Article Habitat Characteristics Coincidence of Dead and Living Long-Tailed Gorals (Naemorhedus caudatus) According to Extreme Snowfall

Hee-Bok Park 1 and Sungwon Hong 2,3,*

1 Restoration Research Team (), Research Center for Endangered , National Institute of Ecology, Yeongyang 36531, ; [email protected] 2 Department of Science and Biotechnology, Kyungpook National University, Sangju 37224, Korea 3 Department of Horse, Companion, and Wild Animal Science, Kyungpook National University, Sangju 37224, Korea * Correspondence: [email protected]; Tel.: +82-54-530-1221; Fax: +82-54-530-1959

Simple Summary: The long-tailed goral (Naemorhedus caudatus Milne-Edwards) is a critically endan- gered herbivore in South Korea. From March to June in 2010, 24 animals were found to have died due to heavy snowfall in the Wangpi Stream basin. In this study, we hypothesized that gorals that died due to snowfall are low-status individuals that lived in the sub-optimal or non-suitable areas. The results suggested that the sites where dead gorals were found were highly related to typical goral habitats. The optimal goral habitats could become uninhabitable following heavy snowfall. Most of the dead animals were pregnant females or were young, implying that they could not escape their primary habitats due to lower mobility. Thus, when there is a climate catastrophe, the optimal

 goral habitats should be considered for rescue and artificial feeding.  Abstract: The long-tailed goral (Naemorhedus caudatus) is a critically endangered herbivore in South Citation: Park, H.-B.; Hong, S. Habitat Characteristics Coincidence Korea. Despite government efforts to recover the population through reintroduction programs, the of Dead and Living Long-Tailed animal remains vulnerable to heavy snowfall. From March to June 2010, 24 animals were found dead Gorals (Naemorhedus caudatus) due to heavy snowfall in the Wangpi Stream basin. In this study, we hypothesized that gorals that According to Extreme Snowfall. died due to snowfall are low-status individuals that lived in the sub-optimal or non-suitable areas. Animals 2021, 11, 997. https:// Using the occurrence data from extensive field surveys from 2008 to 2010 in the Wangpi Stream and doi.org/10.3390/ani11040997 the carcass location data, we (1) defined the goral habitat characteristics and (2) compared the habitat characteristics between dead and living gorals using ensemble species distribution modeling. The Academic Editor: Luciano Bani results suggested that the sites where dead gorals were found were highly related to typical goral habitats. These results implied that the optimal goral habitats could become uninhabitable following Received: 1 March 2021 heavy snowfall. Most of the dead animals were pregnant females or were young, implying that Accepted: 31 March 2021 they could not escape their primary habitats due to lower mobility. Thus, when there is a climate Published: 2 April 2021 catastrophe, the optimal goral habitats should be considered for rescue and artificial feeding.

Publisher’s Note: MDPI stays neutral Keywords: survival; climate change; population viability analysis; ensemble species distribu- with regard to jurisdictional claims in published maps and institutional affil- tion model iations.

1. Introduction Extreme winter seasons can affect the foraging behaviors of herbivores and can result Copyright: © 2021 by the authors. Licensee MDPI, Basel, Switzerland. in increased fatality [1]. Food deprivation causes a deficiency in energy in herbivores [2]. This article is an open access article Moreover, low temperatures during winter lead to high-energy requirements [3]. Besides distributed under the terms and this, warm spells in the winter with rainfall (rain-on-snow) can occasionally cause “icing,” conditions of the Creative Commons which restricts access to food, resulting in starvation and lower survival and fecundity Attribution (CC BY) license (https:// rates [4]. creativecommons.org/licenses/by/ Long-tailed goral (Naemorhedus caudatus) belongs to four Naemorhedus species of Capri- 4.0/). nae, which are endangered and are one of the least studied montane bovid in the world.

Animals 2021, 11, 997. https://doi.org/10.3390/ani11040997 https://www.mdpi.com/journal/animals Animals 2021, 11, x 2 of 12

which restricts access to food, resulting in starvation and lower survival and fecundity rates [4]. Long-tailed goral (Naemorhedus caudatus) belongs to four Naemorhedus species of , which are endangered and are one of the least studied montane bovid in the world. The animal is distributed in Eastern and Northern Asian mountains, including Russia, , and Korea [5,6]. This species is listed as a vulnerable species by the Inter-

Animals national2021, 11, 997 Union for Conservation of Nature and Natural Resources (IUCN) [7]. Whereas,2 of 13 in South Korea, the species has been protected as a critically endangered species by the Ministry of Environment and natural monument no. 217 by the Cultural Heritage Admin- istration. As of 2018,The the animal estimated is distributed population in Eastern was andonly Northern 679 individuals Asian mountains, in representative including Russia, habitats [8]. This smallChina, popu and Korealation [ 5has,6]. This survived species isas listed they as in ahabit vulnerable deep species valleys by theand International cliffs where humans andUnion predators for Conservation cannot easily of Nature approach and Natural them Resources[9]. (IUCN) [7]. Whereas, in South Korea, the species has been protected as a critically endangered species by the Ministry of The species hasEnvironment been protected and natural under monument the reintroduction no. 217 by the Culturalpolicy Heritageof the Korean Administration. gov- As ernment. However,of they 2018, still the estimatedface several population threats, was such only 679as road-kills individuals inand representative climate change habitats [8]. [10,11]. In particular,This heavy small populationsnowfall hasrepresents survived asa mortality they inhabit risk deep [12]. valleys Gorals and cliffs evolved where humansto have shorter forelegsand than predators hind-legs cannot in easily order approach to quickly them [climb9]. cliffs; however, this means The species has been protected under the reintroduction policy of the Korean govern- that, when they are ment.trapped However, by heavy they still snow, face severalthey cannot threats, escape such as road-killsand are andlikely climate to die change from [10 ,11]. starvation [13]. In particular, heavy snowfall represents a mortality risk [12]. Gorals evolved to have shorter Two notable incidentsforelegs than occurred hind-legs due in order to toabno quicklyrmal climb heavy cliffs; snowfalls however, this in meansSouth that, Korea. when they The first was, from Marchare trapped 1964 by to heavy February snow, they1965, cannot approximately escape and are 6000 likely gorals to die were from starvation trapped [13]. by snow and died. The Twosecond notable was, incidents from occurredMarch to due June to abnormal 2010, 24 heavy gorals snowfalls were infound South dead Korea. The first was, from March 1964 to February 1965, approximately 6000 gorals were trapped by in Uljin, a small countrysidesnow and died. county The [12,14]. second was, In 2010, from Marchthe fresh to June snow 2010, cover 24 gorals depth were was found the dead second highest for in10 Uljin, years, a small and countrysidethe volumes county of sunshine [12,14]. In 2010,and air the freshtemperature snow cover for depth six was months (from Januarythe secondto June) highest were forremarkably 10 years, and lower the volumesthan those of sunshine between and 1971 air temperatureand 2000 for (Figure 1). In recentsix years, months strict (from policies January reduced to June) were their remarkably fatality from lower human than those activities, between 1971re- and 2000 (Figure1). In recent years, strict policies reduced their fatality from human activities, sulting in severe climaticresulting conditions in severe climatic being conditions the most being crucial the most factor crucial decreasing factor decreasing the goral the goral population currently.population Thus, it currently. is vital Thus,to understand it is vital to why understand the gorals why the die gorals due dieto dueabnormal to abnormal climatic conditions climaticto reduce conditions the fatality to reduce of future the fatality catastrophes. of future catastrophes.

FigureFigure 1. Climatic 1. Climatic conditions conditions in 2010 in compared 2010 compared to those in otherto those years. in ( aother) Monthly years. average (a) Monthly fresh snow average cover depth fresh (cm) forsnow three monthscover depth (January: (cm) dark for bar, three February: months intermediate (January: dark dark bar, bar, and March:February: light bar)intermediate from 2001 to dark 2010, bar, (b) monthly and averageMarch: sunshine light bar) (hr) andfrom (c) 2001 air temperature to 2010, ( forb) sixmonthly months (fromaverage January sunshine to June) (hr) of 2010 and (dotted (c) air line), temperature compared to for that betweensix months 1971 and (from 2000 January (line) and to the June) monthly of number2010 (dotte of deadd line), gorals. compared to that between 1971 and 2000 (line) and the monthly number of dead gorals. Gorals live in flexible, hierarchical social groups [13,15]. They follow a polygyny mating system and live in groups that can comprise over 11 animals, depending on Gorals live in flexible,resource availability;hierarchical however, social thegroups males [13,15]. usually live They in solitude, follow a except polygyny during mat- the rutting ing system and liveseason in groups [16,17 ].that Gorals can prefercompri tose live over on exposed 11 animals, rocks ordepending on the south-facing on resource regions of availability; however,steep the valleys males where usually snow live melts in solitude, easier and except can access during their the preferred rutting food. season However, gorals that are lower in the social hierarchy or yearlings could be expelled to live in areas [16,17]. Gorals prefer to live on exposed rocks or on the south-facing regions of steep val- more vulnerable to heavy snowfall [18]. Besides, in areas with soil with a weak draining leys where snow meltscapacity, easier snow and does can not meltaccess quickly, their which preferred can be fatalfood. for However, gorals. Thus, gorals we hypothesized that are lower in the socialthat hierarchy the characteristics or yearlings of both could the preferred be expelled habitats to of live gorals in andareas the more locations vul- where nerable to heavy snowfalldead gorals [18]. were Besides, found in would areas be with different. soil Inwith this a study, weak we draining aimed to capacity, (i) identify the snow does not melthabitat quickly, characteristics which can of be gorals, fatal (ii)for compare gorals. habitatThus, we characteristics, hypothesized and (iii) that predict the the vulnerable areas. characteristics of both the preferred habitats of gorals and the locations where dead gorals were found would be different. In this study, we aimed to (i) identify the habitat charac- teristics of gorals, (ii) compare habitat characteristics, and (iii) predict the vulnerable areas. Animals 2021, 11, x 3 of 12

2. Materials and Methods 2.1. Study Area Extensive field surveys were conducted in the Wangpi River basin in the eastern part of South Korea (Figure 2a). The selected area was the remote not-industrialized country- side (urban area percentage: 1.00%) [17], which had a high forest cover (over 92.17%) and Animals 2021, 11, 997 3 of 13 many streams. A small portion of the land is used for agriculture (approximately 4.96%). In 2018, the Korean red pine (Pinus densiflora Siebold and Zucc.; Pinaceae) was the domi- nant flora (approximately 46%), while Oak (Fagaceae spp.) occupied 37.66%. Over 90% of trees in the area are2. Materialsaged between and Methods 30 and 60 years (Korean National Spatial Data Infra- 2.1. Study Area structure Portal, http://www.nsdi.go.kr (1 April 2021). The area has one ofExtensive the highest field surveysgoral popula were conductedtions in South in the Korea Wangpi [14], River and basin gorals in the in eastern part of South Korea (Figure2a). The selected area was the remote not-industrialized countryside the area have distinct genotypes from other populations, probably due to the natural cli- (urban area percentage: 1.00%) [17], which had a high forest cover (over 92.17%) and many matic and geographicstreams. barriers, A small such portion as a of hi thegh land mountainous is used for agriculture plateau [19]. (approximately However, 4.96%).the In 2018, study area does containthe Korean areas red with pine different (Pinus densiflora populationSiebold densi andties. Zucc.; The Pinaceae) northern was areas the (red dominant flora polygon in Figure (approximately2b of the Gwangcheon 46%), while St Oakream (Fagaceae watershed spp.) have occupied higher 37.66%. population Over 90% den- of trees in the sities than the southernarea are areas aged (violet between polygon 30 and 60of yearsFigure (Korean 1b, evidenced National Spatialby the Datadensity Infrastructure of the Portal, traces such as feces,http://www.nsdi.go.kr marking, etc. [20]. (accessed on 1 April 2021).

FigureFigure 2. (a) Land2. (a) use Land in Southuse in Korea South and Korea (b) study and ( sitesb) study (red: sites urban, (red: yellow: urban, agriculture, yellow: agriculture, dark green: forest, dark light green: grassland,green: violet: forest, wetland, light green: emerald: grassland, bare land, violet: and blue: wetland, water). emerald: The emerald bare polygon land, and represents blue: water). the protected The area. Light and darkemerald red points polygon represent represents the locations the protected of goral carcassesarea. Light and and hair dark samples, red respectively.points represent the locations of goral carcasses and hair samples, respectively. The area has one of the highest goral populations in South Korea [14], and gorals in the 2.2. Field Survey andarea Carcass have distinctSurvey genotypes from other populations, probably due to the natural climatic and geographic barriers, such as a high mountainous plateau [19]. However, the study area Surveys were conducted based on the line transect method. The survey sites were does contain areas with different population densities. The northern areas (red polygon in determined basedFigure on remoteness2b of the Gwangcheon from urban Stream areas watershed and the haveproportion higher populationof the rocky densities re- than the gions by reviewingsouthern the literature areas (violet on goral polygon habi oftats. Figure Most1b, of evidenced the sitesby were the densitysteep rocky of the val- traces such as leys with a small humanfeces, marking, population etc. [ 20[13].]. The surveys were conducted five times (March, April, May, June, and September) in 2008, three times in 2009, and twice in 2010. To elucidate the2.2. habitat Field Survey of the and gorals, Carcass we Survey searched for hairs of the gorals indicating their preferred habitats.Surveys When weregorals conducted spend more based time on thein a line given transect area method.or mark Thetheir survey terri- sites were tory by using theirdetermined horns, they based are likely on remoteness to release from hairs urban at the areas sites and or the have proportion hair caught of the rockyin regions by reviewing the literature on goral habitats. Most of the sites were steep rocky valleys branches [9]. We used hair and feces to determine the preferred habitats of the gorals, as with a small human population [13]. The surveys were conducted five times (March, April, footprints can be confusedMay, June, with and September)those of ot inher 2008, herbivores three times such in 2009,as Siberian and twice roe in 2010. (Cap- reolus pygargus Pallas) Toand elucidate feral the habitat ( of hircus the gorals, Linnaeus) we searched [9,20]. for The hairs location of the goralsof all indicating hair samples was recordedtheir preferred using habitats. a GPS recor Whender gorals (Garmin, spend 60csx). more timeThe forest in a given rangers area and or mark their residents found theterritory carcasses by using of dead their gorals, horns, they and are they likely were to release only hairsfound at in the the sites northern or have hair caught parts of the study inarea branches [21]. [9]. We used hair and feces to determine the preferred habitats of the gorals, as footprints can be confused with those of other herbivores such as Siberian ( pygargus Pallas) and feral goats (Capra hircus Linnaeus) [9,20]. The location of all hair samples was recorded using a GPS recorder (Garmin, 60csx). The forest rangers Animals 2021, 11, 997 4 of 13

and residents found the carcasses of dead gorals, and they were only found in the northern parts of the study area [21].

2.3. Environmental Factors Influencing the Habitats of Goral Based on previous studies, we tentatively selected 27 environmental factors to char- acterize the habitat conditions of the dead and living gorals (Table1). Previous studies only considered a few landscape factors, such as altitude, slope, aspect, and anthropogenic factors, such as distance to the road and trail. However, in this study, we considered more factors, such as distance to agricultural fields, factories, human residences, and soil characteristics to explore the primary goral habitats (Table1)[ 14,22–24]. We considered the density of the anthropogenic, landscape, and forest variables within the male home range (1.46 km2; radius: 682 m measured by GPS collars) [25]. We primarily considered the male home range because typically, the male home range is larger than that of the female due to their solitariness, as such, their range would often encompass the female home range. As soil characteristics might affect the mortality or habitat of the gorals due to the differing snow melting ability related to the thermal cover or vegetation distribution, we added these characteristics as important variables [26]. The study sites are relatively small fragmented habitats, and we did not consider the meteorological factors to differentiate the habitats. The resolution of all variables was resampled to 30 x 30 m using the resample function of the raster R package [27].

Table 1. Description, abbreviation, related references, and source map (website, resolution, and making year) of the 27 selected variables.

Description and No Classification Factors Hypothesis from Reference Source Map Literature 1 Distance to factory (m) 2 Distance to road (m) 3 Road density (m2/1.46 km2) EGIS Anthropo Human activity can Human population density (https://egis.me.go.kr genic fac negatively affect the goral [22–24,28] (n/1.46 km2) (accessed on 1 April tors habitats. 4 Distance to rice paddy (m) 2021), 30 × 30 m, 2007) 5 Distance to field (m) 6 Field density (m2/1.46 km2) 7 Distance to trail (m) National Spatial Data Infrastructure Portal Distance to deciduous 8 (http://nsdi.go.kr forest (m) (accessed on 1 April Types of forest can 2021), 5 × 5 m, 2018) Forest Density of deciduous forest influence goral’s feeding [13,15,23] 9 (m2/1.46 km2) and marking sites. Density of coniferous forest 10 (m2/1.46 km2) Density of forest 11 (m2/1.46 km2) Age class 12 (1 (young) to 9 (old)) Diameter class 13 (0 (young) to 3 (large)) Crown density 14 (1 (sparse) to 3 (dense)) EGIS (https://egis.me.go.kr 15 Distance to stream (m) Gorals used to live near (accessed on 1 April [14,15] the waterway. 2021), 10 × 10 m, 2013) Landscape Stream density 16 (m2/1.46 km2) Lake density 17 (m2/1.46 km2) Gorals used to live at Biz-gis (www.biz-gis.com high altitudes to avoid 18 Altitude (m) [17,24] (accessed on 1 April anthropogenic 2021), 30 × 30 m) disturbance. Animals 2021, 11, 997 5 of 13

Table 1. Cont.

Description and No Classification Factors Hypothesis from Reference Source Map Literature As snow can be more Biz-gis (www.biz-gis.com easily melt, the gorals 19 Aspect (1 to 360) [14,15,29] (accessed on 1 April prefer the habitats in 2021), 30 × 30 m) southern regions. Gorals prefer to live in Biz-gis (www.biz-gis.com 20 Slope (◦) the deep valleys to avoid [22,28] (accessed on 1 April humans. 2021), 30 × 30 m) If water cannot be displaced, snow becomes National Spatial Data stacked. The large Infrastructure Portal Soil drainage 21 quantity of snow can [14] (http://nsdi.go.kr (1 (bad) to 4 (very good)) negatively affect the (accessed on 1 April goral’s escape from the 2021), 25 × 25 m, 2018) Soil map snow. Gorals prefer to live in front of large rocks to Degree of rock exposure escape from predation. 22 (1 (little) to 4 (a lot)) Besides, stacked snow [13,15,26] melts more quickly than in other soil. Strongly eroded soil Soil type 23 stacked by snowfall could (1 (no) to 3 (many)) make gorals fall easily. Exposure to wind can Degree of wind exposure 24 cause snow to remain for [4,14] (1 (exposed) to 3 (closed)) a long period. 25 Distance to wet soil (m) Moist soil delays the rate [14] Wet soil density of snow melting. 26 (m2/1.46 km2) Slant type Gorals prefer to live in 27 (1 (rising slope) to 3 convex slopes to escape [13] (falling slope)) from predation.

2.4. Definition of Habitat Characteristics of Gorals We built different species distribution models (SDMs) to define the habitat suitability and relationship between goral occurrence and environmental factors [30,31]. We used goral presence data from the line transect method so that the data could be spatially auto- correlated. Thus, before building the models, we cleaned some occurrence points within the home range of goral using the spThin package [32]. We adapted eight model algo- rithms [33]: two envelopes and distance-based approaches (BIOCLIM and Domain [34]), two regression-based approaches (generalized linear models (GLMs [34]) and generalized additive models (GAMs [35]), and four machine-learning methods (classification and re- gression trees (CART [36]), random forests (RFs [37]), boosted regression trees (BRTs [38]), and Maxent [39]). The Fine-tuning of each model was conducted according to the instruc- tions from the practices by Dr. Zurell (https://www.damariszurell.io/HU-GCIB/5_SDM_ algorithms.html) (accessed on 1 April 2021). Before building the model, we evaluated the collinearity and removed the collinear variables by assessing the variance inflation factors (cut-off value = 5) with the vifstep function of the “usdm” packages in R Studio v1.1.456 [40]. Then, we excluded eight variables, especially the distance to a rice paddy, soil drainage, tree age class, degree of rock exposure, distance to a factory, stream density, aspect, and tree diameter class. We explored the relationships between the goral presence and each environmental variable in order to identify the relevant variables. First, we validated the predictive power by adding the variables using 10-fold cross-validation procedures to estimate the area under the ROC curve (AUC ≥ 0.80) and True Skill Statistic (TSS) (≥0.6). The procedure was repeated 50 times, and the results were averaged [41,42]. If the added variables did not improve the model accuracy (i.e., average variable importance ≤ 5), we extracted them Animals 2021, 11, 997 6 of 13

and added other ones. Second, if the plotted models did not show reasonable relationships, we did not consider those variables [43]. For example, when the relationship between the presence of gorals and the distance to streams exhibited quadratic shapes, which could not be explained, we considered the corresponding variable as a non-influential factor. Finally, after determining all the influential variables, we conducted 10-fold cross-validation and selected the model algorithms that showed a high goodness of fit (ROC curve ≥ 0.80 and TSS ≥ 0.60). We then used the median of the model performance [30]. We used the Animals 2021, 11, x 6 of 12 varImp function of the caret R package [44] to define the variable importance. For the algorithms for which the R packages do not provide importance (BIOCLIM, Domain, and Maxent from maxnet R package), we calculated the importance based on the ROC curve (importancevariables [44,45]. = 1 − AUCThen,) bythe excludingrelative importan each includedce was variablecalculated from by thedividing full combination the variable ofimportance variables [ 44by, 45the]. total Then, importance. the relative importance was calculated by dividing the variable importance by the total importance. 2.5. Definition of Site Characteristics of Dead Gorals 2.5. DefinitionWhen we of Sitedetermined Characteristics the goral of Dead presence Gorals sites, we clipped the areas in the upper studyWhen regions we determined(red polygon the in goral Figure presence 2b due sites,to considerably we clipped different the areas goral in the densities upper study [20]. regionsThen, we (red made polygon 50 background in Figure2b points due to in considerably the polygon different to differentiate goral densities the habitat [20]. charac- Then, weteristics made of 50 dead background gorals using points the in therandomPoints polygon to differentiatefunction of the the dismo habitat R characteristics package [27]. ofBy deadfollowing gorals the using previous the randomPoints procedures (identification function of the of dismo relevant R package variables) [27 of]. Byvariable following selec- thetion previous using ensemble procedures SDMs, (identification we defined of the relevant habitat variables) characteristics of variable of dead selection gorals (Figure using ensemble3). SDMs, we defined the habitat characteristics of dead gorals (Figure3).

FigureFigure 3. 3.( a(a)) Procedures Procedures of of ensemble ensemble species species distribution distribution modeling modeling to to predict predict the the potential potential distribu- distri- bution of gorals (orange polygon) and areas with a high mortality rate (blue polygons). (b) An tion of gorals (orange polygon) and areas with a high mortality rate (blue polygons). (b) An example example of the predicted habitat of the gorals and areas with a high mortality rate. of the predicted habitat of the gorals and areas with a high mortality rate.

3.3. Results Results 3.1.3.1. Habitat Habitat Characteristics Characteristics of of Gorals Gorals in in Uljin Uljin TheThe models models that that reached reached fair fair levels levels of of goodness-of-fit goodness-of-fit were were RFs RFs (AUC (AUC = = 1.00 1.00± ±0.00 0.00 andandTSS TSS == 1.001.00 ±± 0.00;0.00; Average Average ±± StandardStandard error) error) and and BRTs BRTs (AUC (AUC = 0.81 = 0.81 ± 0.002± 0.002 and TSSand = TSS0.64 = ± 0.64 0.003;± 0.003 Figure; Figure S1). The S1). ensemble The ensemble model modelhad high had levels high of levels goodness-of-fit of goodness-of-fit (AUC = (AUC0.87 and = 0.87 TSS and = 0.69) TSS =and 0.69) suggested and suggested that the that distance the distance to the to field the field(m), (m),coniferous coniferous tree treeden- 2 densitysity (m (m2), and), and distance distance to tothe the trail trail (m) (m) influenced influenced the the goral goral habitats habitats (Figure (Figure 4).4). Distance Distance to tothe the field field (m, (m, 0.59 0.59 ± 0.09)± 0.09) was was identified identified as the as themost most influential influential factor factor (Figure (Figure 4a). 4Thea). Theprob- probabilityability of the of thegoral goral increased increased in regions in regions with with a high a high density density of ofconiferous coniferous trees trees and and in- increasedcreased as as the the distance distance to to fields fields and and trails trails increased increased (Figure (Figure 44b).b).

Figure 4. Three selected variables of ensemble model with the algorithms of Random Forest and Boosted Regression Tree: (a) relative variable importance (%) and (b) relationship between goral presence probabilities and the three relevant factors. Animals 2021, 11, x 6 of 12

variables [44,45]. Then, the relative importance was calculated by dividing the variable importance by the total importance.

2.5. Definition of Site Characteristics of Dead Gorals When we determined the goral presence sites, we clipped the areas in the upper study regions (red polygon in Figure 2b due to considerably different goral densities [20]. Then, we made 50 background points in the polygon to differentiate the habitat charac- teristics of dead gorals using the randomPoints function of the dismo R package [27]. By following the previous procedures (identification of relevant variables) of variable selec- tion using ensemble SDMs, we defined the habitat characteristics of dead gorals (Figure 3).

Figure 3. (a) Procedures of ensemble species distribution modeling to predict the potential distri- bution of gorals (orange polygon) and areas with a high mortality rate (blue polygons). (b) An example of the predicted habitat of the gorals and areas with a high mortality rate.

3. Results 3.1. Habitat Characteristics of Gorals in Uljin The models that reached fair levels of goodness-of-fit were RFs (AUC = 1.00 ± 0.00 and TSS = 1.00 ± 0.00; Average ± Standard error) and BRTs (AUC = 0.81 ± 0.002 and TSS = 0.64 ± 0.003; Figure S1). The ensemble model had high levels of goodness-of-fit (AUC = 0.87 and TSS = 0.69) and suggested that the distance to the field (m), coniferous tree den- sity (m2), and distance to the trail (m) influenced the goral habitats (Figure 4). Distance to

Animals 2021the, 11, 997field (m, 0.59 ± 0.09) was identified as the most influential factor (Figure 4a). The prob-7 of 13 ability of the goral increased in regions with a high density of coniferous trees and in- creased as the distance to fields and trails increased (Figure 4b).

Animals 2021, 11, x 7 of 12

FigureFigure 4. Three 4. selected Three variablesselected of variables ensemble modelof ensemble with the model algorithms with of Randomthe algorithms Forest and of Boosted Random Regression Forest Tree:and (a) relative variable importance (%) and (b) relationship between goral presence probabilities and the three relevant factors. Boosted Regression Tree: (a) relative variable importance (%) and (b) relationship between goral presenceIt was probabilities observed andItthat was ththere observede three were relevant that large there clusters factors. were large located clusters in the located central in thelongitudinal central longitudinal areas areas (Figure 5), whereas(Figure there5), whereaswere a few there optimal were a fewareas optimal in the areaswestern in the regions. western However, regions. However, the the suitability was distinctivelysuitability was lower distinctively in eastern lower areas in eastern near the areas seashore near the (Figure seashore 5a). (Figure When5a). When thresholds were thresholdsdivided into were presence divided and into abse presencence areas, and absencethe clusters areas, in thelongitudinal clusters in cen- longitudinal ters were fragmentedcenters and were were fragmented weakly connected and were weakly in some connected areas (dotted in some circles areas in (dotted Figure circles in 4b). The westernFigure regions4b). were The westernremarkably regions fragmented were remarkably (Figure fragmented 5b). (Figure5b).

Figure 5.FigureHabitat 5. Habitat suitability suitability (a) and predicted (a) and predicted presence-absence presence-absence areas (b) within areas the(b) studywithin sites. the study (a) Areas sites. with (a) high suitabilityAreas are representedwith high suitability in green, whereas are represented low suitability in green, areas whereas are represented low suitability in pink. areas (b) Predicted are represented presence areas are representedin pink. in(b green,) Predicted whereas presence absence areas areas are representedrepresented in in light green, gray. whereas The dotted absence circles areas indicate areareas repre- where the connectivitysented between in light fragmented gray. The areas dotted is low. circles indicate areas where the connectivity between fragmented areas is low. 3.2. Site Characteristics of Dead Gorals 3.2. Site CharacteristicsThe of Dead models Gorals that had fair levels of goodness-of-fit were RFs (AUC = 1.00 ± 0.00 and The modelsTSS that = had 1.00 fair± 0.00), levels Maxent of goodness-of-fit (AUC = 0.81 ±were0.003 RFs and (AUC TSS = = 0.65 1.00± ±0.01), 0.00 and Domain TSS = 1.00 ± 0.00),(AUC Maxent = 0.80 (AUC± 0.002 = 0.81 and ± TSS 0.00 =3 0.64 and± TSS0.004; = 0.65 Figure ± 0.01), S2). The and ensemble Domain model (AUC scored = high 0.80 ± 0.002 and levelsTSS = of0.64 goodness-of-fit ± 0.004; Figure (AUC S2). = The 0.88 andensemble TSS = 0.65)model and scored suggested high thatlevels the of influential factors were similar to those of the preferred habitats of the goral, such as distance to goodness-of-fit (AUC = 0.88 and TSS = 0.65) and suggested that the influential factors were the field (m) and coniferous tree density (m2). However, aspect (◦) was newly selected similar to those of(Figure the preferred6 ). The distance habitats to of the th fielde goral, (0.41 such± 0.03%) as distance was identified to the field as the (m) most and influential 2 coniferous tree densityfactor. Other(m ). factorsHowever, were aspect also severely (°) was influential newly selected (Figure (Figure6a ). The 6). probability The dis- of goral tance to the field (0.41 0.03%) was identified as the most influential factor. Other factors were also severely influential (Figure 6a). The probability of goral mortality due to snow- fall generally increased with an increase in distance to fields and areas with higher conif- erous tree densities, as found in south-east facing areas (Figure 6b).

Figure 6. Three selected variables of the ensemble model with the algorithms of Random Forest, Maxent, and Domain. (a) Relative variable importance (%) and (b) relationship between goral presence probabilities and the three relevant factors. Animals 2021, 11, x 7 of 12

It was observed that there were large clusters located in the central longitudinal areas (Figure 5), whereas there were a few optimal areas in the western regions. However, the suitability was distinctively lower in eastern areas near the seashore (Figure 5a). When thresholds were divided into presence and absence areas, the clusters in longitudinal cen- ters were fragmented and were weakly connected in some areas (dotted circles in Figure 4b). The western regions were remarkably fragmented (Figure 5b).

Figure 5. Habitat suitability (a) and predicted presence-absence areas (b) within the study sites. (a) Areas with high suitability are represented in green, whereas low suitability areas are represented in pink. (b) Predicted presence areas are represented in green, whereas absence areas are repre- sented in light gray. The dotted circles indicate areas where the connectivity between fragmented areas is low.

3.2. Site Characteristics of Dead Gorals The models that had fair levels of goodness-of-fit were RFs (AUC = 1.00 ± 0.00 and TSS = 1.00 ± 0.00), Maxent (AUC = 0.81 ± 0.003 and TSS = 0.65 ± 0.01), and Domain (AUC = 0.80 ± 0.002 and TSS = 0.64 ± 0.004; Figure S2). The ensemble model scored high levels of goodness-of-fit (AUC = 0.88 and TSS = 0.65) and suggested that the influential factors were similar to those of the preferred habitats of the goral, such as distance to the field (m) and Animals 2021, 11, 997 8 of 13 coniferous tree density (m2). However, aspect (°) was newly selected (Figure 6). The dis- tance to the field (0.41 0.03%) was identified as the most influential factor. Other factors were also severely influential (Figure 6a). The probability of goral mortality due to snow- fall generally increasedmortality with due an to increase snowfall in generally distance increased to fields withand anareas increase with inhigher distance conif- to fields and erous tree densities,areas as withfound higher in south-east coniferous facing tree densities, areas (Figure as found 6b). in south-east facing areas (Figure6b).

Animals 2021, 11, x 8 of 12

FigureFigure 6. Three 6. selectedThree selected variables variables of the ensemble of the ensemble model with mo thedel algorithms with the algorithms of Random of Forest, Random Maxent, Forest, and Domain. (a) RelativeMaxent,variable and importanceDomain. (a (%)) Relative and (b) relationshipvariable importance between goral (%) presence and (b) probabilitiesrelationship and between the three goral relevant factors. presence probabilities and the three relevant factors. A large cluster ofA areas large with cluster a high of areas goral with mortality a high goral rate mortalityis located rate in isthe located central in lon- the central gitudinal regions (Figurelongitudinal 7a). regionsWithin (Figure the clus7a).ter, Within some the areas cluster, were some fragmented areas were fragmentedbecause the because map excluded thethe south-east map excluded facing the areas south-east (Figure facing 7b). areas (Figure7b).

FigureFigure 7. Mortality 7. Mortality probability probability due to heavy due snowfall to heavy (a) andsnowfall predicted (a) and mortality predicted areas ( bmortality) within the areas study (b sites.) within (a) Areas withthe high study mortality sites. are (a represented) Areas with ingreen, high mortality whereas low are mortality represen areasted arein representedgreen, whereas in pink; low (b )mortality areas of predicted ar- deatheas are are represented represented in green. in pink; (b) areas of predicted death are represented in green.

4. Discussion 4. Discussion The long-tailed goral is a critically endangered species and is one of the least studied The long-tailedspecies goral of is montane a critically endanger [6,7]. Ased definingspecies theand distribution is one of andthe least habitat studied characteristics species of montanewas ungulates conducted [6,7]. prior As the defining study to the make distribution conservation and plans habitat for the characteristics endangered species, was conducted priorinvestigating the study the to ecology make of conser the habitatvation of goralsplans isfor urgently the endangered needed [46]. species, Several studies investigating the ecologyexplored of the the habitat habitat characteristics of gorals ofisgorals. urgently Most needed of them [46]. focused Several on landscape studies factors, explored the habitatsuch characterist as slope, altitude,ics of gorals. aspect, andMost distance of them to streamsfocused [14 on,22 landscape,23,25]. However, factors, as gorals can be flexible regarding these factors, the results of these studies cannot be used to such as slope, altitude,generalize aspect, ideal and habitat distance characteristics. to streams For [14,22,23,25]. example, individual However, herbivores as gorals of the same can be flexible regarding these factors, the results of these studies cannot be used to gen- eralize ideal habitat characteristics. For example, individual herbivores of the same spe- cies, including gorals, can adapt to different altitudes based on food abundance and hu- man disturbances [47]. Therefore, the results of previous studies on goral habitats did not have a complete view [13,14]. The results of our habitat analysis suggest that human dis- turbance was the primary factor influencing the goral habitat. By exploring the relevant variables of the 27 tentative environmental factors, the hab- itat characteristics of gorals suggest that the species could be critically susceptible to hu- man activity (Figures 4 and 5) [24]. Large distances from fields and trails and the high density of coniferous trees indicate habitats that are less disturbed by humans and are more remote [13]. Areas with a high density of broadleaf trees provide optimal living con- ditions for gorals. However, these areas also have a high human population [48]. Thus, in our study, we found that gorals preferred to live in areas with a high density of coniferous trees, even though areas with broadleaf trees offered continuous food availability [49]. It is difficult for gorals to avoid human activity as activities often occur in goral habitats, as the areas of optimal habitat for gorals are shrinking and fragmented in terms of the edge effect. Therefore, it is crucial to review the ongoing construction and land use changes within the habitats of the gorals. In Uljin, as highways were constructed, the government planned for local roads to be restored as non-paved roads. While the restoration could provide more suitable habitats, the construction and restoration impacts should be thor- oughly investigated initially. The cable car construction in the Seorak national mountain, which has the highest goral population, should be reconsidered. Animals 2021, 11, 997 9 of 13

species, including gorals, can adapt to different altitudes based on food abundance and human disturbances [47]. Therefore, the results of previous studies on goral habitats did not have a complete view [13,14]. The results of our habitat analysis suggest that human disturbance was the primary factor influencing the goral habitat. By exploring the relevant variables of the 27 tentative environmental factors, the habitat characteristics of gorals suggest that the species could be critically susceptible to human activity (Figures4 and5)[ 24]. Large distances from fields and trails and the high density of coniferous trees indicate habitats that are less disturbed by humans and are more remote [13]. Areas with a high density of broadleaf trees provide optimal living conditions for gorals. However, these areas also have a high human population [48]. Thus, in our study, we found that gorals preferred to live in areas with a high density of coniferous trees, even though areas with broadleaf trees offered continuous food availability [49]. It is difficult for gorals to avoid human activity as activities often occur in goral habitats, as the areas of optimal habitat for gorals are shrinking and fragmented in terms of the edge effect. Therefore, it is crucial to review the ongoing construction and land use changes within the habitats of the gorals. In Uljin, as highways were constructed, the government planned for local roads to be restored as non-paved roads. While the restoration could provide more suitable habitats, the construction and restoration impacts should be thoroughly investigated initially. The cable car construction in the Seorak national mountain, which has the highest goral population, should be reconsidered. Habitat selection by gorals could be dependent on their rank within the hierarchy [50]. Gorals of low-status may suffer from reduced access to resources, whereas those in the higher-order generally have priority access [51]. Within the group, the location of territory and its suitability can also be determined by rank [52]. For example, the optimal habitats to survive heavy snowfall could be occupied by the individual with the highest ranking. Thus, analyzing the spatial positions related to the ranks could suggest some reasons why gorals have been found dead in abnormal climates during the winter season. The difference between the preferred habitats of gorals and high mortality areas was rejected, which disagrees with our hypothesis. We found that the characteristics of areas with a high mortality rate also represented the preferred habitats of gorals. Variables such as large distances to the field and high density of coniferous trees are the same variables that explain the preferred habitats of gorals. For example, south-east facing areas are the preferred habitats of gorals because snow melts quicker due to more sunlight exposure, as the preferred diet grows more easily on these aspects [15,49]. However, our results suggested that a high mortality rate also characterized these areas. Thus, we concluded that mobility might be the most critical factor affecting the survival of gorals. Most of the dead gorals (89.4%) were yearlings or pregnant females (Table2)[ 21]. The autopsy report from the dead gorals suggested that they were starved but had no diseases. Generally, their mobility is less than that of males and nongravid females [53,54]. In low temperatures, pregnant females spend more energy maintaining their body temperature for their cubs [55], whereas yearlings usually have less physical ability than adults. When heavy snowfall occurs, most gorals tend to migrate to lower altitudes [12,25,56]. However, if gorals are unable to escape from stacked snow, their home ranges become smaller, and they are forced to stay in solitude in their preferred areas, such as rocky south-facing ar- eas [15,25,57]. While these areas offer food, as the weather becomes extreme, the vegetation cover decreases. The lower food availability causes gorals to become exhausted and can potentially starve. Animals 2021, 11, 997 10 of 13

Table 2. Carcass information (observed and collected day, sex, age class (yearling: age < 1, juvenile 1 < age < 3, adult > 3), estimated ages, and pregnancy; CGRB, 2010).

No Observed Day Collected Day Sex Age Class Estimated Ages Pregnant 1 2010.03.16 2010.03.19 M Juvenile >1 2 03.23 04.01 F Adult >15 Y 3 03.25 03.26 M Yearling <1 4 03.28 04.01 Adult 5–6 5 03.31 04.02 Juvenile >1 6 04.07 04.07 F Adult >10 Y 7 04.03 04.08 F Adult >10 Y 8 04.09 04.09 M Juvenile >1 9 04.09 04.09 Yearling <1 10 04.10 04.10 Juvenile <1 11 04.10 04.10 M Juvenile >1 12 04.11 04.11 F Adult 5–8 Y 13 04.10 04.10 M Juvenile <1 14 04.11 04.11 F Juvenile >1 15 04.11 04.11 Juvenile <1 16 04.20 04.21 F Adult 17 04.20 04.21 Juvenile 18 04.21 04.29 19 04.21 04.21 Juvenile 20 04.21 04.21 Juvenile

If the “age” and the “dominant rank” of animals are proportional, intraspecific com- petition (i.e., male > female and adult > juvenile) might also affect mortality. However, our previous research showed that the age and dominant rank of the gorals were not proportional [17]. The number of adults (5 ± 3) was higher than those of juveniles (3 ± 2). Besides, the number of females (4.5 ± 0.5) was higher than males (2.5 ± 0.5). However, even though the sites were the representative habitats for the goral, the number of sites in the study area was only two. While the information was limited, any study reporting that intraspecific competition of gorals can cause high mortality for females and juveniles is unknown to us. The survival of pregnant females and yearlings is crucial to increase the goral popu- lation [10]. The remote countryside offers gorals an optimal habitat (Figure4). However, when extreme snowfall occurs, these areas become the regions with the highest mortality rates (Figure5). As most gorals migrate to lower altitudes due to heavy snowfalls, it was believed that the priority rescue areas should be at lower altitudes. However, our results suggest that high altitude remote areas with high goral population density should be focused on, as rescuing gorals belonging to vital age classes is necessary to increase their population. Thus, the primary sites for rescue and artificial feeding should be considered and established at optimal goral habitats in a climate catastrophe event.

5. Conclusions Extreme winter seasons can affect the foraging behaviors of herbivores and result in an increased fatality. This study hypothesized that gorals that die due to abnormal weather conditions could be low-status individuals who live in sub-optimal or non-suitable areas. The habitat comparisons between dead and living long-tailed gorals using SDMs suggested that the sites where carcasses were identified were highly related to typical goral habitats, thus contradictory to our hypothesis. The results implied that optimal goral habitats could become deadly following heavy snowfall. The dead animals were mainly pregnant females or young individuals who could not escape their primary habitats due to their low mobility. Thus, we concluded that primary sites for rescue and artificial feeding should be established in the optimal goral habitats when there is a climate catastrophe. Animals 2021, 11, 997 11 of 13

Supplementary Materials: The following are available online at https://www.mdpi.com/article/ 10.3390/ani11040997/s1, Figure S1: Box plots of the average and standard error of AUC and TSS values of 8 model algorithms (BIOCLIM, Domain, GLMs, GAMs, CART, RFs, BRTs, and Maxent) to predict goral habitats. The dotted circles indicate the selected models of the eight models using the ensemble model. Figure S2: Box plots of the average and standard error of AUC and TSS values of 8 model algorithms (BIOCLIM, Domain, GLMs, GAMs, CART, RFs, BRTs, and Maxent) to predict areas with high mortality rate due to heavy snowfall. Please see my above comment. Author Contributions: Conceptualization, H.-B.P., and S.H.; methodology, S.H.; software, S.H.; validation, S.H.; formal analysis, S.H.; investigation, H.-B.P.; resources, H.-B.P.; data curation, H.-B.P.; writing—original draft preparation, S.H.; writing—review and editing, S.H.; visualization, S.H.; supervision, S.H.; project administration, H.-B.P.; funding acquisition, H.-B.P., and S.H. All authors have read and agreed to the published version of the manuscript. Funding: This research was funded by National Research Foundation of Korea, grant number 2021R1C1C2004162 and was supported by the National Institute of Ecology (NIE-C-2021-46). Institutional Review Board Statement: The study was conducted in accordance with the guidelines of the Institutional Review Board of the Kyungpook National University. Data Availability Statement: Data available on request due to restrictions. Conflicts of Interest: The authors declare no conflict of interest.

References 1. Schaefer, J.A.; Messier, F. Habitat selection as a hierarchy: The spatial scales of winter foraging by muskoxen. Ecography 1995, 18, 333–344. [CrossRef] 2. Tews, J.; Ferguson, M.A.D.; Fahrig, L. Modelling density dependence and climatic disturbances in caribou: A case study from the Bathurst Island complex, Canadian High Arctic. J. Zool. 2007, 272, 209–217. [CrossRef] 3. Hong, S.; Cowan, P.; Do, Y.; Gim, J.S.; Joo, G.J. Seasonal feeding habits of coypu (Myocastor coypus) in South Korea. Hystrix 2016, 27, 123–128. 4. Albon, S.D.; Irvine, R.J.; Halvorsen, O.; Langvatn, R.; Loe, L.E.; Ropstad, E.; Veiberg, V.; van der Wal, R.; Bjorkvoll, E.M.; Duff, E.I.; et al. Contrasting effects of summer and winter warming on body mass explain population dynamics in a food-limited Arctic herbivore. Glob. Chang. Biol. 2017, 23, 1374–1389. [CrossRef] 5. Grubb, P. Order Artiodactyla. In Species of the World, 3rd ed.; Wilson, D.E., Reeder, D.M., Eds.; The Johns Hopkins University Press: Baltimore, MD, USA, 2005; pp. 637–722. 6. Zaumyslova, O.Y.; Bondarchuk, S.N. The use of camera traps for monitoring the population of long-tailed gorals. Achiev. Life Sci. 2015, 9, 15–21. [CrossRef] 7. Bragina, E.; Kim, S.; Zaumyslova, O.; Park, Y.S.; Lee, W. Naemorhedus caudatus. IUCN Red List Threat. Species 2020, 2020, e.T14295A22150540. Available online: https://iucnredlist.org/species/14295/22150540 (accessed on 1 April 2021). 8. SRTI. Annual Report; Species Restoration Technology Institute (SRTI): Gurye, Korea, 2018. 9. Park, H.B.; Park, H.C.; Cowan, P.E.; Hong, S. Comparison of hair-trapping methods for the long-tailed goral. Wildl. Soc. Bull. 2018, 42, 310–313. [CrossRef] 10. Kim, B.J.; Lee, B.K.; Lee, H.; Jang, G.S. Considering threats to population viability of the endangered Korean long-tailed goral (Naemorhedus caudatus) using VORTEX. Anim. Cells Syst. 2016, 20, 52–59. [CrossRef] 11. Lee, S.; Kim, B.J.; Bhang, K.J. Habitat analysis of endangered Korean long-tailed goral (Naemorhedus caudatus raddeanus) with weather forecasting model. Sustainability 2019, 11, 6086. [CrossRef] 12. Won, B.H. Illustrated Encyclopedia of Fauna and Flora of Korean Mammals; Samhwa: Seoul, Korea, 1967. 13. Yang, B.G. Systematics, Ecology and Current Population Status of the Goral, Naemorhedus caudatus, in Korea. Ph.D. Thesis, Chungbuk National University, Cheongju, Korea, 2002. 14. Park, H.B. The Habitat Using Characteristics of Long-Tailed Goral (Naemorhedus caudatus) in the Northern Region of Gyeongbuk Province and the Effect of Climate Change. Master’s Thesis, Kyungpook National University, Dae-gu, Korea, 2011. 15. Myslenkov, A.I.; Voloshina, I.V. Ecology and Behaviours of the Amur Goral; Nauka: Moscow, Russia, 1989. 16. Chaiyarat, R.; Laohajinada, W.; Kutintara, U.; Nabhitabhata, J. Ecology of the Goral in Om Koi Wildlife Sanctuary, . Nat. Hist. Bull. Siam Soc. 1999, 47, 191–205. 17. Park, H.B.; Han, C.W.; Hong, S. Camera trapping of long-tailed goral (Naemorhedus caudatus) in BaekAm and Geumjong Mountains, South Korea. J. For. Environ. Sci. 2018, 34, 71–76. 18. Mishra, C.; Johnsingh, A.J.T. On habitat selection by the goral Nemorhaedus goral bedfordi (, Artiodactyla). J. Zool. 1996, 240, 573–580. [CrossRef] Animals 2021, 11, 997 12 of 13

19. Choi, S.K.; Chun, S.; An, J.; Lee, M.Y.; Kim, H.J.; Min, M.S.; Kwon, S.W.; Choi, T.Y.; Lee, H.; Kim, K.S. Genetic diversity and population structure of the long-tailed goral, Naemorhedus caudatus, in South Korea. Genes Genet. Syst. 2015, 90, 31–41. [CrossRef][PubMed] 20. CHA. Survey and Monitoring Long-Tailed Goral (Naemorhedus caudatus) in Uljin, Samcheok, Bonghwa Areas; Cultural Heritage Administration: Daejeon, Korea, 2007. 21. CGRB. Autopsy Report of the Dead Long-Tailed Goral (Naemorhedus caudatus) in Uljin; Conservation Genome Resource Bank for Korean Wildlife: Seoul, Korea, 2010. 22. Choi, T.Y.; Park, C.H. Korean goral potential habitat suitability model at Soraksan National Park using fuzzy set and multi-criteria evaluation. J. Korean Inst. Landsc. Archit. 2004, 32, 28–38. 23. Seo, C.W.; Choi, T.Y.; Choi, Y.S.; Kim, D.Y. A study on wildlife habitat suitability modelling for goral (Nemorhaedus caudatus rddeanus) in Seoraksan National Park. J. Korea Soc. Environ. Restor. Technol. 2008, 11, 28–38. 24. Kim, D.B.; Koo, K.A.; Kim, H.H.; Hwang, G.Y.; Kong, W.S. Reconstruction of the habitat range suitable for long-tailed goral (Naemorhedus caudatus) using fossils from the Paleolithic sites. Quat. Int. 2019, 519, 101–112. [CrossRef] 25. Cho, C.; Kim, K.; Kwon, G. Habitat altitude and home range of the endangered long-tailed goral (Naemorhedus caudatus): Seasonal and monthly home range and altitude change. Mammalia 2015, 80, 481–489. [CrossRef] 26. Cook, J.G.; Irwin, L.L.; Bryant, L.D.; Riggs, R.A.; Thomas, J.W. Relations of forest cover and condition of : A test of the thermal cover hypothesis in summer and winter. Wildl. Monogr. 1998, 141, 3–61. 27. Hijmans, R.J. Package “Raster”: Geographic Data Analysis and Modeling. The Comprehensive R Archive Network, CRAN. 2014. Available online: https://CRAN.R-project.org/package=raster (accessed on 1 April 2021). 28. Won, C.M.; Smith, K.G. History and current status of mammals of the Korean peninsula. Mamm Rev. 1999, 29, 3–36. [CrossRef] 29. Chen, W.; Hu, J.C.; Lu, X. Habitat use and separation between the Chinese (Capricornis milneedwardsi) and the (Naemorhedus griseus) in winter. Mammalia 2009, 73, 249–252. [CrossRef] 30. Araujo, M.B.; New, M. Ensemble forecasting of species distributions. Trends Ecol. Evol. 2007, 22, 42–47. [CrossRef][PubMed] 31. Hong, S.; Kim, J.Y.; Kim, Y.M.; Kim, D.K.; Joo, G.J. Factors influencing initial population establishment and habitat expansion of introduced nutrias (Myocastor coypus) in South Korea. Ecol. Inform. 2020, 59, 101111. [CrossRef] 32. Aiello-Lammens, M.E.; Boria, R.A.; Radosavljevic, A.; Vilela, B.; Anderson, R.P. spThin: An R package for spatial thinning of species occurrence records for use in ecological niche models. Ecography 2015, 38, 541–545. [CrossRef] 33. Guisan, A.; Thuiller, W.; Zimmermann, N.E. Habitat Suitability and Distribution Models: With Applications in R; Cambridge University Press: Cambridge, UK, 2017. 34. Hijmans, R.J.; Phillips, S.; Leathwick, J.; Elith, J. Dismo: Species Distribution Modeling; R Package Version 1.1-4; 2017. Available online: https://cran.r-project.org/web/packages/dismo/index.html (accessed on 1 April 2021). 35. Hastie, T. Gam: Generalized Additive Models; R Package Version 1.20; 2020. Available online: https://CRAN.R-project.org/ package=gam (accessed on 1 April 2021). 36. Therneau, T.; Atkinson, B. Rpart: Recursive Partitioning and Regression Tress; R Package Version 4.1-15; 2019. Available online: https://CRAN.R-project.org/package=rpart (accessed on 1 April 2021). 37. Liaw, A.; Wiener, M. Classification and Regression by randomForest. R News 2002, 2, 18–22. 38. Greenwell, B.; Boehmke, B.; Cunningham, J.; Developers, G.B.M. GBM: Generalized Boosted Regression Models; R Package Version 2.1.8; 2020. Available online: https://CRAN.R-project.org/package=gbm (accessed on 1 April 2021). 39. Philips, S. Maxnet: Fitting ‘Maxent’ Species Distribution Models with ‘Glmnet’; R Package Version 0.1.2; 2017. Available online: https://CRAN.R-project.org/package=maxnet (accessed on 1 April 2021). 40. Naimi, B.; Hamm, N.A.S.; Groen, T.A.; Skidmore, A.K.; Toxopeus, A.G. Where is positional uncertainty a problem for species distribution modelling? Ecography 2014, 37, 191–203. [CrossRef] 41. Pearce, J.; Ferrier, S. Evaluating the predictive performance of habitat models developed using logistic regression. Ecol. Model. 2000, 133, 225–245. [CrossRef] 42. Allouche, O.; Tsoar, A.; Kadmon, R. Assessing the accuracy of species distribution models: Prevalence, kappa and the true skill statistic (TSS). J. Appl. Ecol. 2006, 43, 1223–1232. [CrossRef] 43. MacNally, R. Regression and model-building in conservation biology, biogeography and ecology: The distinction between—And reconciliation of—‘predictive’ and explanatory’ models. Biodivers. Conserv. 2000, 9, 655–671. [CrossRef] 44. Kuhn, M. Caret: Classification and Regression Training; R Package Version 6.0-86; 2020. Available online: https://CRAN.R-project. org/package=caret (accessed on 1 April 2021). 45. Biecek, P. DALEX: Explainers for complex predictive models in R. J. Mach. Learn. Res. 2018, 19, 1–5. 46. Krebs, C. Ecology: The Experimental Analysis of Distribution and Abundance, 6th ed.; Pearson: San Francisco, CA, USA, 2009. 47. Pepin, D.; Adrados, C.; Janeau, G.; Joachim, J.; Mann, C. Individual variation in migratory and exploratory movements and habitat use by adult ( elaphus L.) in a mountainous temperate forest. Ecol. Res. 2008, 23, 1005–1013. [CrossRef] 48. Bhattacharya, T.; Bashir, T.; Poudyal, K.; Sathyakumar, S.; Saha, G.K. Distribution, occupancy and activity patterns of goral (Nemorhaedus goral) and Serow (Capricornis thar) in Khangchendzonga Biosphere Reserve, Sikkim, . Mamm. Study 2012, 37, 173–181. [CrossRef] 49. Ilyas, O.; Khan, J.A. Food habits of barking deer (Muntiacus muntjak) and goral (Naemorhedus goral) in Binsar Wildlife Sanctuary, India. Mammalia 2003, 67, 521–532. [CrossRef] Animals 2021, 11, 997 13 of 13

50. Ekman, J.B.; Askenmo, C.E.H. Social rank and habitat use in willow tit groups. Anim. Behav. 1984, 32, 508–514. [CrossRef] 51. Barroso, F.G.; Alados, C.L.; Boza, J. Social hierarchy in the domestic : Effect on food habits and production. Appl. Anim. Behav. Sci. 2000, 69, 35–53. [CrossRef] 52. Krause, J. Differential fitness returns in relation to spatial position in groups. Biol. Rev. 1994, 69, 187–206. [CrossRef][PubMed] 53. Fritzell, E.K. Habitat use by prairie raccoons during the waterfowl breeding season. J. Wildl. Manag. 1978, 42, 118–127. [CrossRef] 54. Malagnino, A.; Marchand, P.; Garel, M.; Cargnelutti, B.; Itty, C.; Chaval, Y.; Hewison, A.J.M.; Loison, A.; Morellet, N. Do reproductive constraints or experience drive age-dependent space use in two large herbivores? Anim. Behav. 2021, 172, 121–133. [CrossRef] 55. Schmidt, N.M.; Grondahl, C.; Evans, A.L.; Desforges, J.; Blake, J.; Hanse, L.H.; Beumer, L.T.; Mosbacher, J.B.; Stelvig, M.; Greunz, E.M.; et al. On the interplay between hypothermia and reproduction in a high arctic . Sci. Rep. 2020, 10, 1514. [CrossRef] 56. Mitra, S. Gorals (Nemorhaedus goral hodgsoni) in Mahananda wildlife sanctuary, West Bengal, India. Zoos Print J. 2003, 18, 1199–1201. [CrossRef] 57. Pendharkar, A.P.; Goyal, S.P. Group size and composition of the gray goral in Simbalbara Sanctuary and Darpur reserved forest, India. J. Mammal. 1995, 76, 906–911. [CrossRef]