I WSF, Brasília - Alan MacLeod - Pest Risk Analysis Research in Europe :Developments from EU project PRATIQUE
1. Pest Risk Analysis Research in Europe
:Developments from EU project PRATIQUE
Alan MacLeod
Pest Risk Analyst
PRA Workshop
Brasilia
March 2012
2. Outline
• Introduction to Fera
• PRATIQUE
– Datasets
– Consistency
– Mapping
– Factors that ease eradication
– Computer Assisted PRA (CAPRA)
• Work in China & Russia
3. What is Fera?
• A government science agency which
provides the UK’s food and environment
sectors with:-
• expert scientific advice
• regulatory services
• applied research facilities and
• emergency responsiveness
4. National Assembly for Wales, Agriculture Department
Forestry Commissio n
- Forest Research
UK
•York
London
5. Plant Protection
Phytophthera
Contingency
Response
Policy
Containment
and
Eradication
Plant health
and Seeds
Inspections
Bee Disease
Pest Risk
Analysis
Diagnosis
and
Taxonomy
Seed Listing
and
marketing
Pollinator
Research
National
reference
laboratory
Plant Clinic
National Bee
Unit
Plant
Breeders
Rights
7. Wildlife Management
Badgers and
TB
Bird Strike
Control
NonNative
Species
Secretariat
Rabies in
Wildlife
Vaccine
Deployment
Wildl ife
Damage
Control
Methods
Fertility
Control
Welfare
Bird Radar
Wind Farm
EIA
Disease
Dynamics
Invasive
Species
Population
Monitoring
Eradication
Programmes
10. PRATIQUE: Permission granted to a ship or boat to
use a port on satisfying the local quarantine
regulations or on producing a clean bill of health
11. Acknowledgments
6 European research institutes (Fera, CIRAD, INRA, JKI, LEI, PPI)
5 European universities (IBOT, Imperial, UNIFR, UPAD, WU)
2 international organisations (CABI & EPPO)
2 partners from outside Europe (CRCNPB & Bio-Protection)
Food and Environment Research Agency
13. PRATIQUE Aims
• To enhance PRA techniques for the EU / EPPO
(i) by assembling datasets required for PRA for
the whole EU (27 countries)
(ii) by conducting multi-disciplinary research to
enhance techniques
(iii) by providing a user friendly decision support
system
15. Why did PRA in Europe need
enhancing?
1. PRA is a young area of study (first schemes
developed only in 1990s)
2. Lack of data to analyse the risks posed by pests to
countries in the EU or EPPO
3. Developments outside of PRA can be applied in PRA
4. PRA procedures are complex* for the risk analysts
and the decision makers. Tools needed which brings
all factors together
*EPPO PRA scheme (2009): over 50 questions, 5 level risk rating, 3 levels
of uncertainty - no mechanisms to combine ratings and derive risk
16. 1. Young discipline
• Whilst there is a history of plant health*, formal
pest risk analysis is relatively young
• ISPM No. 2 Guidelines for PRA (1996)
• ISPM No. 11 PRA (more detail)(2004)
– tells us what to do but not how
– “Climatic modelling systems may be used…” (2.2.2.2)
– “There are analytical techniques which can be used in
consultation with experts in economics….” (2.3.2.3)
• As well as standards, need tools and resources
* MacLeod et al. (2010) Food Security, 2, 49-70
17. 2. Lack of Data for PRA
• PRA quality is highly dependent on data
• EU and EPPO need to produce PRAs
relevant for all member states
• Data from some member states difficult to
obtain
• Language barriers
• Crop, pathway, and impacts-related data
often very difficult to obtain
18. Wrote to EU Member States
• Collected electronic / web accessible data
sources (e.g. Crop / pest distribution)
• Import data, other economic datasets, yields…
• PRA area data e.g. land use, climate data, soil
types, …
• Pest management data
• Reviewed datasets
21. Dataset quality and usefulness evaluations
Dataset
Categories
Total
evaluated
Data rating
(overall)
Total
retained
A B C D U
Pests in the current area
of distribution
236 50 61 53 70 2 166
Pathways and economic
datasets
118 5 37 38 16 22 96
Area under consideration
for the PRA (land use
etc)
266 30 105 91 27 13 239
Pest management 155 24 66 28 8 29 147
Score Definition
A Essential, high quality and widely applicable
B Good quality but applicable to specific regions
C Narrow or very limited usefulness or overlap with categories A or B.
D Unreliable, contain too many errors or are generally irrelevant
U Cannot currently be assessed due to a language barrier
24. Consistency
• Reviewed 43 schemes & guidelines seeking best
practice on ensuring consistency:
– Biosecurity and plant health standards
– PRA schemes
– Weed risk analysis schemes
– Animal health schemes
• Consistency in risk rating more likely if:
– use a clear and structured framework
– ask unambiguous questions
– obtain responses from groups of assessors
– provide examples to help guide risk rating, e.g. CFIA
– mechanism to combine risk elements (risk matrices)
25. EPPO (2009) PRA Scheme - Format
• Series of questions:
Categorisation (19)
Entry (14)
Establishment (15)
Spread (3)
Impacts (16)
Risk management (44)
• Explanatory Notes
• Responses required:
5 level risk rating
3 level uncertainty score
Written justification
• No method for summarising
each section or overall risk
and uncertainty
26. Consistency
Revised EPPO scheme
• To improve structure
• Reword some questions = clearer meaning
• Provide biological examples for rating guidance at 5
levels for each question
• A visualiser developed to review questions
• Mechanism to combine risk elements
• Matrix models provided to summarise risk and
uncertainty from many questions and sub-questions
28. Qualitative Impact Assessment Methods: Visualiser to
review responses to questions
• Each question’s risk
rating from very low (1)
to very high risk (5) is
put on the graph as a
bubble
• The larger the size of
the bubble, the greater
the uncertainty
• Each cluster of
questions has the same
colour
• A bar marks the
summarised rating
(here for entry) of the
expert(s)
• Visualisation of the
author’s judgment, no
modelling!
29. Qualitative Impact Assessment Methods: Visualiser to
review responses to questions
• Each question’s risk
rating from very low (1)
to very high risk (5) is
put on the graph as a
bubble
• The larger the size of
the bubble, the greater
the uncertainty
• Each cluster of
questions has the same
colour
• A bar marks the
summarised rating
(here for entry) of the
expert(s)
• Visualisation of the
author’s judgment, no
modelling!
30. Consistency
Was no mechanism to combine factors that
contributed to risk (risk elements)
Examined the concept of risk matrix
Used in USA & Australia
31. Risk matrix
Likelihood of
introduction
Establishment
Low Medium High
Entry
Low Low Low Medium
Medium Low Medium High
High Medium High High
33. Risk matrix with uncertainty
Likelihood of
introduction
Establishment
Low Medium High
Entry
High
High Medium Low
Establishment
Low Medium
Low Low Low Medium
Entry
Medium Low Medium High
High Medium High High
34. Very Unlikely / Minimal (Score / rating of 1)
The distributed scores/ratings corresponding to the three levels of uncertainty
Uncertainty distributions
Very Unlikely
Unlikely
Very Unlikely / Minimal (Score / rating of 1)
Low Medium High
Unlikely / Minor (Score/ rating of 2)
Low Medium High
Unlikely / Minor (Score/ rating of 2)
Low Medium High
Likely / Major (Score / rating of 4)
Moderately Likely / Moderate (Score / rating of 3)
Moderately Likely / Moderate (Score / rating of 3)
Moderately likely
Likely
Very likely
The distributed scores/ratings corresponding to the three levels of uncertainty
Very Unlikely / Minimal (Score / rating of 1)
Low Medium High
Unlikely / Minor (Score/ rating of 2)
Low Medium High
Likely / Major (Score / rating of 4)
Low Medium High
Low Low Medium High
High
Moderately Likely / Moderate (Score / rating of 3)
Very Likely / Massive (Score / rating of 5)
Low Medium High
Low Medium High
Low Medium High
Very Likely / Massive (Score / rating of 5)
Low Medium High
Uncertainty rating
Low Medium High
Question/ risk element score
Low uncertainty: 90%
confidence that rating is
correct
Medium: 50% confidence
that rating is correct
High uncertainty: 35%
confidence that rating is
correct
(after Intergovernmental
Panel on Climate Change,
2005)
Assignment based on the
beta & truncated normal
distribution
36. Matrix models
Have generic models for
• Entry
• Establishment
• Spread
• Impact
Could combine likelihood of entry, establishment,
spread and impact to show overall pest risk
Loss of detail when combine all elements
Can be difficult to agree how to combine elements
(low likelihood : high impact)
38. Maps can help risk assessors
Global Annual Degree Days base 10°C
(from Baker, 2002)
World Potato Production (from Monfreda et al., 2008)
39. Why do we need a DSS for risk
mapping?
• General maps of climate, current pest distribution, crop
distribution or other factors do not directly indicate pest risk
• Risk maps can be very useful in PRA but guidance is
needed :
– To advise when appropriate to map (may not be needed)
– May be inappropriate to map predictions (data problems)
– Mapping requires significant modelling and mapping
skills, resources and time
– Maps can be created by a confusingly wide variety of
methods
– Maps can produce misleading results
41. Climatic Mapping DSS
Asks questions to help decide if should map, and is so what
technique to use
Is it appropriate to map climatic suitability? (sub questions)
What type of organism is being assessed and what are the key
climatic factors affecting distribution?
How much information is available on the climatic responses of
the pest?
What category of location data is available?
Based on the type of organism, the information available on its
climatic responses and the category of location data, how
well is each climatic mapping method likely to perform?
42. Pest location data category
N Pest location data category Availability
1 Native range locations only
2 Native plus exotic range locations
3 Locations biased to the periphery of the range
4 Locations biased to the centre of the range
5 Few location data points
6 Very few location data points
7 Erroneous locations included
8 Locations influenced by natural barriers
9 Locations influenced by seasonal invasion
10 Distribution constrained by hosts
11 Regional distribution data only
12 Locations influenced by climate change
13 Location category unknown
43. “Traffic Lights” to summarise performance of
different model based on availability of data on
the pest’s distribution and responses to climate
Climate
Response
Information
Availability
Location Data Category
Methods + ++ +++ 1 2 3 4 5 6 7 8 9 10 11 12 13
Phenology
models
CLIMEX
match
CLIMEX
compare
Regression
models
KEY
Climatic response rating or location data category irrelevant to model
functioning
Method poorly adapted to climatic response or location data category - results
very difficult to interpret
Method moderately well adapted to climatic response or location data
category - results moderately difficult to interpret
Method well adapted to climatic response or location data category - results
relatively straightforward to interpret
44. Area of potential establishment for
Diabrotica virgifera virgifera
Climatic suitability Hosts Area of potential
establishment
& =
45. Area at highest risk
Host distribution Sandy soils Maize output not on
sandy soils
Total maize output
Climate suitability
Maize output not on
sandy soils
Area at highest risk
& =
& =
46. Climatic Mapping: Tutorials and
manuals
• How to run several models,
e.g. Diva-GIS, Maxent,
Openmodeller Desktop and
CLIMEX,
• How to compare model
outputs
• How to interpret the results
47. Risk Mapping Conclusions
• The PRATIQUE DSS enables assessors to create
and combine maps to display:
– the area of potential establishment
– the area where plants are at highest risk (i.e areas
most suitable for the pest and of highest "value")
• useful for prioritising surveillance programmes
• Link to spread models
• Link to economic models
49. Spread models for Diabrotica
Radial expansion model
Dispersal kernel model
Diabrotica v. virgifera spread 1992-2011
50. Spread – example result
Dispersal kernel model
Showing A. glabripennis spread
from 4 outbreak sites over 30
years.
Based on Climex model
Colours: % population
abundance
white < 10-6 %,
yellow,
orange
red > 10%
2
1
2
2
2
r
p
1
1
1
p
2
1
2
1
1
p
u
p
u p
f r
51. Establishment: A. glabripennis
(Years for development, Climex)
4+ or not possible Years for development 4 years 3 years 2 years 1 year
52. Economic modelling
• Simple qualitative approach
• More complex quantitative approach
– Partial budgeting
– markets based (partial equilibrium)
53. Analysis of previous eradication
efforts
• > 170 campaigns (102 species)
(41 invertebrate species, 26 pathogens, 27 plant species)
• For each campaign ask 96 questions
• Seek to identify factors for eradication success
• Linear mixed effect models (LMMS) & classification and
regression trees (CART) applied
FINDINGS
• Small infestations (< 4,000 ha) are easier to eradicate
• Eradications in man-made habitats are more successful
• Natural habitats provide a major challenge
• Fungi most difficult to eradicate
Pluess et al. 2012. Biological Invasions, DOI 10.1007/s10530-011-0160-
54. 4. Provide a user friendly DSS
• Previous EPPO Scheme (2009) difficult to use
• For the analyst
– Many questions (most detailed system)
– Some seem repetitive
– Difficult interface
– Difficult to make consistent judgements
– Difficult to summarise
• For the decision maker
– Lengthy documents produced
– Difficult to focus on key elements
55. User friendly DSS
• PRATIQUE provided
• a computerised EPPO PRA scheme
incorporating PRATIQUE outputs
• Revised structure
• Reworded questions
• Rating guidance
• Links to datasets
• Guidance documents
• Can share PRA document (for group work)
58. Sentinel trees in Asia
• To produce a dataset of potential Asian pests of
selected woody plants not yet introduced into Europe
Beijing suburban area
Continental conditions
Fuyang, nr. Hangzhou
Warm and humid climate
(Dr. Fan Jian-tin;
Zhejiang Forestry University)
59. Sentinel trees in Asia
Beijing
• 400 seedlings of 4 species exported
• 177 seedlings survived after a long stay in customs
• planted in a semi-urban nursery 5th May 2007
Abies alba- 60
Quercus suber- 50
Quercus ilex- 48
Cupressus sempervirens- 19
•Monthly survey
•No serious insect damage
observed until June 2008
•Alternaria sp. found on Abies
•Unidentified fungi on
Quercus and Cupressus
60. Sentinel trees in Asia
Hangzhou
• 598 young trees of 7 species planted
• Each 1m – 1.5m tall
• planted in a forestry region May 2008
•More than 50 species of
insects during summer 2000
Most yet unidentified
Some highly damaging
e.g. tussock moth on oaks
61. Arboreta Surveys
• Far East Russia (Siberia): surveys of pests on
European trees and shrubs in arboreta
Harsh Siberian climate
not suitable for many
European plants
Maritime climate
62. Novel method to obtain lists of potential
plant pests before introduction
• Sentinels in China colonised by:
97 insect species
24 symptomatic infections
• Russian arboreta
Of the many insect species, 30 high risk species
identified
106 symptomatic infections and 75 fungal species
on 56 woody plants
• BUT significant identification problems
• Future International Plant Sentinel Network?
63. Comparison of methods
Arboreta Sentinel trees
Logistics - “Simple” - Complicated
No. of plant species - Many - Few
Statistics - Poor - Robust
Weaknesses - No seedling pests - No mature tree pests
- Mostly foliage pests
- Lethal pests - Travel and plantation
difficult to assess stress
Complementary methods
Both require strong local links !
67. Ecological activities occur across various
temporal and spatial scales
Millennia
Centuries
Years
Months
Days
Hours
Landscape
evolution
Forests
develop
Impacts of Invasive
species
Climate change
El Nino events
Trees
grow
Local land use
Annual
crops Where risk assessors
change
All year aim to inform
round crops
The scale at which
much field research
is performed
cm m km 100 km 1,000 km
Adapted from Turner, Dale & Gardner (1989) Landscape Ecology 3 (3/4) 245-252
Infection