erin St.
1395 White Oaks Blv.
Oakville's ID
#
91076
1365
105181
86833
97951
31192
12505
27348
88759
84564
71591
105623
111234
18989
47503
Location
605581 - 4814608
605582 - 4814613
605602 - 4814603
605607 - 4814597
605550 - 4814448
605546 - 4814436
605577 - 4814441
605573 - 4814429
605643 - 4814447
605512 - 4814170
605525 - 4814147
605535 - 4814126
605541 - 4814114
605549 - 4814101
605556 - 4814092
605582 - 4814070
605531 - 4814172
Initial EAB Infestation
•
Process/Method:
- Only applied to ash trees by buffering street/park trees and merging
with HSI-based ash layer
- Man-made features and other signatures produce spurious results
•
Results/Observations:
- Upper Index (black labels) = Gallery Count / Area (square meter)
directly from spreadsheet provided
- With this index, the cutoff appears between 10-50
- Lower Index (purple labels) = Total Gallery Count (different number in
spreadsheet from above) / Area (based on dbh)
- With this index, the cutoff appears to be around 100
- With dbh index, some initial higher values decrease with influence of
larger dbh and some medium values increase due to influence of
smaller dbh.
AMEC Earth & Environmental
IntTerra
EAB Infestation
Red = high EAB infestation Blue = no or low EAB infestation
EAB Infestation
Red = high EAB infestation Blue = no or low EAB infestation
Comparing Milwaukee and Oakville Projects
• No LiDAR data, slightly different delivery format
• Similar ash classification accuracy (slightly >80%)
• Both provided an opportunity to assess UTC
• More false alarms in Milwaukee but higher capture
of ash
• No door-to-door verification yet in Oakville
• EAB stress not tested in Milwaukee (no known
locations of EAB yet)
• Public dissemination via website in Oakville
AMEC Earth & Environmental
IntTerra
What’s Next with Hyperspectral / Conclusions
- Need more studies, new species and locations to
-
-
evaluate accuracy
- Collection of spectral signatures would identify
which species can be mapped accurately and
what image specifications are needed
Cost of collection is high for city forestry
applications but cost-effective compared to an onthe-ground approach
I.D. Partners (invasive species, water quality, etc)
Map multiple species
Do partial surveys and/or citizen web-inventories
AMEC Earth & Environmental
IntTerra
Thank You. Contact:
Ian Hanou
AMEC Earth & Environmental
[email protected]
(303) 742-5320
(303) 503-4846