Acta Herpetologica 15(2): 73-85, 2020 DOI: 10.13128/a_h-7771 Estimating abundance and habitat suitability in a micro-endemic snake: the Walser viper Gentile Francesco Ficetola1,2,*, Mauro Fanelli3, Lorenzo Garizio3, Mattia Falaschi1, Simone Tenan4, Sam- uele Ghielmi5, Lorenzo Laddaga6, Michele Menegon7,8, Massimo Delfino3,9 1 Dipartimento di Scienze e Politiche Ambientali, Università degli Studi di Milano, Milano, Italy 2 Laboratoire d’Écologie Alpine, Univ. Grenoble Alpes, Univ. Savoie Mont Blanc, CNRS, LECA, Grenoble, France 3 Dipartimento di Scienze della Terra, Università di Torino, Torino, Italy 4 National Research Council, Institute of Marine Sciences (CNR-ISMAR), Venezia, Italy 5 MUSE - Museo delle Scienze, Trento, Italy 6 Società di Scienze Naturali del Verbano Cusio Ossola, Museo di Scienze Naturali, Collegio Mellerio Rosmini, Domodossola Italy 7 Division of Biology & Conservation Ecology, School of Science & the Environment, Manchester Metropolitan University, Manchester, UK 8 PAMS Foundation, P.O. Box 16556, Arusha, Tanzania 9 Institut Català de Paleontologia Miquel Crusafont, Universitat Autònoma de Barcelona, Cerdanyola del Vallès, Barcelona, Spain *Corresponding author. E-mail: francesco.ficetola@unimi.it Submitted on: 2020, 1st January; revised on: 2020, 12th June; accepted on: 2020, 26th September Editor: Dario Ottonello Abstract. Recently described species suffer lack of information that hampers setting up appropriate conservation strat- egies. The situation is particularly complex with micro-endemic snakes, for which detection and monitoring are par- ticularly challenging. The Walser viper Vipera walser is a recently described snake inhabiting a small area of the SW Italian alps. We combined information on species distribution with repeated monitoring to identify the areas most suitable for the species, and to obtain estimates of species abundance. Species distribution models were used to identi- fy the topographical, climatic, and land-cover features related to the occurrence of vipers. Furthermore, repeated tran- sects and N-mixture models were used to estimate abundance and to identify factors related to the variation of abun- dance. The available data suggested that the species has a disjunct range, with a Northern range of ~45 km2, and a southern range of ~225 km2. Distribution models suggested that vipers are associated with areas with open vegetation, altitude between 1300 and 2300 m, high precipitation, low forest cover, low slope, and southern aspect. N-mixture models confirmed very low detection probability of these vipers, and suggested that the species has a low abundance, with the highest abundance in south-facing plots. We provide the first quantitative information on habitats and abun- dance variation for Walser vipers. The broad confidence intervals of abundance estimates exemplify the complexity of providing range-wide measures of abundance for secretive species. Given the narrow range of these vipers, continuous monitoring is required to understand how they respond to ongoing environmental changes in mountainous areas. Keywords. Alpha-hulls, detectability, endemism, habitat suitability models, land cover, Vipera walser. INTRODUCTION 2014). In the last decades, the integration of new genetic and morphological data has greatly expanded our knowl- Italy is among the European countries with the high- edge of Italian biodiversity, with the identification of sev- est endemism of amphibians and reptiles (Sillero et al., eral reproductively isolated lineages, that have been pro- ISSN 1827-9635 (print) © Firenze University Press ISSN 1827-9643 (online) www.fupress.com/ah 74 Gentile Francesco Ficetola et alii posed as new candidate species (e.g., Cornetti et al., 2015; on species range. For instance, Raxworthy et al. (2003) Dufresnes et al., 2018; Senczuk et al., 2019). developed SDM for chameleons in Madagascar, and then The Walser viper has been a striking addition to the performed targeted surveys in areas suggested to be suit- Italian herpetofauna. In NE Piedmont, the occurrence able by models. In several cases, these targeted surveys of vipers morphologically assumed to be adders (Vipera revealed new localities of occurrence, suggesting that berus) has long been recognized (Andreone and Sinda- SDM can greatly improve our knowledge on species liv- co, 1998; Sindaco et al., 2006). However, recent genetic ing in remote areas, or for which information on the dis- data revealed strong genetic differences between these tribution is limited. “adder-like” vipers and true adders; the closest relatives Second, we performed repeated visits in a large num- of these vipers belong to the cluster regrouping V. dinniki, ber of patches within the species range, to assess spatial V. kaznakovi and V. darevskii, in the Caucasus area. As a variation of abundance. Variation of abundance is a key consequence, these adder-like vipers have been described parameter to assess the threat status of species, still, accu- as a new species, the Walser viper (Vipera walser, Ghiel- rate measurement of abundance can require extremely mi et al., 2016), a species endemic of a small area from extensive workload (Pollock et al., 2002; Dodd, 2010). In the mountains between Biella and the Ossola valley, with the last years, approaches have been proposed to obtain a possible range < 500 km2 (Fig. 1). estimates of abundance from repeated counts, without Recently described, micro-endemic species often marking or capturing individuals (Royle and Nichols, suffer a tremendous lack of biological data. This is par- 2003; Royle, 2004). The number of individuals counted at ticularly problematic, because the small range inher- fixed sites on multiple occasions can be used to estimate ently exposes these species to a high risk of extinction. the detection probability, and the size of populations can Therefore, sound biological and ecological information be estimated on the basis of N-mixture models (Royle is required to set up appropriate monitoring and conser- and Nichols, 2003; Royle, 2004; Kéry et al., 2009; Dail vation programs. Habitat suitability and spatial variation and Madsen, 2011). Despite several limitations (Barker of abundance represent critical information to assess the et al., 2018; Link et al., 2018), such models can provide conservation status of species. In the last years, advanc- cost-effective estimates of abundance while account- es in analytical tools have greatly improved our ability ing for imperfect detection (Ficetola et al., 2018a; Kéry, to provide sound biological information even in species 2018), as detection probability is typically low in snakes. for which only a limited amount of data is available (e.g., In addition, N-mixture models can provide insights on Raxworthy et al., 2003; Peterson et al., 2011; Mazerolle, the factors determining spatial variation of population 2015). abundance and, if monitoring is repeated through time, In this paper, we combine modelling and field data they can provide reliable estimates of population trends to improve our knowledge of the ecology of the Walser (Ficetola et al., 2018c). These data are particularly critical viper. First, we used habitat suitability models to evalu- for recently described species, for which information on ate the relationships between viper distribution, land abundance is nearly absent. cover, and topographical features at a fine spatial scale. Correlative species distribution models (SDM) allow to identify relationships between localities of presences and METHODS spatial variation of environmental variables, and can provide multiple key information (Peterson et al., 2011). Species range and distribution models SDM can reveal how the species respond to broad-scale variation of environmental features, thus providing first In March 2019, we combined bibliographic data (Ghielmi et al., 2006; Ghielmi et al., 2016) with new field observations information on the habitats that are more suitable for performed in 2016-2018 by the authors, and personal com- the target species (Guisan and Thuiller, 2005; Peterson munications by local naturalists, to gather an exhaustive data- et al., 2011). These models often have a coarse spatial set of V. walser occurrences. We used the alpha-hull approach scale and sometimes lack accurate information on spe- for an accurate definition of the species range. The alpha-hull cies microhabitat or on habitat features that can only approach is a procedure based on Delauney triangulation that be measured in the field (Beck et al., 2012; Potter et al., uses presence points to estimate species ranges, and can allow 2013; Ficetola et al., 2018b). Still, if reliable maps of habi- for the exclusion of unoccupied areas within a species range tats or land cover are available, it is possible to obtain (Burgman and Fox, 2003). Simulations showed that the alpha- relatively detailed information on species responses to hulls provide better approximations of species ranges compared to minimum convex polygons, particularly when the ranges major land cover categories. Furthermore, SDM can pro- have discontinuities (Burgman and Fox, 2003). Alpha-hulls duce spatially-explicit maps, that can refine information were built using the alphahull package in R (Pateiro-Lopez and Abundance and habitat of Vipera walser 75 Rodrjguez-Casal, 2010), following the procedure detailed in ability. Since we have a-priori expectation on the frequency of Ficetola et al. (2014). observations (i.e., we expect more observations in high-suitabil- Land cover data were obtained from the geoportal of the ity areas), we used a one-tailed test (Warren and Seifert, 2011). Piedmont region (http://www.geoportale.piemonte.it/) at the resolution of 2 m. We considered four land cover categories: agricultural, pasture, natural open vegetation, and forest. As Field activities a measure of land cover, we calculated the percentage of each land cover category in 100 × 100 m cells. Furthermore, we con- In the period 2016-2019, we established 71 fixed plots sidered three topography variables: altitude (m a.s.l.), aspect using visual encounter surveys (Crump and Scott, 1994). Each (northness), and slope. As a climatic parameter, we considered plot was visited during one year only (range: 5-26 plots per summed annual precipitation, obtained through the Chelsa- year); plots were visited multiple times (average: 5.7 surveys clim dataset (Karger et al., 2017). This dataset is available at a per plot; range: 3-12 surveys) from late May to early October, coarser resolution than the other predictors (30 arc-seconds, i.e., during the period of highest activity of vipers. The size approx. 650 × 900 m in the study area). We therefore used the of plots ranged between 600 and 23,000 m2 (mean: 7600 m2), B-spline interpolation to downscale it at the 100-m resolution because of logistic and accessibility constraints. Visits were (see Karger et al., 2017). Temperature variables were not includ- performed from 7 am to 17 pm (solar time), but most of them ed because they were strongly collinear with altitude (r > 0.9). (69%) were performed in late morning (between 8.30 and 12.00 The pairwise correlation between the remaining environmental am). Plots were placed across the whole range of the species, variables was always |r| < 0.65, suggesting limited collinearity mostly nearby areas with previous records of Walser vipers. In issues (Dormann et al., 2013). 2019, the location of plots was selected after the development We used maximum entropy modelling (MaxEnt) (Phil- of species distribution models, in order to validate the SDM lips et al., 2006; Elith et al., 2011) to build species distribution and identify eventual new locations of the viper. Out of the 17 models relating the occurrence of vipers to land cover and top- plots surveyed in 2019, three were in areas with low SDM suit- ographical data. MaxEnt is a presence-background approach ability (suitability below the 10th percentile training presence that evaluates the suitability of a given cell on the basis of envi- threshold; range: 0.06-0.22) but nearby areas with high suit- ronmental features (habitat, climate, topography…) in that cell. ability. Ten were in areas with high SDM suitability (> 0.45) and Comparative studies showed that MaxEnt is among the most nearby localities where the species is known to be present, and efficient approaches to build SDM (Elith et al., 2006; Elith et four were in areas with high suitability but out of the known al., 2011). MaxEnt is well suited to evaluate complex or non- species range (Mombarone; roughly 3 km SE of the southern linear relationships between species and environmental features, limit of the species range, see Fig. 1b). During surveys, one to and produces an output representing the suitability of a specific four observers carefully patrolled the plots, searching for active area. We run models with linear, quadratic, and hinge features. vipers. In other areas, the search under artificial shelters greatly To identify the best regularization parameter (b), we build five improved the detection of reptiles (De Leo et al., 2006; Joppa et models with increasing values of b: 1, 2, 5, and 10 (Warren and al., 2010; Sewell et al., 2012). Therefore, in 2019, visual encoun- Seifert, 2011). The model with b = 2 showed the highest cross- ters were integrated with the use of artificial cover objects (shel- validation performance (see below), and was selected as model ters). In each plot, we placed three corrugated bitumen 70 × 90 with highest generality (Warren and Seifert, 2011). cm shelters that were checked during each survey. We used two approaches for the validation of species distri- bution model. First, the model was tested using a 5-fold cross- validation. Presence records were split into five sub-sets; the Assessment of species abundance model was built using 80% of data (calibration data), and we tested predictive performance on the remaining 20% (test data). We used N-mixture models to estimate the abundance of We repeated this procedure five times, each time using a differ- vipers on the basis of repeated counts at plots, and to iden- ent test dataset (Nogués-Bravo, 2009). As a measure of model tify environmental variables related to variation of abundance. performance, we calculated the area under the curve of the N-mixture models allow the joint estimation of animal abun- receiver operator plot (AUC), averaged over the five runs. AUC dance and detection probability on the basis of repeated sur- is an imperfect measure of the performance of SDM (Lobo et veys at fixed sites, without the need of capturing and marking al., 2008), thus we also used a binomial test to evaluate if our individuals for identification (Royle, 2004). Some analyses high- models predict presence records better than expected under lighted that jointly estimating detection probability and abun- randomness, comparing the observed frequencies of correct dance could be problematic, and these models are sensitive to and incorrect predictions. In this test, we assumed that a cell is violations of their assumptions (Barker et al., 2018; Duarte et suitable for vipers if it has suitability larger than 10th percentile al., 2018), still analyses of real-world data showed that N-mix- training test threshold (Pearson et al., 2007). Second, the model ture models can provide reliable estimates of the abundance of was developed in late winter 2019. Therefore, we used data col- wild vertebrates (Ficetola et al., 2018a; Kéry, 2018; Costa et al., lected during the 2019 field activities to confirm the reliability 2019). We assumed that each plot, sampled during one single of model predictions. Specifically, we used a likelihood ratio test season, represented a closed population. We thus used a static to assess whether viper observations are more frequent in areas (i.e., non-dynamic) N-mixture model formulation. The average with high SDM suitability, compared to areas with low suit- distance between each plot and the closest one was 170 m (SE 76 Gentile Francesco Ficetola et alii = 19.6), a distance much longer than the typical movements of 47% of records were obtained through recent surveys related species (e.g., Zuffi et al., 1999; Nash and Griffiths, 2018). by the authors; we also obtained some personal com- N-mixture models were fitted using a Poisson error distribu- munications by local naturalists (3% of records). Spe- tion, as in preliminary analyses Poisson models showed a lower cies records were clustered in two main areas: a northern Akaike’s Information Criterion (AIC) (Burnham and Anderson, group between the Strona and the upper Sesia Valley, and 2002) than the corresponding zero-inflated Poisson models. In N-mixture models, we set the upper bound to approximate a southern group between the lower Sesia Valley and the the infinite summation of the likelihood (K) at the maximum Biella Alps. The lack of known records between these two observed species abundance +100, since simulations suggested areas suggests the possibility of a disjunct distribution, that this value provides stable and robust estimates (Ficetola et with a northern range of approx. 45 km2, and a southern al., 2018c). range of approx. 225 km2. We considered four variables potentially influencing the The species distribution model suggested that vipers detection of vipers: date (day of the year), hour, length of the were associated to areas with open natural vegetation, survey (in minutes, log transformed), and the number of peo- altitude between 1300 and 2300 m, high annual precipi- ple participating in the survey. For date and hour, we also test- tation, low forest cover, low slope, and with a southern ed quadratic terms to take into account potential non-linear relationships. To identify the model best explaining variation aspect (Fig. 2). The presence of agricultural land cover of viper detection, we built models including all the potential and pastures showed very limited importance (relative combinations of independent variables, and ranked them on importance < 1%). The cross-validation procedure sug- the basis of their AIC values. The model with lowest AIC value gested excellent performance, with average AUC on test explains more variation with a limited number of parameters, data = 0.924 (SD = 0.038). Using the 10th percentile train- and was selected as the minimum adequate model (Burnham ing presence threshold (suitability threshold = 0.36), the and Anderson, 2002). First, we performed model selection on model correctly predicted occurrence at 86.3% of test variables affecting detectability, then we performed model selec- data, a performance significantly better than expected by tion on variables affecting abundance. chance (binomial test: P << 0.001; success rate expected We considered six variables potentially affecting the abun- dance of vipers at plots: plot surface, aspect, altitude, slope, and under randomness = 13.5%). suitability predicted by the species distribution model. SDM The model identified several patches of highly suit- suitability was included because previous studies suggested that able habitats through the whole species range, in moun- environmental suitability models also help to predict parame- tainous areas at altitude of 1300-2200 m, with gentle, ters such as population density and fitness-related traits (Bram- south-facing slopes and covered by open vegetation. billa and Ficetola, 2012; Weber et al., 2017; Lunghi et al., 2018). Within the species range, the total suitable surface (i.e., Variables representing land cover were not included because with suitability higher than the 10th percentile thresh- 85% of plots were in areas with open natural vegetation, which old) was 81.9 km2. Several areas with good suitability is the most important land cover variables determining viper were detected outside the known range of the species, for distribution (see results). We built models with all the possible combination of variables potentially affecting abundance, and instance, in the SW of the study area (e.g., the Mombar- ranked them using AIC, while keeping constant the observa- one area) (Fig. 1b). Potentially suitable areas were also tional component of the model (i.e., including the variables best detected between the northern and southern populations. explaining variation in detection probability). The correlation coefficient between site covariates was weak (r < 0.4 for all the considered variables). Model validation with the 2019 data N-mixture models were run with the package Unmarked in R (Fiske and Chandler, 2011). Before analyses, independent Four plots surveyed in 2019 were outside the known variables were scaled at mean = 0 and SD = 1 to improve model range of V. walser, but in areas showing high suitability convergence. A goodness-of-fit test showed limited overdisper- according to the SDM (Mombarone area; Fig. 1b). Each sion and confirmed that the model was appropriate to the data of them received six surveys; we never detected vipers at (χ2 = 432.6, permutation P = 0.12, c-hat = 1.1). these plots. Thirteen plots surveyed in 2019 were inside the known range of V. walser; four were in areas with low suitability (suitability ≤ 0.2) and nine in areas with RESULTS high suitability (> 0.45). Vipers were never detected in the plots with low suitability, while they were detected Distribution data, species range, and distribution model in 56% of plots with high suitability. The possibility to Overall, we gathered 117 distribution records (Fig. obtain at least one detection was significantly higher 1a). Localities from the literature (Ghielmi et al., 2006; in plots with high suitability (likelihood ratio test: χ21 = Ghielmi et al., 2016) accounted for 50% of records, while 4.96, P = 0.013), confirming that the SDM can success- fully predict viper occurrences. Abundance and habitat of Vipera walser 77 a) b) c) Fig. 1. a) Distribution records used to build species distribution models, and species range estimated using alpha-hulls. b) Location of fixed plots where we performed repeated counts. The arrow indicates the plots located in possibly suitable sites outside the known range of the species (Mombarone). c) Suitability map, estimated using species distribution models. 0.37 is the 10th training presence threshold. To limit the risk of poaching, points plotted on the maps have a random spatial error of up to 2 km (Lunghi et al., 2019). d) An adult male of V. wal- ser (photo by GFF). 78 Gentile Francesco Ficetola et alii Repeated counts and abundance estimations DISCUSSION Overall, we obtained 39 viper detections from 24 For recently described species, rapid biological and out of the 67 plots within the species range. Vipers were ecological studies are pivotal for a prompt assessment of detected across the whole study period (from June to early the species status. By combining distribution records with October). The best-AIC model suggested that the number ecological modelling and repeated surveys we provide of detection was only affected by the length of surveys, key ecological data on the endemic, poorly known, Wal- longer surveys allowing a higher probability of detecting ser viper. vipers (B ± SE = 0.73 ± 0.16, z = 4.45, P < 0.001). Con- The application of alpha-hulls to species distribution versely, the detection rate was unrelated to date or hour data suggested that this viper has a disjunct range, with of the survey, nor to the number of observers (Table 1). a northern range comprising the Strona and the upper The detection probability of vipers was approx. 0.036 in Sesia Valley, and a southern range between the Biella 35-min. surveys (35 min. was the average length of sur- Alps and the lower Sesia Valley. Alpha-hulls have an veys), but the incertitude of these estimates was large excellent capacity to reveal discontinuities in the specie (95% CI of detection probability after 35 min: 0.007-0.16). range (Burgman and Fox, 2003), still they can be affected The abundance of vipers was negatively related to the by the lack of biological records caused, for instance, by northness of plots, with higher abundances in south-fac- limited surveys. The available genetic data did not detect ing plots (B = -0.70 ± 0.30, z = -2.35, P = 0.019). We did strong differences among Walser vipers (Ghielmi et al., not detect any relationship between the remaining varia- 2016), still additional genetic or genomic studies, using bles and the abundance of vipers (Table 1). The estimated highly variable or fast-evolving markers, are needed to abundance per plot ranged between zero and five indi- understand the fine-scale genetic variation among the viduals; the best estimate of total abundance across the 67 different populations. plots was 175 vipers, but confidence intervals were wide (95% CI: 38-383 vipers). Therefore, at the surveyed plots, the average density was 3.4 individuals / ha. Habitat suitability Species distribution models showed that viper dis- Table 1. a) Relationships between detections of Vipera walser and tribution is related to the interplay between topographic, variables potentially influencing detection. The table reports the climatic, and habitat parameters. First, vipers are asso- outcome of univariable N-mixture models, in which only each of ciated with gentle, south-exposed slopes at altitudes these variables was related to the viper detections. All the models between 1500 and 2300 m. This study was performed on with more than one variable showed higher AICc values than the a rather small geographical extent, and at this scale, the one with length of survey only. b) Relationships between detections of Vipera walser, and variables potentially influencing abundance. correlation between altitude and the available tempera- The table reports the outcome of N-mixture models, in which viper ture data is almost perfect (r > 0.9), therefore altitude detection probability was related to length of survey, and viper can be considered as a proxy of temperature. The narrow abundance was related to each of these variables. All the models altitudinal range suggests a very narrow thermal niche, with more than one variable showed higher AICc values than the which is typical of many micro-endemic species (Quin- model with aspect only (Appendix 1). tero and Wiens, 2013; Slatyer et al., 2013; Cunningham et al., 2016). This inherently exposes micro-endemic spe- AICc z P cies to a high risk of extinction, and is particularly alarm- a) variables potentially influencing detection ing under scenarios of climate change (Botts et al., 2013; Date* 262.9 -0.321 0.749 Slatyer et al., 2013; Böhm et al., 2016). Until recently, Hour* 261.6 -1.14 0.256 Walsers vipers were assumed to be adders Vipera berus N observers 262.8 0.51 0.610 (Andreone and Sindaco, 1998; Sindaco et al., 2006). Length of survey 244.4 4.48 <0.001 However, it should be remarked that the altitudinal range b) variables potentially influencing abundance of Walser’s vipers does not match the one observed for Altitude* 246.7 0.13 0.895 adders in the Southern Alps. Present-day adder popula- Aspect 238.6 -2.35 0.019 tions living in Lombardy (approx. 80 km E of the study Slope 245.2 1.24 0.217 area) have a broad altitudinal range (550-2500 m) and Plot area 246.7 -0.22 0.830 26% of records are above 2000 m (Bernini et al., 2004). Precipitation 245.7 1.03 0.302 Conversely, all available Walser viper data are restrict- SDM suitability 243.5 1.66 0.097 ed at altitudes of 1300-2300 m, with just 4% of records * Preliminary tests did not show any effect also for quadratic terms. above 2000 m. This suggests that Walser vipers have a Abundance and habitat of Vipera walser 79 0.9 0.9 0.6 Suitability 0.6 0.3 0.3 0.0 0 1000 2000 3000 4000 0 1 2 3 Altitude (m) Aspect (northness) 1.0 0.9 0.8 Suitability 0.6 0.6 0.3 0.4 0.0 0.2 0 0.4 0.8 1.2 600 1000 1400 1800 2200 Slope Annual precipitation (mm) 0.9 0.9 Suitability 0.6 0.6 0.3 0.3 0 50% 100% 0 50% 100% Forest cover Open vegetation cover Fig. 2. Relationship between environmental variables and suitabil- the average response across the cross-validated runs; the confidence ity, as estimated by species distribution models. The dark lines are bands indicate ± one standard deviation. narrower niche than adders, still, additional studies are ence of a relatively humid environment with open vegeta- required to compare the niches of the two species. tion, which is the main habitat of this species (Ghielmi et Second, vipers were associated to areas with relatively al., 2016). Given the very small surface of suitable habi- high precipitation levels and high cover of natural, open tats, ensuring the long-term persistence of these environ- vegetation. The Walser viper range includes some of the ments will be essential for the long-term survival of Wal- valleys with the highest rainfall in the Alps (Mercalli et ser vipers. al., 2008). High precipitation levels can allow the exist- 80 Gentile Francesco Ficetola et alii Our plots covered well the whole range of the spe- cies (Fig. 1b), still performing extrapolations of abun- dance across a species range is extremely complex (Santini et al., 2018b). If we assume that the relation- ship between viper density and plot aspect is constant across the whole range, we can perform projections of the spatial variation of potential density (Fig. 3) that, in turn, could be used to obtain rough estimates of abun- dance across the species range. Such a projection would lead to a best unbiased prediction of 26,000 vipers, but the associated confidence intervals are extremely wide (95% intervals: 5500-130,000 vipers). These figures can be useful to obtain a first approximation of abundance when no other data are available, but must be taken with extreme caution. First, model extrapolation outside the sampled areas is always challenging, because we cannot be sure that the relationship between habitat and abun- dance is constant across the range. Importantly, species distribution models often overestimate species distribu- tion (Guisan and Rahbek, 2011). This occurs because additional factors can influence species abundance, for instance, when a given patch is unsuitable because of the presence of a limiting factor not considered in this anal- ysis. Second, the reliability of estimates obtained through Fig. 3. Spatial variation of the density of Vipera walser: best predic- N-mixture heavily depends on the verification of model tions of the N-mixture model, assuming that abundance is affected by slope. The map indicates the averaged density (individuals / ha); assumptions (Barker et al., 2018; Link et al., 2018). For 95% confidence intervals of the estimates are available in the leg- instance, unmodelled heterogeneity of detection proba- end. Projections have only been performed for the most suitable bility can heavily bias total estimates of abundance (Link areas within the species range, according to the results of the spe- et al., 2018), and it is unlikely that our model took into cies distribution model. account all the potential factors affecting viper detec- tions. Third, the reliability of abundance estimates is sen- sitive to variation of detection probability, and low val- Variation in abundance ues of detection probability always challenge models that rely on unmarked individuals (Ficetola et al., 2018c). Repeated counts and N-mixture models can provide Unfortunately, the detection probability of vipers was estimates of the absolute abundance of individuals, and extremely low. are particularly useful when no data from other sources Many conservation agencies require quantitative are available. Our models showed that, for individu- measures of abundance (e.g., IUCN, 2001; Stoch and als of Walser viper, detection probability is low (around Genovesi, 2016), but obtaining reliable estimates, with 4% in standard surveys), and the species generally has a high accuracy and limited uncertainty, can be challeng- limited abundance (average: 3.4 individuals / ha). A low ing. Our study showed that uncertainty can be large even detection probability has been often observed in snakes for micro-endemic species for which a very large number (Luiselli et al., 2011; Rodda, 2012), and density estimates of surveys is performed. Given the low detectability, an are comparable to the values observed in other studies alternative approach for the monitoring of Walser vipers (reviewed in Santini et al., 2018a). For instance, Neu- could rely on occupancy modelling (MacKenzie et al., meyer (1987) used capture-mark-recapture to assess the 2017). Modelling co-occurrence in an occupancy frame- abundance of Vipera berus in subalpine environments, work or in a joint SDM framework may help improv- and found a density of approx. 3 individuals / ha. Trans- ing inference on species occupancy when data for other lating modelled abundance in density estimates is often species are available (e.g., common lizard). In addition, complex. Nevertheless, the home range size of the study distribution may be better inferred using spatially and species is likely small compared to plot size, thus the temporally replicated data in an occupancy framework, derived density probably does not suffer by temporary instead of presence-only data alone in a SDM framework. emigration (Kery and Royle, 2016). Abundance and habitat of Vipera walser 81 One solution would be integrating opportunistic infor- information is available on the decline of his habitat, and mation (i.e., presence-only data, probably the most abun- unrecorded habitat loss is possible. Remote sensing data dant and widely distributed source of information for the are an efficient approach to assess habitat trends through species) with detection-non detection data from an occu- time, when ground information is lacking (Tracewski et pancy design (Dorazio, 2014). al., 2016). Data from Corine land cover suggest that the cover of suitable habitats within the species range has remained stable during the last decade, with approx. Recommendation for monitoring and conservation 124.6 km2 of natural open vegetation in 2006, and 125.6 km2 in 2018. However, during the last decades, the cover For data deficient species, a key goal for the near of open vegetation has decreased in several areas of the future is certainly improving knowledge on distribu- Alps because of the abandonment of traditional livestock tion, abundance, and threats. Walser vipers are rare and farming (Falcucci et al., 2007), and this could cause habi- elusive, thus improving information on total abundance tat loss for Walser vipers in the future. If future monitor- and population trends will be essential. Repeated counts ing will identify declines of abundance, occupancy, or of at fixed sites are an excellent strategy for a consistent suitable habitat, Walser vipers can be classified as Endan- assessment of temporal variation in abundance, and have gered according to the IUCN redlist criteria. been recommended for the monitoring of many rep- Climatic change and the loss of open natural vegeta- tiles (Stoch and Genovesi, 2016). However, a large num- tion are the strongest threats to the biodiversity of mon- ber of sites and surveys are required for a correct esti- tane environments of Europe (Brambilla et al., 2010; mate of trends in elusive species. For instance, several Brambilla et al., 2017), and will likely also affect Walser tens of sites, each surveyed multiple times per year, are vipers in the next future. Despite only recently described, required if we want to obtain reliable estimates of spe- these vipers already are one of the most iconic animals of cies trends with dynamic N-mixture models (Ficetola et the Alps, and could serve as a flagship species for the pri- al., 2018c). The plots surveyed in this study can provide oritization of conservation management in these increas- the needed baseline for the development of a long-term ingly threatened environments. monitoring program. Alternatively, approaches involv- ing the capture of individuals can be used to collect indi- vidual longitudinal data simultaneously to occupancy ACKNOWLEDGEMENTS data. Capture-mark-recapture information would allow estimating population growth rate, the factors affecting it We thank M. Kéry for discussions on statisti- and its temporal variability. This can be done without the cal issues; the comments of two anonymous reviewers need to estimate population size and with the freedom improved a previous draft of the manuscript. R. Sindaco to simultaneously test effects on drivers of demographic and SHI-Piemonte kindly provided the artificial shelters. rates (e.g., survival, recruitment, and fecundity) (Tenan S. Ursenbacher, R. Onay, M. Galuppi, D. Monteferraio, F. et al., 2014). Marovino and A. Dalpasso helped during field sampling Despite the importance of future monitoring, con- and / or provided field data. This research did not involve servation biologists must identify priorities and plan the manipulation of animals, therefore no permits were management actions even if information is incomplete required. (Soulé, 1985). Our data can already be useful for the con- servation planning of Walser vipers. Distribution data and species distribution models suggest a very restricted REFERENCES distribution, with a total extent of occurrence < 300 km2, and an estimated area of occupancy = 86.5 km2. A lim- Andreone, F., Sindaco, R., editors. (1998): Erpetologia del ited geographic range is a key criterion for redlist assess- Piemonte e della Valle d’Aosta, Atlante degli Anfibi e ment. For instance, the IUCN (2001) classifies as Threat- dei Rettili. Museo Regionale di Scienze Naturali, Tori- ened under the criterion B species that have small geo- no. graphic range (extent of occurrence < 5000 km2 and / or Barker, R.J., Schofield, M.R., Link, W.A., Sauer, J.R. area of occupancy < 500 km2), and also satisfy at least (2018): On the reliability of N-mixture models for two of these three conditions: a) severely fragmented or count data. Biometrics 74: 369-377. restricted number of locations; b) continuing decline of Beck, J., Ballesteros-Mejia, L., Buchmann, C.M., Dengler, the species or his habitat; c) extreme demographic fluc- J., Fritz, S.A., Gruber, B., Hof, C., Jansen, F., Knapp, tuations. The Walser viper has a very small range and is S., Kreft, H., Schneider, A.K., Winter, M., Dormann, known from a limited number of locations. Until now, no 82 Gentile Francesco Ficetola et alii C.F. (2012): What’s on the horizon for macroecology? Cunningham, H.R., Rissler, L.J., Buckley, L.B., Urban, Ecography 35: 673-683. M.C. (2016): Abiotic and biotic constraints across rep- Bernini, F., Bonini, L., Ferri, V., Gentilli, A., Razzetti, E., tile and amphibian ranges. Ecography 39: 1-8. Scali, S., editors. (2004): Atlante degli Anfibi e dei Ret- Dail, D., Madsen, L. (2011): Models for estimating abun- tili della Lombardia. Provincia di Cremona, Cremona. dance from repeated counts of an open metapopula- Böhm, M., Williams, R., Bramhall, H.R., McMillan, K.M., tion. Biometrics 67: 577-587. Davidson, A.D., Garcia, A., Bland, L.M., Bielby, J., De Leo, F., Donelli, O., Scali, S. (2006): Use of natural and Collen, B. (2016): Correlates of extinction risk in artificial shelters in a snake community of northern squamate reptiles: the relative importance of biology, Italy. In: Atti del V convegno nazionale Societas Her- geography, threat and range size. Global Ecol. Bioge- petologica Italica, pp. 25-31. Zuffi, M.A.L., Ed, Firenze ogr. 25: 391-405. University Press, Firenze. Botts, E.A., Erasmus, B.F.N., Alexander, G.J. (2013): Small Dodd, C.K.J., editor. (2010): Amphibian ecology and con- range size and narrow niche breadth predict range servation. A handbook of techniques. Oxford Univer- contractions in South African frogs. Global Ecol. Bio- sity Press, Oxford. geogr. 22: 567-576. Dorazio, R.M. (2014): Accounting for imperfect detection Brambilla, M., Caprio, E., Assandri, G., Scridel, D., Bas- and survey bias in statistical analysis of presence-only si, E., Bionda, R., Celada, C., Falco, R., Bogliani, G., data. Global Ecol. Biogeogr. 23: 1472-1484. Pedrini, P., Rolando, A., Chamberlain, D. (2017): A Dormann, C.F., Elith, J., Bacher, S., Buchmann, C., Carl, spatially explicit definition of conservation priorities G., Carre, G., Marquez, J.R.G., Gruber, B., Lafour- according to population resistance and resilience, spe- cade, B., Leitao, P.J., Munkemuller, T., McClean, C., cies importance and level of threat in a changing cli- Osborne, P.E., Reineking, B., Schroder, B., Skidmore, mate. Divers. Distrib. 23: 727-738. A.K., Zurell, D., Lautenbach, S. (2013): Collinearity: Brambilla, M., Casale, F., Bergero, V., Bogliani, G., a review of methods to deal with it and a simulation Crovetto, G.M., Falco, R., Roati, M., Negri, I. (2010): study evaluating their performance. Ecography 36: Glorious past, uncertain present, bad future? Assess- 27-46. ing effects of land-use changes on habitat suitability Duarte, A., Adams, M.J., Peterson, J.T. (2018): Fit- for a threatened farmland bird species. Biol. Conserv. ting N-mixture models to count data with unmod- 143: 2770-2778. eled heterogeneity: Bias, diagnostics, and alternative Brambilla, M., Ficetola, G.F. (2012): Species distribution approaches. Ecol. Model. 374: 51-59. models as a tool to estimate reproductive parameters: Dufresnes, C., Mazepa, G.O., Rodriguez, N., Brelsford, a case study with a passerine bird species. J. Anim. A., Litvinchuk, S.N., Sermier, R., Betto-Colliard, C., Ecol. 81: 781-787. Blaser, O., Borzée, O., Cavoto, E., Fabre, G., Ghali, K., Burgman, M.A., Fox, C.W. (2003): Bias in species range Grossen, C., Horn, A., Lavanchy, G., Leuenberger, J., estimates from minimum convex polygons: implica- Phillips, B.C., Saunders, P.A., Savary, R., Maddalena, tions for conservation and options for improved plan- T., Stöck, M., Dubey, S., Canestrelli, D., Jeffries, D.L. ning. Anim. Conserv. 6: 19-28. (2018): Genomic evidence for cryptic speciation in Burnham, K.P., Anderson, D.R. (2002): Model selection tree frogs from the Apennine Peninsula, with descrip- and multimodel inference: a practical information- tion of Hyla perrini sp. nov. . Frontiers in Ecology and theoretic approach. Springer Verlag, New York. Evolution doi: 10.3389/fevo.2018.00144 Cornetti, L., Ficetola, G.F., Hoban, S., Vernesi, C. (2015): Elith, J., Graham, C.H., Anderson, R.P., Dudík, M., Fer- Genetic and ecological data reveal species bounda- rier, S., Guisan, A., Hijmans, R.J., Huettmann, F., ries between viviparous and oviparous lizard lineages. Leathwick, J.R., Lehmann, A., Li, J., Lohmann, L.G., Heredity 115: 517-526. Loiselle, B.A., Manion, G., Moritz, C., Nakamura, M., Costa, A., Oneto, F., Salvidio, S. (2019): Time-for-space Nakazawa, Y., Overton, J.M., Peterson, A.T., Phillips, substitution in N-mixture modeling and population S.J., Richardson, K.S., Scachetti-Pereira, R., Schapire, monitoring. J. Wildlife Manage. 83: 737-741. R.E., Soberon, J., Williams, S., Wisz, M.S., Zimmer- Crump, M.L., Scott, N.J. (1994): Visual encounter sur- mann, N.E. (2006): Novel methods improve predic- veys. In: Measuring and monitoring biological diversi- tion of species’ distributions from occurrence data. ty: standard methods for Amphibians, pp. 84-92. Hey- Ecography 29: 129-151. er, W.R., Donnelly, M.A., McDiarmid, R.W., Hayek, Elith, J., Phillips, S.J., Hastie, T., Dudík, M., Chee, Y.E., L.C., Foster, M.S., Eds, Smithsonian Institution Press, Yates, C.J. (2011): A statistical explanation of MaxEnt Washington. for ecologists. Divers. Distrib. 17: 43-57. Abundance and habitat of Vipera walser 83 Falcucci, A., Maiorano, L., Boitani, L. (2007): Changes in lution for the earth’s land surface areas. Sci. Data 4: land-use/land-cover patterns in Italy and their impli- 170122. cations for biodiversity conservation. Landscape Ecol. Kéry, M. (2018): Identifiability in N-mixture models: a 22: 617-631. large-scale screening test with bird data. Ecology 99: Ficetola, G.F., Barzaghi, B., Melotto, A., Muraro, M., 281-288. Lunghi, E., Canedoli, C., Lo Parrino, E., Nanni, V., Sil- Kéry, M., Dorazio, R.M., Soldaat, L., Van Strien, A., va-Rocha, I., Urso, A., Carretero, M.A., Salvi, D., Sca- Zuiderwijk, A., Royle, J.A. (2009): Trend estimation li, S., Pennati, R., Andreone, F., Manenti, R. (2018a): in populations with imperfect detection. J. Appl. Ecol. N-mixture models reliably estimate the abundance of 46: 1163-1172. small vertebrates. Sci. Rep. 8: 10357. Kery, M., Royle, J.A. (2016): Applied Hierarchical Mod- Ficetola, G.F., Lunghi, E., Canedoli, C., Padoa-Schiop- eling in Ecology: Analysis of distribution, abundance pa, E., Pennati, R., Manenti, R. (2018b): Differences and species richness in R and BUGS: Volume 1:Prelude between microhabitat and broad-scale patterns of and Static Models. Academic Press, Burlington, MA. niche evolution in terrestrial salamanders. Sci. Rep. 8: Link, W.A., Schofield, M.R., Barker, R.J., Sauer, J.R. 10575. (2018): On the Robustness of N-mixture models. Ficetola, G.F., Romano, A., Salvidio, S., Sindaco, R. Ecology 99: 1547-1551. (2018c): Optimizing monitoring schemes to detect Lobo, J.M., Jiménez-Valverde, A., Real, R. (2008): AUC: a trends in abundance over broad scales. Anim. Con- misleading measure of the perfromance of predictive serv. 21: 221-231. distribution models. Global Ecol. Biogeogr. 17: 145- Ficetola, G.F., Rondinini, C., Bonardi, A., Katariya, V., 151. Padoa-Schioppa, E., Angulo, A. (2014): An evaluation Luiselli, L., Rugiero, L., Capula, M. (2011): Are commu- of the robustness of global amphibian range maps. J. nal nesting counts as useful as mark-recapture data Biogeogr. 41: 211-221. for estimating population size in snakes? Herpetol. J. Fiske, I., Chandler, R. (2011): unmarked: An R package 21: 73-81. for fitting hierarchical models of wildlife occurrence Lunghi, E., Corti, C., Manenti, R., Ficetola, G.F. (2019): and abundance. J. Stat. Softw. 43: 1-23. Consider species specialism when publishing datasets. Ghielmi, S., Bergò, P.E., Andreone, F. (2006): Nuove seg- Nature Ecol. Evol. 3: 19. nalazioni di Zootoca vivipara Jaquin e di Vipera berus Lunghi, E., Manenti, R., Mulargia, M., Veith, M., Corti, Linnaeus, in Piemonte, Italia nord-occidentale (Novi- C., Ficetola, G.F. (2018): Environmental suitability tates Herpetologicae Pedemontanae II). Acta Herpe- models predict population density, performance and tol. 1: 29-36. body condition for microendemic salamanders. Sci. Ghielmi, S., Menegon, M., Marsden, S.J., Laddaga, L., Ursen- Rep. 8: 7527. bacher, S. (2016): A new vertebrate for Europe: the dis- MacKenzie, D.I., Nichols, J.D., Royle, J.A., Pollock, K.H., covery of a range-restricted relict viper in the western Bailey, L.A., Hines, J.E. (2017): Occupancy estima- Italian Alps. J. Zool. Syst. Evol. Res. 54: 161-173. tion and modeling: inferring patterns and dynamics Guisan, A., Rahbek, C. (2011): SESAM - a new frame- of species occurrence. 2nd edition. Academic Press, work integrating macroecological and species distri- Burlington, MA. bution models for predicting spatio-temporal patterns Mazerolle, M.J. (2015): Estimating detectability and bio- of species assemblages. J. Biogeogr. 38: 1433-1444. logical parameters of interest with the use of the R Guisan, A., Thuiller, W. (2005): Predicting species dis- environment. J. Herpetol. 49: 541-559. tribution: offering more than simple habitat models. Mercalli, L., Cat Berro, D., Acordon, V., Di Napoli, G. Ecol. Lett. 8: 993-1009. (2008): Cambiamenti climatici sulla montagna pie- IUCN. (2001): IUCN Red List Categories and Criteria: montese. Rapporto tecnico realizzato da Società mete- Version 3.1. IUCN, Gland, Switzerland and Cam- orologica Subalpina per conto di Regione Piemonte. bridge, UK. Società Meteorologica Subalpina Castello Borello, Joppa, L.N., Williams, C.K., Temple, S.A., Casper, G.S. Bussoleno (TO), Italy. (2010): Environmental Factors Affecting Sampling Nash, D.J., Griffiths, R.A. (2018): Ranging behaviour of Success of Artificial Cover Objects. Herpetol. Con- adders (Vipera berus) translocated from a develop- serv. Biol. 5: 143-148. ment site. Herpetol. J. 28: 155-159. Karger, D.N., Conrad, O., Böhner, J., Kawohl, T., Kreft, Neumeyer, R. (1987): Density and seasonal movements of H., Soria-Auza, R.W., Zimmermann, N.E., Linder, the adder (Vipera berus L. 1758) in a subalpine envi- H.P., Kessler, M. (2017): Climatologies at high reso- ronment. Amphibia-Reptilia 8: 259-275. 84 Gentile Francesco Ficetola et alii Nogués-Bravo, D. (2009): Predicting the past distribution Global drivers of population abundance in terrestrial of species climatic niche. Global Ecol. Biogeogr. 18: vertebrates. Global Ecol. Biogeogr. 27: 968-979. 521-531. Senczuk, G., Castiglia, R., Böhme, W., Corti, C. (2019): Pateiro-López, B., Rodrjguez-Casal, A. (2010): Generaliz- Podarcis siculus latastei (Bedriaga, 1879) of the west- ing the convex hull of a sample: The R package alpha- ern Pontine islands (Italy) raised to the species rank, hull J. Stat. Softw. 34: 1-28. and a brief taxonomic overview of Podarcis lizards. Pearson, R.G., Raxworthy, C.J., Nakamura, M., Peterson, Acta Herpetol.: doi: 10.13128/Acta_Herpetol-25159. A.T. (2007): Predicting species distributions from Sewell, D., Guillera-Arroita, G., Griffiths, R.A., Beebee, small numbers of occurrence records: a test case using T.J.C. (2012): When is a species declining? Optimizing cryptic geckos in Madagascar. J. Biogeogr. 34: 102- survey effort to detect population changes in reptiles. 117. Plos One 7: e43387. Peterson, A.T., Soberon, J., Pearson, R.G., Anderson, R.P., Sillero, N., Campos, J., Bonardi, A., Corti, C., Creemers, Martinez-Meyer, E., Nakamura, M., Araujo, M.B. R., Crochet, P.A., Isailović, J.C., Denoël, M., Ficetola, (2011): Ecological niches and geographic distribu- G.F., Gonçalves, J., Kuzmin, S., Lymberakis, P., de tions. Princeton University Press, Princeton. Pous, P., Rodríguez, A., Sindaco, R., Speybroeck, J., Phillips, S.J., Anderson, R.P., Schapire, R.E. (2006): Maxi- Toxopeus, B., Vieites, D.R., Vences, M. (2014): Updat- mum entropy modeling of species geographic distri- ed distribution and biogeography of amphibians and butions. Ecol. Model. 190: 231-259. reptiles of Europe based on a compilation of country- Pollock, K.H., Nichols, J.D., Simons, T.R., Farnsworth, wide mapping studies. Amphibia-Reptilia 35: 1-31. G.L., Bailey, L.L., Sauer, J.R. (2002): Large scale wild- Sindaco, R., Doria, G., Razzetti, E., Bernini, F., editors. life monitoring studies: statistical methods for design (2006): Atlas of Italian Amphibians and Reptiles. and analysis. Environmetrics 13: 105-119. Polistampa, Firenze. Potter, K.A., Arthur, W.H., Pincebourde, S. (2013): Slatyer, R.A., Hirst, M., Sexton, J.P. (2013): Niche breadth Microclimatic challenges in global change biology. predicts geographical range size: a general ecological Global Change Biol. 19: 2932-2939. pattern. Ecol. Lett. 16: 1104-1114. Quintero, I., Wiens, J.J. (2013): What determines the cli- Soulé, M.E. (1985): What is conservation biology? Biosci- matic niche width of species? The role of spatial and ence 35: 727-734. temporal climatic variation in three vertebrate clades. Stoch, F., Genovesi, P. (2016): Manuali per il monitoraggio Global Ecol. Biogeogr. 22: 422-432. di specie e habitat di interesse comunitario (Direttiva Raxworthy, C.J., Martinez-Meyer, E., Horning, N., Nuss- 92/43/CEE) in Italia: specie animali. ISPRA, Roma. baum, R.A., Schneider, G.E., Ortega-Huerta, M.A., Tenan, S., Pradel, R., Tavecchia, G., Igual, J.M., Sanz‐Agui- Peterson, A.T. (2003): Predicting distributions of lar, A., Genovart, M., Oro, D. (2014): Hierarchical known and unknown reptile species in Madagascar. modelling of population growth rate from individual Nature 426: 837-841. capture–recapture data. Methods Ecol. Evol. 5: 606-614. Rodda, G.H. (2012): Population size and demographics. Tracewski, L., Butchart, S., Di Marco, M., Ficetola, G.F., In: Reptile biodiversity. Standard methods for inven- Rondinini, C., Symes, A., Wheatley, H., Beresford, A., tory and monitoring, pp. 283-300. McDiarmid, R.W., Buchanan, G. (2016): Toward quantification of the impact Foster, M.S., Guyer, C., Gibbons, J.W., Chernoff, N., of 21st century deforestation on the extinction risk of ter- Eds, University of California Press, Berkley and Los restrial vertebrates. Conserv. Biol. 30: 1070-1079. Angeles, CA. Warren, D.L., Seifert, S.N. (2011): Ecological niche mod- Royle, J.A. (2004): N-mixture models for estimating pop- eling in Maxent: the importance of model complexity ulation size from spatially replicated counts. Biomet- and the performance of model selection criteria. Ecol. rics 60: 108-115. Appl. 21: 335-342. Royle, J.A., Nichols, J.D. (2003): Estimating abundance Weber, M.M., Stevens, R.D., Diniz-Filho, J.A.F., Grelle, from repeated presence-absence data or point counts. C.E.V. (2017): Is there a correlation between abun- Ecology 84: 777-790. dance and environmental suitability derived from Santini, L., Isaac, N.J.B., Ficetola, G.F. (2018a): TetraDEN- ecological niche modelling? A meta-analysis. Ecogra- SITY: A database of population density estimates in phy 40: 817-828. terrestrial vertebrates. Global Ecol. Biogeogr. 27: 787- Zuffi, M.A.L., Macchia, M., Ioale, P., Giudici, F. (1999): 791. Winter activity in a coastal population of Vipera aspis Santini, L., Isaac, N.J.B., Maiorano, L., Ficetola, G.F., Hui- (Reptilia, Viperidae). Revue d’Ecologie-La Terre Et La jbregts, M.A.J., Carbone, C., Thuiller, W. (2018b): Vie 54: 365-374. Abundance and habitat of Vipera walser 85 APPENDIX Appendix S1. N-mixture models relating the detections of Vipera walser to variables potentially influencing abundance. The table reports the regression coefficient of each variable included in each model; models are ranked on the basis of the corrected Akaike’s information cri- terion. Only models with weight >0.001 are shown. k: N of parameters in the model. Length of Model Aspect Altitude Plot area Precipitation Slope k AICc weight survey 1 0.69 -0.70 - - - - 4 238.6 0.312 2 0.69 -0.68 - - 0.06 - 5 240.8 0.103 3 0.70 -0.70 - -0.02 - - 5 240.9 0.098 4 0.69 -0.70 -0.01 - - - 5 240.9 0.097 5 0.69 -0.70 - - - 0.00 5 240.9 0.097 6 0.70 -0.71 - - 0.10 -0.07 6 243.2 0.032 7 0.70 -0.68 - -0.02 0.06 - 6 243.2 0.031 8 0.69 -0.68 -0.01 - 0.06 - 6 243.2 0.031 9 0.70 -0.70 -0.01 -0.02 - - 6 243.3 0.029 10 0.70 -0.70 - -0.02 - 0.00 6 243.3 0.029 11 0.69 -0.70 -0.01 - - 0.00 6 243.4 0.029 12 0.73 - - - - - 3 244.4 0.017 13 0.72 - - - - 0.21 4 245.2 0.012 14 0.71 -0.71 - -0.02 0.10 -0.07 7 245.7 0.009 15 0.74 - - - 0.17 - 4 245.7 0.009 16 0.70 -0.71 0.00 - 0.10 -0.07 7 245.7 0.009 17 0.70 -0.68 0.00 -0.02 0.06 - 7 245.7 0.009 18 0.70 -0.70 -0.01 -0.02 - 0.00 7 245.8 0.008 19 0.74 - - -0.04 - - 4 246.7 0.006 20 0.72 - 0.03 - - - 4 246.7 0.006 21 0.72 - - - 0.06 0.17 5 247.4 0.004 22 0.73 - - -0.04 - 0.21 5 247.5 0.004 23 0.72 - 0.00 - - 0.21 5 247.5 0.004 24 0.75 - - -0.03 0.16 - 5 248 0.003 25 0.74 - 0.02 - 0.17 - 5 248 0.003 26 0.71 -0.71 0.00 -0.02 0.10 -0.07 8 248.2 0.003 27 0.74 - 0.03 -0.04 - - 5 249 0.002 28 0.73 - - -0.03 0.06 0.17 6 249.8 0.001 29 0.72 - 0.01 - 0.06 0.17 6 249.8 0.001 30 0.73 - 0.00 -0.04 - 0.21 6 249.9 0.001 31 0.75 - 0.03 -0.03 0.17 - 6 250.4 0.001