Uncovering environmental , land-use and fire effects on the distribution of a low-dispersal species , the Hermann ’ s tortoise Testudo hermanni

Habitat destruction has resulted in the fragmented distribution of numerous terrestrial species, which poses a challenge for conservationists. Furthermore, species management can be further compounded by life history constraints such as low dispersal, hindering the ability of species to recolonize areas they formerly occupied. For these species, a thorough understanding of the local threats and factors that limit their distribution is crucial for effective management. We used occupancy models to examine which factors at landscape and habitat scales (i.e. land uses, fire history, and vegetation structure) explain the presence of terrestrial tortoises within the range of the westernmost isolated population of the endangered Testudo hermanni hermanni in the Albera Range (NE of the Iberian Peninsula). We randomly surveyed 25 sites (75% of the area known with presence of tortoises) of natural woodlands with 5 to 8 replicates per site in spring 2012-2014. From a sampling effort of 148 hours, we only detected 52 tortoises in 12 of 25 transects. These low numbers are evidence of low population densities. Sites with presence of tortoises were spatially aggregated although the species was absent from apparently adequate sites on the edges of its distribution range. Current and historic land-use primarily explained the presence of tortoises. Besides, wildfires and reduction of habitat complexity also participate to explain the distribution of Hermann’s tortoises. We also discuss some aspects of the conservation of Testudo hermanni in relation to our results.


Introduction
The current distribution of a species is shaped by historical, environmental and biotic-interactions (MacArthur, 1972;Brown, 1984;Gaston, 2000;Wisz et al., 2013), as well as anthropogenic causes in regions intensely modified by humans (Blondel et al., 2010;Vimal et al., 2012;Fattorini, 2014).Our capacity to track how these causes have modified species distribution is limited, and mostly derived from the analysis of current distributions and long-term monitoring in selected taxonomic groups such as birds and mammals (Barbet-Massin, Thuiller and Jiguet, 2012;Pekin and Pijanowski, 2012).For example, historical changes in land-use over the past century have affected landscape structure and consequently the composition of animal communities (Barbero et al., 1990;Farina, 1991;Kiss, Magnin and Torre, 2004).In the northern rim of the Mediterranean basin, these changes were characterized by two opposing trends since the twentieth century.First, land-use intensification and abandonment of marginal areas, generally situated in the mountains, became economically nonviable over time (Debussche, Lepart and Dervieux, 1999).Second, scrubs and forest have expanded and semi-natural open habitats have decreased (Chauchard, Carcailet and Guibal, 2007).As a consequence, intensive wildfires have replaced disturbance caused by burn-beating and grazing in the northern Mediterranean mountains (Barbero et al., 1990).
The collective effects of land-use transformation and fire in the northern rim of the Mediterranean basin contribute to shape the distribution of species, and produced a significant reduction and fragmentation of their distribution range (Pausas et al., 2009;Blondel et al., 2010;Kozlowski, 2012;Rodríguez-Caro et al., 2016).Highly modified landscapes may prevent species from expanding or even reoccupying their former distribution range (Wilcove, McLellan and Dobson, 1986;Sisk, Haddad and Ehrlich, 1997).In addition, populations that occupy very small areas due to fragmentation are more at risk of local extinction (Lacy, 2000;McCarthy and Thompson, 2006), which is exacerbated in organisms with reduced dispersal ability (Fahrig and Merriam, 1994;Cushman, 2006).A single intensive disturbance could drive an entire population to extinction, or at least significantly reduce population size (Kallimanis et al., 2005).In this scenario, designing conservation programs to protect these populations requires two complementary sources of information: (i) knowledge of threats affecting the population demography; and (ii) knowledge of factors currently shaping species distribution.
In this study, we investigated the effects of fire, habitat structure (habitat scale) and landuse (landscape scale) on the Hermann's tortoise Testudo hermanni.The expansion of urban and agricultural areas (i.e.vineyards) has resulted in a drastic reduction and fragmentation of this species range (Guyot and Clobert, 1997;Bertolero et al., 2011), particularly in the western Mediterranean coastal belt from Italy to the Iberian Peninsula (IUCN, 2013).The Albera Range (NE of the Iberian Peninsula) hosts the westernmost population of T. hermanni, which occupies a reduced area of 134 km 2 (Bertolero, 2007) and is located more than 400 km away from the nearest population in the Massif des Maures (France).Natural dispersal between both populations is therefore impossible.
The Albera Range is prone to summer fires.In 1986, 2000 and 2011, fire events totally or partially burned the range occupied by the tortoise population.Tortoise mortality by fires was relatively low in 1986 (16% of adults and 39% of juveniles; Felix et al., 1989) since the fire affected the area at night when most tortoises were hidden.In contrast, mortality rate was high in 2000 (66%) and 2011 (89%; Vilardell-Bartino et al., 2011a) in the burnt areas.Although the total range of the tortoise population was not apparently diminished, population density decreased from 4.6-10.95tortoises/ha (Felix et al., 1989) to only 0.3 tortoises/ha (Bertolero, 2007).It is currently considered the most threatened population in the whole range of T. hermanni (Bertolero et al., 2011), with other identified threats including illegal collecting, increased number of predators (i.e.wild boars Sus scrofa, European badgers Meles meles, and beech martens Martes foina), and high rates of nest predation (Budó et al., 2002;Bertolero et al., 2011;Capalleras, Budó and Vilardell-Bartino, 2011b).
The aim of our study is to examine how landuse, habitat structure, and fire affect the distribution of T. hermanni within the Albera Range.Specifically, we aimed to investigate what factors at the habitat and landscape scales explain the presence/absence of the Hermann's tortoise, and, to discuss what conservation measures are required to manage, to protect and to recover this endangered population.

Study area
The study was conducted in the lowland of the Albera Range in North Catalonia, at the foothill of the Pyrenees.It is characterized by a typical Mediterranean climate with hot-dry summers and moderate rainfall in spring and autumn.The main natural vegetation is the oak Quercus forest -specifically the cork oak Q. suber and shrubby open landstypical of Mediterranean vegetation.By the beginning of the 20th century, vineyards dominated the landscape, and after the devastating Phylloxera plague, fields were abandoned and partially replaced by woodlands (Poyatos, Latron and Llorens, 2003).At present, the landscape consists of a patchy mosaic of different land-use types, including agricultural land (cereals and vineyards), woodland and pine plantations.The Albera region is partially protected by the Paratge Natural d'Interes Nacional de la Albera and Natura Web 2000 since 1986 and 2006 respectively.This area has a long history of intense fires, with most fires occurring during summer.Information (i.e.size, location and date) of fires is included in this analysis since 1950, with the largest summer fires registered being in 1978, 1986 and 2012.

Study species
The Hermann's tortoise T. hermanni (Gmelin, 1789) is distributed across the Mediterranean basin, from Turkey to Spain (Bertolero et al., 2011), and the main Mediterranean islands.There are two described subspecies: T. h.boettgeri and T. h.hermanni respectively from the eastern and the western part of Europe, with the Po Valley being the geographical limit between the two subspecies.The distribution of T. h.boettgeri is continuous and large, whereas T. h.hermanni has a fragmented and restricted distribution that is a relict of a general and continuous distribution in the past (Bertolero et al., 2011).This species mostly occupies semiopen habitats with low canopy (e.g.open forests, scrublands) that are suitable for thermoregulation, breeding (nesting), and foraging.They avoid marshy areas, dense forests and intensive agricultural areas (Bertolero et al., 2011).
Mean home range of T. h.hermanni is 1 or 2 hectares (Bertolero et al., 2011) and annual adult survival rate is 0.85-0.97 in Spain and France (Fernandez-Chacon et al., 2001).Juveniles have a lower survival rate than adults due to their susceptibility to predators (Hayley and Goutner, 2002).Females produce from 1 to 7 eggs during spring and hatchlings emerge at the end of summer.Predation on eggs (by wild boars, European badgers, and beech martens; Vilardell-Bartino et al., 2009) can be severe, leading to the destruction of 95% of nests in some regions (Swingland and Stubbs, 1985).
Part of this population is located within the Paratge Natural d'Interes Nacional de l'Albera (www.gencat.cat/parcs/albera), a protected reserve managed by the Government of Catalonia (Generalitat de Catalunya).However, the extremely low density of this population suggests that it is much endangered and its future is very uncertain (Bertolero et al., 2011).Since the 1990s there have been several conservation projects whose suggestions have been essentially: (1) manage and protect the few existing natural populations; (2) augment natural populations by reintroducing individuals in natural habitats through a conservation breeding program; and (3) use education to inform the public about the tortoises' plight in general (Bertolero et al., 2011).

Tortoises' survey
We surveyed tortoises by visual encounter in transects at 25 sites with natural vegetation, excluding fields, plantations and other human-modified habitats in order to maximize sightings (Couturier et al., 2014).The 25 transects covered 100 km 2 , representing 75% of the known distribution of this species in the Albera range (fig.1).Transect length averaged 1252 ± 64 m, and were located in Albera lowlands (mean altitude 199 m, and range 83-388).The average distance between transects was 2.3 km.Each transect was systematically sampled over one hour, and visited 5-8 times for a total of 148 hours of sampling.Transects were walked by one researcher during April-June 2012, April-May 2013, and April 2014, when tortoise activity is maximum due to reproduction.Surveys were during sunny and warm days when temperature was higher than 20°C and tortoise activity is at its peak.All individuals sighted were scored (sex, age category, identification), geo-referenced using a GPS, and the perpendicular distance to the transect line recorded.Most tortoises were already marked as part of an ongoing capture-mark-recapture study (Capalleras Budó and Vilardell-Bartino, 2011b); new individuals were marked in order to avoid re-sights of the same individuals in consecutive visits of the same transect.

Ecogeographical variables
At each transect, we collected information on several ecogeographical variables (EGV) that can potentially explain presence/absence of T. hermanni (Couturier et al., 2014).These EGVs included: (1) Fire history.The number of fires and the time since last fire at each transect were obtained from the Prevention Plan for the Fire Forest in the Albera (Agriculture Department in Catalonia).Both fire variables have been identified as important for understanding the impact of fire on Mediterranean reptile communities (Santos and Cheylan, 2013;Smith, Bull and Driscoll, 2013;Santos, Badiane and Matos, 2016) and particularly on T. hermanni (Couturier et al., 2014).
(2) Spatial aggregation (distance).To examine the clustering of transects with and without tortoises, we calculated the distance of each transect to the Sant Quirce de Colera Monastery (X = 4 695 995 and Y = 504 844).The selection of this point was not random: the examination of the first available aerial photograph (taken in 1956) suggests that the area around this monastery was the only point with forested natural vegetation in the early XX century (i.e.cork oak forest, see online supplementary fig.S1) within the current T. hermanni range, the rest being mostly dedicated to agriculture fields.Due to its habitat preferences (Bertolero et al., 2011;Couturier et al., 2014;Vilardell-Bartino, Capalleras and Budó, 2015), we speculate that these forested areas would potentially support a stable population of tortoises 100 years ago.Width and path distance from and between transects center points were also calculated.Path distance tool analyses the least accumulative cost distance to a nearest source (monastery) for each cell, while accounting for surface distance, and horizontal and vertical cost factors.High values represented high accumulative cost distance between transects and the monastery.
(3) Habitat complexity.Plant diversity and habitat structure of transects were characterized by recording abundance of plant species and vegetation height along a 100-m transect within the tortoise transect.Plant species and vegetation height were recorded at points 1 m apart along the 100-m transect; thus, we recorded 100 points that characterized the abundance of plant species and habitat structure at each one of the 25 transects.Plant species included the extent of grass and relative abundance of shrub and tree species.Vegetation height was measured with a tape measure and classified in 1-m height classes.From the matrix of relative abundances of plant species and height classes, we calculated the evenness of each transect (evennessSP and evestruc100 respectively).Evenness refers to how close in proportion classes/species are in a single transect.Evenness quantifies how similar each transect is numerically; for this reason, higher evenness scores indicate more complex (heterogeneous) transects in terms of plant composition and habitat structure.
(4) Land-use heterogeneity.Land-use cover was obtained from the Mapa de Cultivos y Aprovechamientos on the study area between the years of 1979-1980 at 1:50000 resolution.From this spatial layer, we extracted 12 landuse types.We created 1000-m buffers around each transect when calculating each land-use type.From the matrix of relative abundances for each land-use type, we calculated the evenness of each transect (evenness1000).As for habitat complexity, higher evenness scores indicate more complex (heterogeneous) transects in terms of land uses.

Statistical procedures
Measures of species abundance and species site occupancy may be confounded when detectability is low (MacKenzie et al., 2002;Royle, 2004;Royle, Nichols and Kéry, 2005).Some models have been specifically developed to address this problem in occupancy probability estimation.They are based on spatially (several sites) and temporally (several visits) repeated collection of presence/absence data (MacKenzie et al., 2002) or counts (Royle, 2004;Royle, Nichols and Kéry, 2005).In our study, we used siteoccupancy species distribution modelling that accounted for imperfect detection as part of model fitting (Royle and Dorazio, 2008;Kéry, 2010).
Model selection was based on AIC scores (Burnham and Anderson, 2002) of models combining different covariates potentially influencing detection (i.e.cloud cover, observer) and/or occupancy (EGVs) parameters.Most tortoises were detected by the sound produced by the shell rasping as they move.Therefore, only covariates affecting the observer and the animal's activity could possibly affect the detection probability, e.g.meteorological conditions, observer or the hour of the day (Couturier et al., 2014).To overcome this detectability bias, each transect was replicated at different times of day by different trained observers.Fieldwork was performed during the reproductive season, when tortoises are the most active.Models were run with the "unmarked" package v0.8-7 (Fiske and Chandler, 2011) in R software (R Development Core Team, 2007).
In order to reduce the number of variables to be fitted in the models, we performed a PCA on the following variable: Number of fires, Time since last fire, Evestruc100, and EvennessSP.Then, we ran 34 models with the variables cloud cover and observer as predictors for detection, and distance, evenness1000, PC1, PC2, and PC3 as predictors for occupancy.The occupancy covariates have to be sitespecific and cannot vary between visits whereas detection covariates have to be visit-specific, meaning they can differ between each site and visit.We retained the covariates of the best models (i.e.AIC < 2).We also calculated the evidence ratio, the AIC weights and the relative likelihood for each model.In order to assess the relative importance of the covariates of the best models, we calculated the average estimated value obtained.

Results
A total of 52 tortoises were spotted in 12 of 25 transects (naïve occupancy: 48%).The abundance varied from 0 to 15 individuals per transect and the mean number of tortoises encountered per visit was 0.325.
From the PCA (supplementary fig.S2), we obtained 4 principal components (PC) from which we retained the first three: PC1, which explained 48% of the variance, corresponds to the covariates number of fires (toward the negative values) and time since last fire (toward the positive values); PC2, which explained 23% of the variance, mostly captures the habitat complexity with evestruc100 and evennessSP (both toward the negative values); PC3, which explained 23% of the variance, mostly captures the variability of evennessSP (toward the negative values).
From the 34 models, 4 were considered as the best fitted with AIC below 2. From the 5 occupancy predictors examined, all pointed to differences between transects with and without tortoises (table 1).Regarding occupancy, best models included the following covariates: distance to the monastery (distance; mean estimate = −96.02;shorter distances in transects with presence of tortoises), surrounding habitat heterogeneity (evenness1000; mean estimate = −14.57;lower heterogeneity around transects with presence of tortoises), fire history (PC1; mean estimate = −3.82;lower number of fires and/or longer time since last fire in transects with presence of tortoises), and habitat complexity (PC2; mean estimate = −0.332/PC3;mean estimate = −0.23;higher habitat complexity in transects with presence of tortoises).

Discussion
Hermann's tortoises in the Albera Range constitute the last natural population of the Iberian Peninsula, and the most endangered population in the whole species.It is therefore important to collect information about the threats and the factors shaping their current distribution to improve their conservation, and maybe to extrapolate to other endangered populations.By combining different ecogeographical variables, including those related to fire history, habitat complexity and land-use heterogeneity, this study presents the main factors explaining detectability and occupancy of the Hermann's tortoise at the Albera Range.There is substantial evidence that many reptile species (Santos, Badiane and Matos, 2016), and in our case the Hermann's tortoise, are affected by habitat structure, wildfires, and land-use (Couturier et al., 2014).Here we provide proof of the impact of current and historical land-use on the current distribution of the Hermann's tortoise in the Albera Range.Our results are based on a low number of tortoise sights.However, we trust these results to be reliable since our three-years sampling effort was high and covered a significant part of the range.Thus, the low sample size is evidence for extremely low population density, which has been already detected in more extensive field studies (Felix et al., 1989;Bertolero, 2007;Couturier et al., 2014).
Indeed, our results revealed that the most important variable predicting species presence was the distance to the monastery, since most sites with presence of tortoises were aggregated around this point.Historic land-use information at the Albera is key to understanding the spatial aggregation of transects with tortoise's presence: during the early XX century, most pre-Pyrenean lowlands were occupied by Table 1.Occupancy models ranked according to their Akaike's Information Criterion (AIC) showing differences between AIC scores ( AIC), the number of parameters (K), variables fitted for the occupancy parameter, including the distance to the monastery (distance), surrounding habitat heterogeneity (evenness1000), fire history (PC1), and habitat structure (PC2 and PC3), variables fitted for the detection parameter including cloud cover and observer, the evidence ratio, the AIC weight (AICw), the accumulated AIC weight (Acc.AICw), and the relative likelihood.agriculture land (Poyatos, Latron and Llorens, 2003).The aerial photograph from 1956 confirms this pattern on the Albera lowlands as, except the monastery area, most of the surrounding landscape was transformed to agricultural use, mainly vineyards.After the phylloxera plague, vineyards were abandoned and replaced by forest up to the current uses.Our capacity to understand the land-use legacy on the current distribution of species is limited (Foster et al., 2003).However, given the tortoise habitat preferences we suspect that the monastery area could have been a relict population and therefore the starting point for tortoise recolonisation.This would explain why no tortoises were found in unburnt suitable habitats far from the monastery area.We acknowledge that the low population densities could have precluded detecting the presence of some tortoises in negative (absence) sites.These individuals might be residual populations located in the limit of the distribution Albera range (personal communication with X. Capalleras and J. Budó), although molecular markers are necessary to further examine this point.
Our second most important variable, land-use heterogeneity of the surrounding habitat, seems to negatively affect tortoise occupancy and part of this result could be explained by site selection criteria.Although transects were selected in areas with natural vegetation, land-use heterogeneity increased with the presence of agriculture (e.g.vineyards) in the surrounding 1000-m buffer.Thus, large patches of natural vegetation seem to favor tortoise establishment whereas patchy landscapes do not.This conclusion suggests that transformation of natural vegetation to crops can negatively impact tortoises as already suggested for populations in other areas within the distribution range (Popgeorgiev et al., 2014).
Our study suggests that in this region, the presence of T. hermanni is primarily determined by landscape scale variables (i.e.historic and current land use) whereas the effects of fire his-tory and habitat structure only appear to be secondary.This result contrasts with previous studies that identified wildfires as having a strong negative effect on the presence and abundance of Hermann's tortoises (Felix et al., 1989;Couturier et al., 2011Couturier et al., , 2014;;Luiselli et al., 2014).It suggests that in landscapes supporting a long and intense human activity, factors related to land-use history can outshine other ecological factors, which are usually important in other regions.

Hermann's tortoise conservation
Given the critical status of the Hermann's tortoise population in the Albera Range, conservation action is urgently required in order to bolster this population and avoid local extinction.Different strategies have been proposed, developed and applied to mitigate these threats and thereby increase the survival probability of the last natural tortoise population from the Iberian Peninsula (Capalleras, Budó and Vilardell-Bartino, 2011;Vilardell-Bartino, Capalleras and Budó, 2011b).We clearly show that the existence of large patches of natural woodlands is one of the main factors affecting the presence of T. hermanni.Therefore, there is a need to improve habitat management, contributing to the species' conservation by restoring the habitat heterogeneity (i.e.mixture of forested and open areas) within this Mediterranean area.
Being a species with low dispersal capacity, habitat loss and fragmentation as a consequence of fire and other human actions pose a significant threat (Bertolero et al., 2011).Preventing the extension of human activities in the Albera range is necessary to protect Hermann's tortoises.Improving connectivity between the different patches by creating corridors would allow natural populations to colonize new habitats.In addition, introduction of captive individuals seems adequate (Lepeigneul et al., 2014) once there is suitable habitat available without tortoises in the study area.Before considering such conservation programs, it is essential to identify the threats and causes of the species' decline and any management decisions should be supported by proper field studies (Lepeigneul et al., 2014;Vilardell-Bartino, Capalleras and Budó, 2015).
In conclusion, besides confirming the negative impact of repeated wildfires and reduction of habitat complexity on the Hermann's tortoise reported elsewhere in the Mediterranean region (Popgeorgiev, 2008;Bertolero et al., 2011;Couturier et al., 2011Couturier et al., , 2014)), our study reveals the essential role of historic and current land-use in explaining the presence/absence of this species.It would be valuable to consider this factor in the future, especially when starting new breeding and reintroduction programs.Moreover, it should be included in future studies aiming to identify distribution factors, especially for threatened species in regions with a long human history.Centre de Reproducció de Tortugues for their valuable logistic support and transmission of historic information of the herpetofauna, vegetation and economic development of the Albera region.Leonardo Bejarano (Generalitat de Catalunya) supplied GIS shapes of habitat use at the study area, and Núria Nadal (Forestal Catalana, S.A.) supplied information about the fire history within the project « Pla de Prevenció d'Incendis Forestals (PPIF) del massís de l'Albera » granted by Departament d'Agricultura, Ramaderia, Pesca, Alimentació i Medi Natural (Generalitat de Catalunya).We are especially grateful to Thibaut Couturier and Albert Bertolero, who provided us with excellent advice about our statistical procedures.A. Badiane is funded by an IMQRES Ph.D. scholarship.X. Santos was funded by a post-doctoral grant (SFRH/BPD/73176/2010) from Fundação para a Ciência e a Tecnologia (FCT, Portugal).

Figure 1 .
Figure 1.Geographical information about (a) the location of the Albera Range in the Iberian Peninsula, (b) the location and number of fires that occurred within the studied area, and (c) the species observation area (in grey, modified from Budó et al., 2002) where black dots correspond to sampled transects.