ECOGRAPHY 24: 659670.

Copenhagen 2001

Methods for comparison of biotic regionalizations: the case of pteridophytes in the Iberian Peninsula
Ana L. Marquez, Raimundo Real and J. Mario Vargas

Marquez, A. L., Real, R. and Vargas, J. M. 2001. Methods for comparison of biotic regionalizations: the case of pteridophytes in the Iberian Peninsula. Ecography 24: 659670. We made several regionalizations of the Iberian Peninsula based on the distributions of the pteridophyte ora to see whether the regionalization depended on the type and scale of lattice or the similarity index considered. We used ve types of lattice in which the scale also varied: river basins, mountains and plains, natural regions, physiographic and geological regions, and administrative provinces; and two similarity indices: those of Jaccard and of Baroni-Urbani and Buser. The regionalizations varied according to the type of lattice, the grain size, and the similarity index used. To assess the different regionalizations we used four methods: 1) the coefcient of variation of the size of sites in each lattice, 2) the bestblock method, which considers as the best lattice that which maximizes the number of matches between presences over all pairwise site comparisons, 3) the Mantel test, to assess the statistical signicance of the regionalizations obtained, and 4) mapability, which considers the most contiguous regionalization to be the best. The best regionalization according to our four criteria was that based on administrative provinces and Jaccards index. This yielded a small central region and three large regions: northern, western, and eastern. A. L. Marquez (rrgimenez@uma.es), R. Real and J. M. Vargas, Dept de Biologa Animal, Fac. de Ciencias, Uni6. de Malaga, E-29071 Malaga, Spain.

A major objective of descriptive biogeography is to simplify the complex patterns of contemporary species distributions by classifying areas based on their biotic composition. That areas of the world can be characterized on the basis of the presence of certain taxa and the absence of others has long been recognised by biogeographers (Buffon 1761), and is a deeply entrenched component of much contemporary biogeographical thinking. In addition, classifying geographical areas into groups with different species composition is valuable for nature conservation planning (Brown et al. 1993), so as to ensure that all groups are represented in the selection of natural reserves (Margules 1986), and to evaluate the biological resources of an area in a regional and global context (Carey et al. 1995). However, currently administrators are often confused by the number of different classications of their country proposed by ecologists and phytogeographers. In addition,
Accepted 6 March 2001 Copyright ECOGRAPHY 2001 ISSN 0906-7590 Printed in Ireland all rights reserved
ECOGRAPHY 24:6 (2001)

as Thaler and Plowright (1973) pointed out, phytogeographers have disagreed about whether oristic areas represent real entities in nature or whether they are simply convenient, subjective constructs that lack any objective and unambiguous basis. Operationally, it is possible to identify biotic boundaries when a group of areas with similar biota shares fewer species than expected at random with other group of areas with similar biota. Then the Operational Geographic Units (OGUs; Crovello 1981) of this territory can be grouped into biotic regions (Real et al. 1992a, Myklestad and Birks 1993, McLaughlin 1994). Clustering methods represent one approach to this problem (Birks 1987, Legendre 1990), but only when a probabilistic procedure allows the researcher to distinguish between statistically signicant clusters and those indistinguishable from random expectation (Marquez et al. 1997). Area clusters may be

659

considered as biotic regions when a group of OGUs is statistically signicant. Legendre (1987) advocated constraining the clustering to respect spatial contiguities. However, in some cases, biotic regions may reect a climatic or ecological pattern that may be not strictly contiguous in space, and then they may be better detected with clusters only dened by the similarity of their species composition. Then, area clusters can be related to environmental variables by techniques such as regression or canonical variates analysis (Carey et al. 1995, Olivero et al. 1998) to investigate possible causes of the regionalization pattern. However, to analyse biogeographic patterns, any territory must be subdivided initially into different types of OGUs (Crovello 1981, Kenkel et al. 1989, Mandrak 1995). This establishes a lattice that inuences the biogeographic regions subsequently obtained, because any biotic boundary necessarily has to follow along the edges of the sites used to partition the study area. Cressie (1993) dened a lattice as a countable collection of spatial sites that are spatially regular or irregular, represent points or regions, and are indices for continuous or discrete variables. In biogeographic regionalization the concept of lattice corresponds to any kind of area subdivision, irrespective of its size and the criteria used to dene it, and the sites correspond to the OGUs. The lattices can be natural (e.g., watersheds, Vargas et al. 1998), arbitrary (e.g., latitude-longitude cells, Schall and Pianka 1977, or equal-area quadrats, Marquez et al. 1997). The sites can have the same size when the lattice is regular, such as those based on grid-squares, or different sizes when the lattice is irregular, such as those made according to, for example, physical, geological, or political viewpoints. Variation in size and shape may affect the species richness of each site and hence the results of the subsequent regionalization. Square equal-area quadrats are considered not to be subject to these problems as quadrat size and shape are uniform (Mandrak 1995), although coastal quadrats may vary in area when terrestrial organisms are studied, and mountainous areas have more surface area than at areas in a equal-area quadrat. There are also problems in dening the scale of the grain used (Jardine 1972, Birks 1987, Carey et al. 1995). Scale is dependent on both the extent of the study area and the size of organism studied (Wiens 1989). If too coarse a lattice is used, biotic regions will be correspondingly coarse. Conversely, the use of too ne a lattice may fail to reveal major biotic regions and generate many spurious small regions where recordings may have been uneven. Results obtained using small sites may be different from those obtained using larger sites, although these different patterns might be interrelated (Blondel 1985, Vargas 1993). However, this does not mean that there are no criteria for comparing the results obtained using different types of sites or different scales (Levin 1992). These criteria, however, need proper denition and testing. 660

Ferns could provide a valuable source of biogeographical information that is distinct from, or at least complementary to, that provided by other types of plant (Marquez et al. 1997). Pteridophytes are a group of species with a great dispersal capacity due to the lightness of their spores (Jermy 1984). However, most fern species are restricted in distribution and only a limited number have wide ranges (Kato and Iwatsuki 1983), because their distribution is highly related to climate and substratum (Barrington 1993, Given 1993, Kornas 1993). The 113 Iberian pteridophyte species are well recorded and well understood taxonomically in the Iberian Peninsula (Salvo et al. 1984, Castroviejo et al. 1986), and are an important number of species in a European context. In this paper we have used several criteria to select the best biotic regionalization among several partitions obtained using a quantitative method of classication. With this aim, we have used fern distributions in the Iberian Peninsula a) to establish whether the Iberian Peninsula may be divided into pteridogeographic regions or whether the pteridophyte species distributions follow a mainly gradual pattern; b) to verify whether the resulting regionalization patterns are different according to the type of lattice, the number of sites considered, and the similarity index used; and c) to apply several criteria to select the best regionalization pattern.

Material and methods

Study area
The Iberian Peninsula is situated on the south-western end of Europe, and presents a stronger peninsular condition than other large Mediterranean peninsulas because its isthmus is relatively narrow. In addition, the presence of the Pyrenees in the isthmus hinders the spread of propagules between the Iberian Peninsula and the rest of the European continent. The interchange of biota with Africa is also hindered by the Straits of Gibraltar. Climatically, the Iberian Peninsula is placed in the southern band of the temperate zone.

Species distribution and Operational Geographic Units (OGUs)

The distribution of the 113 species of pteridophytes were obtained from Salvo et al. (1984), supplemented with new records published later. We considered ve types of OGUs in the Iberian Peninsula: river basins, mountains and plains, natural regions, physiographic and geological regions, and administrative provinces (Anon. 1980) (Fig. 1). The province lattice is political-administrative, although is
ECOGRAPHY 24:6 (2001)

partially based on natural partitions, whereas the other four lattices have natural borders. Natural regions and physiographic and geological regions are partially based on vegetation, and so some circular reasoning may be involved in their use as phytogeographic regions (Birks 1987, Real and Ramrez 1992, Vargas 1993).

Classication analysis
Initially we constructed a presence-absence matrix for the 113 species in each lattice considered in the area. We thus had ve presence-absence matrices, in which OGU size and number varied: river basins (11113),

mountains and plains (12113), natural regions (17 113), physiographical regions (22 113), and administrative provinces (56 113). We applied two similarity indices to each pair of OGUs: Jaccards (1908) index and Baroni-Urbani and Busers (1976). The Jaccard index considers only double presences, whereas the Baroni-Urbani and Buser index incorporates double absences. Jaccard values may be considered as absolute measures of biotic similarity, because they are not affected by other species absent from the two OGUs compared, while the Baroni-Urbani and Buser coefcient measures similarity in the context of the Iberian Peninsula, because species absent from the two OGUs but present in other Iberian OGUs increase the similarity between the two OGUs analysed

Fig. 1. The ve different lattices considered in the study (Anon. 1980).

ECOGRAPHY 24:6 (2001)

661

(Real et al. 1992b). However, the possibility that two OGUs are considered similar exclusively because of double absences is avoided in the Baroni-Urbani and Buser index by multiplying double absences by double presences. We used UPGMA (Unweighted Pair-Group Method using Arithmetic Averages) as our clustering method (Sneath and Sokal 1973). Agglomerative techniques are preferred to divisive techniques because the former maximizes the similarity within a group, which denes biotic regions. We represented the results as dendrograms. To test the groups obtained in any association analysis several authors use a null model based on a Monte Carlo randomization procedure (Connor and Simberloff 1979, Oden et al. 1993). However, this model uses the observed data matrix to generate a null distribution, and so observed and null distributions lack statistical independence (Grant and Abbot 1980). Notwithstanding, it is feasible to determine all the possible distributions of N species in any of the OGUs compared for each value of N, and so an exact randomization test (Sokal and Rohlf 1981, p. 788) can be performed to determine whether an observed value of the Jaccard index or the Baroni-Urbani and Buser index is signicantly different from those expected at random. We obtained the statistical signicance of the observed similarity values, according to the exact randomization tests in Real and Vargas (1996) for the Jaccard index, and in Baroni-Urbani and Buser (1976) for the Baroni-Urbani and Buser index. In this way, the similarity matrices were transformed into matrices of statistical signicance (Marquez et al. 1997) that consisted of pluses, for similarity values signicantly higher than those expected at random, minuses, for similarity values signicantly lower than those expected at random, and zeros, for similarity values compatible with random expectation. Our null hypothesis is that biotic boundaries do not exist, so that we started searching for boundaries at the level of the dendrogram with the lowest similarity value, and then moved to the higher similarity nodes, nishing when no more signicant boundaries appeared. This divisive procedure does not contradict the use of an agglomerative classication, because signicant boundaries divide the space into biotic regions that require not only the existence of a biotic boundary, but also a biotic afnity between OGUs in the region, which is favoured by the agglomerative method. We tested at every node of the dendrogram for the existence of signicant biotic boundaries between the OGUs separated by the node using the method of McCoy et al. (1986), as modied by Real et al. (1992b) for application to dendrograms, as indicated in Appendix 1. 662

To determine which lattice and similarity index best dene the biotic regions we used four methods: 1) Coefcient of variation of the OGUs size in each lattice (Sokal and Rohlf 1981). Differences in the size of the OGUs within a lattice produce area-related distortions, such as differences in the number of species in different parts of the study area. So, the lattice with the lowest coefcient of variation of the OGUs size is considered the best one. This method selects the best lattice regardless of the distribution data. 2) Bestblock (Phipps 1975), which considers the best lattice to be that which maximizes the number of matches between presences over all pairwise OGU comparisons. This method selects the best lattice after taking into account the distribution data regardless of the regionalization obtained. 3) We used the Mantel test of matrix comparison (Mantel 1967) to compare each original similarity matrix with each matrix of hypothetical similarities according to the resulting regionalizations. The matrices of hypothetical similarities were obtained by considering all the similarities within a biotic region to be 1, all the similarities between OGUs separated by weak boundaries to be the random expectation (0.33 for Jaccards index and 0.5 for Baroni-Urbani and Busers index), and all the similarities between OGUs separated by strong boundaries to be 0. Only signicant regionalizations were considered. The most signicant regionalization is that which is most consistent with the original similarities and is thus considered the best. 4) Mapability (Harper 1978). Contiguity is preferred to discontiguity of the biotic regions obtained using each similarity index in each lattice. We measured the mapability in terms of the minimum number of quadrats UTM 50 50 km necessary to join together the parts of all fragmented regions. The lower this number of quadrats is, the more mapable the regionalization is. We applied the two former criteria to the ve lattices and the two latter criteria to the ten regionalizations resulting from using each similarity index in each lattice.

Results
We found signicant biotic boundaries for all types of OGU using both similarity indices. All the regionalization patterns obtained using the Jaccard index (Table 1) and most of the regionalizations obtained using Baroni-Urbani and Busers index (Table 2) were statistically signicant according to the Mantel test of matrix comparison (Table 3). Only the regionalization obtained using mountains and plains and the BaroniUrbani and Buser index was found to be non-signicant (p\0.01) using the Mantel test.
ECOGRAPHY 24:6 (2001)

Table 1. Statistically signicant boundaries between groups of OGUs formed by UPGMA using the Jaccard similarity index. DW \0 and signicant GW indicate weak boundary between the groups; DS\0 and signicant GS indicate a strong boundary between the groups. N.S.: p\0.05; *: pB0.05; **: pB0.01; ***: pB0.005. Groups set up by UPGMA Weak Group A River basins Mountains and plains 35 29 25 Natural regions 213 Physiographical regions 72 717 716 Administrative 1654 provinces 1640 1424 4943 1412 149 2333 1415 3736 2029 Group B 98 4 109 49 1314 42 2017 27 1454 4954 3754 2324 12 2110 209 4254 29 DW (AA) 0.206 0.562 0.153 0.219 0.372 0.239 0.201 0.189 0.439 0.387 0.522 0.254 0.320 0.407 0.202 0.126 0.080 DW (BB) 0.168 0.627 0.153 0.209 0.571 0.280 0.222 0.707 0.356 0.348 0.592 0.191 0.392 0.389 0.162 0.126 0.080 DW 0.187 0.595 0.153 0.214 0.472 0.259 0.211 0.448 0.397 0.367 0.557 0.222 0.356 0.398 0.182 0.126 0.080 GW 9.074 20.249 8.573 24.756 28.758 28.422 16.107 27.311 208.640 176.987 45.891 38.099 8.727 13.508 6.812 6.328 5.555 p *** *** *** *** *** *** *** *** *** *** *** *** *** *** ** * * DS 0.501 0.079 0.555 0.457 0.099 0.349 0.486 0.183 0.011 0.049 0.033 0.351 0.316 0.300 0.505 0.581 0.627 Biotic boundary Strong GS 0 0 0 0 0.781 0.606 0 19.153 80.326 18.986 0 1.365 0 0 0 0 0 p N.S N.S. N.S. N.S. N.S. N.S. N.S. *** *** *** N.S. N.S. N.S. N.S. N.S. N.S. N.S.

Table 2. Statistical signicant boundaries between groups of OGUs formed by UPGMA using the Baroni-Urbani and Buser similarity index. DW\ 0 and signicant GW indicate weak boundary between the groups; DS\ 0 and signicant GS indicate a strong boundary between the groups. N.S.: p\ 0.05; *: pB0.05; **: pB0.01; ***: pB0.005. Groups set up by UPGMA Weak Group A River basins Mountains and plains Natural regions Physiographical regions Administrative provinces 35 24 614 714 1612 1640 4954 2024 1656 Group B 98 15 511 42 27 4912 2012 12 1840 DW (AA) 0.116 0.307 0.114 0.140 0.513 0.193 0.096 0.099 0.062 DW (BB) 0.116 0.307 0.114 0.147 0.008 0.091 0.099 0.072 DW 0.116 0.307 0.114 0.144 0.575 0.101 0.093 0.099 0.067 GW 6.98 24.015 22.100 43.864 94.044 161.909 89.621 9.930 9.620 p ** *** *** *** *** *** *** *** *** DS 0.591 0.400 0.593 0.551 0.089 0.456 0.611 0.608 0.635 Biotic boundary Strong GS 0 0 0 0.007 28.972 23.214 0 0 0 p N.S N.S N.S. N.S. N.S. *** N.S. N.S. N.S.

Table 3. Results of the application of the four criteria used to determine which is the best regionalization. CV: Coefcient of variation of the OGUs size; TM: Total matches between presences in all pair-wise comparisons; SQ: Number of squares 5050 km necessary to join together each fragmented region. OGU Systems Coefcient Best of variation block Jaccard CV River basins Mountains and plains Natural regions Physiographical regions Administrative provinces 0.481 0.445 0.360 0.447 0.359 TM z p(z obsBz exp) 0.002 0.001 0.001 0.001 0.001 Mantel test Baroni-Urbani and Buser z 5.33 3.91 12.36 20.06 212.9 p(z obsBz exp) 0.002 0.014 0.001 0.001 0.001 Mapability Jaccard Baroni-Urbani and Buser SQ 0 0 0 1 0 SQ 0 0 0 0 3

20 983 10.17 21 754 13.72 26 885 24.92 28 977 60.75 47 881 544.9

ECOGRAPHY 24:6 (2001)

663

Area classication using the Jaccard similarity index

Figure 2 shows the biotic boundaries found using each lattice and the Jaccard similarity index. The lattices of river basins and natural regions present only a weak biotic boundary that divides the Iberian Peninsula into a northern and a southern region, whereas two or more boundaries are obtained using the other lattices (Fig. 2, Table 1). In the administrative provinces lattice, three strong boundaries divide the Iberian Peninsula into four supraregions: northern supraregion, western supraregion, eastern supraregion, and a small central supraregion. The western and eastern supraregions are divided by weak biotic boundaries into four and three regions, respectively (Fig. 2).

Which lattice denes best the biotic regions?

The administrative provinces have the lowest coefcient of variation for the size of the sites (Table 3). That means that using the provinces the Iberian territory is more evenly distributed than using the other lattices. According to the bestblock method, the administrative provinces lattice maximizes the number of positive matches over all pairwise species comparisons (Table 3). The signicance values associated with the Mantel text are lower for river basins and nonsignicant for mountains and plains when using the Baroni-Urbani and Busers index, whereas the other regionalizations all have the same high probability value (Table 3). The regions obtained from physiographical regions using the Jaccard index and from administrative provinces using the Baroni-Urbani and Buser index are fragmented (Table 3, Figs 2 and 3), whereas the other lattices yield continuous regions. Therefore, the only regionalization that obtains the best scores according to the four criteria is the one obtained using administrative provinces and the Jaccard index (Fig. 2).

Area classication using the Baroni-Urbani and Buser similarity index

Figure 3 shows the biotic boundaries found using each lattice and the Baroni-Urbani and Buser index. In four lattices (river basins, mountains and plains, natural regions, and physiographical regions) there are only weak biotic boundaries (Table 2) that divide the Iberian Peninsula into northern and southern regions, although using mountains and plains the regionalization was not statistically signicant (Table 3). However, using the provinces there are three big regions (northern, western, and eastern) and three small regions (Fig. 3).

Discussion
Although there is no absolute regionalization of the Iberian Peninsula according to the distribution of pteridophytes, the concept of the biotic region is not an useless artefact. Regionalizations vary according to the type of OGU, the grain size, and the similarity index used. In this way, any regionalization must indicate

Fig. 2.

664

ECOGRAPHY 24:6 (2001)

Fig. 2. Dendrograms and biotic regions for pteridophytes in the Iberian Peninsula using the Jaccard similarity index on each lattice considered. Numbers of the units as in Fig. 1. W: weak boundary; S: strong boundary; *: pB 0.05; **: p B 0.01; ***: pB 0.005.

those characteristics, which are inseparable from the pattern obtained. This is especially true given the exploratory nature of current approaches to biotic regionalization (Carey et al. 1995). However, the biogeographic regions and boundaries are useful for understanding the mechanisms underlying the distribution of organisms, provided that each regionalization reects a distribution pattern that is statistically signicant and may be related to environmental conditions,
ECOGRAPHY 24:6 (2001)

although detected through the lters of the OGUs, the scale of grain, and the similarity index.

The indices of similarity

Oden et al. (1993) modied the method of Womble (1951) to detect intervals of marked change in categorical variables between contiguous sites. When the vari665

able analysed is the presence or absence of species this method could be used to obtain biotic boundaries and regions. However, the basic statistic that results from the method of Oden et al. (1993) is the complement of the simple matching coefcient, which is not recommended for biogeographical regionalizations, since regions may be dened only on the basis of shared absences, even with no single species in common (Baroni-Urbani and Buser 1976). We opted for using the Jaccard and Baroni-Urbani and Buser similarity indices to estimate the associations between the OGUs and, later, to test the statistical signicance of such associations, because they are commonly used and are somewhat complementary. The Jaccard index only takes into account the double presences of species in the two OGUs compared, and the similarity between two OGUs is not affected by other species present in other OGUs in the study area. The Baroni-Urbani and Buser index takes into account double absences, which offers potentially useful information because if a species is absent from two OGUs, it may be due to ecological or historical factors impacting in a similar way on the species in the context of the area studied. If the pattern obtained considering the similarities between each pair of OGUs only (using the Jaccard index) is maintained when the similarities take into account all the study area (using the Baroni-Urbani and Buser index), then the lattice used may be considered suitable for revealing a consistent distribution pattern of the species. The regionalizations obtained using the two indices are identical when based on river basins and on natural regions, partially similar using physiographical regions

and provinces, and very different using mountains and plains. The differences seem to be more related to the type of OGU than to the scale of grain. It may be that when using river basins and natural regions the distribution of pteridophytes is very well dened, because the double absences do not modify the overall relationship pattern obtained using only the presences, whereas mountains and plains are very unsuitable units for recording the distribution of pteridophytes, so that double absences greatly alter the relationship pattern between the OGUs, with physiographical regions and provinces in an intermediate position. The number of boundaries obtained using the Baroni-Urbani and Buser coefcient tends to be lower than that obtained using the Jaccard index, probably because the double absences considered in the BaroniUrbani and Buser index tend to homogenize the regions. This may be seen especially in the south, that has a lower number of species than in the north. Perhaps this tendency should be taken into account when a researcher decides which index to use to reveal a regionalization pattern, according to the underlying knowledge and assumptions.

The criteria to select the best regionalization

Even when using an numerical method of classication and a probabilistic procedure to assess the groups of OGUs obtained, several regionalizations may result from the use of different types of OGUs or different similarity indices, so that several criteria are needed to compare the suitability of the regionalizations obtained.

Fig. 3.

666

ECOGRAPHY 24:6 (2001)

Fig. 3. Dendrograms and biotic regions for pteridophytes in the Iberian Peninsula using the Baroni-Urbani and Buser similarity index on each lattice considered. Numbers of the units as in Fig. 1. W: weak boundary; S: strong boundary; *: p B0.05; **: pB 0.01; ***: pB0.005.

We used two criteria prior to the classication procedure, and two criteria after the classication. The coefcient of variation (Sokal and Rohlf 1981) indicated that the size of each OGU in the provinces is more similar than in the other lattices and is thus to be preferred independently of the distribution data. The bestblock method (Phipps 1975) showed that administrative provinces provide the best lattice for establishing the biotic boundaries of the Iberian Peninsula according to its pteridoora (Table 3), because the ner scale of the partition produced more units, and because
ECOGRAPHY 24:6 (2001)

the presences are more equitably distributed through the OGUs, thus producing more matches of presences in pairwise comparisons. The regionalization obtained using provinces and the Jaccard index was one of ve regionalizations that were suitable according to both the Mantel test and the mapability criterion, because it had the highest statistical signicance and was totally mapable. Mantel and mapability tests were applied to the nal regionalization patterns, and showed whether the patterns were in accordance with the original similarities between OGUs and with the within667

region continuity recommendable for a biotic region, respectively. The higher number of units in the provinces lattice is not the only cause of this partitioning being selected. These four criteria are not biased towards partitionings with a large number of units nor more isodiametricallyshaped units. The coefcient of variation selects a partitioning of more equal-sized units, but not more circular in shape, and irrespective of the number of units. The Phipps algorithm not always chooses the partitioning with the greatest number of units (see Phipps 1975). Although the provinces lattice has more units it is not more likely to produce statistically signicant results after using Mantels test, and, in fact, the partitioning obtained using provinces has the same high signicance level than other partitionings. As regards the criterion of mapability, a partitioning based on more units is more likely to produce fragmented regions, although this is not the case for provinces when using Jaccards index.

human inuence on natural distributions, because in most countries human activities follow political-administrative borders.
Acknowledgements We thank H. J. B. Birks, S. P. McLauglhin and P. L. Nimis for revisions of the manuscript and their valuable suggestions. We are also grateful to M. A. Rendon, J. C. Guerrero and J. Olivero for their help with the statistical procedures.

References
Anon. 1980. Divisiones territoriales de Espana. CEOMAT, Madrid. Baroni-Urbani, C. and Buser, M. W. 1976. Similarity of binary data. Syst. Zool. 25: 251 259. Baroni-Urbani, C. and Collingwood, C. A. 1976. A numerical analysis of the distribution of British Formicidae (Hymenoptera, Aculeata). Verhandl. Naturf. Ges. Basel 85: 51 91. Barrington, D. 1993. Ecological and historical factors in fern biogeography. J. Biogeogr. 20: 275 280. Birks, H. J. B. 1987. Recent methodological development in quantitative descriptive biogeography. Ann. Zool. Fenn. 24: 165 178. Blondel, J. 1985. Biogeographie. Approche ecologique et evo lutive. Masson. Brown, A., Horseld, D. and Thompson, D. B. A. 1993. A new biogeographical classication of the Scottish Uplands. I. Descriptions of vegetation blocks and their spatial variation. J. Ecol. 81: 207 230. Buffon, G. L. L. 1761. Histoire Naturelle, Generale et Parti culiere. Vol. 9. Imprimerie Royale, Paris. Carey, P. D. et al. 1995. An environmentally dened biogeographical zonation of Scotland designed to reect species distributions. J. Ecol. 83: 833 845. Castroviejo, S. et al. 1986. Flora Iberica. Plantas vasculares de la Pennsula Iberica y Baleares. Serv. Publ. C.S.I.C., Madrid, pp. 2 158. Connor, E. F. and Simberloff, D. 1979. The assembly of species communities: chance or competition? Ecology 601: 1132 1140. Cressie, N. A. C. 1993. Statistics for spatial data. John Wiley. Crovello, T. J. 1981. Quantitative biogeography: an overview. Taxon 30: 563 575. Given, D. R. 1993. Changing aspects of endemism and endangerment in Pteridophyta. J. Biogeogr. 20: 293 302. Grant, P. R. and Abbot, I. 1980. Interspecic competition, island biogeography and null hypotheses. Evolution 34: 332 341. Harper, C. W. Jr 1978. Groupings by locality in community ecology and paleoecology: tests of signicance. Lethaia 11: 251 257. Jaccard, P. 1908. Nouvelles recherches sur la distribution orale. Bull. Soc. Vaud. Sci. Nat. 44: 223 270. Jardine, N. 1972. Computational methods in the study of plant distributions. In: Valentine, D. H. (ed.), Taxonomy, phytogeography and evolution. Academic Press, pp. 381 393. Jermy, A. C. 1984. Origin and distribution of Pteridophytes in the Mediterranean area. Webbia 38: 397 416. Kato, M. and Iwatsuki, K. 1983. Phytogeographic relationships of pteridophytes between temperate North America and Japan. Ann. Miss. Bot. Gard. 70: 724 733. Kenkel, N. C., Juhasz-Nagy, P. and Podani, J. 1989. On sampling procedures in population and community ecology. Vegetatio 83: 195 207. Kornas, J. 1993. The signicance of historical factors and ecological preference in the distribution of African pteriophytes. J. Biogeogr. 20: 281 286.
ECOGRAPHY 24:6 (2001)

Administrative provinces as operational geographic units

In the Iberian Peninsula, provinces provide the lattice that yields the most appropriate regionalization according to our four statistical criteria when using Jaccards index, and is thus the regionalization to be preferred in attempting to identify the biogeographical processes affecting the distribution of pteridophytes. One reason for this may be that provinces are administrative units of ancient origin and are not completely arbitrary. This is likely to be true elsewhere, at least in Europe, where administrative units are partially based on natural partitions perceived by local people, as is the case, for example, with vice-counties in the British Isles (BaroniUrbani and Collingwood 1976). Different types of arbitrary areas within a study region can be used, such as countries, counties, provinces, river basins, latitude/longitude blocks, or equal size U.T.M. squares (Birks 1987). In the last decade regular geographical units, like networks of regular squares, have been commonly used (e.g. Martn and Gurrea 1990, Lausini and Nimis 1991, Myklestad and Birks 1993, Carey et al. 1995, Moreno-Saiz et al. 1998). However, regular squares have articial borders and, therefore, may include several fragments of different geographic or environmental units, so hindering the search for biogeographical processes that may only act on natural units (Palomo and Antunez 1992). The division in provinces is also arbitrary, but the administrative borders between provinces follow, in part, natural boundaries such as geographical ranges, so that they may be more appropriate to assess the environmental explanations for species distributions. In addition, provinces may be a suitable type of OGU for analysing 668

Lausini, D. and Nimis, P. L. 1991. Ecological phytogeography of the southern Yukon territory (Canada). In: Nimis, P. L. and Crovello, T. J. (eds), Quantitative approaches to phytogeography. Kluwer Academic Publ., pp. 35 122. Legendre, P. 1987. Constrained clustering. In: Legendre, P. and Legendre, L. (eds), Developments in numerical ecology. Springer, pp. 289307. Legendre, P. 1990. Quantitative methods and biogeographic analysis. Evolutionary biogeography of the marine algae of the North Atlantic. In: Garbary, D. J. and South, G. R. (eds), NATO ASI Series. Vol. G22. Springer, pp. 9 34. Levin, S. A. 1992. The problem of pattern and scale in ecology. Ecology 73: 19431967. Mandrak, N. E. 1995. Biogeographic patterns of sh species richness in Ontario lakes in relation to historical and environmental factors. Can. J. Fish. Aquat. Sci. 52: 1462 1474. Mantel, N. 1967. The detection of disease clustering and a generalized regression approach. Cancer Res. 27: 209 220. Margules, C. R. 1986. Conservation evaluation in practice. In: Usher, M. B. (ed.), Wildlife conservation evaluation. Chapman and Hall, pp. 298314. Marquez, A. L. et al. 1997. On identifying common distribu tion patterns and their causal factors, a probabilistic method applied to pteridophytes in the Iberian Peninsula. J. Biogeogr. 24: 613631. Martn, J. and Gurrea, P. 1990. The peninsular effect in the Iberian butteries (Lepidoptera: Papilionoidea and Hesperioidea). J. Biogeogr. 17: 8596. McCoy, E. D., Bell, S. S. and Walters, K. 1986. Identifying biotic boundaries along environmental gradients. Ecology 67: 749 759. McLaughlin, S. P. 1994. Floristic plant geography: the classication of oristic areas and oristic elements. Progr. Phys. Geogr. 18: 185208. Moreno-Saiz, J. C., Castro-Parga, I. and Sainz-Ollero, H. 1998. Numerical analyses of distributions of Iberian and Balearic endemic monocotyledons. J. Biogeogr. 25: 179 194. Myklestad, A. and Birks, H. J. B. 1993. A numerical analysis of the distribution patterns of Salix L. species in Europe. J. Biogeogr. 20: 132. Oden, N. L. et al. 1993. Categorical wombling: detecting regions of signicant change in spatially located categorical variables. Geographical Anal. 25: 315336. Olivero, J., Real, R. and Vargas, J. M. 1998. Distribution of breeding, wintering, and resident water-birds in Europe: biotic regions and the macroclimate. Ornis Fenn. 75: 153 175. Palomo, L. J. and Antunez, A. 1992. Los atlas de distribucion de especies. In: Vargas, J. M., Real, R. and Antunez, A. (eds), Objetivos y metodos biogeogracos. Aplicaciones en herpetologa. Monogr. Herpetol. 2. Valencia, pp. 39 50. Phipps, J. B. 1975. Bestblock: optimizing grid size in biogeographic studies. Can. J. Bot. 53: 14471452. Real, R. and Ramrez, J. M. 1992. Fundamento losoco de los objetivos y metodos de la Biogeografa. In: Vargas, J. M., Real, R. and Antunez, A. (eds), Objetivos y metodos biogeogracos. Aplicaciones en herpetologa. Monogr. Herpetol. 2. Valencia, pp. 221230. Real, R. and Vargas, J. M. 1996. The probabilistic basis of Jaccards index of similarity. Syst. Biol. 45: 380 385. Real, R., Guerrero, J. C. and Ramrez, J. M. 1992a. Identica cion de fronteras bioticas signicativas para los anbios en la cuenca hidrograca del Sur de Espana. Donana Acta Vertebrata 19: 5370. Real, R., Vargas, J. M. and Guerrero, J. C. 1992b. Analisis biogeograco de clasicacion de areas y de especies. In: Vargas, J. M., Real, R. and Antunez, A. (eds), Objetivos y metodos biogeogracos. Aplicaciones en herpetologa. Monogr. Herpetol. 2. Valencia, pp. 7384.
ECOGRAPHY 24:6 (2001)

Salvo, A. E., Cabezudo, B. and Espana, L. 1984. Atlas de la pteridoora Iberica y Balear. Acta Bot. Malacit. 9: 105 129. Schall, J. J. and Pianka, E. R. 1977. Species densities of reptiles and amphibians on the Iberian Peninsula. Donana Acta Vertebrata 4: 27 34. Sneath, P. H. A. and Sokal, R. R. 1973. Numerical taxonomy. The principles and practices of numerical classication. Freeman. Sokal, R. R. and Rohlf, F. J. 1981. Biometry. The principles and practice of statistics in biological research. Freeman. Thaler, G. R. and Plowright, R. C. 1973. An examination of the oristic zone concept with special reference to the northern limit of the Carolinian zone in southern Ontario. Can. J. Bot. 51: 1765 1789. Vargas, J. M. 1993. Siete pecados capitales en Biogeografa. Zool. Baetica 4: 39 56. Vargas, J. M., Real, R. and Guerrero, J. C. 1998. Biogeographical regions of the Iberian Peninsula based on freshwater sh and amphibian distributions. Ecography 21: 371 382. Wiens, J. A. 1989. Spatial scaling in ecology. Funct. Ecol. 3: 385 397. Harper, W. H. 1978. Differential systematics. Science 114: 315 322.

Appendix 1. Calculation of parameters used to test the existence of signicant biotic boundaries. For each dendrogram node we established a submatrix of signicant similarities that only included the two groups of OGUs separated by that node, which we named group A and group B, respectively. This submatrix was divided into three zones: zone A A and zone B B, which corresponded to the signicant similarities between OGUs of group A and between OGUs of group B, respectively; and zone A B, corresponding to the signicant similarities between the two groups of OGUs. We call Pp(AA) the number of pluses within zone AA divided by the total number of pairwise OGU comparisons in zone A A. So, Pp(A A) is the proportion of pluses in zone A A. We call Psp(AA) the number of OGUs in group A that have at least one plus divided by the total number of OGUs in group A. We can then compute d1(AA) as follows: if the number of pluses in A A is zero, then d1(A A)= 0; otherwise, d1(A A) = Pp(AA) Psp(A A)
(Pp(A A))2 +(Psp(AA))2

The values of d1(A A) range from 0 to 0.707, estimating to what extent similarities higher than expected at random (+ ) predominate within zone AA. We dene Pm(A A) and Psm(A A) as the proportion of minuses in zone AA and the proportion of OGUs in group A with at least one minus, respectively, and these are computed in the same way as Pp(A A) and Psp(A A), but taking into account the minuses. We then dene d2(A A) in the following way: if the number of minuses in zone AA is zero, then d2(A A)=0; otherwise, d2(A A) = Pm(A A) Psm(A A)
(Pm(A A))2 +(Psm(A A))2

The values of d2(A A) range from 0 to 0.707, estimating to what extent similarities lower than expected at random ( ) predominate within zone A A. We dene Pp(A B) and Psp(A B) in a similar way to Pp(AA) and Psp(AA), but with reference to zone AB. So, d3 is zero when the number of pluses in AB is zero; otherwise,

669

d3 =

Pp(A B) Psp(A B)
(Pp(A B)) + (Psp(A B))
2 2

The values of d3 range from 0 to 0.707, estimating to what extent similarities higher than expected at random (+) predominate within zone A B. The parameter DW(A A) measures to what extent the similarities that are higher than expected (+ ) tend to be in zones AA but not in A B (see McCoy et al. 1986), where DW(A A) =d1(A A) d2(A A) d3. Similarly, DW(B B)= d1(B B) d2(B B) d3, where d1(B B) and d2(BB) are calculated as d1(A A) and d2(A A), but computing the pluses and minuses in zone B B. The average of DW(AA) and DW(B B), named DW, measures to what extent similarities that are higher than expected (+ ) tend to be in either zones AA or B B but not in A B. We dene d4 in the same way as d3, but compute the minuses in A B. We then compute the parameter DS= d4 d3 d2(A A) d2(B B), which gives a measure of whether the similarities that are lower than expected () tend to be located in A B, but not in A A or B B.

The statistical signicance of a node was assessed using a G-test of independence (Sokal and Rohlf 1981, McCoy et al. 1986) of the distribution of the signs + , and 0 in the three zones of the submatrix, and so we obtained the parameters GW, for weak boundaries, and GS, for strong boundaries. If similarities higher than expected (+) tend signicantly to be in zones AA or B B, but not in A B, that is, if DW\ 0 and GW is statistically signicant, then there is at least a weak biotic boundary between both groups of OGUs. In this case, if DW(AA) \ 0 then the group of OGUs A constitutes a biotic region, and the same applies for DW(BB) and the group of OGUs B, because then a group of OGUs with similar biota shares a number of species compatible with random expectation with another group of OGUs that also has a similar biota. If similarities signicantly lower than expected () tend to be located in AB, but not in A A or B B, that is, if DS \ 0 and GS is signicant, then a strong biotic boundary exists between the groups of OGUs. In this case a group of OGUs shares fewer species than expected at random with another group of OGUs. The areas delimited by strong boundaries are biotic supraregions. A supraregion can consist of several biotic regions separated by weak boundaries.

670

ECOGRAPHY 24:6 (2001)