INTRODUCTION

Body size is a key ecological trait reflecting diet, locomotion, and life history strategies (Peters 1983; Schmidt-Nielsen 1984). For poikilotherms including reptiles, body size plays an additional role in thermoregulation, specifically through its influence on the ratio of surface area to volume which governs heat exchange, and influences metabolic rates (Ashton & ­Feldman 2003; Makarieva et al. 2005; Rodrigues et al. 2018). Therefore, body size trends within poikilothermic groups such as reptiles provide insight into ecological niche availability over time and can potentially be used as paleoenvironmental proxies (Head et al. 2009, 2013; Godoy et al. 2019: Stockdale & Benton 2021; Parker et al. 2023;). Body size distributions within reptile communities may be useful indicators of paleohabitats, but they have rarely been measured in the fossil record because environment reconstructions based on the terrestrial vertebrate fossil record have relied mainly on mammalian faunal proxies (e.g. Hernández Fernández & Vrba 2006; Cerling et al. 2011; Liu et al. 2012; Bibi & Kiessling 2015; Plummer et al. 2015; Žliobaitė et al. 2016). The Shungura Formation of the Omo Group (Lower Omo River Valley, southwest Ethiopia) presents an excellent natural laboratory for using the reptile fossil record to characterize the abiotic and biotic conditions underpinning body mass changes in terrestrial and aquatic poikilotherms over time.

The Shungura Formation is a chronostratigraphically well-constrained and nearly continuous section of fluviolacustrine sedimentary sequences that preserves a rich, high-resolution record of faunal and environmental change during the Plio-Pleistocene (Arambourg 1948; Howell & Coppens 1974; Heinzelin 1983; Boisserie et al. 2008). The formation is divided into 12 members identified as Basal, then A to L (excluding I) from oldest to youngest. The depositional lithologies of these members represent changes in fluviolacustrine architecture on the paleolandscape, most notably expansion and reduction in lake extent (Heinzelin 1983). While mammalian evolution and diversity in the Shungura Formation have been described in detail, the only previous descriptions of the non-mammalian vertebrate faunas have been in taxonomic works on fishes, turtles, and crocodylians (Arambourg 1948; de Broin 1979; Tchernov 1986; Stewart & Murray 2008). Despite this limitation, the sedimentological interpretations and densely-sampled isotopic and mammalian faunal records of the Shungura Formation provide a unique opportunity to place changes in local reptile body size maxima in the context of environmental change, as informed by geochemical and mammalian fossil records. Here, we estimate the masses of the best-preserved reptile taxa from the Shungura Formation – crocodylians, turtles, and snakes – and investigate their potential relationships to biotic and abiotic factors in the environment of the Shungura Formation. We use the detailed time series of environmental and mammalian faunal changes in the Shungura sequence (Alemseged 2003; Bobe & Behrensmeyer 2004; Hernández Fernández & Vrba 2006; Cooke 2007; Boisserie et al. 2010; Passey et al. 2010; Levin 2015; Plummer et al. 2015; Blondel et al. 2018; Negash et al. 2020; Bibi 2023) to place body size histories of Shungura Formation reptiles within the context of local climate and community composition. We compare the maximum size time series for reptiles with mass estimates for various large mammal groups from Bibi & Cantalapiedra (2023) to test whether co-occurring reptiles and mammals exhibited common size trends, indicating shared environmental pressures, or inverse relationships, which could indicate competition or opposing habitat preferences.

Shungura Formation turtle body size histories at community scales were previously examined to estimate paleotemperature and paleoprecipitation using models relating modern turtle community body size to climate (Parker et al. 2023). These results indicated that, although climate variables are only loosely linked to size in modern communities, these models can provide information about relative environmental change through time based on the Shungura record. The tests for correlation with geochemical proxies and mammal faunas we present here expand on that study to identify factors influencing reptile size maxima, which can be evaluated in other contexts in the fossil record to better understand what conditions are necessary for the evolution of large body sizes.

MATERIAL AND METHODS

The Shungura Formation reptile fossil record

We examined reptile fossils derived from field expeditions, primarily the International Omo Research Expedition (IORE) (Howell & Coppens 1974) and Omo Group Research Expedition (OGRE) (Boisserie et al. 2008, 2010) housed in the National Museum of Ethiopia (NME)/Ethiopian Heritage Authority (EHA, ex-Authority for Research and Conservation of the Cultural Heritage) in Addis Ababa. Some of the largest turtle specimens discovered in the field were not collected due to resource constraints and were measured in situ during OGRE field campaigns.

Although mostly undescribed, the reptile fossil record of the Shungura Formation is composed of similar taxa to the more extensively studied records from more southern Neogene and Pleistocene localities in the Turkana Depression (Harris et al. 2003; Storrs 2003; Wood 2003; Brochu 2020; Head & Müller 2020). The fossil record of turtles consists of semi-aquatic trionychid taxa, including specimens referrable to Cycloderma ‌Peters, 1854 (Fig. 1A) and Trionyx ‌Saint-Hilaire, 1809 (Parker et al. 2023), pelomedusids, including specimens comparable to Pelusios ‌Wagler, 1830 (Fig. 1B), and fully terrestrial testudinids (Fig. 1C), including giant specimens comparable to Centrochelys ‌Gray, 1872 (Parker et al. 2023). Among examined squamates, Python ‌Daudin, 1803 (Fig. 1D) is represented by precloacal vertebrae. Other squamates, including Varanus ‌Merrem, 1820, and small-bodied snakes, are not included here due to their sparse record with prohibitively small sample sizes. Crocodylians from the Shungura include cranial remains referable to the extremely tubulirostrine genus Euthecodon ‌Fourtau, 1920 (Fig. 1F) and members or close relatives of the extant longirostrine genus Mecistops ‌Gray, 1844 (Fig. 1E). We refer to the latter as cf. Mecistops here, as some specimens were previously Crocodylus ‌cataphractus ‌Cuvier, 1824, a synonym of Mecistops ‌cataphractus ‌Cuvier, 1825, the extant African sharp-nosed or slender-snouted crocodile (McAliley et al. 2006), but note that they are morphologically distinct from modern Mecistops and their taxonomic assignment is tentative pending phylogenetic analysis (Brochu et al. personal communication). Additionally, numerous brevirostrine specimens (Fig. 1D) are similar to species of Crocodylus from other eastern African sequences (Brochu 2020). Species diagnoses within the genus Crocodylus of the Plio-Pleistocene are subject to ongoing revision (Brochu 2001, 2020; Storrs 2003; Brochu et al. 2010; Brochu & Storrs 2012). Cranial specimens that we identify as Crocodylus here include specimens comparable to the robust species C. ‌thorbjarnarsoni, but further taxonomic study is required to determine whether other species from the genus are also present in the Shungura Formation.

Occurrences and abundances of reptile taxa are not uniformly distributed throughout the members of the Shungura Formation. Member G, spanning 2.27 to 1.91 Ma, is divided into “lower G” and “upper G”, which have yielded distinct fossil assemblages, likely due to the lacustrine system present during the deposition of upper G. Therefore, we have considered these two levels separately alongside the other members. Including all specimens cataloged from 1967-1976 (IORE) and 2006-2018 (OGRE), the highest sampling is from Member E (2.39-2.33 Ma), with 699 reptile specimens in total (nearly 12 % of the total fossil record for this member). Over 100 specimens have also been collected from members C (2.94-2.54 Ma), F (2.32-2.29 Ma), and lower G (2.27-2.06 Ma). Sampling of turtles is also sufficient for the oldest members A (3.60-3.44 Ma) and B (3.44-2.94 Ma), with pelomedusid specimens recovered from every member from B through upper G (2.06-1.91 Ma). However, trionychid fossils complete enough for body size estimation have only been collected from four of those members, B, E, F and upper G. Only 1-2 turtle specimens are known from each member from H (1.91-1.78 Ma) through L (1.38-1.09 Ma). Abundances of crocodylians collected vary greatly across members, with 572 specimens from Member E, but zero from A and under 10 for B, D (2.53-2.39 Ma), H, and J (1.76-1.56 Ma). No occurrences of Euthecodon are recorded from members A and D, while Crocodylus is absent from members A, B, and H. Cf. Mecistops specimens are known from members C, F through J, and L.

Body size estimation

We estimated body length and body mass for each specimen and determined the maximum reconstructed body mass for turtle and crocodylian groups in each member of the Shungura Formation. For turtles, we used carapace length as the metric for body length. For incomplete specimens, we estimated carapace length using linear regressions from other element measurements based on complete specimens (Appendix 1). We used clade-specific linear regressions to estimate body mass from these carapace lengths. These linear regressions (Table 1) are derived from the data in Regis & Meik (2017).

For Python, we used the regression for total body length from the width between the prezygapophyses from ­McCartney et al. (2018); this regression is based on a sample of extant snakes, and the equation is ln(Total Body Length) = 1.095*ln(x) + 4.528, where x is the trans-prezygapophyseal width in mm. Then, we used the regression for body mass from total length for Pythonidae ‌Fitzinger, 1826 from Feldman & Meiri (2013).

For crocodylians, we used regressions relating skull length to total body length or body mass for extant taxa (Webb & Messel 1978; Hutton 1987; Thorbjarnarson 1988; Verdade 2000; Sereno et al. 2001; Iijima et al. 2016). For incomplete specimens, we estimated skull length using regressions from postcranial or mandibular measurements (Appendix 2), or, in the case of vertebral specimens, estimated total length using the regressions from vertebral centrum length of Iijima & Kubo (2020; Appendix 2).

For cf. Mecistops, identifiable cranial fragments most frequently comprise the anterior premaxilla and maxilla or the mandibular symphysis of the dentary. We relate the measurement of the narrowest width of the premaxilla, across the diastema between the premaxillary and maxillary alveoli, to the total skull length, based on the ratio observed in the complete specimen NME L 398 2508A (Appendix 2). Similarly for mandible specimens, we estimate skull length using the ratio between the length of the mandibular symphysis and skull length observed in NME L 398 2508A and 2508B (Appendix 2). Because Shungura specimens identified here as cf. Mecistops have broader snouts relative to skull length than either of the two extant species of Mecistops (M. ‌cataphractus and M. ‌leptorhynchus; Brochu et al. personal communication), this method for total skull length measurement is better practice than estimating from a regression on modern specimens (for reference, the equation SL = 10.719*x + 10.225 relates the narrowest premaxilla width of five specimens of Mecistops housed at the Natural History Museum, London to total skull length).

For Euthecodon, most specimens from the Shungura Formation are rostral fragments. We developed an equation relating the spacing between maxillary or mandibular tooth alveoli to total skull length (Fig. 2). We excluded measurements between the anteriormost two alveoli because those teeth are spaced further apart than those along the rest of the snout, which may be due to ontogenetic exclusion of the second premaxillary tooth by the large third tooth, which occurs early in ontogeny, leaving four premaxillary teeth (Brochu 2021). We then estimated total body length for Euthecodon based on the relationship between dorsal cranial length (DCL) and total body length (TL) for the extant tubulirostrine crocodylian Gavialis ‌gangeticus, ‌TL = –69.369 + 7.4*(DCL) (Sereno et al. 2001).

Crocodylus cranial specimens from the Shungura Formation have intermediate snout lengths, with similar snout to skull length ratios to the extant saltwater crocodile C. ‌porosus, so we used the equation trained on that extant species from Sereno et al. (2001): TL = –20.224 + 7.717*(DCL). Using these two different equations to estimate total body length takes into account the difference in the proportion of body length made up by the skull in crocodylians with varying snout morphologies. We used a single regression equation (Table 1) to estimate body mass from body length (Slavenko et al. 2016). No species-specific body mass regressions have been published for extant tubulirostrine crocodylians, but the mass regression for general Crocodylia performed consistently and conservatively in estimating crocodylian mass in comparison to species-specific regressions (Webb & Messel 1978; ­Chabreck & Joanen 1979; Hutton 1987; Thorbjarnarson 1996).

Mammal and environment data sources

We compiled mammal body size and environmental proxy datasets from the literature to compare with the trends of Shungura Formation maximum reptile size, including maximum size for mammalian taxa, reconstructed lake level in the Turkana Depression, paleosol carbon and oxygen isotopes, faunal composition of mammalian herbivores, and herbivore carbon and oxygen isotopes. We used member-level stratigraphic resolution for all data.

We took estimates of maximum mammalian body masses for each Shungura member using the methods and data in Bibi & Cantalapiedra (2023), which use loglinear regressions between tooth length and body mass trained on extant species with masses in the Pantheria database (Jones et al. 2009). We applied regressions specific to each order of mammals and each tooth position (Bibi & Cantalapiedra 2023). We applied these regressions to obtain mass estimates for all of the Shungura specimens in the Mammal Dental Metrics Database (Bibi 2023). Because this database includes no proboscidean specimens from the Shungura Formation, we avoided the issues with order-specific regressions for ­Proboscidea noted by Bibi & Cantalapiedra (2023). We also followed their method for removing duplicates by averaging mass estimates for individual specimens for which multiple teeth were measured. From the dataset, we removed specimens that were not attributable to a specific Shungura member, and subsequently took the maximum mass from each group in each member. For Equidae ‌Gray, 1821, most specimens were described prior to the differentiation of Member G into upper/lower units (Hooijer 1975); we assigned these specimens to lower G, to which the majority of the terrestrial mammal specimens in Member G date (Bibi 2023). The bovid tribe Tragelaphini was well-sampled across members, unlike other tribes, so we also tested for correlation with its maximum sizes. We excluded Rhinocerotidae ‌Gray, 1821 due to low sampling in this dataset. For Hippopotamidae ‌Gray, 1821, we instead used unpublished mass estimates based on astragali measurements (522 specimens measured by JRB) using Martinez & Sudre (1995)’s equation for terrestrial cetartiodactyls. They belong to three distinct lineages: aff. Hippopotamus ‌protamphibius ‌(Arambourg, 1944) and its likely successor aff. Hippopotamus ‌karumensis ‌(Coryndon, 1977) (see Harris 1991), the dominant hippopotamus lineage in the Shungura Formation, which is endemic to the Turkana Depression; aff. Hippopotamus ‌aethiopicus ­­Coryndon & ­Coppens, 1975, a pygmy hippopotamid (Coryndon & Coppens 1975) possibly related to aff. Hip. protamphibius/karumensis but of much smaller size, known only in the upper part of the sequence (from Member G to Member L); and Hippopotamus sp., less frequent that the other hippopotamids in the Omo Valley during the Plio-Pleistocene and distinguished by particular massive and wide astragali differing from the more slender build of the postcranials belonging to the other lineages. Today Hippopotamus shares semi-aquatic habitats with crocodylians and turtles. The largest Shungura specimens belong to Hippopotamus sp. and to aff. Hippopotamus ‌karumensis, which also displays morphological features (high orbits) and a geochemical signature (δ18O; Harris et al. 2008) indicating semiaquatic behavior.

We used the reconstruction of lake levels in the Turkana Depression by Nutz et al. (2020), which is based on study of the sedimentary formations of the Nachukui Formation, south of the Shungura Formation on the western bank of Lake Turkana. This lake level record is suited for comparison to ecological and environmental changes in the region, although lake level change is at least partially diachronous between the Shungura and Nachukui Formations (Lepre 2014). We used member average lake level values on a relative scale based on the curve presented in figure 15 of Nutz et al. (2020). Differential patterns of lake extent between the west and north shores of paleo-lakes in Turkana are minimal relative to the temporal durations included in these averaged member ages. For an additional metric of Shungura Formation-specific lake extent, we quantified the depositional environments of each member using the coding provided by Heinzelin (1983), which scored each unit’s facies in the following scheme: 0 for ephemeral streams, 1-4 for fluviatile facies (from channel to levee, floodplains, and swamps), 5 for mudflats, 6 for deltaic, and 7-8 for lacustrine (nearshore to deeper water). For each member, we used the maximum score in this scheme across units as a metric for the presence of lake conditions in the member.

Paleosol isotope data for both carbon and oxygen comes from the measurements of pedogenic carbonates compiled by Levin et al. (2011). This dataset includes 49 samples of Shungura paleosols dated between 3.2-1.18 Ma. We calculated the average δ18O and δ13C values for each member. We additionally drew on published data for mammalian tooth enamel δ18O and δ13C, based on isotopic ratios measured for over 1,000 dental specimens from nine herbivore families by Negash et al. (2020). We use member-average values for all specimens as a metric for the dietary composition of all mammalian herbivores, derived from a subsample of this dataset only including stable carbon isotopic values of specimens for which element identification is documented in the Omo Database, and attributed to a M2 or M3 tooth.

We used two Shungura Formation mammalian faunal composition datasets. Bobe & Behrensmeyer (2004) defined a subset of mammals that they took to be grassland indicators, including bovids in the families Alcelaphini ‌Simpson,1945 and Antilopini ‌Gray, 1821, the suid Metridiochoerus ­­Hopwood, 1926, Equus ‌Linnaeus, 1758, and Theropithecus ‌oswaldi ‌(Andrews, 1916). We used the proportion of mammal fossils from each member falling in this group as a metric for grassland habitat, although these raw fossil abundances may not accurately represent the relative abundances of taxa in the true paleocommunity due to filtering in taphonomy and collection biases (Bobe & Behrensmeyer 2004). However, with relatively constant taphonomic conditions and collections procedures, changes in abundance values across time bins represent shifts in the relative proportion of taxa, regardless of their absolute abundances. The second dataset (Bobe et al. 2007) provided an alternative faunal proxy for the presence of grasslands, measuring separately the percentage of mammal fossils from each member belonging to reduncin bovids and bovids in the clades Antilopini, Alcelaphini, and Hippotragini (“AAH bovids”). Reduncin bovids indicate open but moist habitats with fresh grass, while the AAH bovids today live only in at least seasonally dry grasslands or bushlands (Levin 2015).

Finally, we used two numeric estimates of paleoenvironmental variables sampled across Shungura members. Passey et al. (2010) provided estimates of soil temperatures based on measurements of clumped isotopes in pedogenic carbonates. Hernández Fernández & Vrba (2006) reconstructed mean annual precipitation at sites including individual Shungura members based on models of mammal community structures trained on modern sites. Figure 3 summarizes all data series used as paleoenvironmental proxies to which we compared the maximum body size time series.

Correlation analyses

To test for coordinated patterns of maximum size change over time between groups and with environmental variables, we used correlation tests. For all body mass variables, we used log10 transformation of mass estimates in kilograms. We employed standard Pearson product-moment correlation tests (rcorr function from the R package Hmisc). We also ran Spearman’s rank correlation tests, appropriate for non-normal data, for comparisons including data series whose values were not normally distributed. We also tested for correlation between maximum size in each reptile group and fossil abundance by member for all reptiles and for only turtles or crocodylians. Abundance counts came from the catalog of specimens identified to these groups collected in the Shungura Formation between 1967 and 2023. These correlation tests reveal whether maximum size is higher in members where reptiles are more frequently preserved or more intensely sampled.

Because we ran many tests for pairwise correlation between the same set of data series, it was necessary to correct for multiple comparisons (Wright 1992). For each set of correlation tests whose results are presented in Tables 2-6, we used sequential Bonferroni correction to adjust the significance cutoff for the resulting p-values. This correction takes into account that some of a large set of pairwise statistical tests will have p-values under 0.05 by chance, rather than due to true significance. It adjusts the significance level for the pairwise test with the lowest p-value by dividing it by the number of tests, then adjusts the significance level for the next lowest p-value to (0.05 / n(comparisons) –1), and so on, until the next lowest p-value does not fall under the adjusted significance level (Holm 1979).

It is important to note that correlation tests do not take into account the order of the data points and, therefore, cannot test hypotheses of causal effects of one variable on another over time. Tests differentiating correlative and causal relationships between time series require more data points than are available from sampling at the member level within the Shungura Formation (Sugihara et al. 2012; Reitan & Liow 2019). However, we were able to test whether two time series change together between members using correlation tests between the first differences of the series’ values. These tests use the value of a variable in one member subtracted from its value in the next member, calculated iteratively for each data series. Correlation between first differences shows that the magnitude and direction of change between bins for two series is coordinated across the time series. The results for first difference correlation are more informative than the correlation tests on raw values because first differencing detrends the data series; if independent directional trends exist in two series over time, significant correlation in raw variable values is likely even without any causal relationship between the variables.

Abbreviations

AAH bovids Antilopini, Alcelaphini, and Hippotragini;

DCL dorsal cranial length;

IORE International Omo Research Expedition;

NME/EHA National Museum of Ethiopia/Ethiopian Heritage Authority (ex-Authority for Research and Conservation of Cultural Heritage);

OGRE Omo Group Research Expedition;

TL total body length.

RESULTS

Size estimates

While some body size estimates reported here fall within the range of local extant faunas, there are extraordinary size maxima present in the Shungura Formation. For crocodylians (Euthecodon) and tortoises, the Turkana Depression is home to the largest representatives of those groups anywhere in the world during the Early Pleistocene. Notably, the estimated mass of Euthecodon sp. from Shungura Member L of 2 300 kg makes it the largest Early Pleistocene reptile globally. The large body size of ­Euthecodon in the Turkana Depression contributes to a global trend of increased mean crocodylian body size during the Plio-Pleistocene (Godoy et al. 2019; Godoy & Turner 2020). The high diversity of crocodylians occurring in Turkana during this period indicates the availability of multiple distinct niches, including macro-predatory ones, for these semi-aquatic carnivores within a large lake system (Scheyer et al. 2013; Drumheller & Wilberg 2020). Of the specimens we attribute to Crocodylus, the largest has an estimated mass just over 1 000 kg (Member F), while the largest cf. Mecistops measured is slightly smaller, at an estimated 860 kg (Member L).

The in situ tortoises from Shungura Members E and H, with preserved carapace lengths of 110 and 100 cm, respectively, are the largest known testudinids from the Early Pleistocene of Africa. Tortoises from 1-1.7 m in length are present continuously in eastern Africa from the Early Miocene through the Middle Pleistocene (de Lapparent de Broin 2000). Thereafter, these massive tortoises survived only in Madagascar into the Holocene (Bour 1984). Freshwater turtles in Africa have maintained a maximum body size of 40-60 cm since the Oligocene (Meylan et al. 1990; Pérez-García 2019). Most soft-shelled turtle specimens (Trionychidae ‌Gray, 1825) from the Shungura Formation fall within this range, but the largest partial specimen, from Member B, is larger, with an estimated carapace length of 71 cm. Modern species of Pelusios range in size from 12-55 cm carapace length (Itescu et al. 2014), so the Shungura Formation pelomedusids, whose maximum sizes center around 30 cm, fall around the middle of this range. Similar to Euthecodon and cf. Mecistops, the largest pelomedusid specimen is from Member L, with a carapace length of 42 cm.

Size estimates for Python are consistent with maxima of extant African P. ‌sebae ‌(Gmelin, 1789) and P. ‌natalensis ‌Smith, 1840 (e.g. Pitman 1974; Alexander 2018). A single vertebra from Member F (OMO 33-3613), with trans-prezygapophyseal width 52.1 mm, represents an estimated body length of 7 m, larger than reliable estimates for modern African pythons and equivalent to the largest verifiable lengths of any extant snake species (Murphy & Henderson 1997).

Size patterns through time

The three crocodylian groups measured (Appendix 4) have relatively similar body length estimates in the early Members of the Shungura Formation (452 cm for cf. Mecistops, ­Member C, 300-500 cm for Crocodylus from members A-D, and c. 600 cm for Euthecodon from Members B-F). All three groups increase in size in Member F, where the largest Crocodylus specimen, a robust partial skull (OMO 221-1973-2716) was found. From 2-1 Ma, the maximum size pattern for Crocodylus diverges from that of Euthecodon, as Euthecodon increases in size into Member L, while Crocodylus body length decreases to under 5 m in members K and L. In the final member, cf. Mecistops (OMO 341-10040) exceeds the Crocodylus in size, which is not observed outside of Member L. Differences in postcranial morphologies, which could indicate how mass per body length varies between groups, have not been studied in these fossil taxa; therefore, we applied the same length to mass regressions for all three groups. Such unknown variation in mass per length between crocodylians makes trends within each of these groups over time more reliable than absolute differences between groups (Fig. 4).

Testudinids have estimated size maxima over 50 kg in all members where they are recorded: B, C, D, E, G, and H (Appendix 3). Tortoises are figured as absent in Member F because no specimens complete enough for size estimation have been collected; the only tortoise material known from F is small fragmentary material from an excavation in unit F-0 (2.34-2.321 Ma). In lower G, the only known tortoise specimen comes from unit G-13 (2.062-2.072 Ma), so there is an apparent c. 200 000 year gap in the presence of large terrestrial turtles during that interval, despite high sampling of reptiles in members F and G (Fig. 4).

Aquatic turtle maximum size is greatest in the members where trionychids occur – B, E, F, and upper G – with the overall largest aquatic turtle estimated at 28 kg in Member B. Pelomedusid maximum size remains approximately constant across the Formation, with its minimum occurring at around 2 kg in upper G and its maximum of 6.8 kg in Member L.

The largest estimated body masses for Python come from the middle of the Shungura sequence, in Member F and upper G (Appendix 5). Python maximum size is lowest, under 3 m in length (or 10 kg in mass), in Members B, E, and J. In the remaining members, typical Python maximum length is between 4-5.5 m.

Correlation with sampling

Maximum body size across the Shungura Formation is not correlated to reptile sampling across members (Fig. 4). Table 2 lists correlation coefficients testing for relationships between maximum size in each group and the number of occurrences per member of all reptiles, crocodylians, and turtle specimens, as well as correlation between maximum size and member age. No significant correlation is observed with time; maximum size in all groups both increases and decrease across the sequence. Crocodylus has near-significant positive correlation with crocodylian sampling. However, there is a highly negative correlation coefficient for the relationship between trionychid maximum size and turtle sampling.

Correlation between groups and with paleoenvironmental variables

No tests for pairwise correlation between reptile groups’ sizes show result in significant correlation after correction for multiple comparisons (Table 3). The observed correlation coefficient between the maximum sizes of terrestrial tortoises and Python (0.997) is the highest. High negative correlation coefficients are observed between trionychid turtle size and the size of both Crocodylus and terrestrial tortoises.

After correction for multiple comparisons, none of the relationships between reptile size and paleoenvironmental or mammalian faunal variables are significant. The sequential Bonferroni correction reduces the significance cutoff for correlation tests within Table 4 (comparing reptile and mammal sizes) to 0.0009; none of the observed p-values are below this threshold which applies to the pairwise test with the highest significance. Therefore, the correlations denoted with * in Tables 4-6 do not exceed the number that are expected to exhibit that level of correlation by chance alone. However, they do identify which reptile and mammal taxa show the relatively strongest covariance in maximum size over time. The highest correlation coefficient in Table 4A is for a negative relationship between maximum size of Crocodylus and aff. Hippopotamus ‌aethiopicus (the pygmy hippos), while both cf. Mecistops and Python maximum size are positively related to aff. Hip. aethiopicus. The next strongest relationship is positive, between size in pelomedusid turtles and Hippopotamus sp.

No significant correlation is observed between change between members in maximum size of reptiles and mammals (Table 4B). The highest positive correlation coefficient there is between aquatic turtles and aff. Hippopotamus ‌protamphibius/karumensis.

Among the comparisons with paleoenvironmental variables (Table 5), there is one correlation test that is significant after sequential Bonferroni correction: a positive relationship between the maximum size of aff. Hippopotamus ­protamphibius/karumensis and paleosol δ18O. Among the other correlation tests with high correlation coefficients, paleosol δ18O displays a positive relationship to size in both Pelomedusidae ‌Cope, 1868 and Hippopotamus sp. Pelomedusid size is negatively related to paleosol δ13C. Three groups display high negative correlation coefficients with lake level in Table 5A: pelomedusids, Hippopotamus sp., and Equidae. Equid size is also negatively related to the depositional environment lake scores of Heinzelin (1983). Finally in Table 5A, soil temperature estimates (Passey et al. 2010) are negatively related to the size of Crocodylus and aquatic turtles and positively related to tragelaphin bovid size.

Table 5B shows correlation in change between consecutive members in size series and environmental proxies. There, the highest observed positive correlation coefficients are between size in Python and paleosol δ18O, and between aquatic turtle size and lake level (Fig. 8). The greatest negative correlation coefficient is between pelomedusid size and δ13C.

There are several groups whose maximum size is nearly correlated with mammalian faunal environmental proxies (Table 5A), though not after Bonferroni correction. Mammalian herbivore δ18O is positively related to body size in ­Cercopithecidae ‌Gray, 1821, Pelomedusidae, and Hippopotamus sp. aff. Hippopotamus ‌protamphibius/karumensis size is higher when the AAH % indicating open-habitat bovids is also high. Also, aff. Hip. protamphibius/karumensis size is strongly positively related to the grassland indicator proportion of Bobe & Behrensmeyer (2004). Mean annual precipitation from Hernández Fernández & Vrba (2006) is negatively related to size in Cercopithecidae.

As for correlation in change between members (Table 6B), we observed high negative correlation coefficients between mammalian herbivore δ13C and size in aquatic turtles and aff. Hip. protamphibius/karumensis. The relationship to herbivore δ13C was positive for size in aff. Hip. aethiopicus. Pelomedusid size increased when herbivore δ18O increased (Fig. 8). Changes in percentage of AAH bovids were negatively related to size change in Pelomedusidae, positively related to size change in aff. Hip. protamphibius/karumensis, and negatively related to size change in aff. Hip. aethiopicus. The latter two relationships were reversed with respect to the Reducine bovid proportion; the smaller aff. Hip. aethiopicus increased in size as the moist/open-habitat-specialist reduncins became more numerous, while the larger endemic hippopotamid lineage increased in size as the more dry-adapted AAH bovids came to predominate. Aff. Hip. aethiopicus size change had a highly negative correlation coefficient vs. the proportion of grassland indicator mammals, as did size change in Pelomedusidae. Size change in testudinid tortoises had a correlation coefficient of 0.999 with change in the mean annual precipitation estimates. Pelomedusid size change was also positively related to those precipitation changes. Although none of these relationships are significant, each instance of high correlation indicates where two series change over time in a coordinated manner, possibly due to similar environmental factors affecting both.

DISCUSSION

Body size maxima for these groups are not artefacts of uneven sampling across members because there are no cases of significant positive correlation between size and member-specific occurrence counts of reptile specimens. The negative correlation between trionychid size and sampling increases our confidence that the absence of this taxon from some members is not due to insufficient sampling of turtles in those intervals. However, positive correlation between Crocodylus size and crocodylian sampling indicates that maximum size in this group, which includes size estimates based on postcrania that may have been inconsistently collected, may be partially related to sampling effort. Apart from this case, low correlation with sampling metrics suggests that sampling throughout the formation, while very uneven for reptiles, is not driving the size trends observed across members. Collection bias against reptiles does not appear to be size-selective. If anything, larger reptiles are generally more likely to be collected due to ease of identification relative to, for example, small squamate specimens (although we note that the very largest specimens, such as intact tortoise carapaces and one cranium of Euthecodon, are left in situ due to resource constraints). However, the very low sample counts, particularly in the most recent members, mean that the maximum sizes reconstructed there may not accurately reflect the actual maximum sizes in the paleocommunities of every member.

Maximum size trends in the crocodylians and turtles of the Shungura Formation point to local-scale environmental influences on body size in these reptiles. The two dominantly terrestrial reptile taxa, Testudinidae and Python, have high (though non-significant) correlation in change in size maxima between members (Table 3B). Figure 5 plots these size changes against paleosol carbon isotopes and mammalian herbivore faunal metrics, both providing information about the openness of vegetation over time. With respect to habitat preferences, extant tortoises are frequently found in open areas within tropical forest or savannas (Brattstrom 1961). Modern giant tortoises, which survive only on oceanic islands, have a grazing diet and help to maintain open habitats known as “tortoise turf” (Hansen & Galetti 2009; Falcón & Hansen 2018). However, it is unknown whether their ecologies shifted after colonization of those islands from continental mainlands (Hansen et al. 2010). The presence of large tortoises in Member C of the Shungura Formation, which is reconstructed as having a relatively higher tree cover compared to later members, suggests that these testudinids were not restricted to open habitats in the Pleistocene. However, their maximum size increased into Member H coincident with increasing paleosol δ13C values and the proportion of grazing mammals, which suggests that newly open environments were also then suitable for giant terrestrial tortoises, as they are today. Maximum size change in tortoises is positively correlated with change in the faunally-derived estimates of paleoprecipitation by member (Hernández Fernández & Vrba 2006; Table 6B). Across the sequence, tortoise size increased as mammalian herbivore faunas shifted away from being arid-adapted, a result which undermines the putative open-habitat association of tortoises, although increased evaporative water loss indicated by δ18O records may have maintained open habitats during intervals with higher estimated paleoprecipitation. Modern African testudinids require at least somewhat open habitats because they use basking behavior for thermoregulation, so we ­interpret this pattern as indicating that, in the Shungura sequence, tortoise size most likely increased during intervals with higher rainfall but also some open environments available.

Python is a habitat generalist within African ecosystems, and the relationship between size and environment is not well understood for its largest species. Here, the strongest environmental correlate for size change in Python is a positive relationship with pedogenic δ18O (Table 5B). Pedogenic δ18O is influenced by the isotopic composition of rainwater and increases to less negative values with more evaporative water loss from soils. δ18O changes over time with different sources of water input to the Omo River, including direct precipitation (low δ18O), and high levels of evapotranspiration increase δ18O. Therefore, this result represents a contrast between precipitation histories reconstructed by faunal proxies, which we observed to change with the tortoise and snake body size series, and isotopic proxies, which indicate Python size increase concurrent with shifts to more evaporative water loss from the landscape. Tests of association between tortoise and snake size and isotopic values at other eastern African sites could confirm whether the largest representatives of these groups preferentially inhabited wetter or drier environments.

Tortoises over one meter in length were present in the ­Turkana Depression from the Late Miocene (at Lothagam, Wood 2003) until the Middle Pleistocene (personal observations in West Turkana, Turkana Basin Institute Turkwel collections). However, no tortoises of this size are documented between 2.3 Ma and 2.1 Ma; this age range is well-sampled for turtles in the Shungura Formation, yet no large tortoises at all have been discovered in Member F or lower G below unit G-13 (though lower G has nearly 16 000 vertebrate remains collected, making it the richest sequence of the formation). The NME/EHA collections include very fragmentary material likely attributable to Testudinidae collected during a 1971-1973 excavation in unit F-0; while too fragmentary for use in any body size regression, these specimens appear to be of tortoises much less than 1 m in length. This temporal gap in the record of giant tortoises aligns with the timing of the Shungura Formation’s archaeological record, which is concentrated in Member F and lower G (noted in Fig. 5; Delagnes et al. 2011; Maurin et al. 2014). These stone tools indicate hominin occupations at several locations in the vicinity of the Omo River. The earliest presence of archeological lithic remains is in F-1, after which no tortoises have been recovered from Member F. The next tortoise in age appears in unit G-12. In situ lithics are known from unit G-13, but not thereafter, when tortoise size rebounds to over 1 m/100 kg (Member H, Fig. 5). In the Early Pleistocene and thereafter, there is evidence of consumption of tortoise meat by hominins (Stiner et al. 2000; Blasco et al. 2011; Klein & Cruz-Uribe 2016), with reduction in tortoise body sizes observed globally coincident with the spread of Pleistocene humans (Joos et al. 2022). Large tortoises are easy hunting for humans, and individually contain plentiful meat, so may have been preferentially butchered by humans, who display preferences toward catching larger prey (Ben-Dor & Barkai 2021). However, without any direct evidence of tortoise butchery in eastern Africa from this time period, we cannot currently determine whether the gap in the record of large tortoises in F and (most of) lower G relates to hominin occupation. Testudinid fossils from diverse localities await further study, from which a high-resolution regional time series could then be assembled to reveal what environmental conditions and interactions restrict tortoise size.

Among aquatic turtles, trionychids are consistently larger than pelomedusids (Fig. 6). Both trionychid and pelomedusid turtles today inhabit permanent bodies of water and derive their diets primarily from aquatic animals and plants (Akani et al. 2001), though the pelomedusid species Pelusios ‌niger, with a maximum carapace length of around 35 cm, comparable to many of the Shungura Formation pelomedusid specimens, incorporates a higher proportion of terrestrial vertebrate prey items than smaller-bodied species of Pelusios ‌(Luiselli et al. 2021). Extant Trionyx ‌triunguis has also been observed scavenging carcasses of herbivorous mammals, though fish and frog meat are much larger components of their omnivorous diets (Akani et al. 2001). With functional links to terrestrial and freshwater habitats, trait structure observed in terrestrial and aquatic turtles can inform environmental conditions where fossil communities lived (Conley & Samuels 2022; Parker et al. 2023).

Across the Turkana Depression, trionychid specimens of around half a meter in length have been collected at many Plio-Pleistocene sites. Extant Trionyx ‌triunguis which inhabits the Omo and Lake Turkana can reach over 100 cm in carapace length (Taskavak & Akcinar 2009), but body sizes in the 40-60 cm range reported here from the Shungura Formation would also be common in extant populations. To understand whether maximum body size in this species, or size distributions within populations, have changed over time, more comprehensive collection of trionychid fossil material is necessary. Past field collections have had a strong bias against collection of fragmentary turtle material, although it is relatively abundant. If the gaps in the trionychid record from all members except for Basal, B, E, F, and upper G are true absences, and not due to collection bias, then the lack of large trionychids could provide information about the depositional environments preserved from those ages.

The positive correlation coefficients between overall aquatic turtle maximum size and lake level (Table 5B; Figs 6; 8) suggest that large size in these turtles is dependent on habitat availability. Aquatic turtle size maxima are driven by the presence of trionychids in members B, E, F, and lower G, all of which sample lake high stands. Where lacustrine environments in the Lower Omo Valley had smaller volumes, large-bodied trionychids may have avoided those areas in favor of other areas of the Turkana Depression where lake volume or food resources were higher. In contrast, pelomedusid body size displays negative correlation with lake level, potentially suggesting that the smaller-bodied aquatic turtle group was more successful in smaller water bodies (Table 5A). Trionychid body size is also negatively related to the size of terrestrial tortoises and Crocodylus, the latter of which is a possible predator of turtles. Pelomedusid size change, while not dramatic across the sequence, also shows links to terrestrial environmental proxies. It is negatively correlated with paleosol δ13C (Table 5B) and shows positive correlation with herbivore δ18O and the faunal precipitation estimates (Table 6B). Again, these non-significant correlation results present mixed signals as to whether reptile size maxima relate to more open/arid environments and mammal faunas, or the opposite. However, the relationship between δ18O and pelomedusid size (Figs 6; 8) indicates that this trait is responding to changes in regional hydrology. The changes in sources of water input reflected in soil δ18O could correspond to differences in nutrient availability or water temperature relevant to these aquatic turtles’ niches. Overall aquatic turtle size is higher in members where Passey et al. (2010)’s soil temperature estimates are lower.

The hippopotamid lineage aff. Hip. protamphibius-­karumensis has size maxima significantly correlated to paleosol δ18O (Table 5), providing evidence that these semi-aquatic mammals were responding to regional hydrologic changes. Additionally, this hippopotamid lineage’ size maxima occurred concurrently with higher AAH percentage in mammal ­communities, which reflects the abundance of dry grass-adapted bovids (Table 6). This indicates that these hippopotamids increased in size during intervals with higher evaporative water loss and coincident with the spread of grassier habitats. This echoes the proposed relation between C4 grassland developments and the success of hippopotamines (Boisserie et al. 2011), and the dental enamel δ13C results from Harris et al. (2008) indicating a nearly pure C4 diet for this lineage. However, with respect to δ13C, the lineage aff. Hip. protamphibius-karumensis has a negative correlation coefficient in our results, while the smaller aff. Hip. aethiopicus has high size maxima in more C4-enriched members (Table 6B). Such opposite relationships for these two hippopotamid groups are also seen in correlation with overall mammal herbivore δ18O (Table 6B), suggestive of niche partitioning between the hippopotamids. Hippopotamus sp. is not consistently present in the Shungura sequence, but where it is, size maxima are negatively related to lake level and positively related to paleosol δ18O (Table 5A). This result suggests that the third group of hippopotamids, who are generally present in the region during lake high stands, seem to decrease in size when lake area expands and evaporation decreases.

Another semi-aquatic group whose evolution to large size is of interest in the Shungura Formation are the otters, a very large (c. 200 kg) species of which was present in members B and C and which may have competed with crocodylians for aquatic or terrestrial prey (Grohé et al. 2022). The gigantism observed in both Euthecodon and this otter suggests the presences of high resource and habitat availability for aquatic vertebrate carnivores in the Omo Valley during several intervals of the Shungura Formation’s deposition. In particular for Euthecodon, large aquatic range sizes would be necessary to support a population of these extremely large carnivores (Scheyer et al. 2013). We propose that these body size increases are due to indeterminate growth exhibited by these reptiles; in permissive habitats with abundant food and/or limited mortality, individuals reached larger sizes.

The three crocodylian morphotypes considered, the extremely tubulirostine Euthecodon, Crocodylus, and cf. Mecistops, have both similarities and differences in their maximum size changes over time (Fig. 7). Both Euthecodon and cf. Mecistops attain their overall size maxima in the most recent Shungura member. Both Crocodylus and Euthecodon increase in size from members E-F. From Member F onwards, there is an apparent decrease in Crocodylus size. Specimens from members B, D, and F (OMO 28-1968-3213, L 40-29, and OMO 221-1973-2716) are all extremely robust cranial fragments, likely comparable to C. ‌thorbjarnarsoni ‌(Brochu & Storrs 2012), while specimens from later members have unresolved taxonomy and attain smaller size maxima. Looking at high-resolution temporal sequences like this with multiple crocodylians present can help untangle dynamics of their niche partitioning and faunal turnover in the context of paleoenvironments and prey species presence (Gardin et al. 2024). In our tests, there were few instances where high correlation coefficients were observed between crocodylian size maxima and environmental proxies. Crocodylus size showed negative correlation with the size of terrestrial tortoises and aff. Hip. aethiopicus (Fig. 8), as well as negative correlation with estimated soil temperatures. The latter is of interest in light of the theoretical expectation that ectothermic reptiles can attain the largest body sizes only at high temperatures (Head et al. 2009; Parker et al. 2023).

The assembly of high-resolution time series datasets for vertebrate traits through particular stratigraphic sections like this one enables the testing of hypotheses of differential responses to environmental change. The use of regressions like those deployed here (see Appendices) based on isolated skeletal elements makes estimating body mass tractable for large sample sizes of reptile specimens. In thoroughly-studied depositional contexts like the Turkana Depression, comparison of time series data from vertebrate, invertebrate, plant, and geochemical metrics can identify processes influencing evolution in both community composition and individual taxa’s traits. Beyond correlation tests, future studies with only a small increase in the number of time steps used for sampling can test for causal relationships between such variables (Sugihara et al. 2012; Reitan & Liow 2019). Such analyses have previously been carried out mainly in marine settings, but their application to temporal sequences in the terrestrial fossil record has high potential to reveal how co-evolution and trait-based responses to environmental change have occurred in the past (Liow et al. 2015; Hannisdal & Liow 2018; ­Reitan & Liow 2019; Lidgard et al. 2021). Identifying such trait changes within faunas can, in this case, provide information about habitat shifts that occurred where early humans lived, and more broadly, allow us to build predictive models of faunal responses to future climate changes (Polly et al. 2011; Parker et al. 2023).

CONCLUSION

We situated reptile body size histories in their environmental context through time in the Shungura Formation in order to characterize the abiotic and biotic conditions underpinning body mass changes in these groups. The observed patterns of maximum body size through time in relation to environmental proxy data demonstrate how local environmental shifts can drive body size evolution. While correlation tests were non-significant, we identify several reptile size-environment relationships, both matching and challenging previous expectations, that can be developed as proxies through testing in other contexts.

Several of the reptile groups studied display relationships between their body size and estimates of paleo-temperature, paleo-precipitation, and vegetation openness, all of which could be clarified through comparison to other sequences in the Turkana Depression. The presence and size of aquatic trionychid turtles is positively related to lacustrine conditions. Both crocodylians and hippopotamids also attain size maxima during lake high stands. The associations between body size in these reptile groups and habitat that we identify here can serve in future to assist in reconstructing climate and vegetation patterns in the eastern African record, if they prove generalizable to other sedimentary sequences. Reptile size is simple to measure and the previously-understudied fossil record of turtles, snakes, and crocodylians can be tapped into as a proxy to inform habitat availability in both terrestrial and aquatic habitats.

Acknowledgements

We are grateful to the National Museum of Ethiopia/­Ethiopian Heritage Authority (ex-ARCCH, Ministry of Tourism) for allowing to access collections and perform research. We deeply thank the NME staff (T. Getachew, S. Melaku, and G. Tekle Yemanebirihan) for guidance and support with collection study and Blade Engda Redae for assistance with specimen numbers and databasing. We thank the many members of the Omo Group Research Expedition for helpful discussion that contributed to the content of this paper. We thank F.K. Manthi and E. Ndiema (National Museums of Kenya) for access to comparative specimens. This work was funded by a Herchel Smith Fellowship, Emmanuel College Panton Trust grant, and a Cambridge University Worts Travelling Scholars grant to A.K.P., NERC NE/W007576/1, NSF 2124836, and a Cambridge Africa ALBORADA grant to J.J.H. and J.M., with additional support from the Integrated Climate Change Biology and Conservation Paleobiology in Africa programmes of the IUBS to J.J.H. We are deeply indebted to the hundreds of people who participated to the fieldwork missions of the IORE, of the OGRE and of other research programs, who managed collections and databases, who prepared specimens, who contributed to their study, who provided financial support, who helped with administrative processes, and who provided advice and moral support. The OGRE is a joint program of PALEVOPRIM, the CFEE and the EHA principally funded by the Ministry of Europe and Foreign Affairs, the National Research Agency, the Région Nouvelle-Aquitaine, CNRS INEE, PALEVOPRIM, and the Fyssen Foundation. The OGRE is extremely grateful to the EHA, the SNNPR, the South Omo Zone, the Nyangatom and Dassanetch Weredas and their people for their help and reception. We would also like to thank the editor-in-chief, Michel Laurin, the associate editor, Aurélien Mounier, and the reviewers for their work on the manuscript.

Data Availability

All data and code used are available at: https://doi.org/10.5281/zenodo.18403027

APPENDICES

Appendix 1Appendix 1. — Linear regression equations used to estimate straight line carapace length for turtle specimens. Abbreviations: CL, carapace length; EHA, National Museum of Ethiopia/Ethiopian Heritage Authority; KNM, National Museums of Kenya; NHML, Natural History Museum (London); PPHM, Panhandle Plain Historical Museum; UMZC, Cambridge University Museum of Zoology.TaxonMeasurement, x (cm)Regression to CL (in cm)Regression trained on N =Sources for regression data Trionychidae Skull lengthCL = 3.5595*x-5.38693UMZCHyohypoplastron maximum lengthCL = 2.25*x2KNM-LT 28483, EHA OMO 229Costal width at midline (average)CL = 10.051*x -2.982710NHML, KNM, & EHA collections Pelomedusidae Plastron lengthCL = 0.985*x1KNM-WS 14376Midline length of posterior lobe of plastronCL = 4.2047*x -54.5575OMO 57/5-1972-324, OMO 38-1968-3640, L 3-10, F 164-10079, F 256-10009, L 182-100002 Testudinidae Humerus lengthCL = 1.7523*x + 49.7033NHML & Hay 1908Maximum width of acetabulumCL = 13.32*x2NHM 3097, PPHM 1534Midline length of neuralCL = 5.234*x1KNM-FM 21225

Appendix 2Appendix 2. — Linear regression equations used to estimate skull length for crocodylian specimens; total body length for vertebral regressions from Iijima & Kubo (2020). * measurement for final two rows, vertebral centrum length, is in mm. Abbreviations: KNM, National Museums of Kenya; N, number of specimens; NHML, Natural History Museum (London); SL, skull length; TBI, Turkana Basin Insitute; TL, total body length.TaxonMeasurement, x (cm*)Regression to SL (in cm) and TL (in mm)Regression trained on N =Sources for regression data Euthecodon Mandible lengthSL = 0.8754*x + 1.21134KNM-ER 757, KNM-KP 18330, KNM-ER 8260, KNM-LT 26306Mandible length to posterior of dentary articulationSL = 0.8754*x + 1.21135NMK-ER 757, KNM-KP 18330, KNM-ER 8260, KNM-LT 26306, KNM-KP 66227Average tooth spacing (excluding two anteriormost tooth sockets)SL = 19.966*x + 30.4464KNM-ER 757, KNM-LT 26306, KNM-ER 8260, KNM-ER 18330cf. MecistopsNarrowest width of premaxillaSL = 9.97*x1NME L 398 2508ALength of mandibular symphysisSL = 3.8706*x1NME L 398 2508A Crocodylus Mandible lengthSL = 0.6829*x +6.438123NHML collectionsMandible length to posterior of dentary articulationSL = 0.8206*x + 4.579119NHML collectionsLength of mandibular symphysisSL = 3.1846*x + 17.78920NHML collectionsHeight of dentary at posterior end of mandibular symphysisSL = 13.158*x2NME AL 126-11, L.449-4 (Hadar, Shungura Crocodylus)Length between centers of d9-d12 alveoliSL = 5.1384*x + 8.31818NHML collectionsFemur circumferenceSL = 4.8793*x + 1.28574NHML collectionsD3-D10 vertebral centrum lengthlog(TL) = 1.906 + 0.993*log(x)5Iijima & Kubo 2020C8-D2 vertebral centrum lengthlog(TL) = 2.015 + 0.956*log(x)2Iijima & Kubo 2020

Appendix 3Appendix 3. — Specimen data of Testudines by taxa for specimens with maximum body mass estimates for each member of the Shungura Formation.TaxonShungura MemberMidpoint Age (Ma)Specimen numberMeasurementDimension (cm)Carapace length (cm)Mass reconstruction (kg) Trionychidae Basal3.672OMO 80-1974-903Costal width4.744.314.6B3.188L 729-10005Costal width7.471.427.7E2.356OMO 57/5-10025Costal width4.340.212.8F2.305OMO 306-1976-386Costal width4.845.315.0upper G1.986F 164-NC3Carapace length585820.9 Pelomedusidae B3.188OMO 28-1967-960Carapace length33.533.53.54C2.734OMO 3/1-10038Length of posterior lobe of plastron21.837.14.74D2.457L 824-13Carapace length33.133.13.42E2.356OMO 38-1968-3640Carapace length35354.01F2.305L 182-10002Carapace length34.534.53.85lower G2.167OMO 257-1973-5315Carapace length36364.35upper G1.986F164-10079Carapace length27271.91L1.239OMO 346-10082Length of posterior lobe of plastron2342.26.83 Testudinidae B3.188OMO 28-1967-958Neural length15.480.666.1C2.734L 823-3Plastron length103108.9151.3D2.457[measured in field]Maximum width of acetabulum6.687.983.9E2.356OMO 70-NC1Carapace length110110155.6lower G2.167L 626-105Neural length16.284.876.0upper G1.986F 164-10068Humerus length15.476.757.7H1.84OMO VE 3-NC1Carapace length100100119.7

Appendix 4Appendix 4. — Specimen data of Crocodylidae by taxa for specimens with maximum body mass estimates for each member of the Shungura Formation. For Crocodylus ‌Laurenti, 1768, measurement type “Dentary height” refers to the height of the mandible at the posteriormost level of the mandibular symphysis. For cf. Mecistops, “narrowest width” refers to the width of the narrowest part of the premaxilla, posterior to the premaxillary alveoli. For Euthecodon ‌Fourtau, 1920, the measurement “Tooth spacing” is the average distance between the center of alveoli (dentary or maxilla) for all alveoli present in each specimen, excluding the anteriormost two.TaxonShungura Formation Member Age (Ma) Specimen numberMeasurement Dimension (cm) Estimated skull lengthLength reconstruction (cm) Mass reconstruction (kg) Crocodylus A3.517OMO 127-1973-4466Length between d9-d12 alveoli7.245.3329.4226.1B3.188OMO 28-1968-3213Dentary height5.167.1497.6714.9C2.734L 449-4Skull length52–381.1339.6D2.457OMO 5/2-1967-617D3-D10 vertebral centrum length4.3–342.8252.8E2.356L 40-29Dentary height5.673.7548.4937.7F2.305OMO 221-1973-2716Dentary height5.977.6578.91080.4lower G2.167OMO 310-1976-549C8-D2 vertebral centrum length5.5–477.3636.5upper G1.986OMO 2-1967-209Length between d9-d12 alveoli11.768.4507.6755.9J1.658OMO 358-10043D3-D10 vertebral centrum length6.6–516.2792K1.458P 995-1aSkull length53–381.1339.6L1.239OMO 393-10133Femur circumference11.557.4422.7453.6C2.734OMO 3-1967-910Length of mandibular symphysis15.861.1451.7545.8cf. MecistopsF2.305L 398-2508ASkull length65.8–487.6675.4lower G2.167OMO 150-1972-1Width of narrowest part of snout5.554.8402.9396.8upper G1.986OMO 372-10016Narrowest width6.160.8449.1537.1J1.658OMO 358-10041Narrowest width5.351.8379.9336.6L1.239OMO 341-10040Length of mandibular symphysis18.571.6532.3863.2 Euthecodon B3.188L1-151Tooth spacing3.1292.7616.91302.3C2.735OMO 18-1968-3215Tooth spacing3.5103.16731660.6D2.547L 64-34Tooth spacing3.1593.3621.31328.6E2.356OMO 38-1973-4629Tooth spacing2.889.6569.61042.6F2.305OMO 129/a-1972-4Tooth spacing3.0594.4606.51242.3lower G2.167OMO 6-1967-381Tooth spacing3.54103.96791701.6upper G1.986F 164-10102Tooth spacing3.65106695.21817.6H1.84F 161-22Tooth spacing3.1696.5622.81337.4J1.658OMO 394-10046Tooth spacing3.0995.2612.51276.4K1.458OMO 339-NC1Skull length92–611.41270.3L1.239OMO K 7-1969-4410Tooth spacing4.07111.6756.82303.1

Appendix 5Appendix 5. — Specimen data of Python ‌Daudin, 1803 for specimens with maximum body mass estimates for each member of the Shungura Formation.TaxonShungura MemberMidpoint Age (Ma)Specimen numberMeasurementDimension (mm)Total length (cm)Mass reconstruction (kg) Python B3.188L1-32bPrezygapophyseal width20.42515.59C2.735L 47-67Prezygapophyseal width4154041.13D2.547L 824-12Prezygapophyseal width40.152738.6E2.356L 82-30Prezygapophyseal width243008.9F2.305OMO 33-3613Prezygapophyseal width52.170281.6lower G2.167OO 75-1971-2862Prezygapophyseal width3748330.67upper G1.986F 163-11Prezygapophyseal width44.859552.99J1.658OMO 379-10008Prezygapophyseal width22.92847.68L1.239OMO 389-10055Prezygapophyseal width3444024.08

Fig. 1. — Representative specimens of examined reptile taxa from the Shungura Formation: A, Cycloderma sp. (OMO 229) carapace in dorsal view; B, ­Pelusios sp. (L 3-10, F-3, c. 2.31 Ma) carapace in dorsal view; C, Testudinidae indet. (OMO 18/inf-10074, C-4-8, c. 2.57-2.76 Ma) right humerus in dorsal/capitular view; D, ­Python sp. (OMO 340-10193, L-2, c. 1.34 Ma) precloacal vertebra in anterior view; E, cf.Mecistops ‌Gray, 1844 (OMO 372-10016, upper G, c. 2.057-1.911 Ma) partial rostrum in palatal view; F, Euthecodonbrumpti ‌(Joleaud, 1920) (IORE collections unnumbered) skull in dorsal view; G, Crocodylus sp. (L 875-1, F-3, c. 2.32 Ma) partial skull in dorsal view. Scale bars: A-C, E-G, 5 cm; D, 1 cm.10.5281/zenodo.18760446

Table 1. — Regression equations used to estimate body mass (in kg, unless noted otherwise) from body lengths for turtles and crocodylians.TaxonMeasurement, x (cm)Regression to body massRegression source Testudinidae Carapace lengthBM = 2.751*log10(x) -3.424Regression derived from data in Regis & Meik (2017) Trionychidae Carapace lengthlog10(BM) = 1.344*log10(x) -1.049Regression derived from data in Regis & Meik (2017) Pelomedusidae Carapace lengthlog10(BM) = -3.814 + 2.861*log10(x)Regression derived from data in Regis & Meik (2017) Crocodylia Total lengthlog10(BM) = -4.67 + 2.79*log10(x)Slavenko et al. (2016) Python Total lengthlog10(BM) = -5.131 + 2.611*log10(x) [mass unit: g]Feldman & Meiri (2013)

Fig. 2. — Estimation of skull length for fragmentary specimens of Euthecodon ‌Fourtau, 1920: A, partial mandible of E. cf. brumpti (L 32-203, C 5-7, c. 2.6-2.7 Ma) in dorsal view; B, tooth spacing along the rostra of four complete skulls used to train the regression, plotted as points, with smoothed curves showing how alveolar spacing changes along each snout, with the anteriormost alveoli the greatest distances apart; C, the linear regression equation used to relate that average tooth spacing value (excluding the anteriormost two alveoli) to dorsal skull length, based on the best-fit line through points representing the average tooth spacing of each dentary and maxilla in the four complete skulls. Abbreviations: Ant, anterior; ER, East Rudolf; KNM, National Museums of Kenya; KP, Kanapoi; LT, Lothagam; Pos, posterior. Scale bar: A, 5 cm.10.5281/zenodo.18989721

Fig. 3. — Schematic summarizing all paleoenvironmental variables used in analyses of correlation with reptile body sizes (colored ovals), excluding paleo-lake level (Nutz et al. 2020) and depositional lake scores (Heinzelin 1983), which were estimated independently through sedimentological synthesis. The data series are color-coded by their sources: blue, Levin 2015; green, Bobe & Behrensmeyer 2004; pink, Hernández Fernández & Vrba 2006; orange, Bobe et al. 2007; purple, Negash et al. 2020; yellow, Passey et al. 2010. Faded lines run from each variable (colored ovals) to the concept for which it acts as a proxy. Components influencing the paleosol isotopic ratios are linked to them by solid lines, as are the processes which filter those isotopic ratios in fossil material relative to paleosol values.10.5281/zenodo.18989723

Table 2. — Correlation tests between the maximum sizes of Shungura Formation reptiles and sampling (overall reptile or order-specific) or member age. Each cell contains the Pearson correlation coefficient and the p-value for the test. For the sampling comparisons (data series with non-normal distributions), each cell also shows the Spearman rank correlation coefficient and corresponding p-value. Significant results after sequential Bonferroni correction are marked with **. P-values under 0.05 which are not significant after Bonferroni correction are marked with *, and correlation coefficients over 0.75 are marked in bold.TaxonReptile samplingTurtle sampling Crocodylian samplingSquamate samplingTimeAquatic turtles0.217, p = 0.605 0.071, p = 0.8670.24, p = 0.567 0.171, p = 0.686–––0.229, p = 0.586 Trionychidae –0.933, p = 0.067 –0.8, p = 0.2–0.907, p = 0.093 –0.738, p = 0.262–––0.578, p = 0.422 Pelomedusidae 0.132, p = 0.755 0.548, p = 0.160.109, p = 0.798 0.366, p = 0.373––0.284, p = 0.495 Testudinidae 0.595, p = 0.159 0.45, p = 0.310.697, p = 0.082 0.414, p = 0.355––0.024, p = 0.96 Euthecodon –0.406, p = 0.215 –0.073, p = 0.831––0.448, p = 0.167 –0.087, p = 0.8–0.336, p = 0.313 Crocodylus 0.574, p = 0.065 0.45, p = 0.165–0.552, p = 0.078 0.626, p = 0.04*–0.238, p = 0.48cf. Mecistops0.4, p = 0.433 0.543, p = 0.266–0.426, p = 0.399 0.429, p = 0.397–0.189, p = 0.719 Python 0.003, p = 0.994 0.25, p = 0.516––0.227, p = 0.5580.141, p = 0.718

Fig. 4. — Maximum body size of crocodylians (A), turtles (B), and Python (C) by member in the Shungura Formation in comparison with sampling based on specimen counts by member from Omo collections.10.5281/zenodo.18989725

Table 3. — Correlation between maximum size of reptile groups sampled in the Shungura Formation across the members of the formation. Each cell contains the Pearson correlation coefficient and the p-value for the test. A shows results for pairwise correlation tests between log10 mass values and B shows results for pairwise correlation tests for the first differences in those values between members, not including results for Trionychidae because the sampled specimens come from too few consecutive members to calculate correlation in their first differences. NA indicates comparison with too few observations of change between bins in common between the two series to run correlation test. P-values under 0.05 which are not significant after Bonferroni correction are marked with *, and correlation coefficients over 0.75 are marked in bold. Crocodylus cf. Mecistops Pelomedusidae Testudinidae Trionychidae Python A Euthecodon –0.276 p = 0.440.513 p = 0.2980.25 p = 0.551–0.464 p = 0.2950.555 p = 0.4450.38 p = 0.313 Crocodylus ––0.209 p = 0.691–0.257 p = 0.538–0.139 p = 0.793–0.85 p = 0.15–0.22 p = 0.569cf. Mecistops––0.41 p = 0.4920.283 p = 0.817NA0.547 p = 0.262 Pelomedusidae –––0.674 p = 0.142–0.401 p = 0.599–0.172 p = 0.684 Testudinidae –––––0.88 p = 0.315–0.12 p = 0.82 Trionychidae ––––––0.286 p = 0.714B Euthecodon –0.239 p = 0.568NA0.238 p = 0.650.102 p = 0.87NA0.086 p = 0.914 Crocodylus –NA–0.079 p = 0.8810.141 p = 0.859NA–0.216 p = 0.784 Pelomedusidae –––0.826 p = 0.174NA0.165 p = 0.835 Testudinidae ––––NA0.997 p = 0.051Table 4. — Correlation statistics for reptile masses vs mammal masses by taxon. A shows results for pairwise correlation tests between log10 mass values and B shows results for pairwise correlation tests for the first differences in those values between members. The columns marked S have non-normally-distributed data, so the correlation test results shown in A are the Spearman rank correlation coefficient and corresponding p-value. The cercopithecid, aff. Hip. aethiopicus, Hippopotamus sp., and trionychid data series are excluded from B because they include too few consecutive members sampled. NA indicates other comparisons with too few observations of change between bins in common between the two series to run correlation test. P-values under 0.05 which are not significant after Bonferroni correction are marked with *, and correlation coefficients over 0.75 are marked in bold. Mammal data from Bibi (2023), Bibi & Cantalapiedra (2023), except for Hippopotamidae (this study). Abbreviations: aHa, aff. Hip. aethiopicus; aHpk, aff. Hip. protamphibius/karumensis; Cerco, cercopithecids; Hsp., ­Hippopotamus sp.; Trag, tragelaphins. Equidae Bovidae Trag (S)Cercoa Hpk (S)a HaH sp.A Euthecodon –0.42 p = 0.349–0.559 p = 0.074–0.37 p = 0.2930.215 p = 0.6830.027 p = 0.937–0.143 p = 0.819–0.227 p = 0.714 Crocodylus –0.575 p = 0.1770.407 p = 0.2140.541 p = 0.1060.473 p = 0.4210.36 p = 0.277–0.965 p = 0.035*–0.158 p = 0.8cf. Mecistops0.427 p = 0.573–0.557 p = 0.2510.1 p = 0.8730.587 p = 0.4130.143 p = 0.7870.809 p = 0.4–0.074 p = 0.906Aquatic turtles–0.292 p = 0.5260.238 p = 0.57–0.536 p = 0.2150.487 p = 0.4060.119 p = 0.779NA–0.489 p = 0.511 Trionychidae 0.16 p = 0.840.208 p = 0.792–0.6 p = 0.4NA–0.2 p = 0.8NANA Pelomedusidae 0.522 p = 0.229–0.262 p = 0.5320.357 p = 0.4320.326 p = 0.5930.119 p = 0.779NA0.949 p = 0.051 Testudinidae 0.392 p = 0.443–0.31 p = 0.4980.179 p = 0.702–0.255 p = 0.745–0.071 p = 0.879NANA Python –0.222 p = 0.632–0.114 p = 0.7710.095 p = 0.823–0.151 p = 0.8090.083 p = 0.8310.914 p = 0.267–0.722 p = 0.168B Euthecodon 0.127 p = 0.81–0.468 p = 0.172–0.073 p = 0.852–0.048 p = 0.895–– Crocodylus –0.558 p = 0.250.326 p = 0.3920.42 p = 0.301–0.274 p = 0.476––Aquatic turtle–0.373 p = 0.4660.061 p = 0.9080.112 p = 0.833–0.719 p = 0.108–– Pelomedusidae –0.132 p = 0.8030.296 p = 0.5690.122 p = 0.819––0.629 p = 0.181–– Testudinidae –0.323 p = 0.677–0.006 p = 0.9930.035 p = 0.955––0.498 p = 0.393–– Python 0.642 p = 0.358–0.644 p = 0.3560.696 p = 0.304–0.472 p = 0.528––Table 5. — Correlation statistics of reptile and mammal maximum sizes with a) environmental variables across members of the Shungura Formation. A shows results for pairwise correlation tests between log10 mass values and B shows results for pairwise correlation tests for the first differences in those values between members, not including results for Trionychidae, Cercopithecidae, or soil temperature because those series have too few consecutive members sampled to calculate correlation in their first differences. Each cell contains the Pearson correlation coefficient and p-value for the correlation test. The row marked S has non-normally-distributed data, so the correlation test results shown are the Spearman rank correlation coefficient and corresponding p-value. The cercopithecids, Hippopotamus sp., and soil temperature data series are excluded from B because they include too few consecutive members sampled. NA indicates other comparisons with too few observations of change between bins in common between the two series to run correlation test. P-values under 0.05 which are not significant after Bonferroni correction are marked with *, p-values which are significant after Bonferroni correction are marked with **, and correlation coefficients over 0.75 are marked in bold. Data from Heinzelin (1983), Passey et al. (2010), Levin (2015) and Nutz et al. (2020).Paleosol δ 13 CPaleosol δ 18 OLake level Soil temperatureDepositional lake scoreA Euthecodon –0.082 p = 0.8220.373 p = 0.2890.224 p = 0.5330.202 p = 0.7450.685 p = 0.02* Crocodylus –0.129 p = 0.742–0.05 p = 0.8980.192 p = 0.594–0.824 p = 0.1760.378 p = 0.252cf. Mecistops–0.302 p = 0.6210.307 p = 0.6150.028 p = 0.964NA0.053 p = 0.921 Testudinidae 0.019 p = 0.9680.252 p = 0.63–0.721 p = 0.067–0.142 p = 0.82–0.542 p = 0.208Aquatic Turtles0.028 p = 0.952–0.108 p = 0.8180.654 p = 0.111–0.834 p = 0.1660.098 p = 0.818 Trionychidae –0.232 p = 0.768–0.667 p = 0.5350.476 p = 0.524NA0.587 p = 0.413 Pelomedusidae –0.751 p = 0.0520.832 p = 0.02*–0.762 p = 0.047*–0.207 p = 0.7930.063 p = 0.883 Python 0.106 p = 0.8030.005 p = 0.990.052 p = 0.9030.76 p = 0.240.003 p = 0.995aff. Hip. protamphibius-karumensis (S)0.588 p = 0.080.952 p<0.0001**0.172 p = 0.6140.2 p = 0.7830.33 p = 0.294aff. Hip. aethiopicus0.292 p = 0.708–0.229 p = 0.7110.133 p = 0.867NA–0.224 p = 0.718Hippopotamus sp.0.323 p = 0.6770.815 p = 0.185–0.901 p = 0.099NA–0.002 p = 0.997 Equidae –0.297 p = 0.518–0.522 p = 0.288–0.825 p = 0.022*0.186 p = 0.814–0.805 p = 0.029* Cercopithecidae –0.455 p = 0.4410.629 p = 0.1810.637 p = 0.2480.537 p = 0.6390.232 p = 0.658 Bovidae –0.6 p = 0.067–0.471 p = 0.17–0.108 p = 0.7520.168 p = 0.787–0.411 p = 0.185Tragelaphini (S)–0.273 p = 0.4480.167 p = 0.678–0.418 p = 0.2030.9 p = 0.0830.005 p = 0.989B Euthecodon –0.096 p = 0.8070.032 p = 0.9410.017 p = 0.965–0.671 p = 0.034* Crocodylus –0.092 p = 0.8450.272 p = 0.5550.127 p = 0.765–0.387 p = 0.303Aquatic turtles0.255 p = 0.6250.323 p = 0.5960.813 p = 0.049*–0.024 p = 0.964 Pelomedusidae –0.79 p = 0.061–0.201 p = 0.746–0.704 p = 0.119––0.049 p = 0.926 Python 0.488 p = 0.5120.803 p = 0.406–0.65 p = 0.35––0.707 p = 0.293 Testudinidae –0.582 p = 0.3040.322 p = 0.791–0.737 p = 0.155––0.31 p = 0.611aff. Hip. protamphibius-karumensis0.237 p = 0.540.895 p = 0.003*0.495 p = 0.145––0.034 p = 0.92aff. Hip. aethiopicusNANANA––0.436 p = 0.713 Equidae 0.575 p = 0.2330.633 p = 0.252–0.365 p = 0.477––0.518 p = 0.292 Bovidae –0.455 p = 0.2190.319 p = 0.4420.003 p = 0.993––0.325 p = 0.33 Tragelaphini 0.054 p = 0.890.927 p = 0.003*–0.107 p = 0.768–0.182 p = 0.615Table 6. — Correlation statistics of reptile and mammal maximum sizes with faunal variables across members of the Shungura Formation. A shows results for pairwise correlation tests between log10 mass values and B shows results for pairwise correlation tests for the first differences in those values between members, not including results for Trionychidae and Cercopithecidae because the sampled specimens come from too few consecutive members to calculate correlation in their first differences. Each cell contains the Pearson correlation coefficient and p-value for the correlation test. The row and column marked S have non-normally-distributed data, so the correlation test results shown are the Spearman rank correlation coefficient and corresponding p-value. P-values under 0.05 which are not significant after Bonferroni correction are marked with *, p-values which are significant after Bonferroni correction are marked with **, and correlation coefficients over 0.75 are marked in bold. Data from Bobe & Behrensmeyer (2004), Hernández Fernández & Vrba (2006), Levin et al. (2011), and Negash et al. (2020) and Homo ‌Linnaeus, 1758 in the late Pliocene, as constituents of broader pulses of faunal turnover synchronized by episodes of global climatic change. A more recent concept, the variability selection hypothesis, emphasizes the importance of fluctuating climates and environments, rather than any single trend, in shaping human adaptation and evolution. Here we evaluate these ideas for the Plio-Pleistocene in light of new analyses of fossil mammals from the Turkana Basin of Kenya and Ethiopia. Our results show that between 4 and 1 Ma (million years ago).Herbivore δ 13 C (S)Herbivore δ 18 OAAH % (S)Reduncin %Grassland indicator proportion Faunal mean annual precipitationA Euthecodon 0.591 p = 0.061–0.054 p = 0.8750.1 p = 0.7690.238 p = 0.4820.157 p = 0.6450.258 p = 0.622 Crocodylus –0.077 p = 0.8210.048 p = 0.8890.061 p = 0.8670.036 p = 0.921–0.038 p = 0.916–0.179 p = 0.734cf. Mecistops0.029 p = 10.191 p = 0.7170.086 p = 0.919–0.186 p = 0.724–0.137 p = 0.7960.375 p = 0.755Aquatic turtles–0.619 p = 0.115–0.392 p = 0.3370.539 p = 0.1680.043 p = 0.9190.123 p = 0.7720.324 p = 0.531 Trionychidae –0.4 p = 0.75–0.645 p = 0.3550.4 p = 0.750.915 p = 0.085–0.021 p = 0.9791 p = 0.006* Pelomedusidae 0.31 p = 0.4620.857 p = 0.007*–0.395 p = 0.3320.37 p = 0.367–0.027 p = 0.9490.206 p = 0.695 Python 0.05 p = 0.912–0.194 p = 0.6170.05 p = 0.898–0.38 p = 0.313–0.044 p = 0.911–0.23 p = 0.66 Testudinidae NA0.55 p = 0.201–0.649 p = 0.115–0.104 p = 0.824–0.433 p = 0.3320.085 p = 0.891aff. Hip. protamphibius-karumensis (S)0.462 p = 0.1340.35 p =0.2660.743 p = 0.0089*0.533 p = 0.0910.882 p = 0.0006*–0.543 p = 0.297aff. Hip. aethiopicus0.7 p = 0.233–0.492 p = 0.40.2 p = 0.7830.31 p = 0.611–0.297 p = 0.627NAHippopotamus sp.–0.1 p = 0.950.769 p = 0.128–0.1 p = 0.950.696 p = 0.1920.561 p = 0.326NA Equidae –0.286 p = 0.5560.305 p = 0.507–0.342 p = 0.452–0.656 p = 0.109–0.78 p = 0.038*0.575 p = 0.232 Cercopithecidae –0.029 p = 10.971 p = 0.001*0.628 p = 0.173–0.045 p = 0.9330.553 p = 0.255–0.864 p = 0.136 Bovidae –0.431 p = 0.162–0.092 p = 0.777–0.237 p = 0.482–0.456 p = 0.158–0.247 p = 0.4650.138 p = 0.794Tragelaphini (S)–0.064 p = 0.860.6 p = 0.056–0.334 p = 0.345–0.286 p = 0.424–0.103 p = 0.785–0.543 p = 0.297B Euthecodon 0.098 p = 0.787–0.056 p = 0.8780.356 p = 0.312–0.303 p = 0.3940.371 p = 0.2910.465 p = 0.43 Crocodylus –0.258 p = 0.5020.109 p = 0.7810.342 p = 0.406–0.47 p = 0.240.372 p = 0.3640.247 p = 0.689 Testudinidae 0.638 p = 0.2470.506 p = 0.385–0.532 p = 0.3560.628 p = 0.257–0.611 p = 0.2730.999 p = 0.035*Aquatic turtles–0.825 p = 0.043*–0.432 p = 0.3930.648 p = 0.164–0.517 p = 0.2930.624 p = 0.1850.079 p = 0.9 Pelomedusidae 0.505 p = 0.3060.796 p = 0.058–0.845 p = 0.034*0.693 p = 0.127–0.856 p = 0.029*0.766 p = 0.131 Python 0.553 p = 0.447–0.084 p = 0.9160.227 p = 0.773–0.373 p = 0.6270.211 p = 0.7890.559 p = 0.622aff. Hip. protamphibius-karumensis–0.791 p = 0.004*–0.478 p = 0.1370.86 p = 0.001*–0.79 p = 0.007**0.742 p = 0.014*0.363 p = 0.548aff. Hip. aethiopicus0.755 p = 0.455–0.555 p = 0.626–0.988 p = 0.0970.941 p = 0.219–0.954 p = 0.193NA Equidae 0.175 p = 0.7410.011 p = 0.9830.045 p = 0.933–0.591 p = 0.2170.231 p = 0.66–0.158 p = 0.799 Bovidae –0.382 p = 0.247–0.116 p = 0.734–0.012 p = 0.973–0.072 p = 0.8430.012 p = 0.9750.18 p = 0.772 Tragelaphini –0.514 p = 0.1290.194 p = 0.5910.482 p = 0.189–0.643 p = 0.0620.654 p = 0.0560.333 p = 0.584

Fig. 5. — Maximum testudinid and Python ‌Daudin, 1803 body mass by member in the Shungura Formation vs paleosol δ13C sample values (Levin et al. 2011) and the percentage of Reduncini and Antilopini, Alcelaphini, and Hippotragini ‌(AAH) bovids in each member’s mammal fauna (Bobe et al. 2007).10.5281/zenodo.18760450

Fig. 6. — Maximum body size of aquatic turtles (trionychid or pelomedusid specimens indicated by color) by member in the Shungura Formation vs reconstructed relative lake level in the Turkana Depression (Nutz et al. 2020), δ18O values from paleosol samples (Levin et al. 2011), and δ18O values from mammal herbivore dental samples (Negash et al. 2020).10.5281/zenodo.18760452

Fig. 7. — Maximum body size of crocodylians (Euthecodon ‌Fourtau, 1920, Crocodylus ‌Laurenti, 1768, and cf.Mecistops ‌Gray, 1844) and hippopotamids (brown line series for the local protamphibius-karumensis lineage) by member in the Shungura Formation vs reconstructed relative lake level in the Turkana Depression (Nutz et al. 2020), and the maximum lacustrine facies score by member from Heinzelin (1983).10.5281/zenodo.18760456

Fig. 8. — Alternative visualizations of correlation tests, with the two data series compared on the two axes. The color of the points indicates the age of the member for each observation. All mass data are shown with units log10(kg).10.5281/zenodo.18989727