Fig 1.
Visualization of the assessment region and the data collected.
(A) County map showing the counties designated as infested by the New York State Integrated Pest Management [5] on September 28th 2021 and the 166 county region defined as the convex hull of those infested counties. (B) The number of garden centers present in each county. (C) The number of primary interstate highways transecting each county. (D) The population of each county on a log scale.
Table 1.
Summary table for estimates for linear terms in generalized additive model.
Results are on the logit (log-odds) scale, that is, the scale of the predictor variables.
Table 2.
Components and parameters of the model.
Fig 2.
Comparison of observed and model forecasted infestation in the 166 county model 2014–21.
The Top row displays the actual observed infestation data from 2014–21. The Bottom row displays the corresponding model forecasts for 2015–21. The coloring indicates the proportion of simulations in which a county became infested in a specific year according to the model. More specifically, a county is solid green if all 1000 simulations rendered it infested in a particular year and white if none of the 1000 simulations resulted in it being infested.
Fig 3.
County level accuracy of the model forecast over time in the 166 county region.
Here we display the percentage of simulations that predicted the correct infestation status for each county in each year. The coloring of each county in each year indicates the percentage of simulations that resulted in a correct prediction of its actual infestation status. That is, the percentage of simulations that resulted in it being infested if it indeed was infested, or the percentage of simulations that resulted in it not being infested if it was not infested. More specifically, a county is colored solid blue if all simulations predicted the correct infestation status in a given year (100%) and white if none of the 1000 simulations predicted the correct infestation status (0%).
Fig 4.
Model forecast for 2022–25 in the 581 county region.
We note that the model suggests that the corridors of spread west, southwest, and south will all intensify over the years. In particular, the model suggests that the area around Cleveland, OH, has had an increased and increasing risk of infestation at least since 2021, and it was recently labeled as infested by the New York State Integrated Pest Management Program. The model also forecasts that the infestation in NC will increase substantially over the next few years.
Fig 5.
Estimated coefficients with 95% CIs for the parametric terms in the generalized additive model.
Results are on the scale of odds-ratio, that is, the ratio of the probability of infestation to the probability of no infestation. The likelihood of infestation for a county in 2021 is predicted to increase with the presence of primary interstate highways and larger populations. Interestingly, the likelihood of infestation for a county in 2021 is predicted to increase with an increase in the number of garden centers only when there is also the presence of a primary interstate highway for that county.
Fig 6.
The model space for the 166 county and the 581 county models consist of the combination of their adjacency and primary interstate networks. We note that the adjacency networks in both cases are fully connected, but that the primary interstate networks are not.