interesa, preko procesa posmatranja, mjerenja, interpretiranja ...... modela simultanih jednaÄina, simulacijski pristup u ekonometriji, vrednovanje i optimizacija .... vrijednosti, analiza odnosa troÅ¡kova i prihoda, indikatori performansi, raÄunov
The only limits you see are the ones you impose on yourself. Dr. Wayne Dyer
Idea Transcript
Surveillance for invading plant pathogens: Epidemic modelling to quantify performance and optimise survey design Stephen Parnell Rothamsted Research, United Kingdom
Overview 1. What will a surveillance program tell you? 2. How can we best target our sampling resources? • Modelling & Epidemiology Current applications:
Early-warning surveillance: what will it tell you? Insight from a simple epidemic model: Logistic growth with rate, r
q*
t0
D D D D
t*
Sample N hosts at regular intervals Δ Parnell et al. Journal of Theoretical Biology 305 (2012) 30–36
Early-warning surveillance: what will it tell you?
When an epidemic is discovered for the first time what is its incidence in the population (i.e. detection-incidence)?
Mean detection-incidence is given by:
How well does this “rule of thumb” work in practice?
Citrus canker disease in urban Miami
4 study sites; 17973 trees Disease progress fully observed
Miami Site 4
0.18
incidence (proportion infected)
0.16 0.14 0.12 0.10 0.08 0.06 0.04 0.02 0.00
0
200
400
600
800
1000
1200
time (days)
Fit to logistic curve: epidemic growth rate, r = 0.014
Early-warning surveillance: what will it tell you?
Calculating detection-incidence q* from the data: • •
Simulate random sampling at regular intervals Repeat thousands of times to get mean detection-incidence q*
sampling round 1
sampling round 2
q = 0.005
q = 0.008
Nothing detected (day 30)
Nothing detected (day 60)
sampling round 3
sampling round 4
q = 0.021
q = 0.064
Nothing detected (day 90)
Detection! q*=0.064 (day 120)
Early-warning surveillance: what will it tell you? • How well does the “rule of thumb” work?
rule of thumb 0.10
0.05
0.10
0.05
0.00
0.00 0
10
20
30
40
50
60
sample size (number of trees)
0.15
Miami Site 4 detection-incidence q*
0.15
Miami Site 3 detection-incidence q*
observed
0.15
Miami Site 2 detection-incidence q*
detection-incidence q*
Miami Site 1
0.10
0.05
10
20
30
40
50
60
sample size (number of trees)
0.10
0.05
0.00
0.00 0
0.15
0
10
20
30
40
50
60 0
sample size (number of trees)
10
20
30 30
40 40
50 50
60 60
sample size (number (number of of trees) trees)
Overview 1. What will a surveillance program tell you? 2. How can we best target our sampling resources? • Modelling & Epidemiology Current applications: Citrus disease
Ash dieback
Resistance
Ug99
Cassava viruses
Spatially-targeted sampling Application to Citrus greening disease (HLB) in Florida
Risk-based Sampling Application to Citrus greening disease (HLB) in Florida locations to sample
potential consequences (planting age & size)
X
probability of infection (distance to known outbreaks)
Parnell et al. (2013) Ecological Applications. In Press.
=
risk weighting (where to target samples)
Risk-based Sampling Relative success compared to former strategy proportion observed-positive finds
1.0
0.8
0.6
0.4
0.2
0.0 0.0
0.1
0.2
0.3
0.4
0.5
sample size (proportion of acreage sampled)
Practical output: used in Florida since 2006 to search for multiple pathogens (Multi-Pest Survey) Parnell et al. (2013) Ecological Applications. In Press.
Spatially optimised surveillance Where to sample to maximise the probability of early-warning? A single-run of the epidemic simulation (Individual based model)
Average of thousands of runs of the epidemic simulation disease risk
1
0
Spatial optimisation The answer depends on the question Objective: (pre-invasion) Early warning surveillance
Objective: (post-invasion) Maximising new disease finds
Disease risk Optimal sample placement
Solution: Widespread sampling
Solution: Risk-based sampling
Spatial optimisation Application to Citrus greening disease (HLB) in Florida • Individual based model of invasion and spread of HLB in Florida (Retrospective analysis!) • Estimate of citrus tree distribution at 1km resolution in Florida • Individual-based spread model
Residential trees Commercial trees
Spatial optimisation: HLB in Florida Florida citrus distribution
Spatial optimisation: HLB in Florida Including Risk of Entry: - Travel-census risk map (Tim Gottwald & Tim Riley, USDA) Incorporating into the method: 1. Seed the models runs by probability of entry 2. Run epidemic runs 3. Identify optimal sample locations
Spatial Optimisation Searching for citrus greening (HLB) in Brazil
Dr Francisco Laranjeira Rothamsted International Fellow
Transferred to Embrapa for use to inform regulatory surveillance for HLB in disease-free regions of Brazil
Summary • Take home messages Modelling can help to say what a surveillance program will actually tell you (quantification) With epidemiology and modelling we can find optimal surveillance designs Surveillance strategies need to be carefully matched to the specific objective
Acknowledgements • • • • • •
Dr Frank van den Bosch (Rothamsted Research) Dr Tim Gottwald (USDA ARS) Dr Nik Cunniffe (Cambridge University) Prof Chris Gilligan (Cambridge University) Dr Francisco Laranjeira (Embrapa, Brazil) Tim Riley (USDA APHIS)