Elevational gradient and human effects on butterfly species richness in the French Alps

Abstract We examined how butterfly species richness is affected by human impact and elevation, and how species ranges are distributed along the elevational gradient (200–2700 m) in the Isère Department (French Alps). A total of 35,724 butterfly observations gathered in summer (May–September) between 1995 and 2015 were analyzed. The number of estimated species per 100‐m elevational band was fitted to the elevational gradient using a generalized additive model. Estimations were also performed on a 500 m × 500 m grid at low altitude (200–500 m) to test for the human impact on species richness using generalized least squares regression models. Each species elevational range was plotted against the elevational gradient. Butterfly richness along the elevational gradient first increased (200–500 m) to reach a maximum of 150 species at 700 m and then remained nearly constant till a sharp decrease after 1900 m, suggesting that after some temperature threshold, only few specialized species can survive. At low elevation, urbanization and arable lands had a strongly negative impact on butterfly diversity, which was buffered by a positive effect of permanent crops. Butterfly diversity is exceptionally high (185 species) in this alpine department that represents less than 5% of the French territory and yet holds more than 70% of all the Rhopalocera species recorded in France. Both climate and habitat shape the distribution of species, with a negative effect of anthropization at low altitude and strong climatic constraints at high altitude.

species ranges within a bounded biogeographical domain produces a peak of richness at intermediate elevations, due to geometric constraints (Colwell & Hurtt, 1994). Other hypotheses have been proposed to explain the unimodal pattern, including highest productivity at mid-elevations (Mittelbach et al., 2001;Sanders, 2002) or interactions with human activities at lower elevations (Bharti, Sharma, Bharti, & Pfeiffer, 2013;McKinney, 2002;Nogues-Bravo, Araujo, Romdal, & Rahbek, 2008). Elevational gradients are tightly interconnected with human activities, and both climate and local factors (e.g., land use) are likely interacting to explain the species richness patterns observed along elevational gradients.
Butterfly diversity is known to be particularly high in mountain regions, presumably because elevational gradients encompass several gradients in climatic and environmental factors (especially temperature and moisture) and vegetation assemblages vary along elevational gradients, contributing to environmental heterogeneity (Pellissier et al., 2013). Butterfly populations are also considered good indicators of environmental changes as they react more sensitively to habitat loss and decline more rapidly than birds and plants in regions with high human pressure (Thomas et al., 2004). Human effects on butterfly diversity can be negative through urbanization (habitat loss) and habitat conversion to arable lands but, on another hand, human activities such as traditional agriculture maintain some open habitats (such as permanent croplands or extensive pastures) and promote environmental heterogeneity (Uchida, Hiraiwa, & Ushimaru, 2016) that can represent favorable habitats for butterflies (Bartonova, Benes, Fric, Chobot, & Konvicka, 2016;Botham et al., 2015;Horak & Safarova, 2015;Jew, Loos, Dougill, Sallu, & Benton, 2015).
The aim of the present study is to analyze the pattern of elevational richness of butterflies in an area of high human pressure in the French Alps, based on a set of bioclimatic, land cover and >35,000 occurrence data of butterfly. More precisely, we asked whether butterfly richness decreases monotonically with increasing elevation or exhibits a peak at intermediate elevations. We further examined the effects of anthropization (urbanization and arable land) and other landscape characteristics (forest, permanent crops, grassland, sparse vegetation) on species richness.

| Study area
This study was carried out in the northwestern Alps, Isère department, France ( Figure 1). The department of Isère extends c. 130 km from north to south and c. 120 km from east to west with a total area of 7,431 km 2 and is densely populated (157.37 inhabitant/km 2 ). The southeastern half of Isère consists in a mountainous region with four main mountain ranges, namely Chartreuse, Vercors, Belledonne and F I G U R E 1 Topographic map of the Isère department and distribution of all the 4,776 sites with at least one observation (in blue) and of the 75 sites (in red) with species estimation more than 70%, of which 64 were used for analyzing the impact of man at low altitude (<500 m). The city of Grenoble (>500,000 inhabitants) is indicated the northernmost part of the Écrins, each being geographically separated by deep valleys and with an elevational range spanning from 200 m up to >3,500 m in the Écrins (~3,000 m in Belledonne, and ~2,000 m in Vercors and Chartreuse). The northwestern half of Isère is mainly characterized by hills, most not exceeding 700 m. About half the total surface of the department lies between 200 and 500 m.
At low elevations, the landscape is mainly rural or semiurban and consists of patches of agricultural lands, villages or small towns, and deciduous forests. Human impact in this region is not new as there are many traces of human activities back to the Mesolithic in valleys (the roman cities of Vienna, Grenoble (Cularo), and Cremieux were constructed on old prehistorical settlements (Bocquet et al., 1987) and in mountains (Martin, Delhon, Thiébault, & Pelletier, 2012). Traditional agriculture and pasture have shaped the landscape since thousands of years. Urbanization and intensive agriculture (mainly corn crops) have risen dramatically during the last decades in the three main river beds (Rhône, Isère, Drac) at the expense of swamps and wetlands that constituted the main landscape elements in the bottom of these large glacier valleys, and Grenoble is now the tenth city in France with more than 500,000 inhabitants. Rural areas (small cities, traditional extensive agriculture) can reach up to c. 1,500 m in the southeastern part of the department. Human impact is not restricted to lowlands in Isère: forestry represents a major resource for this mountainous department since the sixteenth century, and it was the first department in France to develop hydroenergy-based industries during the nineteenth century; finally, it is one of the most attractive touristic areas in Europe (many recreative areas including ski resorts). The intermediate elevations are dominated by mixed and coniferous forests (mostly managed) which are replaced at higher elevations (>1,700 m) by grasslands (alpine meadows, used as pastures and harboring ski resorts) and finally by bare rocks and/or glaciers (~>2,300 m). Human activities generate a mosaic of habitats more or less favorable to butterflies (urban, arable lands, pastures, permanent crops, sparse vegetation). In addition to human activities largely shaping the landscape, climatic conditions are particularly diverse in Isère. The influence of a continental, oceanic and mediterranean climate offers a wide range of environmental conditions at lower elevations which gradually turn into an alpine climate at higher elevations. For the 1995-2015 period, the mean annual temperature varies between +11 and −0.5°C, and the mean annual precipitation between 700 and 2,200 mm is depending on the elevation and the massif.  (Table 1).

| Species richness estimation and study design
To estimate the number of species at a given site, and to link it with land cover characteristics, the department of Isère was divided into 500 × 500 m cells, and each observation falling into a cell was attributed to this site: the 35,724 observations corresponded to a total of 6,447 sites distributed across the department ( Figure 1). This dataset was strongly biased toward low elevation observations: most of the sites that were sampled several times were located below 500 m. We therefore performed two types of analyses of species richness. First, we evaluated variation of species richness along the 200-2,700 m gradient by dividing it in twenty-six 100-m elevational bands, and we analyzed the relationships between species richness per elevational band and environmental parameters (altitude, climate, main vegetation type). Second, we focused on human impact at lower altitude (200-500 m) on species richness in 500 × 500 m grid cells.
The number of observed species in a given area is always lower than the real number of species because of undetected species, and several unbiased estimators of species richness have been developed (Gotelli & Colwell, 2011). The estimated number of species at a site relies on sampling intensity, which in turn depends on multiple variables such as accessibility, proximity to urban areas, perceived interest for researchers, or presence of research institutions (Ficetola, Bonardi, Sindaco, & Padoa-Schioppa, 2013;Stolar & Nielsen, 2015;Yang, Ma, & Kreft, 2014). Comparative analyses showed that the first-order jackknife is one of the best performing approaches for biodiversity estimates (Chazdon, Colwell, Denslow, & Guariguata, 1998). Species richness in each elevational band of 100 m (broad scale analysis) and in each grid cell (low elevation analysis) was estimated on the basis of occurrence data using the first-order jackknife estimator (Colwell & Coddington, 1994), as implemented in the "vegan" package (Dixon, 2003) in R (R Development Core Team, 2016).

| Environmental data
To assess the potential role of environmental variables in shaping but-

| Human impact on butterfly richness at low elevation (200-500 m)
Estimators of species richness provide more accurate results if the observed species richness is not too far from the estimated one (Chazdon et al., 1998;Colwell & Coddington, 1994). We only considered cells for which the observed species richness was at least 70% of the estimated one (jackknife first-order estimator) and that were visited more than four times: of 6,447 sites, only 75 sites met these requirements, of which 64 were at elevation between 200 and 500 m. Therefore, the analysis of relationships between butterfly richness and landscape features was performed on 64 sites (500 × 500 m cells We used generalized least squares (GLS) to assess the effect of land use on butterfly species richness at lower elevations (200-500 m) while taking into account spatial structure. GLS allows the incorporation of spatial structure into the error of the model and is considered among the techniques with the best performance for the analysis of spatial data (Beale, Lennon, Yearsley, Brewer, & Elston, 2010;Dormann et al., 2007). We built GLS models considering all the possible combinations of independent variables; we then ranked models on the basis of their Akaike's information criterion corrected for small sample size (AICc); the models with lowest AICc values are considered to be the "best models". AICc may select overly complex models; therefore, we considered a complex model as a candidate model only if it had AICc less than the AICc of all its simpler nested models (Richards, Whittingham, & Stephens, 2011). For each candidate model, we calculated the Akaike's weight w (AICc weight), which represents the probability of the different models given the data (Lukacs et al., 2007). We also calculated the relative importance of variables (RI) by summing the AICc weights of models in which each variable is included (Wagenmakers, 2003). The variance inflation factor of all the best AICc models was <2, indicating lack of collinearity issues (Dormann et al., 2013). We used likelihood ratio R 2 (R 2 LR ) as a measure of the variance explained by the model.

| Niche separation
For each species with more than 50 total observations, the mean altitude and standard deviation (SD) was calculated based on all observations. The standard deviation was taken as a proxy for niche width: species with large standard deviation have a broad elevational niche (generalists), and species with small standard deviation are restricted to a small altitude interval (specialists). All statistical analyses were performed in R 3.3.0 (R Development Core Team, 2016).

| RESULTS
The observed species richness was >60% of estimated species richness in every elevational band, with small standard errors, indicating a robust estimation (Table 2)

| Relationships between human impact and butterfly richness at low altitude
Between 200 and 500 m, most of the available area was occupied by arable lands (intensive agriculture, 33%) and by deciduous forests (28%), while urban areas represented 9% of the total area. Grasslands, wetlands and swamps, remains of the most natural lowland habitats, represented less than 5% of the total area; the landscape was also constituted of pastures and orchards (permanent crops, 13%) or sparse vegetation (tree hedges, gloves, isolated trees; 10%) maintained by traditional agriculture practices. A total of 64 models were compared that tested all combinations of six variables: urban, arable, permanent crops, deciduous forest, grasslands, and sparse vegetation; wetlands and swamps were not included because the sum of all land cover variables is constrained to 1. The best AIC model included three variables: species richness was negatively related to the presence of urban and arable lands, while the relationship between species richness and permanent croplands was positive (Tables 3 and 4), with relative importance of 0.97, 0.98 and 0.78, for urban, arable, and permanent crops, respectively. The number of species decreased substantially with increasing anthropization (urban + arable lands), with 2.2 species lost every 10% more anthropized area (Figure 3).

| Species separation along the elevation gradient
The ordination of species along elevational gradient according to their niche width showed that specialist species are restricted to the two extremes of the elevational gradient, below 500 m and above 2,000 m ( Figure 4). Niche width (SD) ranged from 8.9 (Coenonympha oedippus) up to more than 750 (Pyrgus malvoides). Of 106 species, 56 had SD >350 (generalists) and 50 had SD <350 (specialist).
There were significantly more specialist species at low (<500 m) and less specialist species at the intermediate altitude (500-1,000 m) than expected if specialists were randomly distributed across the elevational gradient (Table 5; Figure S2; Fisher exact test, odd ratio = 1.97, p < .05). that is, variation in richness may simply reflect variation in sampling intensity along this elevational gradient (Lomolino, 2001). The correlation between sampling effort and species richness was low (r = .5, p = .01), and the most intensively sampled zone (the 200-500 m elevation belt) was not the most species-rich area, suggesting that variation in sampling effort was not a main predictor of the pattern. The to butterflies (Viljur & Teder, 2016). Interestingly, although nonsignificant, the increase in species richness observed at ~700 and ~1,700 m (153 and 143 estimated species, respectively, Table 2) corresponds to two ecotones: the transition between the foothill and montane zones (from deciduous to coniferous forests), and the montane to subalpine zones (tree line: transition from coniferous forest to grasslands). This "ecotone effect" where species assemblages from two different bioclimatic zones meet is an acknowledged factor of species richness (Lomolino, 2001;McCain & Grytnes, 2010 F I G U R E 3 Relationship between species richness and proportion of anthropized area (urban + arable lands) in 64 sites between 200 and 500 m F I G U R E 4 Niche width (mean altitude and standard deviation) distribution along the elevational gradient for 106 species with more than 50 observations. Dashed lines: 500-m elevational bands common in Isère only a few decades ago and have a large European distribution, but are now restricted to the few remaining protected areas, with little migration between them. These remnant populations are at high extinction risk due to stochastic demographic events and inbreeding (Thomas, 2016), and their situation is similar throughout their distribution range where they face the same habitat loss threat (Celik et al., 2015;Gao, Li, Chen, Guo, & Settele, 2016;van Halder et al., 2008;Jubete & Roman, 2016;Orvossy, Korosi, Batary, Vozar, & Peregovits, 2013). However, this negative effect of anthropization was buffered by a positive effect of permanent crops (extensive pastures, orchards) on butterfly richness. The amount of forest edges and clearings, as well as small-scale agricultural mosaics of fields and forests,

This small northwestern French Alps region is very rich in regard
were found in previous studies to be the most important variables for butterfly diversity (Kivinen, Luoto, Kuussaari, & Saarinen, 2007), possibly due to high oviposition rate and high survival of larvae in those areas (Luoto et al., 2001), in addition to providing nectar resources, efficient sheltered areas to wind and refugees from predators to imagos. At a very local scale, several recent studies have emphasized the role of gardens and urban parks to buffer the negative effect of urbanization on butterflies by providing nectar resources (Fontaine, Bergerot, Le Viol, & Julliard, 2016;Lizee et al., 2016) (Sing, Dong, Wang, & Wilson, 2016). Maintaining sparsely vegetated and semiopen woodlands with glades that constitute important butterfly habitats is a recommended management strategy for conservation goals (Bubova et al., 2015;Nilsson, Franzen, & Pettersson, 2013).

We thank Julien Renaud for his help with GIS data access and Wilfried
Thuiller for helpful advices on data analyses. This work was funded by the Pôle Biodiversité of the Conseil Départemental de l'Isère.