Cambridge Richard Stutt University Nik Cunniffe Erik DeSimone Matt Castle Chris Gilligan RothamstedStephen Parnell ResearchFrank van den Bosch May 2012.

Modelling the spread, control and detection of Ramorum Disease

Cambridge Richard StuttUniversity Nik Cunniffe

Erik DeSimoneMatt CastleChris Gilligan

Rothamsted Stephen ParnellResearchFrank van den Bosch

May 2012

Introduction – Prediction 2010-20

Model must integrate◦ Location of hosts◦ Environmental drivers◦ Pathogen dispersal

Compartmental model

Model – Spatial Stochastic Model

250m x 250m resolution

Combine data on Larch Rhododendron Vaccinium NIWT (other tree hosts)

Weight host types by sporulation/susceptibility

Model – Host Landscape

Pathogen responds totemperature/moisture

Model underlying suitability for each location

Statistical climate model then used to predict future fluctuations about this

Model – Environment

Dispersal kernel describes pathogen spread

Implicitly incorporates many mechanisms

Model – Dispersal

Positive Negative

Spread in the absence of control

Effect of extent of control◦ Felling infected stands◦ Felling infected stands + proactive control

Surveying for P. Ramorum on heathland

Results – Typical Applications

Results – Prediction 2010-20

Results – Need for Control

Results – Effect of Control Radius

Results – Sampling on HeathlandHazard map + known infections = sampling pattern

Continuous model improvement (data driven)

Region specific control

Effect of non compliance

Transition strategies

User friendly models

Current and Future Work

Forestry Commission◦ Bruce Rothnie◦ Joan Webber

FERA◦ Keith Walters◦ Phil Jennings◦ Judith Turner◦ Kate Somerwill

Funding from DEFRA, BBSRC and USDA

Acknowledgements

Extra material for questions

Susceptible hosts in the landscape are divided into a metapopulation at a chosen resolution (250m)

UK Sudden Oak death landscape assembled from:◦ National Inventory of Woodland Trees (NIWT)◦ Forestry Commission commercial Larch data◦ Maximum Entropy suitability models for Rhododendron and

Vaccinium (FERA/JNCC)

Different hosts have different weightings for sporulation and susceptibility

Model - Host

Broadleaved

Young Trees Felled

Coniferous

Construction of Host Landscape

Identify favourable conditions for P. ramorum◦ moisture ◦ temperature

Parameterise using experimental results

Model - Environment

Rela

tive S

poru

lati

on

Temperature

Fit model using historic spread data

Used Maximum Likelihood to assess goodness of fit

Predicted probability of infection by 2010 given starting conditions in 2004

Model - Validation

Survey Positive for P. ramorum

Survey Negative for P. ramorum

Probability of Infection

Risk – Reactive Control

Risk – Proactive Control (250m)

Risk Update – 20 year horizon

Disease Progress – No Control

Disease Progress – Stand Control

Total Infection

Symptomatic

Symptomatic at time of Survey

Disease Progress – 100m Radius

Total Infection

Symptomatic

Symptomatic at time of Survey

Disease Progress – 250m Radius

Total Infection

Symptomatic

Symptomatic at time of Survey

Disease Progress – 500m Radius

Total Infection

Symptomatic

Symptomatic at time of Survey

Disease Progress – Comparisons

Effects of Delay Before Culling

Examine region of South Wales

Cull: no delay after survey 6 month delay

Effect of Delay Before Culling

Key Questions When Surveying for Disease:◦ Where is the disease likely to be?◦ Where is it likely to be most severe and spread

most rapidly?◦ How to optimise the sampling?

Sampling Strategies

Sampling Strategies Uses:

• Currently known outbreaks • Predicted severity of

outbreaks• => Sampling weighting

Survey pattern formed• => sampling from

weightings Map shows a weighting and

a set of survey points (green)