Deer home-range size
Space use by individual deer is an important parameter and we used kernel density estimation (KDE) to estimate home-range size. We applied a smoothed “kernel” of equal-sized grid cells in a systematic pattern across a study region weighted by the density of observations and their locations (Hooten, Johnson, McClintock and Morales 2017). We calculated the 97.5% isopleth – the largest area used by the animal (within the chosen timeframe), excluding the extreme 2.5% of movements. Home range analyses were executed using ArcGIS v10.6 (ESRI, Redlands, CA, USA), using the Geospatial Modelling Environment package and the KDE tool.
Landscape Quantification :
We quantified the landscape at deer locations in ArcGIS 10.6 and R (R Core Team, 2019) from publicly available digital datasets from BC provincial and municipal governments, Capital Regional District (CRD) Land Trust, Habitat Acquisition Trust (HAT) (Blyth 2013); and the US Geological Survey (USGS). Natural landcover features included vegetation greenness and tree cover. We derived Normalized Difference Vegetation Index (NDVI) from cloud-free USGS Landsat-8 2018 imagery for the summer growing season (April to September 2018) at 30m resolution. Landsat-8 imagery was downloaded from the USGS Earth Explorer (https://earthexplorer.usgs.gov/).We averaged pixel values across images from each of the 6 months to determine one greenness value for each pixel on the finished dataset. NDVI measures vegetation greenness and can be used to assess vegetation location and density, infer water and nutrient levels, and infer vegetation structure (for example, irrigated grass is typically greener than coniferous forest). To represent vegetation structure in our models, we used a 100-m resolution tree canopy cover raster dataset derived from high-resolution 2011 CRD aerial imagery by Habitat Acquisition Trust (Taylor 2013). We used this percent tree-cover dataset in our RSF analysis.
We extracted residential zoned lots from zoning maps for the municipalities of Oak Bay, Saanich, and Victoria. We used a BC provincial dataset (https://catalogue.data.gov.bc.ca/dataset/bc-parks-ecological-reserves-and-protected-areas) to quantify parkland and manually digitized golf courses. We discretized residential lots into three area classes – small, medium, and large –using the Natural Breaks (Jenks) classification algorithm in ArcMap. We used a publicly available BC roads dataset (https://catalogue.data.gov.bc.ca/dataset/digital-road-atlas-dra-master-partially-attributed-roads) to quantify roads, and a 100-m resolution impervious surfaces percent-cover dataset from Habitat Acquisition Trust (Blyth 2013) to quantify non-road urban features.
Following Zuur, Ieno and Elphick (2010) we assessed all variables for collinearity (r < 0.7) and variance inflation (VIF < 3.0). Due to high correlation between impervious surfaces and other variables (e.g., small residential lots and roads), we excluded impervious surfaces from our analysis. We scaled all variables (mean=0, s.d. = 1) to allow comparison of effect sizes.