Paleogeography, describing the ancient distribution of highlands, lowlands, shallow seas and deep ocean basins, is widely used in a range of fields including paleoclimatology, plate tectonic reconstructions, paleobiogeography, resource exploration and geodynamics. Global deep-time paleogeographic compilations have been published (e.g., Blakey, 2008; Golonka et al., 2006; Ronov, et al., 1984, 1989; Scotese, 2001, 2004; Smith et al., 1994). However, they are generally presented as static paleogeographic snapshots with varying map projections and different time intervals represented by the maps, and are tied to different plate motion models. This makes it difficult to convert the maps into a digital format, link them to alternative digital plate tectonic reconstructions and update them when plate motion models are improved. It is therefore challenging to use paleogeographic maps to help constrain or interpret numerical models of mantle convection that predict long-wavelength topography (Gurnis et al., 1998; Spasojevic and Gurnis, 2012) based on different tectonic reconstructions, or as an input to models of past ocean and atmosphere circulation/climate (Goddéris et al., 2014; Golonka et al., 1994) and models of past erosion/sedimentation (Salles et al., 2017).

In order to address these issues, we develop a workflow to restore the ancient paleogeographic geometries back to their modern coordinates so that the geometries can be attached to a different plate motion model. This is the first step towards the construction of paleogeographic maps with flexible spatial and temporal resolutions that are more easily testable and expandable with the incorporation of new paleoenvironmental data sets (e.g., Wright et al., 2013). In this study, we use a set of global paleogeographic maps (Golonka et al., 2006) covering the entire Phanerozoic time period as the base paleogeographic model. Coastlines on these paleogeographic maps represent estimated maximum marine transgression surfaces (Kiessling et al., 2003). We first restore the global paleogeographic geometries of Golonka et al. (2006) to their present-day coordinates by reversing the sign of the rotation angle, and then reconstruct them to geological times using a different plate motion model of Matthews et al. (2016). We then use paleoenvironmental information from marine fossil collections from the Paleobiology Database to modify the inferred paleo-coastline locations and paleogeographic geometries. Next, we use the revised paleogeography to estimate the surface areas of global paleogeographic features including deep oceans, shallow marine environments, landmasses, mountains and ice sheets. In addition, we compare the global flooded continental areas since the Devonian calculated from the revised paleogeography with other results derived from other paleoenvironment and paleo-lithofacies maps (Ronov, 1994; Smith et al., 1994; Walker et al., 2002; Blakey, 2003, 2008; Golonka, 2007b, 2009, 2012) or from the strontium isotope record (van der Meer et al., 2017). We estimate the terrestrial areal change over time associated with transferring reconstruction, filling gaps and modifying the paleogeographic geometries based on consistency test. Finally, we test the marine fossil collection data set used in this study for fossil abundances over time using different timescales of the International Commission on Stratigraphy (ICS2016; Cohen et al., 2013, updated) and of Golonka (2000) and discuss the limitations of the workflow we develop in this study.

The data used in this study are global paleogeographic maps and paleoenvironmental data for the last 402 million years (Myr), which originate from the set of paleogeographic maps produced by Golonka et al. (2006) and the Paleobiology Database (PBDB, https://paleobiodb.org), respectively. The global paleogeographic compilation extending back to the Early Devonian of Golonka et al. (2006) is divided into 24 time-interval maps using the timescale of Golonka (2000) which is based on the original timescale of Sloss (1988; Table 1). Each map is a compilation of paleo-lithofacies and paleoenvironments for each geological time interval. These paleogeographic reconstructions illustrate the changing configuration of ice sheets, mountains, landmasses, shallow marine environments (inclusive of shallow seas and continental slopes) and deep oceans over the last ∼ 400 Myr.

The paleogeographic maps of Golonka et al. (2006) are constructed using a plate tectonic model available in the supplement of Golonka (2007a), where relative plate motions are described. In this rotation model, paleomagnetic data are used to constrain the paleolatitudinal positions of continents and rotation of plates, and hotspots, where applicable, are used as reference points to calculate paleolongtitudes (Golonka, 2007a). This rotation model is necessary to restore these paleogeographic geometries (Golonka et al., 2006) to their present-day coordinates so that they can be attached to a different plate motion model. The relative plate motions of Golonka (2006, 2007a) are based on the reconstruction of Scotese (1997, 2004).

Here, we use a global plate kinematic model to reconstruct paleogeographies back in time from present-day locations. The global tectonic reconstruction of Matthews et al. (2016), with continuously closing plate boundaries from 410–0 Ma, is primarily constructed from a Mesozoic and Cenozoic plate model (230–0 Ma; Müller et al., 2016) and a Paleozoic model (410–250 Ma; Domeier and Torsvik, 2014). This model is a relative plate motion model that is ultimately tied to Earth's spin axis through a paleomagnetic reference frame for times before 70 Ma, and a moving hotspot reference frame for younger times (Matthews et al., 2016).

The PBDB is a compilation of global fossil data covering deep geological time. All fossil collections in the database contain detailed metadata, including on the time range (typically biostratigraphic age), present-day geographic coordinates, host lithology and paleoenvironment. Figure 1 represents distributions of the global fossil collections at present-day coordinates and shows their numbers since the Devonian. The recorded fossil collections are unevenly distributed both spatially and temporally, largely due to the differences in fossil preservation, the spatial sampling biases of fossil localities and the uneven entry of fossil data to the PBDB (Alroy, 2010). For this study, a total of 57 854 fossil collections with temporal and paleoenvironmental assignments from 402 to 2 Ma were downloaded from the database on 7 September 2016.

The methodology can be divided into three main steps: (1) the original paleogeographic geometries are restored to present-day coordinates by applying the inverse of the rotations used to make the reconstruction, (2) these restored geometries are then rotated to new locations using the plate tectonic model of Matthews et al. (2016) and (3) the paleo-coastline locations and paleogeographic geometries are adjusted using paleoenvironmental data from the PBDB. Figure 2 illustrates the generalized workflow that can be applied to a different paleogeography model. In order to represent the paleogeographic maps as digital geographic geometries, they are first georeferenced using the original projection and coordinate system (global Mollweide in Golonka et al., 2006), and then re-projected into the WGS84 geographic coordinate system. The resulting maps are then attached to the original rotation model using the open-source and cross-platform plate reconstruction software GPlates (http://gplates.org). Every plate is then assigned a unique plate ID that defines the rotation of the tectonic elements so that the paleogeographic geometries can be rotated back to their present-day coordinates (see example in Fig. 3a, b). We use present-day coastlines and terrane boundaries with the plate IDs of Golonka (2007a) as a reference to refine the rotations and ensure that the paleogeographic geometries are restored accurately to their present-day locations.

When the paleogeographic geometries in present-day coordinates are attached to a new reconstruction model, as Matthews et al. (2016) used in this study, the resulting paleogeographies result in gaps (Fig. 3c, pink) and overlaps between neighboring polygons, when compared to the original reconstruction (Fig. 3a). These gaps and overlaps essentially arise from the differences in the reconstructions described in Matthews et al. (2016) and Golonka et al. (2006). The reconstruction of Golonka et al. (2006) has a tighter fit of the major continents within Pangea prior to the supercontinent breakup. In addition, this reconstruction contains a different plate motion history and block boundary definitions in regions of complex continental deformation, for example along active continental margins (e.g., Himalayas, western North America; Fig. 3c).

The gaps and overlaps cause changes in the total terrestrial or oceanic paleogeographic areas at different time intervals, becoming larger or smaller, when compared with the original paleogeographic maps (Golonka et al., 2006). The gaps can be fixed by interactively extending the outlines of the polygons in a GIS platform to make the plates connect as in the original paleogeographic maps (Fig. 3a, c, d). Changes in the extent of total terrestrial or oceanic area of the paleogeographies with filled gaps are compared with the original paleogeographies in Fig. 3d (Golonka et al., 2006).

Once the gaps are filled, the reconstructed paleogeographic features are compared with the paleoenvironments indicated by the marine fossil collections from the PBDB. These comparisons aim to identify the differences between the mapped paleogeography and the marine fossil collection environments in order to revise the paleo-coastline locations and paleogeographic geometries. Fossil collections belonging to each time interval (Table 1; Golonka, 2000) are first extracted from the data set downloaded from the PBDB. Only the fossil collections with temporal ranges lying entirely within the corresponding time intervals are selected, as opposed to including the fossil collections that have larger temporal ranges. Fossil collections with temporal ranges crossing any time-interval boundary are not taken into consideration. As a result, a minimum number of fossil collections are selected for each time interval. The selected fossil collections are classified into either the terrestrial or marine setting category, according to a lookup table (Table 2).

Marine fossil collections are then attached to the plate motion model of Matthews et al. (2016) so they can be reconstructed at each time interval. Subsequently, a point-in-polygon test is used to determine whether or not the indicated marine fossil collection is within the appropriate marine paleogeographic polygon. The results of these tests are discussed in the following section.

In the next step, we modify the paleo-coastline locations and paleogeographic geometries based on the test (Figs. 4, 5 and Supplement). Modifications are made according to the following rules. (1) Marine fossil collections from the PBDB are presumed to be well dated, constrained geographically, not reworked and representative of their broader paleoenvironments. Their indicative environments are assumed to be correct. (2) Only marine fossil collections within 500 km of the nearest paleo-coastlines are taken into account as most marine fossil collections used in this study are located within 500 km from the paleo-coastlines (see Fig. S1 in the Supplement). (3) The paleo-coastlines and paleogeographic geometries are modified until they are consistent with the marine fossil collection environments and at the same time remain about 30 km distance from the fossil points used (Fig. 5c, f, l). (4) The adjacent paleo-coastlines are accordingly adjusted and smoothed (Figs. 4, 5). (5) The modified area (Fig. 5b, e, k, blue) resulting from shifting the coastline is filled using the shallow marine environment. These rules are designed to maximize the use of the paleoenvironmental information obtained from the marine fossil collections to improve the coastline locations and paleogeography while attempting to minimize spurious modifications.

However, in some rare cases, outlier marine fossil data may be a deceptive recorder of paleogeography. For instance, Wichura et al. (2015) discussed the discovery of a ∼ 17 Myr old beaked whale fossil 740 km inland from the present-day coastline of the Indian Ocean in east Africa. The authors found evidence to suggest that this whale could have traveled inland from the Indian Ocean along an eastward-directed fluvial (terrestrial) drainage system and was stranded there, rather than representing a marine setting that would be implied under our assumptions. Therefore, theoretically, when using the fossil collections to improve paleogeography, additional concerns about living habits of fossils and associated geological settings should be taken into account. In this study, we have removed this misleading fossil whale from the data set. Such instances of deceptive fossil data are a potential limitation within our workflow, which we seek to minimize by excluding inconsistent fossils more than 500 km away from previously interpreted paleo-coastlines described above.

Our study highlights the flexibility of digital paleogeographic models linked to plate tectonic reconstructions in order to better understand the interplay of continental growth and eustasy, with wider implications for understanding Earth's paleotopography, ocean circulation and the role of mantle convection in shaping long-wavelength topography. We present a workflow that enables the construction of paleogeographic maps with variable spatial and temporal resolutions, while also becoming more testable and expandable with the incorporation of new paleoenvironmental data sets.

We develop an approach to revise the paleo-coastline locations and paleogeographic geometries using paleoenvironmental information indicated by the marine fossil collections from the PBDB. Using this approach, the consistency ratio between the paleogeography and the paleobiology records since the Devonian is increased from an average 75 % to nearly full consistency. The paleogeography in the main regions of North America, South America, Europe and Africa is significantly improved, especially in the Late Carboniferous, Middle Permian, Triassic, Jurassic, Late Cretaceous and most portions of the Cenozoic. The flooded continental areas since the Late Devonian inferred from the revised global paleogeography in this study are generally consistent with the results derived from other paleoenvironment and paleo-lithofacies data or from the strontium isotope record in marine carbonates.

Comparing the terrestrial areal change over time associated with transferring the reconstruction and filling gaps, and revising paleogeographic geometries using the paleoenvironmental data from the PBDB, indicates that reconstruction difference is a main factor in paleogeographic areal change when comparing with the original maps, and revising paleogeographic geometries based on PBDB test is secondary.

We provide two sets of digital global paleogeographic maps during 402–2 Ma: (1) the paleogeography reconstructed using the plate motion model of Matthews et al. (2016) and revised using paleoenvironmental information indicated by the marine fossil collections from the PBDB and (2) the original paleogeography of Golonka et al. (2006). We also provide the original rotation file of Golonka et al. (2006), a set of paleogeographic maps illustrating the PBDB test and revision of paleo-coastlines and paleogeographic geometries, a set of GeoTiff files of all revised paleogeographic maps, paleobiology data in shapefile used in this study separated into two sets of consistent marine fossil collections and inconsistent marine fossil collections, an animation for the revised global paleogeographic maps, and a README file outlined the workflow of this study.