Predictor variables included 19 bioclimatic variables, elevation, slope, and average and standard deviation of normalized difference vegetation indices (NDVI) (see Table 1 for full list of variables). Bioclimatic variables can indirectly affect habitat suitability for ticks, elevation impacts microclimate, host presence, and vegetation [50 (link)], whilst slope provides a proxy for the velocity of subsurface water flow and runoff rate, and thus can represent soil moisture content [51 (link),52 ]. Suitability of tick habitat is also affected by the presence and type of vegetation (e.g., [53 (link),54 (link)]), and vegetation greenness/biomass can be described using NDVI. Two different datasets were used to derive bioclimatic “Bioclim” variables: To train and test the models, we used WorldClim 2.0 at 30 arc seconds resolution [55 (link)], whereas the 4 km resolution parameter-elevation regressions on independent slopes model (PRISM) [56 ] dataset was used to predict potential habitat suitability in California. Both WorldClim and PRISM are a set of climate layers, such as precipitation and temperature, that are interpolated from weather station data [55 (link)], which may impact tick survival (e.g., [57 (link)]). The use of two datasets was necessary as WorldClim data only span the 1950–2000 period. Although WorldClim data geographically and temporally matched species observations, they would not provide recent climatic information. For prediction, monthly average minimum and maximum temperature and cumulative precipitation were downloaded from the PRISM website for 2014–2018. Long term monthly averages were then derived for each variable and used to calculate 2014–2018 Bioclim variables with the r.bioclim module in GRASS GIS 7.6 [58 (link)].
Elevation data were downloaded from the processed Consultative Group for International Agricultural Research-Shuttle Radar Topographic Mission dataset (i.e., data gaps filled by interpolation) available at 90 m resolution. Slope was computed from elevation data using the r.slope.aspect module in GRASS GIS version 7.4 [58 (link),59 ]. No noise (smoothed) NDVI 4 km, 7-day composite data were downloaded from the National Oceanic and Atmospheric Administration STAR dataset for the years 1985–2000 (training and testing) and for 2014–2018 (prediction) (https://www.star.nesdis.noaa.gov/smcd/emb/vci/VH/VH-Syst_10ap30.php ). NDVI is the difference between near-infrared light emissions which are reflected by vegetation and visible red light which are absorbed by vegetation. These data were imported into GRASS GIS version 7.64, and the mean and standard deviation were calculated. All training and prediction environmental variables were re-sampled using bilinear spatial interpolation to standardize resolution to 30 arc seconds, imported into R using the raster package (version 2.6-7 [60 ]), and cropped to the extent and shape of the countries from which Amblyomma tick species presence data were obtained.
Predictor variable values were extracted for each presence and background geolocation (30 arc seconds pixel). Many presence and background geolocations near the coast were not well covered by the extent of the environmental predictor raster files; therefore, when a raster did not overlap a geolocation and environmental data could not be extracted a distance matrix was applied to sample the nearest non-empty pixel within a 10 km radius. For data extracted for presence geolocations, collinearity between each pair of predictor variables was assessed using Pearson’s correlation coefficients. When the absolute value of the correlation coefficient was ≥0.80, we retained just one of the two variables in the model. This selection was informed by: the variable with the highest percent contribution to the model (that is, in each iteration of the training algorithm, the increase in regularized gain was added to the contribution of the corresponding variable, or subtracted from it if the change to the absolute value of lambda was negative), and the ecological relevance to the tick species of the variable. Environmental predictor variables were refined until the model consisted of only non-correlated, biologically relevant variables and MaxEnt auto-feature classes were used for all species.
Elevation data were downloaded from the processed Consultative Group for International Agricultural Research-Shuttle Radar Topographic Mission dataset (i.e., data gaps filled by interpolation) available at 90 m resolution. Slope was computed from elevation data using the r.slope.aspect module in GRASS GIS version 7.4 [58 (link),59 ]. No noise (smoothed) NDVI 4 km, 7-day composite data were downloaded from the National Oceanic and Atmospheric Administration STAR dataset for the years 1985–2000 (training and testing) and for 2014–2018 (prediction) (
Predictor variable values were extracted for each presence and background geolocation (30 arc seconds pixel). Many presence and background geolocations near the coast were not well covered by the extent of the environmental predictor raster files; therefore, when a raster did not overlap a geolocation and environmental data could not be extracted a distance matrix was applied to sample the nearest non-empty pixel within a 10 km radius. For data extracted for presence geolocations, collinearity between each pair of predictor variables was assessed using Pearson’s correlation coefficients. When the absolute value of the correlation coefficient was ≥0.80, we retained just one of the two variables in the model. This selection was informed by: the variable with the highest percent contribution to the model (that is, in each iteration of the training algorithm, the increase in regularized gain was added to the contribution of the corresponding variable, or subtracted from it if the change to the absolute value of lambda was negative), and the ecological relevance to the tick species of the variable. Environmental predictor variables were refined until the model consisted of only non-correlated, biologically relevant variables and MaxEnt auto-feature classes were used for all species.
Free full text: Click here
