Page 1

How well can

General Surveillance reveal exotic pests?

(There’s a cow in the room.) Presentation for Science Exchange, 24 May 2012

Samantha Low Choy

Cooperative Research Centre for National Plant Biosecurity @ Mathematical Sciences, Science & Engineering Faculty QUT

Jo Slattery

Plant Health Australia

Sharyn Taylor

Plant Health Australia

Matt Falk

Math Sciences, QUT

WA Surveillance Experts DAFWA

biosecurity built on science


THE PROBLEM biosecurity built on science


Surveillance

Looking for a needle in a haystack

 Seems like an impossible task! - But how does our thinking frame the task? biosecurity built on science


Surveillance

Looking for a needle in a haystack

 Random sample of patches, then randomly select a plant within a patch biosecurity built on science


Surveillance

Looking for a needle in a haystack

- Transect/blob sampling: jump then look - Path/zigzag sampling: traverse the area biosecurity built on science


Surveillance

What are we looking at?  The unit of surveillance is a plant?

biosecurity built on science


Surveillance

What are we looking at?  The unit of surveillance is a cow, n ~ laboratory testing

biosecurity built on science


Surveillance

What are we looking at?  The unit of general surveillance is what the farmer can see during general farm activities

biosecurity built on science


Surveillance

What are we looking at?

 For broadacre crop, the unit of general surveillance comprises many plants - We need to describe how a pest might be revealed by what farmers generally do biosecurity built on science


Surveillance

What are we looking at?

 For broadacre crop, search & detection depends on scale - We need to describe how a pest might be revealed by what farmers generally do biosecurity built on science


WANTED

High priority pest Significant crop loss, wheat and barley ↑complexity of management ↑research for breeding & chemical management

RUSSIAN WHEAT APHID SUNN PEST, HESSIAN FLY

biosecurity built on science


Avenues for detection

 Two-phase sampling but not @ plant scale

(1) Scan from vehicle -

suspicious patch in visible patches

(2) Trigger closer inspection -

suspicious plants target worst symptoms trigger report?

 Incidental sampling biosecurity built on science


0

1

Skill / pass

High

Low

2. Close Inspection

0.00

0.25

0.50

0.75

1.00 0

Skill / pass

FPR

0.25

0.5

0.75

1

I. Vehicle Scan High

Moderate

Detectability @ patch

Low

Life stage of aphid

of worst symptoms consistent with pest

Detectability @ plant

0.75

Moderate

@ paddock, patch, plant

of worst symptoms consistent with pest

0.5

TPR I. Vehicle Scan

Elicited Detectability

Detectability @ paddock

0.25

2. Close Inspection

0.00

0.25

0.50

0.75

1.00

0

0.25

0.5

0.75

1

3. TPR Nymphs

Wingless adults

of the pest 3. Laboratory

3. FPR 0.00

0.25

0.50

0.75

1.00

biosecurity built on science


Zero predictive value

A role for FPs (TNs) and FNs (TPs)  What does it mean when you find nothing? Could the pest still be present? - Evaluate the zero predictive value (aka NPV)  Fielding & Bell (1997) Tony Martin et al (2007)

- Use Bayes’ Theorem  Hilborn & Mangel (1993) Bayes (1786)

 Usual emphasis is on sensitivity of surveillance  Easier to compute!

biosecurity built on science


Zero predictive value A role for FPs and 0s

 What does it mean when you find nothing? Could the pest still be present?

  Missed when present, at LMH levels ZPV = 1 1 + odds No false alarms when absent  

  

TNR(Absent) Weighted by a priori risk of absence (quantify Pest Risk Assessment)

biosecurity built on science


Zero predictive value A role for FPs and 0s

 What does it mean when you find nothing? Could the pest still be present?

  Missed when present, at LMH levels  ZPV = 1 1 + odds  No false alarms when absent    FNR if present @ Low levels + FNR if present @ Medium levels + FNR if present @ High levels

Weighted by a priori risk of prevalence at each level (Each level is not equally likely)

biosecurity built on science


Learning by doing

expert estimates of risk, detectability, biology

 The Bayesian cycle of learning

40

- Supports science via multiple working hypotheses, not fixing on single null hypothesis

10

20

30

EPSS = 3 : Pr(p<.01)= 0.019 , and 95% chance p no bigger than 0.780 EPSS = 10 : Pr(p<.01)= 0.068 , and 95% chance p no bigger than 0.300 EPSS = 30 : Pr(p<.01)= 0.150 , and 95% chance p no bigger than 0.110 EPSS = 100 : Pr(p<.01)= 0.270 , and 95% chance p no bigger than 0.047

0

Plausibility

Elicited p: best estimate = 1%

0.00

0.02

0.04

0.06

0.08

0.10

Probability pest present,biosecurity p built on science


ZPV as %Missed infested plants MIP=0 means area freedom

0.8

Visibility (depth of road) 2% 10% 100%

0.4

1 in 5 chance that MIP under 50%

0.2

0.6

Low plausibility (2.4%) that MIP under 10.

1 in 4 chance of area freedom.

0.0

Cumulative plausibility

1.0

Hardly any chance (1 in 1000) of area freedom.

0

10

20

30

40

50

Percentage of missed infested plants (%)

High plausibility (99%) that MIP under 10%. biosecurity built on science


MODELLING ISSUES biosecurity built on science


Benefit of repeating 0.2

Myrtle Rust

Low-Choy+2011

0.00

0.05

Plausibility 0.10 0.15

Time 1 Time 2

0

5

10

15

20

25

30

After 4 weeks, typical scenario 40 blocks searched • the mean infested #plants doubles (5.97→12.08) •95% sure infested #plants >doubles (17→46) Can harness Bayesian cycle of learning to adapt as information gained & knowledge refined. biosecurity built on science


Complex detectability

Extrapolated a curve

1.0 0.8 0.6 0.4

Elicited three quantiles (like a bioassay curve)

Pr(Inspect| Damage=10%)=0.049 Pr(Inspect| Damage=25%)=0.572 Pr(Inspect| Damage=50%)=0.980 Pr(Inspect| Damage=75%)=0.999 Pr(Inspect| Damage=90%)=1.000 Pr(Inspect| Damage=95%)=1.000 Elicited

0.2

Triggering a close inspection depends on level of damage

Pr(Inspect | Damage=33%)=0.70 Pr(Inspect | Damage=33%)=0.80 Pr(Inspect | Damage=33%)=0.95

0.0

Sunn pest

Probability of follow-up with close inspe

Beyond a single point estimate

0.0

0.2

0.4

0.6 Level of damage

0.8

1.0

biosecurity built on science


Putting it all together

A hierarchical model for assimilated biosecurity in action

Detect browning (T/FP) Presence of the pest

Phase 1: vehicle scan

Missed browning (FN)

Trigger Phase 2: Close-up Phase 2 not triggered

Detect, with browning (T/FP) Missed, with browning (T/FN)

Confirmed (T/FP) Report Rejected (T/FN)

Confirmed no browning (TN)

WHY BOTHER (cf simple 2-stage design)? If you donâ&#x20AC;&#x2122;t want to assume: perfect detection & reporting and a single method of detection, plant scale of detectability, ignore FPR & risk of prevalence â&#x2021;&#x2019; focus on sensitivity not ZPV, random sampling biosecurity built on science


Zeros for detecting pests/disease/resistance • Hu, W., O’Leary, R. A., Mengersen, K., and Low-Choy, S. (to appear). Bayesian classification and regression trees for predicting incidence of cryptosporidiosis, PLoS ONE. • Falk, M., Low Choy, S., Collins, P., Nayak, M. (in prep). Bayesian hurdle models for identifying factors that affect incidence and trends in pest resistance. Designing surveillance – the conceptual model for model-based sampling design  Anderson, C., Low Choy, S., Dominiak, B., Gillespie, P. S.; Davis, R., Gambley, C., Loecker, H., Pheloung, P., Smith, L., Taylor, S.; Whittle, P. (accepted 29 June 2011) “Biosecurity Surveillance Systems, Plants”, In McKirdy, S. (ed), Biosecurity in Agriculture and the Environment, CABI.  Low-Choy, S., Daglish, G., Ridley, A., Burrill, P. (submitted) “Bayesian adjustment of sampling biases for small intensive surveys on farm management practices relevant to biosecurity”  Low-Choy, S., Hammond, N., Penrose, L., Anderson, C., and Taylor, S. (2011b). In Chan et al (eds) Proceedings MODSIM 2011, www.mssanz.org.au/modsim2011/E16/low_choy.pdf  Low-Choy, S., Taylor, S., et al(in prep) “Evaluating general surveillance for early detection of exemplar exotic plant pests”  Low-Choy, S., Whittle, P., and Anderson, C. (accepted 29 June 2011). Quantitative approaches to designing plant biosecurity surveillance, In McKirdy, S. (ed), Biosecurity in Agriculture and the Environment, CABI. Elicitation • Albert, I., Donnet, S., Guihenneuc, C., Low Choy, S., Mengersen, K., and Rousseau, J. (to appear). Combining expert opinions in prior elicitation, Bayesian Analysis. • Johnson, S., Low-Choy, S. and Mengersen, K. (to appear) “Integrating Bayesian networks and Geographic information systems”, Integ Environ Assess Mgmt. onlinelibrary.wiley.com/doi/10.1002/ieam.262/pdf. • Low Choy, S., Murray, J., James, A. and Mengersen, K. (2010) “Indirect elicitation from ecological experts: from methods and software to habitat modelling and rock-wallabies” in O’Hagan, A. and West, M., (eds) The Oxford Handbook of Applied Bayesian Analysis, Oxford University Press: UK, pp 511-544. • O’Leary, R., Fisher, R., Low Choy, S., Mengersen, K., Caley, M. J. (2011) What is an expert? In Chan, F. et al (eds) Proceedings MODSIM2011, www.mssanz.org.au/modsim2011/e9/oleary.pdf  Martin, T. G., Burgman, M. A., Fidler, F., Kuhnert, P. M., Low-Choy, S., McBride, M., Mengersen, K. (2012) Eliciting Expert Knowledge in Conservation Science, Conservation Biology, 26(1): 29-38.  Low-Choy, S. (in press). Priors: Silent or active partners in Bayesian inference? In Alston, C. et al (eds) Case Studies in Bayesian Statistical Modelling and Analysis, John Wiley & Sons, Inc: London.  Fisher, R., O’Leary, R., Low-Choy, S., Mengersen, K., and Caley, J. (to appear). Elicit-n: New method and software for eliciting species richness, Environmental Modelling & Software.

biosecurity built on science


Acknowledgements The experts

        

John Botha Cameron Brumley Rob Emery Darryl Hardie Alan Lord Marc Poole Jeff Russell Dusty Severtson Andy Szito biosecurity built on science

What do we learn from general surveillance  

Surveillance for plant pests is typically resource intensive. Currently a range of quantitative methods is being applied to evaluate the eff...

Read more
Read more
Similar to
Popular now
Just for you