Quantifying the conservation value of Sacred Natural Sites

Abstract Many have asserted that Sacred Natural Sites (SNS) play an important role in nature protection but few have assessed their conservation effectiveness for different taxa. We studied sacred groves in Epirus, NW Greece, where a large number of such SNS have been identified. Based on historical, ethnographic and ecological criteria, we selected eight of these groves and matching control sites and in them we studied fungi, lichens, herbaceous plants, woody plants, nematodes, insects, bats and passerine birds. Our results reveal that the contribution of SNS to species conservation is nuanced by taxon, vegetation type and management history. We found that the sacred groves have a small conservation advantage over the corresponding control sites. More specifically, there are more distinct sets of organisms amongst sacred groves than amongst control sites, and overall biodiversity, diversity per taxonomic group, and numbers of species from the European SCI list (Species of Community Interest) are all marginally higher in them. Conservationists regard the often small size of SNS as a factor limiting their conservation value. The sizes of SNS around the globe vary greatly, from a few square meters to millions of hectares. Given that those surveyed by us (ranging from 5 to 116 ha) are at the lower end of this spectrum, the small conservation advantage that we testified becomes important. Our results provide clear evidence that even small-size SNS have considerable conservation relevance; they would contribute most to species conservation if incorporated in networks.

2016). These groves were established through a range of ritual praxes. Some were dedicated 106 to specific saints, some were little more than community agreements, while others were 107 protected by the threat of excommunication. Different management regimes prevailed 108 through time with some groves being strictly protected, some subjected to controlled 109 management, whereas for others only the protection of mature trees is reported. The groves 110 appear either in the form of protective forests above or close to villages or as groups of 111 veteran trees that accompany outlying churches or icon stands (Stewart, 1993;Nixon, 2006;112 See also Appendix G). Nonetheless, they served in many cases as multifunctional forests for 113 local communities providing among others shaded grazing areas for livestock. Especially in 114 deciduous sacred forests, grazing could be intensive (Papanastasis et al., 2008). 115 Different cultural groups coexisted in Epirus contributing to the variability of the landscape, 116 but they were all associated with sacred groves. Long-term ethnographic research has 117 revealed that of the 80 villages in the mountainous municipalities of Zagori and Konitsa 118 almost all had at least one sacred grove; these groves mostly lie within a narrow range of 119 elevation, typically from 800 to 1200 m (Stara et al., 2016). This is also the zone where most 120 mountain settlements, characterized by a mixed system of agriculture-animal husbandry, have 121 developed historically (Nitsiakos, 2016). 122 Even though the role of SNS in the conservation of biodiversity has long been recognized 123 (Kosambi, 1962;Gadgil and Vartak, 1976;Haridasan and Rao, 1985), they have recently 124 gained more attention amongst conservation biologists because of the many threats to 125 biodiversity due to anthropogenic activities (Pimm et al., 1995;Gao et al., 2013). It has been 126 suggested that incorporating these SNS into existing protected area networks might increase 127 their effectiveness in achieving conservation objectives (Bhagwat and Rutte, 2006;Soury et 128 al., 2007;Corrigan et al., 2013;Ormsby, 2013). 129 Despite the increasing interest in SNS as biodiversity refugia (Dudley et al., 2009), few 130 studies have assessed their effectiveness across taxa, whilst most have focused on specific 131 groups of organisms, such as plants (Boraiah et al., 2003;Khumbongmayum et al., 2006;132 Frascaroli et al., 2016), small mammals (Decher, 1997;Reed and Carol, 2004) or butterflies assessment of biodiversity differences as practiced in other similar studies (Wortley et al. 170 2013, Derhé et al. 2016). We selected control sites (1C-8C) in close proximity; these 171 matched each sacred grove in terms of substrate, topographic position and type of vegetation. 172 In this study, we identified three types of groves in terms of vegetation: those dominated by 173 (i) coniferous, (ii) evergreen broadleaved or (iii) deciduous broadleaved trees. We sampled in 174 these eight pairs of sites over two consecutive years (2013 and 2014) following a sampling 175 protocol that was adapted to the unique characteristics of each taxonomic group (Appendix 176 B). The sampling effort was the same across all sites for any given taxonomic group, so that 177 estimates of biodiversity are comparable.

179
In total, eight taxonomic groups (fungi, lichens, herbaceous plants, woody plants, nematodes, 180 insects, bats and passerine birds) were sampled in each sacred grove and the corresponding 181 control site. All observed organisms of these groups were identified to species level, except 182 for nematodes, which were identified to genus level. The data consist of abundance records 183 per species, except for lichens, herbaceous plants (including ferns) and woody plants, for 184 which only species presence was recorded.  186 The biodiversity we assess here is the total number of species recorded in each site, which we 187 call the species richness of the site. combined. This index is widely used as a measure of multidimensional "distance" between 192 samples for abundance data (e.g. Clarke et al., 2007;Birtel et al., 2015;Nicol et al., 2017); it 193 has the advantage, over some other ordination techniques, that differences in abundance are   198 Sacred groves and control sites were compared in terms of their species richness per site 199 (across all taxa), total species richness per taxon (across all sacred and all control sites) and 200 species richness per site per taxon. 201 Apart from their type (sacred or control), sites are characterized by their location within the 202 region of Epirus (Fig. 1), their vegetation (three forest types) and the area of the site (being 203 the area of the convex hull containing the sample plots within each site) (Table 1) 215 Apart from the species richness per site (alpha diversity) and the species richness across sites 216 (gamma diversity), the sacred and control site communities were compared in terms of their 217 beta diversity or species turnover (Magurran, 2004). Beta diversity between the local scale 218 (sites) and the global scale (union of sites) was measured using Whittaker index and N* index 219 (Lazarina et al., 2013). Both indices give a measure of species turnover in space, which in 220 this case measures the difference in species composition between the local scale (site) and 221 global scale (the union of all sacred or all control sites). N* is roughly defined as the 222 sampling effort (number of samples) above which the samples accumulated will mostly 223 contain species that have already been found. The advantage of the N* index, as opposed to 224 other indices, is that it is independent of the sampling effort, provided that there are enough 225 samples for the index to be calculated (Lazarina et al 2013). The N* index was computed 226 using the R function provided by Lazarina et al (2013). We tested the significance of 227 differences between sacred groves and control sites at the 5% level.

228
All statistical tests and analyses were performed in R 3.2.3 (R Core Team, 2015).  230 By the term "conservation capacity" we refer to the ability of a protected area to conserve 231 biodiversity, assuming that management measures to protect the site are implemented.

232
Conservation capacity involves two components: the number of species that an area of a 233 given size can support at equilibrium, based on the species-area relationship (SAR, see for  Thereafter, the viability of each species is dependent on its population size within the 240 fragment so that current species richness may be a relic of earlier biodiversity levels rather 241 than true conservation capacity. The conservation capacity of the sacred groves was 242 estimated for each taxonomic group, separately, using the Arrhenius SAR: The constant z is typically between 0.2 and 0.3 for islands, while for continental areas it falls 245 within the range of 0.1 to 0.15 (Halley et al., 2013). Calibration of the SAR was achieved by 246 assuming a continental area with exponent 0.15; then c was determined by using the number 247 of species found in the control sites through the formula c = S/A z .

248
The first time-constant of relaxation is the expected time for half the extinction debt to be 249 paid off, which actually is the half-life of extinction debt in a habitat remnant. In the absence 250 of speciation and colonization, the half-life of extinction debt is equal to the time for species 251 richness to fall to half its original value. Based on the models developed in Halley et al.

252
(2016), this is approximately (in years): Here, A is the area of the remnant forest, ρ is the typical total density of individuals of the 255 relevant taxonomic group, τ is the average generation time and S 0 is the initial number of 256 species in the area A at the time of area reduction or isolation. The factor ρA/S 0 is important, 257 being the number of individuals per species. If the initial number of species, S 0 , is not known, 258 the alternative is to use the SAR and substitute Eq. (1) for species number: In order to get ρ and τ, we assume a single average for each taxonomic group (Halley et al.,  (Austad and Fischer, 1991). For insects, the value of τ=1 year was 266 typical of the species in our study, while ρ=7.83×10 4 individuals per ha that we used is 267 clearly a conservative number as it refers to ground-dwelling beetles (Didham et al., 1998). 268 We did not compute curves for lichens or fungi owing to known complications of defining 269 individuals and generation times for these groups.  271 To see how the size of the sacred groves that we studied fits into the global picture, using a 272 literature search, we assembled a database of SNS from various countries, for which we could 273 find the area (

276
In total, across all taxonomic groups studied, 816 species were observed and identified within 277 the eight pairs of sacred groves and control sites (Table C.1). There was great variability in 278 the species richness of the sacred sites relative to their respective control sites for different 279 taxonomic groups: in five of them, the total number of species observed was higher in the 280 sacred groves, and in three groups, it was higher in the control sites (Fig. 2a), but these 281 differences were not statistically significant except for fungi (p=0.001, see Table C.2), for 282 which richness was higher in sacred groves. Combining species across the taxonomic groups, 283 all except two localities had higher species richness in the sacred grove than the trees. There is a strong correlation (Fig. 2b) between the species richness of the sacred groves 288 (x) and control sites (y) in each locality for the six pairs dominated by broadleaved trees, 289 reflecting the success of their matching in the sample design (y=0.727x+30.56, R 2 =0.912, 290 p=0.003). For these localities, there is also a significant difference between overall species 291 richness in the sacred groves and control sites (t-test, p=0.0085). These tests show a 292 consistent trend for greater overall species richness in the sacred groves than the control sites.

293
Ordination shows that the patterns of species composition amongst the three vegetation types  Table C.2) that the site area and type do not affect 299 significantly the total species richness per site (at a 5% significance level). However, their 300 interaction is significant meaning that the relationship between species richness and area 301 differs depending on the type of the site (sacred or control). As sacred sites are mostly 302 smaller in area than control sites ( Table 1). The total species richness is also significantly 303 affected by vegetation type. On a taxonomic group level, the locality is not significant for any 304 group. The type of the site (sacred or control) is significant only for fungi, whereas vegetation 305 type is significant for lichens, herbaceous plants, and woody plants; none of these predictors 306 is significant for nematodes, insects, passerine birds or bats. The interaction between site 307 locality and type is also significant for herbaceous plants and lichens, as was also the case for 308 total species richness.

309
Of the 13 European SCI species that were encountered in the study area, more were found in 310 the sacred groves (eleven) than in their control sites (nine) especially for passerine birds (8 311 versus 4). However, overall the difference was not significant (paired t-test; p=0.30).

312
The Whittaker and N* indices of species turnover reveal significantly greater beta diversity 313 amongst the sacred groves than amongst the control sites (at the 5% level for both indices) 314 (Fig. 3). More specifically, beta diversity is greater in the sacred groves for five taxonomic 315 groups (lichens, herbaceous plants, woody plants, passerine birds and bats); it is slightly less 316 for insects, and very similar between the two site types for nematodes and fungi. Notably, 317 beta diversity is much lower for the nematodes than for all the other taxonomic groups of 318 species, presumably because nematodes were identified only to genus level and, hence, the 319 majority of nematode genera are found in all samples.

320
The area of the sacred groves was small, ranging from 4.9 ha to 115.7 ha with a median size 321 of 18.4 ha. Both the area and the taxonomic group are expected to affect the half-life of 322 species loss following habitat isolation ( Fig. 4a) and, hence, their conservation capacity. The 323 predicted half-life varied greatly amongst taxonomic groups being low for bats and passerine 324 birds, under 100 years for most of the sacred groves, but very high, above 1000 years, for 325 nematodes and herbaceous plants (because of their large populations) and for woody plants 326 (because of large generation times)). However, the general linear modelling analysis did not 327 find a significant relationship between area and species richness.

328
In our literature search, we found 104 SNS for which the area was recorded or could easily be 329 inferred; these occur in all inhabited continents. To these we added the 22 sacred groves in 330 Epirus that we mapped, including the 8 whose biodiversity we studied in detail. The 331 histogram for this ensemble (Fig 4a) shows that the size of SNS varies greatly, ranging from 332 a few square metres to over 100,000 km 2 , with the groves that we studied falling in the 333 smaller part of the range. By contrast, National Parks are always at least 10 km 2 (Fig. 4b).

336
Globally, this is the first study to evaluate the conservation capacity of SNS by use of a large 337 and taxonomically broad set of species. Regarding Hypothesis (I), our study shows that while 338 sacred groves contained more species overall, the difference between them and control sites 339 was not statistically significant unless the north-facing conifer sites were omitted from the 340 analysis. Similar statistical issues have arisen in a previous study comparing protected and 341 unprotected areas for several taxonomic groups (Gray et al., 2016), despite the expected 342 differences between such areas. These results suggest that the advantage of protected over 343 unprotected areas becomes blurred when more than one taxonomic group is examined 344 (Khumbongmayum et al., 2005;Gao et al., 2013). To avoid the bias of masking differences 345 when pooling together data from different taxonomic groups, in the present study, 346 biodiversity was assessed for each group separately. While species richness was higher for 347 most groups in sacred groves, only for fungi was this difference significant. This lends 348 support to Hypothesis (II), except that for lichens, the other taxon that should benefit from the 349 presence of older trees, the differences were not significant. For plants, this lack of strong 350 distinction contrasts with an earlier study (Frascaroli et al., 2016) reporting significantly more 351 species in sacred groves than in reference sites. In contrast to the nuanced difference in species richness between sacred groves and control sites, there was a clear biodiversity 353 benefit when beta diversity was considered (Hypothesis III). Its higher value for sacred 354 groves suggests that there is a greater distinction (in the sets of species) between sacred 355 groves than between control sites. This might be explained by the groves different histories of 356 usage, which have a significant effect on sacred grove's vegetation structure and therefore on 357 the ecological community structure, thus increasing the dissimilarities between groves.

358
Different patterns of land abandonment could also play a role. By contrast, the non-sacred 359 control areas arose largely through natural regeneration in the last 100 years and thus have a 360 more uniform structure.

361
Given the lack of evidence of a strong difference in species richness or composition between 362 sacred groves and control sites, other factors were explored to explain the results found. The

398
Because sacred groves along the mountainsides of Epirus were established for their benefits 399 in terms of cultural and religious beliefs, hill-slope protection, recreation or even scenery 400 (visual amenity), rather than for biodiversity conservation per se, they can be described as 401 suffering from a kind of "rocks and ice syndrome" (Terborgh, 1999). Biodiversity 402 conservation was not the priority in delimiting these areas; this has emerged as a secondary 403 benefit. For that reason, the sites chosen for sacred status were not selected according to 404 conservation criteria. This is especially the case with respect to their size. Size is a major 405 factor limiting conservation capacity (Halpern, 2003;Ramesh et al., 2016), both with respect 406 to the number of species that can be supported in the long-term and in the length of time an 407 extinction debt can be sustained following isolation (Fig. 4). However, people establishing 408 sacred groves might settle for much smaller areas than are necessary in conservation terms, as   However, as the actual sampling area (given any taxonomic group) is the same in each site 413 we expect this to increase only weakly with site area (Phillips et al., 2017). Furthermore, we 414 should not think of these groves as islands of forest in a landscape of cultivation. The groves 415 have always existed in a matrix of habitable or partially-habitable landscape, so for this 416 reason also, it is not so surprising that measurements of diversity fail to show the limiting 417 effect of size expected from Eq. 1. Finally, consistent with historical and photographic 418 evidence, the area of groves is not constant. Most have expanded since 1945 while some were 419 not isolated even in 1945. Also, the variability of areas is not so great (Fig 4a), so that area 420 dependence is not easily detectable if statistical power is low. Thus, while Eqs (1-3), based on 421 isolated fixed-area island models, can illuminate our understanding of conservation capacity 422 and relaxation time, they must be used in conjunction with historical and landscape 423 information when their basic assumptions are not met.

424
These results show a conservation benefit of SNS, which is variable amongst taxa and is          751 We identified sacred grove sites across the landscape based on archival and ethnographic 752 fieldwork. We further identified and mapped the borderline of these groves using ortho- Since our main hypotheses concern biodiversity, we define a control site for each sacred 774 grove so as to assess the biodiversity difference relative to a non-sacred, reference forest.  Four sampling points were chosen at each site (sacred and control) and at each a plot of 100 810 m 2 was established. In each plot, a composite soil sample of five soil cores, 3 cm in diameter 811 and 12 cm in depth, was collected, so that four composite samples were taken from each site.

812
In all cases, the litter layer was removed before sampling. Nematodes were extracted from 813 200 cm 3 of each composite soil sample. For extraction, the modified Cobb's sieving and 814 decanting method (S'Jacob and van Bezooijen, 1984) was employed. After counting total 815 abundance of nematodes, samples were fixed with 4% formaldehyde solution. From each 816 sample, 150 nematodes were selected and identified to the genus level using an identification 817 key (Bongers, 1994). In cases where the number of specimens of a sample was less than 150, 818 we identified them all.   (Clauzade and Roux, 1985;Nimis, 1987;Purvis et al., 1992;Wirth, 1995).   Table   892 C.2. The significance of each predictor variable is judged on a 5% significance level.  To visualize the difference in composition between sites, multidimensional scaling analysis 909 based on Bray-Curtis dissimilarity was conducted for each taxon, separately, and for all taxa    forest name]). Those whose biodiversity we surveyed also are in bold.