Biogeography of European land mammals shows environmentally distinct and spatially coherent clusters

Abstract

Aim To produce a spatial clustering of Europe on the basis of species occurrence data for the land mammal fauna. Location Europe defined by the following boundaries: 11°W, 32°E, 71°N, 35°N. Methods Presence/absence records of mammal species collected by the Societas Europaea Mammalogica with a resolution of 50 × 50 km were used in the analysis. After pre‐processing, the data provide information on 124 species in 2183 grid cells. The data were clustered using the k‐means and probabilistic expectation maximization (EM) clustering algorithms. The resulting geographical pattern of clusters was compared against climate variables and against an environmental stratification of Europe based on climate, geomorphology and soil characteristics (EnS). Results The mammalian presence/absence data divide naturally into clusters, which are highly connected spatially and most strongly determined by the small mammals with the highest grid cell incidence. The clusters reflect major physiographic and environmental features and differ significantly in the values of basic climate variables. The geographical pattern is a fair match for the EnS stratification and is robust between non‐overlapping subsets of the data, such as trophic groups. Main conclusions The pattern of clusters is regarded as reflecting the spatial expression of biologically distinct, metacommunity‐like entities influenced by deterministic forces ultimately related to the physical environment. Small mammals give the most spatially coherent clusters of any subgroup, while large mammals show stronger relationships to climate variables. The spatial pattern is mainly due to small mammals with high grid cell incidence and is robust to noise from other subsets. The results support the use of spatially resolved environmental reconstructions based on fossil mammal data, especially when based on species with the highest incidence.

Biogeographic provincialism in rodent faunas from the Iberoccitanian Region (southwestern Europe) generates severe diachrony within the Mammalian Neogene (MN) biochronologic scale during the Late Miocene

“…These differences are associated with an increase in aridity from northeastern to southwestern Europe Uimenez-Moreno and Sue, 2007;Furi6 et al, 2011). At the continental scale, this area could be considered as a uniform biogeographic lll1it, but when it is examined in detail, two mammalian bioprovinces arise (Alvarez-Sierra et al, 1985;Alherdi and Azanza, 1997;Daams etal, 1998;Morales et al, 1999;Heikinheimo et al, 2007), recognizable since the Eocene (Casanovas-Oadellas and Moya-Sola, 1992;Pelaez-Campomanes, 1993;Badiola et al, 2009). The northern province includes fossil sites from the Rhone, Provence, Cucuron-Basse Ourance and Languedoc-Rousillon basins from southeastern France, and the Valles Penedes basin from Catalonia.…”

Identifying the components

“…grid location of 50 × 50 km) has been assigned to one of the six components based on its items (i.e., the mammals that have been recorded at that location). This is a typical example where normally ad-hoc solutions are used (Heikinheimo et al 2007) as it is difficult to define a meaningful distance measure. Our method only regards the characteristics of the components, the patterns in which the items occur.…”

“…We use a number of UCI datasets (Coenen 2003), the Retail dataset (Brijs et al 1999) and the Mammals dataset (Heikinheimo et al 2007) for experimental validation of our methods. The latter consists of presence records of 121 European mammals in geographical areas of 50 × 50 km.…”

RealKrimp — Finding Hyperintervals that Compress with MDL for Real-Valued Data

“…5: Calculate the coverage of each dimension in H. 6: In order of decreasing coverage, determine if compression improves when letting a dimension go unbounded. 7: Report the resulting hyperinterval if it is interesting according to Equation (14). 8: Until no more interesting hyperintervals can be found, restart from step 3 with two rows not covered by any of the previously reported hyperintervals.…”

“…For every hyperinterval that we could find, every boundary can have infinitely many values leading to the hyperinterval covering exactly the same records, with an arbitrarily small change in the volume of the hyperinterval. To deal with this serious problem in the applicability of real-valued MDL, in this section we introduce the RealKrimp algorithm: a mining scheme that confines its attention to those interesting hyperintervals that locally maximize the inequality of (14); no better compression is obtained by a hyperinterval that is either an extension or a restriction of the considered hyperinterval. The RealKrimp algorithm is given in Algorithm 1.…”