John A' Glaser, Ph'D'

1
Large Area Monitoring for Pesticidal Transgenic
CropsHow Spectral Imaging May Play a Role

• John A. Glaser, Ph.D.
• Biotechnology Research Program Leader
• National Risk Management Research Laboratory
• EPA/OSP Remote Sensing Workshop
• Valdas V. Adamkus Environmental Resource Center
• Lake Michigan Room, 12th Floor
• U.S. EPA Region 5 Office
• 77 W Jackson Boulevard
• Chicago, IL  60604
• November 1 3, 2005

2
(No Transcript)
3
Valuation

• Bt corn is viewed as an environmental asset due
  to the possible avoidance of pesticide
  applications (human health and ecosystem related)
• Hence, its lifetime is important to
  environmentally sustainable considerations.

4
FIFRA

• FIFRA Section 3 requires registration of all
  pesticides
• Registration evaluates
• Human Health
• Ecological Effects
• Plant Incorporated Protectant (PIP) Crops

5
Best Management Criteria

• IRM plan
• Structured refuge
• Secondary pest impacts
• Multiple crop pest impacts
• Monitoring and surveillance
• Grower education
• Coordinated remedial action

6
(No Transcript)
7
Existing Monitoring Strategy
Percent total corn acreage planted to Bt corn in
counties with gt 50,000 total acres of corn
planted (Bt corn industry sales data as compiled
by FSI, Inc., 1999)
8
Current Monitoring Strategy Concerns

• Four limited sections of corn crop used as proxy
  for entire crop.
• Infestations for the corn pests are expected to
  begin as local phenomena.
• Do current proxy samplings provide adequate
  information for resistance detection?

9
How is a proactive monitoring approach
developed?

• Representative sampling of all acreage not
  physically possible
• Current monitoring strategy provides insufficient
  warning of resistance development
• No direct molecular technique is currently
  available to detect resistance
• Need a comprehensive approach to reduce reporting
  time for resistance monitoring

10
(No Transcript)
11
Infrared Photography
Color Infrared Photograph
Unsupervised Classification
Imagery provided by
12
Questions

• Can transgenic corn varieties be distinguished
  from their nearest isolines by spectral
  reflectance imagery?
• Can pest infestations in corn be identified by
  reflectance imagery?

13
Real Time Data Acquisition Camera
System-Hyperspectral (RDACSH)
Cessna 210 used for data collection
In-plane Setup
2004 Spectral Resolution 473 nm to
827 nm
(120 bands) Spatial Resolution
.5 meters Array 640
pixels across, 320 meter coverage 2005 Spectral
Resolution 400 nm to 1,000 nm

(300 bands) Spatial Resolution
.5 meters Array 1,600 pixels
across, 800 meter coverage
Camera
Prism-grating filter
14
RDACSH3 and Radiometer Signatures Same Locations
15
FY 04 Ground TruthingResearch
Test Set
Designed Plantings
16
Selected Corn Hybrids

• Cry1Ab hybrids
• NK N60-B6    107d Bt11
• NK         N60-N2    107d  near iso
• DeKalb DK58-78 YGCB 108d
  Mon810
• DeKalb DK57-01 107d near iso
• Pioneer P34N44 110d Mon810
• Pioneer P34N43 110d near iso
• Cry1F hybrids
• Pioneer P34N42 111d Tc1507
• Pioneer P34N43 110d near iso
• Cry3Bb1 hybrids
• DeKalb    DKC 60-12    110d     Mon863
• DeKalb DKC 60-15     110d    near iso

17
Experimental Design

• Complete random block design replicated five
  times
• Plots 4 rows by 30 ft.
• 30 ft buffer between plot sets
• Two infestations with European Corn Borer
• 1st Infestation at V8-V10 growth stages
• Damage assessed by Guthrie rating
• Imaging of all plots in a two week schedule
• 7 imaging events starting June 22, 2004

18
Corn Growth Stages
Vegetative Stages Emergence VE 1st leaf V1
2nd leaf V2 3rd leaf V3 Nth leaf Vn Tasseling VT
Reproductive Stages Silking R1 Blister R2 Milk
R3 Dough R4 Dent R5 Physiological maturity R6
Iowa State Univ
19
Reflectance Differences for Hybrids Across Time
Box Plots for Percent Reflectance in a Given
Wavelength
June 22
July 12
July 25
August 5
Means
Reflectance at 800 nm
Standard Deviation
58-78
Dekalb Hybrid
57-01
58-78
57-01
58-78
57-01
58-78
57-01
Dekalb 58-78 (Cry1Ab)
Increasing reflectance
Dekalb 57-01 (Near-isoline)
Reflectance of Plots at 800 nm (NIR) for August
20, 2004
Plot Design
20
Corn Hybrid Separation Techniques9 August 2004
aerial imagery of Iowa plots
Logistic regression was used to differentiate
groups of hybrids (i.e., a transgenic hybrid and
its near isoline from other hybrids) Overall
accuracy 92 Plots in white (left) and green
(right) are same hybrid group (hybrids A and B)
as separated from other hybrid groups
21
Temporal Classification Effects NK Bt11 variety
22
Using Imagery to Separate Hybrids
Isodata results from two band classification
(bands with highest separability)
Isodata results
from 120
band classification
Overall accuracy 75
Overall accuracy 70
Legend
Hybrid C- P34N44 (Cry1Ab)
Hybrid D- P34N43

Hybrid E- P34N42 (Cry1F)
Hybrid H- DK58-78 (Cry1Ab)
-
Hybrid I- DKC57-01
Un-supervised Classifications Of Nebraska
Irrigated Plots July 12, 2004
Composite of Bands used For
Classification
Standard NIR Composite
23
Mean Difference in Reflectance Iowa Site July
26, 2004
24
Neural Network Analysis of Multi-temporal Datasets
648 nm (red) from June 22, 2004 759 nm 759 nm
(NIR) from August 20, 2004 474 nm (blue) from
September 17, 2004 Chosen from discriminant
analysis and Used in multi-layer perceptron
neural network 78 Accuracy in separating GMO
from non-GMO varieties
GMO Varieties
Non-GMO Varieties
25
Transgenic and non-transgenic corn
classificationTemporal effects 2004
Southwestern Minnesota Corn Production Areas
5Aug04
21Aug04
8July04
26
Delineation of Infested Plots
Statistic Indicated that Infested Plots were
reflecting lower in a given wavelength
A Simple Threshold was Used to Predict Infested
Plots
Accuracy for infested areas was 90.6 Accuracy
for un-infested areas was 93.7
Legend
Threshold Classification
Un- infested
Increasing reflectance
Infested
27
Delineation of 2nd Generation Infested
Plots Nebraska Imagery, September 3, 2004
Reflectance At 648 nm
Reflectance At 747 nm
Ratio 747 nm/ 648 nm
Infested 2nd gen.
Un-infested
Un-infested
Infested 2nd gen.
Infested 2nd gen.
Un-infested
DKC 57-01
DKC 57-01 Infested 2nd gen
Threshold of Ratio Accuracy 98.6
Plot Layout
28
Separation of Non-infested (all hybrids) and 1st
and 2nd Generation Infested Plots September 17,
2004 Imagery in Nebraska
Confusion Matrix Infested and
Un-infested Plots Overall Accuracy
Pixel-level Plot-level
Un-infested
96 83 1st gen ECB
41
60 2nd gen ECB
95
100
Ratio of Reflectance 747 nm/648 nm
Treatment Legend
Threshold Classification
2.9
Ratio Values
4.7
Un-infested
1st Gen Infested
2nd Gen Infested
29
Separating Infested from Non-infested Plots
using the Ratio (747 nm/648 nm)
F Value/Significant F Value (Bonferroni
corrected)
30
Results for Separation in Infestation Level Plots
Bt Plots
Infestation Level Plots
False Color Composite from Ratio Bands, September
3, 2004
1 separable from 2,3,4,5
September 3, 2004
August 20, 2004
Ratio 747 nm/648 nm
1 separable from 2,3,4,5 2 separable from 4, 5 3
separable from 4, 5 4 separable from 5
Infestation Levels
31
Conceptual Monitoring Design
Targeting Satellite Imagery
Hyperspectral Imagery
Crop Production Area
Potential Hot Spots
32
Stages of investigation

• Proof of concept
• Development of analytical system
• Initial ground truthing
• Compliance vs. resistance monitoring
• Proof of principle
• Statistical testing development
• Data mining
• Proof of practice
• Working system for compliance and/or resistance
  monitoring

33
Information Flow
Imagery Analysis Institute for Technology
Development
EPA Regions
Crop Registrants
Analysis Evaluation EPA National Risk
Management Research Laboratory
Regulatory Applications EPA Office of Pesticide
Programs
Growers
Ground Truthing UDSA ARS, Univ. Neb., Penn State,
Univ. Minn
Public
RESEARCH
USERS
COORDINATION
REGULATORS
Research Partners
User Community
34
Participants

• Ken Copenhaver, George May, Matt Bethel, John
  Fridgen Institute for Technology Development
• Brian Mitchell, NASA Marshall
• Dennis Calvin Penn State, Richard Hellmich
  USDA-ARS-Ames, Thomas Hunt U. Nebraska
• George Moore, USEPA
• David Andow, University of Minnesota

35
Questions?