The declining occurrence of moose (Alces alces) at the southernmost edge of its range raise conservation concerns

Abstract The border region between Austria, the Czech Republic, and Germany harbors the most south‐western occurrence of moose in continental Europe. The population originated in Poland, where moose survived, immigrated from former Soviet Union or were reintroduced after the Second World War expanded west‐ and southwards. In recent years, the distribution of the nonetheless small Central European population seems to have declined, necessitating an evaluation of its current status. In this study, existing datasets of moose observations from 1958 to 2019 collected in the three countries were combined to create a database totaling 771 records (observations and deaths). The database was then used to analyze the following: (a) changes in moose distribution, (b) the most important mortality factors, and (c) the availability of suitable habitat as determined using a maximum entropy approach. The results showed a progressive increase in the number of moose observations after 1958, with peaks in the 1990s and around 2010, followed by a relatively steep drop after 2013. Mortality within the moose population was mostly due to human interactions, including 13 deadly wildlife‐vehicle collisions, particularly on minor roads, and four animals that were either legally culled or poached. Our habitat model suggested that higher altitudes (ca. 700–1,000 m a.s.l.), especially those offering wetlands, broad‐leaved forests and natural grasslands, are the preferred habitats of moose whereas steep slopes and areas of human activity are avoided. The habitat model also revealed the availability of large core areas of suitable habitat beyond the current distribution, suggesting that habitat was not the limiting factor explaining the moose distribution in the study area. Our findings call for immediate transboundary conservation measures to sustain the moose population, such as those aimed at preventing wildlife‐vehicle collisions and illegal killings. Infrastructure planning and development activities must take into account the habitat requirements of moose.


| INTRODUC TI ON
In several parts of the world, particularly in developing countries, populations of large herbivores are decreasing, mainly due to overexploitation and habitat loss (Apollonio et al., 2010(Apollonio et al., , 2017Bragina et al., 2018;Linnel et al., 2020). In Europe and North America, by contrast, large herbivore populations have generally increased over the last several decades (Apollonio et al., 2010(Apollonio et al., , 2017Bragina et al., 2018), primarily due to changes in hunting management, better habitat quality as a consequence of improved land-use practices (e.g., change to multi-species forestry) and land abandonment in rural areas (Boitani & Linnel, 2015). Nonetheless, in recent years, population declines of some species of large herbivores have been reported also in Europe (Loison et al., 2003, Putman et al., 2011.
Changes in large herbivore populations have important implications, as in addition to influencing vegetation composition and patch heterogeneity these animals are important drivers of ecosystem processes (Hobbs, 1996;Ripple et al., 2015Ripple et al., , 2016. Moose (Alces alces, Figure 1) is the largest cervid and, along with the European bison (Bison bonasus), the largest mammal native to Europe (Niedziałkowska, 2017;Schmölcke & Zachos, 2005).
During the Holocene, moose covered almost all of Europe (Niedziałkowska, 2017), but in the Middle Ages, populations in central, western, and southern Europe began to decline significantly, mainly due to human-caused mortality (Kyselý, 2005;Niedziałkowska, 2017;Schmölcke & Zachos, 2005) but later also as a consequence of increased human activity such as infrastructure development, which decreased forest cover and caused landscape fragmentation (Niedziałkowska et al., 2014;Schmölcke & Zachos, 2005). By the beginning of the 20th century, the distribution of moose had reached its smallest extent (Niedziałkowska, 2017), limited to Fennoscandia, parts of Poland, the Baltic states, Belarus, and Russia (Niedziałkowska et al., 2014). Caused by immigration from the former Soviet Union and the expansion of the relict population in the Biebrza National Park amended by reintroductions in Poland, the population has gradually increased since end of World War II (Niedziałkowska, 2017;Świsłocka et al., 2013). Moose can now be found from Scandinavia and Eastern Poland, across the Baltic states to Belarus and Ukraine, and to Russia, up to the Yenisei river (Bauer & Nygrén, 1999;Corbet, 1979;Grubb, 2005). The southernmost range is in Central Europe (Romportl et al., 2017;Schmölcke & Zachos, 2005), as southern populations became established in the south-western region of the Czech Republic. After the fall of the Iron Curtain and into the 1990s, moose dispersed to Austria and Germany (Mrlík, 1995;Schönfeld, 2009). However, since 2010, this south-westernmost European moose population has progressively decreased in size (Romportl et al., 2017).
Moose prefer early seral forest stages with an abundance of deciduous shrubs that provide browsing (Courtois et al., 2002), but they also seek cover in closed forests (e.g., mature coniferous forest) (Beest et al., 2012;Melin et al., 2016). By adapting their temporal and spatial habitat use to changing environmental conditions, moose ensure the fulfillment of their physiological needs (Ball et al., 2001;Dussault et al., 2005). In addition, because moose are sensitive to high temperatures, they rely on a combination of lakes or ponds (Bjørneraas et al., 2011;Dussault et al., 2004) and closed or mature forests to mitigate heat stress (Borowik et al., 2020;Melin et al., 2016). These water bodies also provide a nutrient-rich supply of food (Borowik et al., 2018;Tomek, 1977). In winter, moose prefer forests and natural grasslands, are the preferred habitats of moose whereas steep slopes and areas of human activity are avoided. The habitat model also revealed the availability of large core areas of suitable habitat beyond the current distribution, suggesting that habitat was not the limiting factor explaining the moose distribution in the study area. Our findings call for immediate transboundary conservation measures to sustain the moose population, such as those aimed at preventing wildlife-vehicle collisions and illegal killings. Infrastructure planning and development activities must take into account the habitat requirements of moose.

K E Y W O R D S
Bohemian Forest Ecosystem, Habitat suitability modelling, Moose (alces alces) F I G U R E 1 Moose (Alces alces in our study area, photo by Thomas Engleder) regenerating and young forests, which offer both woody browsing and shelter (Beest et al., 2012;McNicol & Gilbert, 1980;Melin et al., 2016). Moose habitats in Central Europe often comprise forested landscapes (Schönfeld, 2009), with a mosaic of pastures and a diverse landscape structure (Romportl et al., 2017). In Poland, moose inhabit lowland forests with shrubs but also bogs as well as mire habitats with sedge-moss and reed communities (Borowik et al., 2018;Tomek, 1977). Overall, moose are flexible in their habitat use and will adapt to different conditions as long as abundant foraging opportunities and forested shelters are present (Bjørneraas et al., 2011).
Moose are highly mobile animals and their home ranges are large, between 10 and 60 km 2 depending on the age, sex and reproductive status of the animal, and environmental characteristics such as forage availability and climatic conditions (Beest et al., 2011;Cederlund & Sand, 1994;Murray et al., 2012). Migrations are common in seasonal habitats (Ball et al., 2001;Rolandsen et al., 2017).
In Scandinavia, for example, moose spend the summer at higher elevations because of better food quality and lower temperatures and then migrate in autumn, once snow limits their movement and food accessibility, to reach lower elevations (Andersen, 1991;Bunnefeld et al., 2011;Melin et al., 2013Melin et al., , 2016. Similarly, moose can disperse over long distances, for example, from central Poland to southern Bohemia (Niedziałkowska et al., 2014), which has important implications for their potential range expansion. However, during these journeys, moose may block roads and otherwise interfere with infrastructure, such that they are notoriously susceptible to traffic-associated mortality (e.g., Seiler, 2005). Compared to moose distribution in Scandinavia and Canada, Central Europe has a much higher road density, and the absence of suitable crossing points on motorways and main roads hinders the safe movements of large animals, including moose (Strnad et al., 2012). Central European motorways extend over several thousand kilometers, and traffic intensity has substantially increased. Consequently, wildlife-vehicle collisions have become an important cause of death for large herbivores (Bragina et al., 2018;Mrlík, 1995), and large patches of suitable habitat for moose have become increasingly isolated (Ree et al., 2015).
In this study, we examined moose populations within the transboundary region of Austria, the Czech Republic, and Germany.
Although moose are listed as endangered and are fully protected by conservation laws in the Czech Republic as well as by hunting laws in Bavaria and Austria, the size and distribution of moose in the study region have been stagnant and possibly declining (Romportl et al., 2017;Schönfeld, 2009). Thus, in this study, to obtain an accurate picture of the history and status of the moose population, we collected all available occurrence data from the region to (a) evaluate changes in moose distribution in the study area; (b) identify the main causes of mortality; and (c) analyze habitat selection to model the extent of suitable habitat. We expected that landscapes with water bodies, wetlands, and a heterogeneous mosaic of shrubs, meadows, and forests provide suitable habitat for the moose population (Beest et al., 2012;Borowik et al., 2018Borowik et al., , 2020Courtois et al., 2002;Melin et al., 2016;Romportl et al., 2017;Tomek, 1977), while anthropogenic structures would reduce the amount and connectivity of available habitat (Bragina et al., 2018;Mrlík, 1995;Ree et al., 2015;Seiler, 2005;Strnad et al., 2012).

| Study area
The border of the study area consisted of a 50-km buffer along the borders of Austria, the Czech Republic, and Germany, between 48.1 N and 51.0 N (44,100 km 2 ; Figure 2) and covering the core area of moose distribution in Central Europe (Homolka, 1998). Forests comprise 40% (85% coniferous), arable land 27.2%, pastures 17.2%, heterogeneous agricultural areas 8%, and artificial surfaces 5% of the study area. From the south-west and south, the study area is naturally bordered by the Danube valley, which includes barriers formed by the Danube itself but also by artificial structures. On the Czech side, the study area is delimited by larger cities (Plzeň and České Budějovice) and landscape with increasing human population density. Šumava National Park (SNP) and Bavarian Forest National Park (BNP) are located in the central part of the study area, together forming one of the largest protected forested areas in Central Europe (Křenová & Hruška, 2012).
The core area is created by mountain ranges along the Austrian, Czech, and German borders, with the highest altitudes occurring in the Bohemian Forest, that is, the Šumava Mts. Forested mountain ranges and highland plateaus with mosaics of peat bogs and extensively used grasslands create the typical landscape of the central SNP and the mountain ranges of the region in general (Janík & Romportl, 2016;Spitzer & Bufková, 2008;Wölfl et al., 2001). The elevation varies between 210 and 1.456 m a.s.l., and the mean annual temperature in the study area ranges from 3 to 9°C (Heurich et al., 2010;Tolasz et al., 2007).
The eastern edge of the study area is in the Třeboňsko Protected Landscape Area (PLA) and adjacent areas (Třeboňsko hereafter), a flat sedimentary basin with acidic soils and an elevation ranging between 400 and 550 m a. s. l. Land cover is mainly characterized by forests, with a smaller amount of agricultural land (e.g., meadows and pastures), human settlements, and various wetland habitats (e.g., peat bogs), including extensive systems of man-made water bodies, such as fishponds and channels (Hanák et al., 2006).
The 30-year-long first period covered the initial phase of moose establishment in the study area, from the very first occurrences to the development of a breeding population and ending with the disappearance of a major migration barrier for large mammals, that is, the fall of the border fence (Iron Curtain) in 1989/1990, which resulted from significant political and socio-economic changes. The following three periods represent moose occurrence patterns during the three decades when movement of the moose population was not hindered by the border fence. The analyses and visualizations of this study were based on 10 km × 10 km squares. Moose occurrence during each period was classified as (a) reproduction, when the presence of female(s) with a calf(calves) within an individual period was recorded; (b) regular occurrence, consisting of >5 occurrences within a single period; and (c) irregular occurrence, when ≤5 occurrences within an individual period were recorded (see also Wölfl et al., 2001). The cause of mortality was categorized as (a) wildlife-vehicle collision, (b) legal culling, (c) poaching, and (d) unknown. approach is based on maximum entropy in estimates of a target probability of species´ distribution (Phillips et al., 2006). In this study, the model was implemented using 10,000 background points, evaluated based on 500 iterations and 10% of the data and assessed using the AUC (area under curve) (Phillips et al., 2006).

| Environmental data and habitat preferences
We calculated several MaxEnt models for each period with corresponding land cover (1958( -1989( = CLC1990, 1990( -1999  Additionally, we evaluated differences in the distances of the moose records from key land cover categories: artificial surfaces as potential barrier and threat, broad-leaved forest as a preferred habitat for foraging and hiding, natural grassland and shrubs for foraging, wetlands, and waterbodies as habitats for hiding and avoiding heat stress (Borowik et al., 2018(Borowik et al., , 2020Beest et al., 2012;Courtois et al., 2002;Melin et al., 2016;Romportl et al., 2017;Tomek, 1977) for each time period. We tested data for normality and used ANOVA and Kruskal-Wallis test depending on normality of the data in R software (R Core Team, 2019).

| Mortality
Between 1958  collisions accounted for the largest proportion (n = 13; 48.2%), followed by legal culling (3; 11.1%) and poaching (1; 3.7%). For the other 10 individuals (37.0%), it was not possible to determine the cause of death. The spatio-temporal distribution of moose mortality is shown in Appendix 1, 2. Vehicle collisions resulting in moose deaths mainly occurred on minor roads (11 from 13 collisions, 84.6%); only two fatal collisions took place on primary roads and highways. The two moose died between 1989 and 1991 due to poaching/legal culling had been recorded in Austria (n = 2); one moose had been shot after hit by a car in Germany.

| Habitat preferences
The AUC calculated for the habitat suitability model was 0.726, suggesting a fair fit (Araujo et al., 2005). Altitude contributed most to the model performance, followed by land cover, slope, and distance to human settlements ( Table 2).
The response curves of the different variables are shown in Human activity was represented by the distance to roads, the distance to settlements and road density. Habitat suitability was highest at intermediate distances to roads, with a slight decrease in the first few hundred meters, followed by an increase in suitability up to 3,000 m away from roads. The shape of the curve was similar to that of the distance from settlements, which were avoided within the first 1,000 m.
Moose avoided areas with a high road density as well. The results of the habitat suitability model showed that suitable moose habitats within the study area were located in three subareas: (a) Třeboňsko (ca. 500 km 2 ), (b) Bohemian Forest Ecosystem (ca. 2,000 km 2 ), especially in the southern SNP and the Šumava PLA near Lipno Reservoir as well as at the transboundary in the eastern part of Bavaria and Austria, and (c) Novohradské hory/Gratzener Bergland/Freiwald/Weinsberger Wald (ca. 900 km 2 ), especially the Austrian part (see Figure 7). Smaller patches of suitable habitat were dispersed at higher altitudes in the western and northern part of the study area, along the Czech-Bavarian border.
The validation with C1 and C2 only data provides similar results and as to expect a higher AUC 0.826 (see Appendix 3 for model). The habitat models for the distinct periods revealed for the first two periods (1958-1989 and 1990-1999) that preferred habitat tended to be at lower altitude with wetlands and water bodies. In contrast, in the periods of 2000-2009 and 2010-2019, suitable habitat shifted to higher altitudes, which also was the most important predictor (see Appendix 4).
In analysis of the distances from moose records to key land cover categories, we used the nonparametric Kruskal-Wallis test, because of non-normal distributed data. Distances from the artificial surfaces are not significantly different across the periods (p = 0.051); however, the distance of the moose from artificial surfaces has been increasing from the second period (1990-1999; mean distance (m) 1958-1989 = 1,906, 1990-1999 = 1,826, 2000-2009 = 2,063, and 2010-2019 = 2,217). We also found that the distance of observations to wetlands (p = 0.000; mean distance (m) 1958-1989 = 11,590, 1990-1999 = 6,597, 2000-2009

| D ISCUSS I ON
Our study provides the first comprehensive assessment of the longterm development of occurrence, mortality, and habitat association The first records of moose occurrence in our study area were from what is now the Czech Republic (1958, although the first record from the Czechia was from 1957; Červený et al., 2001), followed by Austria (1964) and Germany (1976) (Schönfeld, 2009), reflecting a dispersal from Poland (Tomek, 1977  Research from Norway demonstrated the strong effects of environmental stochasticity on moose survival, especially that of calves, and the strong seasonal or regional variation acting upon the survival of adult females (Stubsjøen et al., 2010). Thus, another important factor for the limited population growth may be the vulnerability of the moose population to stochastic events. This might be especially relevant for our study site, as stochastic variation, including environmental and demographic stochasticity, is much higher for small, isolated populations.
Besides the reduction in moose observations, the spatial distribution of the population changed. The core area of moose distri-   Dussault et al., 2006;Rea, 2003). This suggests that moose avoid major and international roads, due to their high traffic intensity and the associated noise and light pollution, both of which discourage moose crossings (Niemi et al., 2017). Our habitat suitability models likewise showed that moose avoid roads. However, considering their large home ranges and dispersal distances, moose will inevitably seek to crossroads and will thus be at risk of being struck by vehicles.
Human-caused mortality, including legally culled and poached animals, was the second most important cause of moose mortality, despite the fact that moose are fully protected by nature conservation and/or hunting laws in Austria, the Czech Republic, and in Germany.
Only in Austria, moose can be shot, but only in exceptional cases, such as when their intense browsing damages young forests (Mrlík, 1995). Although just one moose was confirmed as poached in 1989, undetected illegal killing may have occurred throughout the study area, as previously demonstrated for other protected species in the same region (Heurich et al., 2018). Long-term studies in Scandinavia have shown that variations in human-induced mortality (legal culling, poaching) have substantial effects on moose population dynamics (Solberg et al., 1999). In Poland, hunting and poaching had a strong impact on moose populations, but a 2001 ban on moose hunting resulted in their steady increase (Borowik et al., 2018;Bragina et al., 2018;Tomek, 1977) to the extent that wildlife managers suggested the re-introduction of culling (Dziki-Michalska et al., 2019).
Predation and diseases are leading causes of moose mortality (Okarma et al., 1995;Severud et al., 2019) but there were no reports of diseases and/or severe infections among the moose in the study area, possibly because of the low population density. Predation was also not recorded in the study area, as there have been no bears in the study area and wolf recolonization just occurred recently.
Modeling habitat suitability from observation data is based on several assumptions. The first, and most important, is that there is no observation bias. As moose are obviously easily determined to the species level, we are confident that the sampling coverage of our study area was good and that the observers were highly willing to report observations. In addition, our results are in agreement with other studies of moose habitat selection (Beest et al., 2012;Borowik et al., 2018Borowik et al., , 2020Melin et al., 2016;Romportl et al., 2017), such that the potential bias was probably low. Thus, based on the results of our habitat model, the ideal habitat for moose in the studied area is a mosaic of forest, to provide shelter, more open extensive landscape with natural succession, as a source of food, and water bodies and wetlands for mitigating heat stress (Bjørneraas et al., 2011;Beest et al., 2012). Although this overall landscape preference is similar to moose habitats in Poland (Borowik et al., 2018;Tomek, 1977), within our study area these combined features occur at markedly higher altitudes (700-1,000 m. a.s.l. wetlands (mostly peat bogs) and water bodies in core areas of occurrence. Schönfeld (2009) found that moose occurrence in Bavaria was highest in more forested areas, and in Poland in large forest complexes (Tomek, 1977). Wetlands provide thermoregulation and forage and are thus the most favored land cover type for moose (Bjørneraas et al., 2011;Dussault et al., 2004); however, the distance of moose records to wetlands increased throughout the study period. In winter, coniferous forests offer better conditions for hiding and foraging (Bjørneraas et al., 2011) whereas in summer broad-leaved forests and shrubs enable browsing (Månsson, 2009 A shortcoming of our study is that our model does not provide a seasonal resolution, due to the limited data availability. However, it is suitable in predicting annual moose habitat, including that reached by seasonal migrations (Borowik et al., 2020). Sweanor and Sandegren (1988) determined a migration distance of 320 km in Sweden and migrations in Alaska of up to 280 km were reported by Gasaway et al., (1983). These long migrations may increase the risk of vehicle collisions (Tajchman et al., 2017). Thus, future research on moose should focus on their seasonal habitat selection and their long-distance movements.

| CON CLUS IONS
Our study showed a decrease in the moose population in recent years and a shift in its core area from Třeboňsko to a more fragmented distri- Based on a comprehensive synthesis of available records gathered from three countries, this work is a first step toward a complete assessment of the moose population in the study area. Meanwhile, also samples for genetic analysis are being collected in all three countries. The next step would be to implement a systematic monitoring programme aimed at collecting more detailed information on the spatial-temporal habitat requirements of moose as well as demographic data, such as population density, reproduction, survival, and genetic structure.
The moose population in our study area was isolated and very small, possibly less than 20 animals. Moreover, it seems to be decreasing, with inbreeding and the factors considered in this study likely to drive it further down the extinction vortex. Our results highlight the urgent need for mitigation measures to prevent moosevehicle collisions and illegal killings. Based on the low number of animals in the study area, it can be expected that active measures such as translocation are necessary for the long-term viability of moose population in the study area. Given the large scale spatial requirements of these herbivores, coordinated cross-border management aimed at the conservation of this population is essential.

ACK N OWLED G M ENTS
We would like to thank everyone who contributed data for use in this study, especially Adam Otomar, who gathered a substantial amount of records from the Czech Republic, and Rainer Karsch, who did the same in Bavaria. We are also grateful for the language revision performed by Wendy Ran and for valuable comments from our reviewers.

CO N FLI C T O F I NTE R E S T
The authors have declared that no competing interests exist.

DATA AVA I L A B I L I T Y S TAT E M E N T
Dataset is available on DRYAD https://doi.org/10.5061/dryad.z612j m69s.