Environmental Engineering Reference
In-Depth Information
Table 2. (Continued)
Variable
Description
Bio9
Mean Temperature of Driest Quarter
Bio10
Mean Temperature of Warmest Quarter
Bio11
Mean Temperature of Coldest Quarter
Bio12
Annual Precipitation
Bio13
Precipitation of Wettest Month
Bio14
Precipitation of Driest Month
Bio15
Precipitation Seasonality (Coeffi cient of Variation)
Bio16
Precipitation of Wettest Quarter
Bio17
Precipitation of Driest Quarter
Bio18
Precipitation of Warmest Quarter
Bio19
Precipitation of Coldest Quarter
Because the Worldclim variables are derived from a common set of temperature
and precipitation data, they can exhibit multicollinearity [39]. A Spearman rank cor-
relation matrix was created in JMP (SAS Institute) to explore the relationships be-
tween the WorldClim bioclimatic variables. We removed the four mean temperature
variables (Bio8-Bio11) because they were signifi cantly correlated with minimum and/
or maximum temperature variables, and were less likely to be biologically signifi cant
in contributing to or limiting plague activity. Of the remaining 15 variables, those that
were correlated (Spearman rho > 0.60, p < 0.001) were not used together in the same
model. During model runs, a jackknife manipulation was used to assess the relative
contribution of each variable, and to remove variables that did not contribute signifi -
cantly to the model predictions.
Modeling Current and Future Distribution of Plague in California Using
Maxent
Models of the current potential (i.e., based on climate conditions) distribution of
plague in California were run in Maxent (version 3.1.0). Maxent is a machine learning
program that uses presence-only data to predict distributions based on the principle of
maximum entropy [40]. Maximum entropy [41] is a method to provide the probability
distribution which incorporates the minimum amount of information. Given a set of
constraints determined by environmental variables or functions thereof, Maxent out-
puts the maximum entropy distribution that satisfies these constraints. Among species
distribution models, Maxent has been shown to provide better identification of suit-
able versus unsuitable areas when compared to other presence-only modeling methods
[40, 42]. In place of true absences, Maxent uses background points (pseudo-absences)
to evaluate commission.
Maxent does not need multiple model runs to be averaged together [40]; thus, for
each set of variables, we ran Maxent once. For each Maxent run, 75% of the points
were randomly selected for model training and cross-validation, and 25% of the data
 
Search WWH ::




Custom Search