Third order selection
We ran integrated Step Selection Analysis (iSSA; Avgar et al. 2016) to
determine drivers of within-home range fine-scale resource selection
while accounting for individual movement using the broad and fine
thematic resolution landcover layers. The movement from 59 deer was
modeled using broad thematic resolution. Additionally, the movement from
a subset of deer (n = 27) whose location data spanned all six
landcover types was modeled using the fine thematic resolution layer
during the deer breeding/adult I. scapularis feeding seasons
(2016 – 2020).
To prepare the data for iSSA, we used the amt (Signer et
al. 2019) R package to estimate step lengths and turning angles between
successive steps to fit tentative gamma and von Mises distributions,
respectively. From these distributions, ten random steps were generated
for each observed step (Supporting Information). Habitat attributes were
extracted at the beginning and end of observed and random steps using
the two landcover layers. We fit four iSSA models to data from each deer
using the ‘fit_issf’ function in amt which uses a conditional
logistic regression stratified by start step ID. We included three
covariates expected to influence the movement process in the core model
and modeled habitat selection for all deer using broad thematic
resolution and the subset of 27 deer using fine thematic resolution
(Table 1, model 1). We expected time-of-day to vary deer’s strength of
selection for features within their HR, thus we included day or night as
an interactive term with habitat selection (model 2). Movement
differences driven by the starting habitat were assessed through an
interaction between starting step habitat and movement covariates (model
3). To assess whether deer moved differently depending on habitat and
time-of-day, we included interactions between the ending step landcover,
movement covariates, and time-of-day (model 4).
We assessed model fit for 160 deer-season models by bootstrapping each
individual’s four models independently (n= 1000) to acquire mean
coefficient estimates with a 95% confidence interval and used the ΔAIC
(Burnham & Anderson 2002) to determine the individual’s best fit model
between the null model and model 1 and between models 1 - 4 (Table 1).
We tallied how many deer showed the strongest support for each model by
season and sex to determine the top model (Table 2). The coefficient
estimates from the best fit model for each sex were bootstrapped
(n = 1000) for each individual and the coefficients’ standard
errors were estimated. We used individual movement parameter estimates
from model 4 (where habitat selection was accounted for) to update the
tentative gamma and von Mises distributions and estimated movement rates
and directionality for individual deer over natural and urban
landcovers. These were summarized using boxplots to show deer’s average
and individual movement rates by sex and landcover, faceted by season.
Lastly, to showcase how individual variation in movement and resource
use affects the future probability of vector dispersal, we simulated
spatially-explicit dispersal kernels informed by movement and selection
coefficients estimated from fitted iSSA models for three individuals
that varied in the habitat diversity within their HR and their strength
of selection for fine resolution landcover types using amt(Signer et al. 2017). We utilized the same initialization points
for the simulations across individuals in an area of SI that exemplifies
the juxtaposition of habitat types in residential areas, but where no
individuals were observed occupying. The simulation was informed by the
movement and habitat selection coefficients from model 4 that used a
fine resolution landcover. Rasterized dispersal kernels were constructed
for the first 100 locations for three individuals. The individual
dispersal kernels were mosaicked using the maximum cell value from
overlapping kernel layers to create a single raster mosaic per deer
(Supplemental Information). Because hosts’ movement characteristics may
translate to vector distribution patterns; simulating deer space use
informed by observed movement and habitat selection over a landscape
provides a visualization of this connection.