Transcript 12 Garrett - forecasting

Upscaling disease risk
estimates
Karen Garrett
Kansas State University
Recruitment
• We are seeking a collaborator who can
authoritatively address scaling of
weather/climate data relevant to
pathogens
Outline
• Upscaling disease forecasting models
based on weather
• Network models at national scale
• Network model for soybean rust
incorporating wind speed and direction
Garrett et al. 2006
Garrett et al. 2006
From Sparks, Hijmans, Forbes, and Garrett, in preparation
Adam Sparks
Outline
• Upscaling disease forecasting models
based on weather
• Network models at national scale
• Network model for soybean rust
incorporating wind speed and direction
Many predictive models of plant disease rely upon
fine-scale weather data
This data requirement is a major constraint when
applying disease prediction models in areas of the
world where hourly weather data are unreliable or
unavailable.
We developed a framework to adapt an existing
potato late blight forecast model, SimCast for use
with coarse scale weather data.
Objectives
Develop disease prediction models based on daily
and monthly weather means and compare to
results based on hourly weather data.
Compare risk predictions based on hourly, daily,
and monthly weather data to late blight severity
data sets from several countries.
Predict disease for resistant and susceptible
cultivars under climate change scenarios.
Methods – development of
models
Hourly weather data from the US National Climatic
Data Center were used in SimCast to calculate
blight units, a daily measure of disease risk.
Generalized additive models (GAM) were created
to estimate blight units based on daily or monthly
averages of weather data.
Blight units predicted by SimCast Daily Means susceptible cultivars. “Observed”
blight units are SimCast estimates based on hourly observations.
Blight units predicted by SimCast Monthly Means susceptible cultivars. “Observed”
blight units are SimCast estimates based on hourly observations.
Observed: p>0.01, R2=0.56 Predicted: p>0.01, R2=0.62
Comparison of estimates for blight units at two levels weather data resolution vs.
late blight severity (AUDPC) from 19 cultivar-site-year combinations
Methods – climate change
scenarios
Maps of disease risk were produced using WorldClim
(http://www.worldclim.org/) datasets that include the
IPCC A2a (high growth carbon emission) climate
change scenario for 2080.
We applied SimCast Monthly Means to this data to
compare current and future risk estimates.
We have low confidence in our estimation of relative
humidity – thus seek a collaborator with expertise in this
area
Peru
Peru
Bolivia
Bolivia
Late blight severity for February for current conditions
and 2080 a2a climate scenario
Upshot
Using this approach we have created models that
can quickly estimate late blight risk over large areas
using readily available weather data sets.
Although the models underpredict, they are useful
for evaluating relative levels of risk.
Outline
• Upscaling disease forecasting models
based on weather
• Network models at national scale
• Network model for soybean rust
incorporating wind speed and direction
The connectivity of the American agricultural landscape
Applying graph theory to assess the national risk of
crop pest and disease spread
Peg Margosian, Shawn Hutchinson, and Kim With
Margosian et al. BioScience 2009
The potential for movement through landscapes can be
modeled by evaluating nodes and the edges that
connect them
Node and edge characteristics may influence the
potential for movement
Maize
Soybean
Wheat
Cotton
Outline
• Upscaling disease forecasting models
based on weather
• Network models at national scale
• Network model for soybean rust
incorporating wind speed and direction
Dynamic network models of a
national epidemic: soybean rust
Sweta Sutrave, Caterina Scoglio,
Scott Isard, and Karen Garrett
Sweta Sutrave
Objectives
• Develop a framework for estimating edge weights
using observed epidemic time series in dynamic
network models
• Apply the model to the spread of soybean rust in
the US.
• Evaluate the estimation framework potential for
epidemic modeling.
Data Sets
• Rust status data: 2005 to 2008, from Dr.Scott
Isard.
• Host density data: 2005 to 2008, from US
National Agricultural Statistics Service.
• Wind data: Wind speed and direction, National
Climatic Data Center’s website.
Model
• SI model which classifies nodes as being
susceptible or infected.
• We consider the centroid of each county of the
United states as a node or vertex.
• We assume that the sentinel plot and the area
around it behave in a similar manner and begin by
modeling dynamics within a single season.
Edge weight function
• uji : Edge-weight between two nodes
• A function of the following
- Distance between the sentinel plots.
- Crop density and kudzu density.
- Speed and direction of wind w.r.t the link
Example of Epidemic Prediction
Soybean rust model realization
Green = no rust predicted
Red shading = likelihood of infection
Looking toward the future
• Developing global model for general
environmental-response classes of
pathogens
• Seeking a collaborator who can
authoritatively address scaling of
weather/climate data relevant to
pathogens
Mapping disease risk based on:
-Climate
-Historical disease distribution
-Host availability
Rival models for consideration
Sparks et al., in prep