1. The return of top carnivores to their historical range triggers conflicts with the interests of different stakeholder groups. Anticipating such conflicts is key to appropriate conservation management, which calls for reliable spatial predictions of future carnivore occurrence. Previous models have assessed general habitat suitability for wolves, but the factors driving the settlement of dispersing individuals remain ill-understood. In particular, little attention has been paid to the role of prey availability in the recolonization process. 2. High-spatial-resolution, area-wide relative densities of the wolf’s main ungulate prey species (red deer, roe deer and chamois) were assessed from snow-track surveys and modelled along with wolf presence data and other environmental descriptors to identify the main drivers of habitat selection of re-establishing wolves in the Western European Alps. 3. Prey species abundance was estimated from the minimum number of individuals recorded from snow-tracks along 218 1km transects surveyed twice a year during four successive winters (2012/13–2015/16). Abundance estimates per transect, corrected for species-specific detection probabilities and averaged across winters, were used to model area-wide relative prey density and biomass. 4. Confirmed wolf observations during the same four winters were used to develop a spatially-explicit habitat selection model for establishing wolves, based on our estimates of prey supply and other environmental descriptors of topography, land-use and climate. 5. Detection-corrected ungulate prey abundances and modelled relative densities varied considerably in space (0–2.8, 1.3–4.5 and 0–6.3 per 50ha in red deer, roe deer and chamois, respectively; 1.3–11.65 pooled), while total predicted prey biomass ranged from 23–304kg per 50ha. 6. Red deer density was the most important factor explaining wolf occurrence (31% contribution), followed by roe deer density (22%), winter precipitation (19%) and presence of game reserves (16%), showing that food supply, especially red deer as the most profitable prey in the Western Alps, was the main driver of winter habitat selection during the settlement phase. 7. Synthesis and applications. We demonstrate the crucial importance of including accurate, fine-grained information about prey supply for predicting recolonization patterns of carnivores and thus anticipating areas with potential human-wildlife conflicts where preventive measures should be prioritized.

Ungulate data were collected by means of snow-tracking along 218 1km long transects, during four subsequent winters starting in 2012/13. In order to distribute transects in a stratified manner, the study area was first divided into 34 10x10km squares. In each square, an average of six transects (range: 1-10) were then placed so as to cover the elevational and environmental gradients present in the study area as representatively as possible, while accounting for accessibility, topography and safety (avalanches). The transects were surveyed twice per winter (December to March). To reduce observer effects, each transect was visited by the same person during the whole study period, and the entire fieldwork was conducted by two experienced wildlife-biologists, who trained and standardized their methods for one full season (2011/12) prior to the start of this study. Tracks found in the snow were recorded for the main potential wolf prey: roe deer, red deer, chamois, ibex, mouflon and wild boar. Based on imprint size and track distribution, we estimated the minimal number of individuals present at each visit to a given transect. Multiple individuals were counted if tracks of different size (e.g. from different sexes or age classes) or individuals travelling together in a group could be distinguished. We used a conservative approach: when in doubt we always recorded the lower number.

To model the detection probability of ungulates in relation to the conditions during sampling, we recorded covariates of snow conditions during the transect surveys: average snow height, quality and percentage of snow cover. In addition, the daily amount of fresh snow recorded at all 56 weather stations across the study area was obtained , assigning each transect to the nearest weather station. For each transect walk, we calculated the number of days since the last snowfall. We also transformed this continuous variable into a categorical variable describing snow age. Finally, we recorded the amount of fresh snow that had fallen on the last day with snowfall previous to each survey.

The environmental variables used for predicting relative ungulate densities and wolf occurrence in a spatially explicit way were extracted from existing digital information, from which we produced basic raster layers with a 25x25m cell size. We obtained data on topography (altitude, slope, terrain roughness and topographic position) from the digital elevation model (DEM) of Switzerland swissALTI^3D. Roughness was calculated as the standard deviation of elevation of all cells within a predefined radius (564m), which corresponds to an area of 1km². The topographic position index (TPI) represents the position of a focal cell relative to the surrounding terrain and is calculated as the sum of angles to the ground measured in all eight cardinal directions. Land use and land cover (forest, rock, scree, water bodies, anthropogenic areas, roads and railways, ski lifts) were derived from vector25, a digital vector-map produced and regularly updated by the Swiss Federal Office for Topography swisstopo (https://www.swisstopo.admin.ch), with the exception of grassland cover and forest type information, which were derived from the area statistics dataset of Switzerland and Landsat5 data respectively, both provided by the Swiss Federal office of statistics (https://www.bfs.admin.ch). Winter temperature and precipitation stemmed from the Worldclim dataset (http://www.worldclim.org), which was downscaled from a 1km² raster to a resolution of 100x100m based on the SRTM-V4 digital elevation model and the method described in Zimmermann and Roberts (2001). Estimates of livestock (sheep and goat) densities were obtained by relating the number of livestock per community (as obtained from the Agricultural statistics of Switzerland) to the amount of pastures present in that community. Finally, spatial data on the location of game reserves was obtained from the Federal Office for the Environment (FOEN).

Ungulate data is stored in the files "transects_20xx.csv" for every survey season (2012 means the season 2012/13, 2013 means 2013/14,...)

These are the attributes:

DATE: The date of visiting the transect. Both visits per season were entered separately.

QUAD: Number of the 10x10km square used to distribute transects in a stratified manner

TRANS: Number of transect (orignially in each square many transects were considered, but only few chosen)

TR_ID: The unique ID of each transect (formed from square number and transect number)

CC: Number of observed minimal number of roe deer (Capreolus capreolus)

CE: Number of observed minimal number of red deer (Cervus elaphus)

RR: Number of observed minimal number of chamoix (Rupicapra rupicapra)

SS: Number of observed minimal number of wild boar (Sus scrofa)

CI: Number of observed minimal number of ibex (Capra ibex)

SNOW_H: recorded snow height (in cm)

SNOW_C: recorded snow quality (a>= 5 days old, f<5 days old)

SNOW_FO: snow cover in forested areas (%)

SNOW_OP: snow cover in open areas (%)

RUN: first or second time this transect was visited in the season

STAT: abbreviation of the name of the closest weather station

DSLSF: days since last snow fall

NWSN: height of fresh snow

S_C_C: Snow age (1: 0-1 day old; 2: 2-4 days old; 3: >5 days old)

 

Environmental data is stored in "buffer_data_vs_X00.csv", where 200 or 350 represents the radius of the buffer around the transect for which the data was extracted. For red deer, we used 350m, for roe deer and chamoix 200m.

The attributes are:

TR_ID: the unique ID of each transect

AREA: the area represented by the buffer (different because transects were not straight but curvy)

ALTI: average altitude in the buffer

ROUGH: roughness

SLOPE: average slope

NORTH: average northness

EAST: average eastness

TPI: topographical position index

FOR: percentage of forest cover

ROCK: percentage of rock cover

SCREE: percentage of scree cover

OT_SOIL: percentage of grassland cover

RIV_CR: percentage of rivers and creeks

LAKE_D: average distance to next lake

RIV_CR_D: average distance to next river or creek

WAT_DIS: average distance to next water body

DENSE_FO: percentage of dense forest cover

OPEN_FO: percentage of open forest cover

CONI_FO: percentage of coniferous forest cover

CONIMI_FO: percentage of coniferous-mixed forest cover

DECMI_FO: percentage of decidious-mixed forest cover

DECI_FO: percentage of decidious forest cover

FO_EDG_AL: length of all forest edge

FO_EDG_IN: length of inner forest edge

FO_EDG_OU: length of outer forest edge

RO_RA_D: average distance to roads/rails

SETTLE: percentage cover of settlements

SETTLE_D: average distance to nearest settlement

SKI_CAB_D: average distance to skilifts and cableways

STR_BIG_L: length of main roads (category 1-3)

STR_BIG_S: length of secondary roads (category 4-6)

HUN_RES: percentage cover of hunting ban areas

HUN_RES_PR: presence of hunting ban areas

SHE_GOA: Number of sheep and goats

PREC_S: average precipitation in summer months

PREC_W: average precipitation in winter months

TEMP_S: average temperature in summer months

TEMP_W: average temperature in winter months

DAYS_NO_SNOW: average annual number of days without snowcover