Skip to main content
Dryad logo

Integrating socio-ecological suitability with human-wildlife conflict risk: Case study for translocation of a large ungulate

Citation

McCann, Nicholas et al. (2021), Integrating socio-ecological suitability with human-wildlife conflict risk: Case study for translocation of a large ungulate, Dryad, Dataset, https://doi.org/10.5061/dryad.4b8gthtdc

Abstract

1. Translocations are essential for reestablishing wildlife populations. As they sometimes fail, it is critical to assess factors that influence their success pre-translocation.

2. Socio-ecological suitability models (SESMs) integrate social acceptance and ecological suitability to enable identification of areas where wildlife populations will expand, which makes it likely that SESMs will also be useful for predicting translocation success.

3. To inform site-selection for potential elk (Cervus canadensis) reintroduction to northeastern Minnesota, USA, we developed broad-scale maps of social acceptance from surveys of local residents and landowners, animal use equivalence (AUE) from forage measured in the field, and empirical conflict risk from geospatial data. Resulting SESMs integrated social acceptance favorability scores, AUE, and conflict risk, and weighted SESMs showed the relative influences of acceptance and conflict.

4. Social acceptance was positive for local residents and landowners (mean ≥ 5.4; scale of 1 to 7). AUE (scaled to an elk home range) ranged between 1 and 9 elk/16 km2 during winter, and from 14 to 83 elk/16 km2 during summer. Human-elk conflict risk was low (mean ≤ 0.10; scaled 0 to 1), increasing from north to south. Geographical distributions differed for social acceptance, AUE, and conflict risk, and weighted SESMs revealed unsuitable areas that were otherwise obscured.

5. Synthesis and applications. Integrating human-wildlife conflict risk into SESMs shows where social acceptance of translocated species is likely to erode, even where viewed favorably pre-translocation, to inform translocation planning by highlighting interactions between key factors. Such integrated models supplement existing reintroduction biology frameworks by supporting decision-making and knowledge development. In northeastern Minnesota, natural resource managers who are considering elk reintroductions are using SESMs reported here to identify where human-elk conflict is likely to result in an isolated elk population and where addressing concerns for area residents about conflict risk is essential.

Methods

2.1  |  Study area

We studied elk reintroduction to northeastern Minnesota, USA, where managers are deciding whether to reintroduce elk to 1 of 3 study areas (Cloquet Valley [CLV], Fond du Lac [FDL], and Nemadji [NEM]) that have abundant public land and low road density (mean = 0.96 km/km2) suitable for elk (Lyon 1983; Fig. 1A). In the northern lakes and forests ecoregion (Level III Region 50), the area was characterized by rolling topography, nutrient-poor soils, scattered lakes and rivers (Omernik & Griffith, 2014), and mixed-plant vegetative communities. Forests were coniferous and northern hardwood (often mixed). Maps we developed buffered study areas 20 km (elk dispersal distance; Ryckman et al. 2010).

2.2  |  Social acceptance

We used local resident and landowner surveys to map social acceptance for elk. We mailed 4,000 local residents a questionnaire (stratified random sampling; stratum levels were census blocks, highways, and rivers; see Walberg et al. 2019) about attitudes toward translocating elk and measured attitudes using returned questionnaires (N = 1,521) scored from very unfavorable (1) to very favorable (7). To develop a social acceptance model, we mapped mean attitude scores within a circular moving window (size needed for continuous surface; 4 townships [372 km2]).

We used the same methods for landowners (owners of ≥ 4 ha properties located ≤ 8 km from a study area). Landowner properties were near likely reintroduction sites (outside residential areas and near public lands), making them areas where post-reintroduction human-elk interactions were most likely. We used a stratified random sampling approach to select 4,500 landowners, where stratum was property size (levels: 4-16 ha and >16 ha). Mapped scores were from 2,585 returned questionnaires (including 35 from local residents who met landowner criteria).

2.3  |  Forage

We measured trees, shrubs, and understory vegetation between 14 June and 8 August 2017 (season 1) and 6 June and 8 August 2018 (season 2). Sites we sampled were distributed throughout the study areas in a Geographic Information System (ESRI 2018, R Core Team 2019; Any use of trade, firm, or product names is for descriptive purposes only and does not imply endorsement by the U.S. Government). During season 1 we randomly distributed points on roads that abutted public land, and then randomly sampled 1 point in each vegetated landcover type (Rampi, Knight, & Bauer, 2016) within 50 to 500 m from the road point (to improve logistics). To achieve sampling that was even through space (with respect to roads) and time (date), we systematically sampled study areas and 4 rectangular quadrants overlaid on them. We repeated this process during season 2 while sampling private properties from landowner questionnaire respondents. As elk in eastern North America select regenerating aspen (Populus tremuloides) as winter forage (D. A. Jenkins et al., 2007) we expanded sampling to include randomly selected points in aspen stands harvested ≤ 10 years ago (N = 4 plots per study area; M.P. Westphal, Carlton County; D. Ryan, US Forest Service; J. Kelash, Pine County; and B. Hakala, Saint Louis County; Unpublished data).

We established a circular plot (field plot) centered on each landcover point. Field plots consisted of nested circles (Fig. 1B). In the largest circle (r =11.3 m) we collected ground cover vegetation from 10 150 cm × 10 cm quadrats. In each quadrat we clipped, dried, and massed woody (height < 15 cm) and herbaceous vegetation (classes: grass, forb, sedge, rush, fern, and woody). In 3 medium circles (r = 2.8 m) radiating 5.5 m from the plot center (120° increments) we used a stepped diameter gauge (Paul et al., 2017) to count trees and shrubs with DBH 2.54 cm to 10 cm. In a small circle (r = 1.8 m) centered within each medium circle we used a stepped gauge to count trees and shrubs (≥ 15 cm tall and < 2.54 cm diameter at 15 cm height; D15).

During season 2 we sampled open areas near roads, railroad tracks, and pipelines (right-of-ways; ROWs). We used a stratified random sampling approach to select ROWs; strata were study area and road class (MNDOT, 2017). Railroad and pipeline ROWs were adjacent to tracks and pipeline service roads, and 50 to 500 m (randomly selected) from a road intersection (MNDOT, 2015). At each ROW, we established a 200 m2 rectangular plot (ROW plot). We measured the distance between the road/railroad edge and the nearest tree/shrub line to calculate plot width (maximum of 30 m). Plot length was plot width divided by 200. Once we established a ROW plot, we clipped, dried, and massed ground cover vegetation from 5 quadrats.

We estimated forage biomass (non-avoided species) at each field plot during leaf-on (summer) and leaf-off (winter) because it correlates positively with elk habitat selection across multiple spatial scales (Anderson et al., 2005; Hebblewhite et al., 2008; Merrill et al., 2020). Summer forage was deciduous shrub and tree leaves, forbs, and grasses (not sedges, rushes, and ferns; Schneider et al. 2006, Lupardus et al. 2011). To estimate shrub and sapling leaf biomass, we summed woody stem counts at field plots and used diameter-specific allometric equations (J. C. Jenkins, Chojnacky, Heath, & Birdsey, 2004; Perala & Alban, 1993; Smith & Brand, 1983). Winter forage was deciduous twigs (not grasses; see Jenkins et al. 2007). We used allometric equations to estimate total above ground deciduous shrub and tree biomass (J. C. Jenkins et al., 2004; Perala & Alban, 1993; Smith & Brand, 1983) and estimated twig biomass as the product of total biomass and the proportion that is current year growth consumed by ungulates in Minnesota (0.07; Ohmann et al. 1974, 1976). Forage biomass was for shrubs and trees ≤ 2.54 cm in diameter (D15), corresponding with mean elk foraging height (1.5 m; Rounds 2006, Gehring et al. 2008, VanderSchaaf 2013). We used ANOVAs to determine if forage (square root transformed) differed by landcover and ownership (public or private; stats package in R) and to test for differences in ROW type summer forge. Tukey tests followed significant ANOVAs, α was 0.05, and the variance inflation factor assessed collinearity.

We used random forest (RF) analysis to model field plot forage biomass (Breiman, 2001; Cutler et al., 2007). RF fits a large number of regression trees, with each tree constructed using a random subset of data and predictors. This results in estimates that are essentially cross-validated, making it unnecessary to divide data into training and test sets (Chen et al., 2017; Prasad, Iverson, & Liaw, 2006). RF results in predictions of the dependent variable, and measures of accuracy and variable importance (Liaw & Wiener, 2002). Frequently used in geospatial modeling (Evans & Cushman, 2009; Rodriguez-Galiano, Ghimire, Rogan, Chica-Olmo, & Rigol-Sanchez, 2012), RF models nonlinear relationships and interactions without error distribution assumptions (Cutler et al., 2007), and is robust to missing data (Rodriguez-Galiano et al., 2012) and overfitting (Breiman, 2001), yielding more accurate results than other methods (Chen et al., 2017; Prasad et al., 2006).

To estimate field plot forage biomass, we implemented RF (randomForest package; Liaw and Wiener 2002) using biologically relevant predictors extracted from 15 m resolution rasters (Table 1). We eliminated correlated independent variables (Millard and Richardson 2015; Spearman correlation coefficient |rs| > 0.5; stats package in R) by keeping only the variable that resulted in greater predictive accuracy (Gustafson, Lietz, Wright, & Ecologist, 2003). RF models predicted leaf or total biomass by growing 1,000 regression trees. Using RF forage estimates (not spatially autocorrelated; Moran’s I test, P > 0.25; spdep package in R; Millard and Richardson 2015, Bivand and Wong 2018), we estimated summer and winter forage across the mapped area (Hijmans, 2019). ROW forage was overlaid after assigning mean forb and grass biomass (field measurements). Mean pipeline forage was used for unsampled high voltage power lines (similarly sized openings and vegetation; Minnesota Geospatial Information Office 2016).

2.4  |  Animal unit equivalence

To estimate AUE we used RF forage estimates summed within 16 km2 circular moving windows to match an elk home range (O’Neil & Bump, 2014), using: AUE = (F × C) / (S × M × D); where F was forage, S was dry forage (as % elk mass) required to sustain 1 elk of mass M for 1 day during a season lasting D days, and C was a correction factor reflecting how much forage elk consume. AUE was for cow elk (M = 250 kg; Bender et al. 2006), consuming 2.1% of M daily

(Christianson & Creel, 2009) during a 200 d winter, and 2.2% of M daily (Kuzyk & Hudson, 2007) during a 165 d summer. To account for shrubs and trees not consumed within their use areas, we estimated that elk consume the same proportion of available forage as do moose (Alces; C = 0.03 of forage; Peek et al. 1976, Edenius et al. 2002). Resulting maps estimated elk supported by the surrounding 16 km2, a scale that is biologically significant to elk (home range; (O’Neil & Bump, 2014). We reported the mean and standard deviation (SD) of raster map values in each study area but did not develop statistical comparisons as the large number of raster cells (CLV: 7,841,931 cells; FDL: 3,402,931 cells;  NEM: 4,279,849 cells), made P-values uninformative (Lin, Lucas, & Shmueli, 2013).

2.5  |  Human-elk conflict risk

To estimate empirical human-elk conflict risk we developed a surface reflecting the proportion of area (16 km2 circular window) that was the sum of row crop, hay/pasture, feedlot, and road.

Surface values were expected probabilities of elk traversing these features. Crop and hay/pasture were from landcover data (Rampi et al., 2016). Road surface was centerlines (MNDOT, 2017) buffered by mean road widths (field measurements by class). We estimated feedlot area by buffering cow, horse, and pig feedlot locations by 0.12 km2 (median of 20 random feedlots; MPCA 2007) and added all 4 captive cervid operations within the mapped area (Minnesota Board of Animal Health, unpublished data). Nearly all 923 feedlots had areas where elsewhere in elk range livestock-elk contact occurs and elk raid forage (98% had open lots and pasture and 84% had holding areas; MPCA 2007).

2.6  |  Socio-ecological suitability

To develop integrated socio-ecological suitability maps we calculated mean AUE (summer and winter, weighted equally), social acceptance (residents and landowners, weighted equally), and empirical conflict risk after normalizing map values between 0 and 1 (min-max scalar). We subtracted conflict risk from 1 before normalization to ensure all values corresponded positively with suitability. The resulting map weighted AUE, acceptance, and conflict equally. To assess their relative influence, we developed maps that weighted acceptance and conflict (each weighted 2- and 3-times baseline map value). Using continuous variables enabled us to assess an array of interacting conditions, thereby obviating potentially-influential thresholds where continuous data are discretized into categorical datasets (Behr et al., 2017).

Funding

Minnesota Environment and Natural Resources Trust Fund

Minnesota Cooperative Fish and Wildlife Research Unit

Fond du Lac Resource Management Division

Great Lakes Restoration Initiative

Minnesota Agricultural Research Station, Award: Project# MIN-41-020

Rocky Mountain Elk Foundation