Biogeographical analyses are critical to understanding hominin evolutionary history, and the importance of biogeographic processes has been discussed in the paleoanthropological literature for decades. Dispersal, in particular, has been a central focus of many studies. For example, researchers have discussed the importance of dispersals in hominin diversification (e.g., Lahr and Foley, 1994; Foley, 2003, 2013), environmental and spatial factors affecting hominin dispersal (e.g., Bromage and Schrenk, 1995; Cuthbert et al., 2017; Foley, 2018; Joordens et al., 2019; Mondanaro et al., 2020), and dispersals in contemporaneous fauna (e.g., Turner and Wood, 1993; Dennell et al., 2014; Carotenuto et al., 2016). Some researchers pointed to vicariance (e.g., Vrba, 1992; Trauth et al., 2010) and other biogeographic processes (e.g., sympatry; see below) as being critical in hominin evolution (e.g., Schroer and Wood, 2015; Haile-Selassie et al., 2016; Macho, 2017).

Previous phylogeny-based studies of hominin biogeography used maximum parsimony to infer dispersal events (Strait and Wood, 1999; Strait, 2013). Strait and Wood (1999) were the first to analyze hominin biogeography in a phylogenetic framework. They treated geography as a cladistic character in which geographical regions were considered distinct character states. They inferred the evolution of the geography character implied by several cladograms (Delson, 1986; Walker et al., 1986; Grine, 1988; Skelton and McHenry, 1992; Wood, 1992; Strait et al., 1997) by estimating the ancestral areas at the internal nodes of the trees. According to the parsimony criterion, the optimal ancestral distributions minimize the number of character state changes, and any change in character state was a posteriori interpreted as a dispersal event. Strait and Wood (1999) observed four to seven dispersals between southern, eastern, and South-central Africa. More recently, Strait (2013) used a similar method to suggest that the last common ancestor (LCA) of hominins originated in central Africa and that significant evolutionary events such as postcanine megadontia and increases in brain size took place in eastern Africa.

Strait and Grine's (2004) and Strait's (2013) prior work provides a useful set of hypotheses for testing. Their findings suggest that early human evolution in Africa involved four specific dispersal events. The first dispersal involved the LCA of hominins, except Sahelanthropus, which moved from central to eastern Africa before 6.0 Ma. The second dispersal involved Australopithecus africanus or its ancestor moving from eastern to southern Africa before 3.0 Ma. The third dispersal involved Paranthropus robustus or its ancestor moving from eastern to southern Africa before 1.8 Ma. Finally, Homo habilis dispersed from eastern to southern Africa before 1.8 Ma in the fourth dispersal event.

Since then, additional hominin species have been discovered, and our knowledge of their locations and temporal range have improved (e.g., Berger et al., 2010; Clarke and Kuman, 2019; Herries et al., 2020). Moreover, while previous studies have used phylogenetic data to estimate the importance and directionality of dispersal events in human evolution (e.g., Strait and Wood, 1999; Strait, 2013), other biogeographical processes during human evolutionary history require consideration. The objective of this study is to (1) evaluate Strait's (2013) hypotheses concerning the most likely ancestral range distribution of early hominin taxa, as well as the directionality of dispersals, (2) estimate the biogeographic processes that underlie species distributions (see below), and (3) assess the frequency of biogeographic events by examining alternative biogeographic scenarios.

As noted, phylogeny-based reconstructions of biogeography have relied on parsimony. Parsimony-based reconstructions of hominin biogeography have several shortcomings. These methods have been applied to find dispersal patterns, but dispersal is only one of many biogeographical processes (Ronquist, 1995; Ronquist and Sanmartín, 2011). Moreover, although dispersal can explain any distribution pattern, these hypotheses are not subject to falsification (Morrone and Crisci, 1995; Sanmartín and Ronquist, 2004). For these reasons, Ronquist (1995) referred to these methods as ‘pattern-based,’ which determine evolutionary processes after the results have been obtained (for more information, see Ronquist and Sanmartín, 2011). Additionally, parsimony ignores critical information about time (i.e., branch length). Therefore, longer branch lengths do not have an associated higher probability of change. Finally, the parsimony method used in previous studies (Strait and Wood, 1999; Strait, 2013) assumes that ancestors occur in single areas. However, such an assumption does not reflect widespread hominin taxa discovered from multiple sites. The problem with reconstructing single-area ancestors is that, to invoke vicariance, we need to be able to infer widespread ancestors at the nodes that divide into smaller regions (Lamm and Redelings, 2009; Crisp et al., 2011a, 2011b).

In recent years, historical biogeography has been transformed significantly through the development of statistical approaches and ‘event-based’ methods (Ronquist, 1997; Ree et al., 2005; Landis et al., 2013; Matzke, 2013; Ree and Sanmartín, 2018). These methods estimate geographic range evolution along branches of phylogenetic trees (i.e., anagenetic events) and at internal nodes of the tree (i.e., cladogenetic events; Matzke, 2013). The methodological transition was initiated by dispersal-vicariance analysis (DIVA) in the 1990s (Ronquist, 1997). Dispersal-vicariance analysis is an ‘event-based’ parsimony method that estimates ancestral distribution and biogeographical processes. The two types of anagenetic events allowed in DIVA (Fig. 1) are dispersal and local extinction (see above). Cladogenetic events that are allowed (Fig. 1) are narrow sympatry, as well as narrow and widespread vicariance. Dispersal-vicariance analysis uses a three-dimensional cost matrix that counts biogeographical events (Ronquist, 1997). It assigns a cost of one for extinction and dispersal events, whereas vicariance and duplication within a single area (i.e., narrow sympatry) have zero costs (Ronquist, 1997; for more information, see Sanmartín and Ronquist, 2004). The most parsimonious solution is the estimation that explains the data with the least cost. As a result, DIVA favors hypotheses that minimizes the number of dispersal events.

A maximum likelihood approach to infer geographic range evolution was first proposed by Ree et al. (2005) and was later refined by Ree and Smith (2008). This approach, dispersal–extinction–cladogenesis (DEC) is a parametric model that assigns probabilities, rather than costs, to events. Anagenetic events (i.e., dispersal and extinction) are free parameters used to create an instantaneous rate matrix (i.e., a matrix describing transition rates between discrete geographic ranges along phylogenetic branches). Next, the likelihood of alternative range inheritance scenarios at the cladogenesis events is calculated. Regarding cladogenetic events, both DIVA and DEC allow narrow sympatry and narrow vicariance but disallow widespread sympatry (Fig. 1). However, the two models differ concerning two cladogenetic processes. First, DIVA allows widespread vicariance (see previous text), whereas DEC assumes that one of the daughters must always have a range of only one area. As a result, widespread vicariance is not allowed in DEC. Second, unlike DIVA, DEC allows subset sympatry (Ronquist and Sanmartín, 2011).

One of the shortcomings of the DEC model is its computational limit as the number of areas increases. A Bayesian approach called Bayesian inference of historical biogeography for discrete areas (BayArea; Landis et al., 2013) enhances the computational speed of biogeographic models while enabling biogeographic inference at a finer scale. In BayArea, ancestral ranges are assumed to be identical at cladogenesis events. Consequently, this model allows only widespread and narrow sympatry (see previous text) at cladogenetic events (Fig. 1), where two daughter lineages inherit the same range(s) as their ancestor. Widespread sympatry is only allowed in BayArea and is prohibited in DIVA and DEC models. Here, we use the new approaches to test biogeographic hypotheses and model biogeographic processes in early hominins.