Using network analysis to trace the evolution of biogeography through geologic time: A case study Steffen Kiel* Swedish Museum of Natural History, Department of Palaeobiology, Box 50007, 10405 Stockholm, Sweden ABSTRACT biogeographic provinces, I determined for each The biogeographic distribution of organisms has continuously changed through Earth’s node how many of its genera have their first history as plate tectonics changed the configurations of land masses, ocean basins, and cli- appearance (FA) in the geologic stage to which mate zones. Yet, methods to investigate this dynamic through geologic time are limited. Here, the respective node belongs. Permian occur- network analysis is used to explore and to visualize the biogeographic history of brachiopods rences were added back into the data set when through the entire Triassic period. Many previously recognized biogeographic provinces are counting the number of first occurrences to avoid found, and in addition, the stratigraphic ranges of these provinces were identified. Provinces inflated origination in the oldest Triassic stage. in the Tethys Ocean show the lowest degree of connectedness, which can be linked to higher To analyze the relationship between faunal simi- evolutionary rates in this tropical ocean basin and possibly also to higher habitat heteroge- larity between two nodes on the one hand, and neity. Stratigraphically, the Tethyan provinces are separated largely along the boundaries of their geographic and temporal distance on the the Early, Middle, and Late Triassic. This suggests that the events resulting in faunal changes other, the great-circle distance between the paleo- among the index fossils used to define these sub-periods also affected the brachiopods. How- coordinates of any two nodes was calculated ever, through the ~50 m.y. of the Triassic period, geographic proximity played a greater role using the haversine formula (the paleo-coordi- in producing faunal similarity than proximity in geologic age. Thus network analysis is a nates were originally calculated using PointTrack viable tool to better understand the dynamic evolution of biogeography through geologic time. version 7.0 [Ke et al., 2016]). The time differ- ence between any two nodes was calculated for INTRODUCTION data of the phylum Brachiopoda from the entire the midpoints of their respective geologic stages, Many systems of interacting units in the bio- Triassic period (~50 m.y.) are analyzed as a single using the International Commission on Stratig- logical, social, and technological world can be complex network. The aims are to (1) delimit bio- raphy International Chronostratigraphic Chart seen as complex networks. Obvious examples geographic provinces and their range through time, version 2016/12 (Cohen et al., 2013). include food webs, neural and social networks, (2) evaluate the relative importance of time and the Internet, power grids, and many more (Albert geography in delimiting those provinces, and (3) NETWORK ANALYSIS et al., 1999; Dunne et al., 2002; Newman and assess regional differences in connectivity. The present approach uses weighted net- Park, 2002; Greicius et al., 2003; Pagani and works, in which the localities represent the nodes, Aiello, 2013). Understanding the nature of these DATA and the links between them indicate not only networks is crucial for a broad range of appli- The basis is a recently published data set of the presence of a connection, but are assigned a cations, like assessing the vulnerability of criti- late Permian to Late Triassic brachiopod occur- value (the weight) indicating the strength of the cal infrastructures to failure, or predicting and rences (Ke et al., 2016). This data set was com- connection. These weights were calculated using preventing the spread of epidemics (Rinaldi et piled from published sources, covers all brachio- the Bray-Curtis distance measure. They range al., 2001; Pastor-Satorras et al., 2015). Of par- pod clades, and includes only taxa that had been from zero when the two nodes are identical in ticular interest are methods to detect large-scale systematically described and illustrated; the tax- faunal composition, to 1 when they share no taxa, structures such as communities and modules in onomic unit used herein is the genus. The geo- resulting in the counterintuitive situation that a networks (Newman, 2012), and these approaches graphic unit used in the original study was called link with a high weight indicates low similarity have recently begun to be used to address bio- a “paleogeographic analysis unit” and represents between the nodes, and vice versa. geographic questions (Moalic et al., 2012; Kou- a set of collections from a certain region delim- A thresholding approach was used to illus- gioumoutzis et al., 2014; Kiel, 2016). ited by geographic or tectonic borders (Ke et al., trate and quantitatively analyze relationships Present-day biogeographic distributions have 2016); information on the sampled facies were between nodes (Moalic et al., 2012; Kivelä et a historical basis, and the fossil record is a key to not given. These units span 1.9°–4.6° of longi- al., 2015; Kiel, 2016). The threshold is the maxi- understand both this historical basis and the long- tude on average, and values >12° (up to 41°) mum weight for the links used to construct the term evolution of biogeography (Jablonski et al., were only reached in very high latitudes (i.e., network; links beyond that threshold (indicat- 2006; Renema et al., 2008). Biogeographic anal- northern Siberia); in latitude, they span 1°–2.5° ing lower similarity) are not considered (Kivelä yses through geologic time often use presence- on average, with a maximum extension of 10° et al., 2015). The result is that at low thresh- absence matrices and are typically done in a step- (the Qiangtang block during the Carnian). These olds, the network is broken up into components, wise fashion, whereby each geologic time interval units are also used herein and are referred to which have strong links within each component is analyzed separately and this succession is then as localities or nodes. Excluded from the origi- but only weak links to the remaining network. described in a narrative way (Brayard et al., 2007; nal data set were the Permian records and the This pattern is characteristic for communities Neubauer et al., 2015; Dunhill et al., 2016). But records of Dienerian age (late Induan) because within networks (Newman, 2012), and the com- this approach does not fully capture the dynam- of the very low number of records, and nodes ponents were therefore used as the foundation ics in biogeographic evolution. Here, occurrence with fewer than three genera. to define paleobiogeographic provinces with For a subsequent assessment of the impact both a spatial and a temporal dimension. With *E-mail: [email protected] of evolutionary rates on the longevity of the increasing thresholds, i.e., with the addition of GEOLOGY, August 2017; v. 45; no. 8; p. 711–714 | Data Repository item 2017229 | doi:10.1130/G38877.1 | Published online 19 May 2017 ©GEOLOGY 2017 The Authors.| Volume Gold 45 |Open Number Access: 8 | www.gsapubs.orgThis paper is published under the terms of the CC-BY license. 711 Downloaded from http://pubs.geoscienceworld.org/gsa/geology/article-pdf/45/8/711/2315605/711.pdf by guest on 28 September 2021 successively weaker links, the components will 2015); additional analyses were performed using network plotted onto a series of paleogeographic “grow” by the addition of further nodes and will PAST version 2.17c (Hammer et al., 2001). maps. It illustrates the distinctiveness and lon- merge with other components. The thresholding gevity of the Austrazea province compared to all approach thus allows looking at how patterns RESULTS AND DISCUSSION other provinces, and shows that the main breaks change along a sliding scale of threshold val- in time between the Tethyan provinces are along ues, and networks at different thresholds can Paleobiogeographic Provinces the boundaries of the Early, Middle, and Late Tri- provide different insights into biogeographic The biogeographic provinces derived from assic. These boundaries were originally defined relationships (i.e., Fig. 1). When all components the thresholding approach (Fig. 1) are listed in based on ammonites and microfossils from low merge into one giant component, the so-called Table 1, and many of them correspond to pre- latitudes (Visscher, 1992; Kozur, 2003). Finding percolation point is reached (Kivelä et al., 2015), viously recognized provinces (Ke et al., 2016). this zonation also among Tethyan brachiopod and networks at thresholds beyond this point are In addition, the network analysis retrieved their provinces suggests that the mechanisms result- typically of limited value for delimiting paleo- durations in geologic time as described in the ing in faunal changes among ammonites and the biogeographic provinces. original study (Ke et al., 2016). This is regarded microfossils also affected the brachiopods. Construction and most analyses of the net- here as “proof of concept” that network anal- works were performed using the software pack- ysis is a viable tool to investigate biogeogra- Impact of Geography and Geologic Age on age EDENetworks version 2.18 (Kivelä et al., phy through geologic time. Figure 2 shows the Faunal Similarity Adding stratigraphic duration to the proper- ties of biogeographic provinces raises the ques- Threshold=0.65 Threshold=0.74 tion of whether geography or geologic age is the main force