Ecosystem Collapse in Pleistocene Australia and a Human Role in Megafaunal Extinction

See allHide authors and affiliations

Science  08 Jul 2005:
Vol. 309, Issue 5732, pp. 287-290
DOI: 10.1126/science.1111288


Most of Australia's largest mammals became extinct 50,000 to 45,000 years ago, shortly after humans colonized the continent. Without exceptional climate change at that time, a human cause is inferred, but a mechanism remains elusive. A 140,000-year record of dietary δ13C documents a permanent reduction in food sources available to the Australian emu, beginning about the time of human colonization; a change replicated at three widely separated sites and in the marsupial wombat. We speculate that human firing of landscapes rapidly converted a drought-adapted mosaic of trees, shrubs, and nutritious grasslands to the modern fire-adapted desert scrub. Animals that could adapt survived; those that could not, became extinct.

Humans are thought to have colonized Australia between 55 and 45 thousand years ago (ka) (15), and most of its large animals became extinct between 50 and 45 ka (6, 7). The 60 taxa known to have become extinct include all large browsers, whereas large grazing forms, such as red and gray kangaroos, were less affected. The selective loss of large browse-dependent taxa suggests that ecosystem change may have been important, although animal size may have played a role (8). Inferential evidence of vegetation reorganization and a changed fire regime beginning 45 ka is recorded in terrestrial (9, 10) and marine (11, 12) sediment cores. But no records of ecosystem status through this time interval are available from the vast semiarid zone. We used isotopic tracers of diet preserved in avian eggshells and marsupial teeth (13, 14) to monitor ecosystems before and after human colonization. These dietary reconstructions document ecosystem collapse across the semiarid zone between 50 and 45 ka.

We recovered eggshells of the Australian emu Dromaius novaehollandiae and the extinct giant flightless bird Genyornis newtoni from longitudinal desert dunes and shoreline-marginal dunes around Lake Eyre (LE), the terminal playa of a large interior basin in central Australia (Fig. 1). These collections were supplemented by eggshell remains collected near Port Augusta (PA) and the Darling-Murray lakes (DM) (Fig. 1). Because the eggshells are fragmented, we grouped most shells into collections taken from locations separated by 100 m or more to avoid analyzing multiple fragments from the same egg. Temporal constraints are based on 14C dates on eggshell (LE, n = 111 samples; PA and DM, n = 60), luminescence ages on sand grains enclosing eggshells found in eolian deposits >40,000 years old (n = 28, table S5), and amino acid racemization (AAR) in eggshell organic matter for all eggshells in which we measured isotopic ratios (LE, n = 893; PA, n = 276; DM, n = 220). AAR measurements were converted to calendar ages using an age model derived from 14C and luminescence dates; of 191 LE Dromaius eggshells <45,000 years old, 84 were dated by 14C analyses. All 14C-dated Genyornis eggshells (n = 18) are beyond the reliable age of radiocarbon dating; for LE, they were plotted against calibrated AAR ages.

Fig. 1.

Map of Australia, showing the Lake Eyre Basin (dark gray) and primary collecting localities around Lake Eyre (LE, black), Port Augusta (PA), and the Darling-Murray Lakes (DM), which include the Minindee, Annabranch, and Willandra Lakes; the Perry Sand Hills; and Lake Victoria.

We reconstructed paleodiets for Dromaius and Genyornis based on the carbon isotopic composition (δ13C) of their eggshells (tables S1 to S3). Bird eggshells are a calcite bio-mineral containing 3% organic matter, most of which is sequestered within the calcite crystals of the eggshell, where it is stable in the geological environment for >106 years (15). The isotopic composition of carbon in eggshell organic residues (δ13Corg) and in the calcite matrix (δ13Ccarb) is determined by the δ13C of the birds' diet, offset by systematic biochemical fractionation (16, 17). Most plants use either the C3 or C4 photosynthetic pathways, which around LE yield average δ13C values of –26.3 ± 2.1 per mil (‰) and –13.7 ± 0.7‰ (table S6). Crassulacean acid metabolism plants rarely contribute to Dromaius diet. The offset between eggshell δ13Corg and the δ13C values of food sources is 3‰ (18), which is similar to that observed in controlled feeding experiments (17). The average offset between δ13Ccarb and δ13Corg in Dromaius eggshell is 10.4 ± 2.0‰ (n = 269), whereas the offset is 11.1 ± 1.5‰ (n = 127) for Genyornis, after adjusting for biases due to apparent summer breeding (18). Each eggshell records conditions during a single season. Calcite carbon, derived from blood, reflects food sources in the days to weeks before egg-laying (16). In contrast, eggshell organic residues are derived from protein sources and reflect both recent and potentially older protein reserves, integrating dietary intake over several months.

Dromaius lives across the Australian mainland, nesting in the austral winter throughout its range. Our 140,000-year dietary reconstruction for Dromaius from LE is based on δ13Corg (n = 181) and δ13Ccarb (n = 344) values in individually dated eggshells (Fig. 2). Between 50 and 45 ka, mean dietary δ13C decreased by at least 3.4‰ (95% confidence level), accompanied by an even larger decrease in dietary variance, from 14.8 to 3.8‰ (Fig. 2). Before 50 ka, Dromaius ate a wide range of food sources, ranging from a nearly pure C4 diet to a nearly pure C3 diet, with almost any combination of intermediate feeding strategies. This δ13C distribution is consistent with an opportunistic feeder that lived in an environment with high interannual moisture variability. The isotopic data imply that Dromaius utilized abundant nutritious grasslands in wet years (C4), relying more on shrubs and trees in drier years (C3). After 45 ka, Dromaius utilized a restricted range of food sources, dominated by C3 plants.

Fig. 2.

Time series of Dromaius (A) and Genyornis (B) dietary δ13C reconstructed from δ13Ccarb (light colors) and δ13Corg (dark colors) of individually dated eggshells from the LE region (Fig. 1). The vertical bars (50 to 45 ka) define the megafaunal extinction window, with its estimated uncertainty. Measured δ13Corg and δ13Ccarb values were converted to dietary δ13C by applying biochemical fractionation factors (tables S1 to S3). End-member dietary δ13C values corresponding to 100% C3 and 100% C4 diets (±2σ) are shown on the right. Differences in dietary δ13C values for the preextinction and postextinction windows are shown by the mean (black line), standard error of the mean (dark gray, 95% confidence), and standard deviation (±2σ, light gray). Bar graphs below each time series show the mean and variance calculated for 15,000-year intervals to test whether climate influenced these statistics significantly. White numbers in each bar denote the number of samples (n) in each interval. Because variance is highly dependent on sample size, we have combined adjacent time windows when n < 20.

Frequency histograms of Dromaius dietary δ13C (Fig. 3) provide clues about the nature of past ecosystems around LE. Before 50 ka, δ13C dietary tracers reflect a weakly bimodal pattern, with a broad dominance of C4 dietary sources and a subsidiary peak dominated by C3 plants. Two-thirds of both dietary tracers reflect >50% C4 plant sources, suggesting common nutritious grasslands. In contrast, Dromaius living after 45 ka utilized dominantly C3 dietary sources, always with <50% C4 plants. Although our dietary reconstructions document increased reliance on C3 plants by Dromaius after 45 ka, the landscape around LE included abundant spinifex and cane grass, C4 grasses that are largely inedible and low in nutrition and hence a minor component of recent Dromaius diets.

Fig. 3.

Frequency histograms of dietary δ13C for Dromaius (middle and top rows) and Genyornis (bottom row) from LE, PA, and DM. All sites reveal a shift toward lower mean dietary δ13C values (95% confidence, dark gray) and reduced spread [±2σ (light gray) and variance (var)] after 45 ka. A similar trend is apparent in Vombatidae tooth enamel from PA and DM. Genyornis dietary δ13C always shows less spread and a more consistent proportion of C4 food sources than does that of coexisting Dromaius. Dietary δ13C is derived from measured δ13Corg, δ13Ccarb, and δ13Capat (apat, hydroxyapatite) (tables S1 to S4). The number of analyses (n) is shown.

To evaluate the possible role of climate in the observed dietary shift, we subdivided the data into 15,000-year intervals back to 140 ka, which includes contrasting climates of the Holocene (15 to 0 ka), the last glacial maximum (30 to 15 ka), and the last interglaciation (125 to 110 ka). The mean and variance of each interval (Fig. 2) are not statistically different from the mean and variance of their larger groupings (>50 and <45 ka), suggesting that climate is not the dominant control on dietary δ13C. The restricted, C3-dominated dietary range for Dromaius after 45 ka persists through the cold, dry, last glacial maximum and the Holocene, when temperatures rose and rainfall increased (14, 19).

The δ13C values in eggshells of Genyornis living around LE between 140 and 50 ka (δ13Corg n = 161; δ13Ccarb n = 207; Fig. 2) allow us to compare feeding strategies between a taxon that became extinct and one that survived to the present. Genyornis consumed a more restricted diet, exhibiting only 40% of the isotopic variance observed in contemporary Dromaius. Frequency histograms of Genyornis diet (Fig. 3) are symmetrically distributed and always include some C4 dietary sources, unlike Dromaius, which tolerates a pure C3 diet. There are no large differences in average dietary δ13C or in its variance at 15,000-year intervals (Fig. 2), despite large changes in climate (20). We conclude that Genyornis was a more specialized feeder than Dromaius, targeting a specific set of food resources, and that these resources were frequently available through the range of climates between 140 and 50 ka.

To test the inferences derived from the LE data sets, we developed dietary reconstructions using eggshells from two other regions, PA and DM (Fig. 1); both are cooler and wetter than LE. Although both data sets are large, insufficient samples are available to produce time series spanning the past 140,000 years. Luminescence, 14C, and calibrated AAR dates at both sites demonstrate that Genyornis extinction and ecosystem change occurred 50 to 45 ka. The lack of any time dependence in the mean or variance in δ13C for either taxon from LE (Fig. 2) allows firm comparisons between the diets of Dromaius and Genyornis in age clusters >50 ka and <45 ka for all three regions, without concerns about biasing the statistics by an overrepresentation or absence of data in certain time intervals.

Dietary δ13C values for Dromaius eggshells <45,000 years old from PA and DM reflect almost exclusively C3 food sources, whereas eggshells >50,000 years old exhibit a greater range in δ13C values and a higher proportion of C4 dietary sources (Fig. 3). Genyornis dietary δ13C values exhibit less variance than those of contemporary Dromaius and always include some C4 plants, although in lower proportions (27 to 39%) than at LE (46%). Genyornis eggshell δ13C values show a strong central tendency in both tracers, whereas diets of coexisting Dromaius lack a clear central tendency and have greater variance. The results for all three localities are consistent with a substantial reduction in food sources for Dromaius 50 to 45 ka and a more specialized feeding strategy for Genyornis.

The coeval dietary retraction recorded by Dromaius eggshells from three regions widely separated by geography and climate reflects a major disruption in the range of food sources available to the birds about the time of human colonization. If these changes represent a reorganization of vegetation communities across the Australian semiarid zone, we expect to find similar responses in other animal groups. To test this prediction, we measured δ13C values in hydroxyapatite extracted from wombat tooth enamel from the PA and DM regions (table S4). Wombats are strict herbivores, relying almost exclusively on grasses and reeds. They live, and often die, in sandy deposits used by Dromaius and Genyornis as breeding sites, and their association with eggshells of known age can date their remains (18). Teeth from wombats that lived >50 ka exhibit a wider range of δ13C values and a much larger proportion of C4 plants than do teeth of wombats that lived <45 ka (Fig. 3). This difference suggests that before 50 ka, C4 plants made up 40 to 100% of the wombat's dietary intake, whereas after 45 ka, their diet was dominated by C3 plants, supporting the conclusions derived from Dromaius eggshells.

Our eggshell and tooth δ13C data provide firm evidence for an abrupt ecological shift around the time of human colonization and megafaunal extinction in Australia, about 50 to 45 ka. Climate forcing of the observed vegetation change is unlikely, given that earlier dramatic climate shifts did not result in such a large biotic response and that climate change between 60 and 40 ka was not large, consistent, or sustained. During this interval, the DM region experienced somewhat greater effective moisture, whereas modest drying occurred around LE (2, 19, 20). A persistently weak Australian monsoon after 45 ka may explain the lack of abundant nutritious C4 grasses around LE (14).

A changed fire regime is another plausible mechanism for ecosystem reorganization. Early human colonizers may have altered the timing and frequency of biomass burning. Humans burn landscapes for many purposes, from clearing passageways and hunting along the fire front, to signaling distant bands and promoting the growth of preferred plants. We speculate that systematic burning practiced by the earliest human colonizers may have converted a drought-adapted mosaic of trees and shrubs intermixed with palatable nutrient-rich grasslands to the modern fire-adapted grasslands and chenopod/desert scrub. Nutrient-poor soils (21) may have facilitated the replacement of nutritious C4 grasses by spinifex, a fire-promoting C4 grass that is well adapted to low soil nutrient concentrations. A range of C3 plants may have been lost at the same time, but the isotopic dietary proxy lacks sensitivity to such a loss.

Neither overhunting nor human-introduced diseases, the two most widely cited alternative agents for a human-caused extinction event in Australia, would result in the dramatic changes at the base of the food web documented by our data sets. The reduction of plant diversity apparent in our data, however it came about, would have led to the extinction of specialized herbivores and indirectly to the extinction of their large nonhuman predators. Dietary specialization, rather than feeding strategy (browsing versus grazing), may be the critical extinction predictor. Animals such as Dromaius, with wide dietary tolerances, survived the extinction event, whereas more specialized feeders, such as Genyornis, became extinct.

Supporting Online Material


Figs. S1 to S3

Tables S1 to S6


References and Notes

View Abstract

Navigate This Article