Mid-Holocene aridity recorded in pygmy hippo and giant tortoise bone from southwest Madagascar

Abstract

Aridity can exacerbate threats to endemic biodiversity, and arid intervals during the last couple of millennia may have contributed to endemic large herbivore extinctions on Madagascar. However, regional palaeoclimate records spanning multiple millennia are limited, and the tolerance of extinct taxa to past water scarcity is poorly known. To infer changes in the diet and habitat aridity of extinct pygmy hippos and giant tortoises during approximately 6000-1000 years ago, I used carbon and nitrogen isotope (δ¹³C and δ¹⁵N) data from 49 directly radiocarbon-dated bones collected around Tampolove, southwest Madagascar. Fluctuations in bone δ¹⁵N values through time in both species indicate tolerance of dry habitat during intermittent drying trends, including around a dry period known as the '4.2 ka event'. However, taxon-specific differences in the covariance of bone δ¹³C and δ¹⁵N values suggest that the diets of pygmy hippos and giant tortoises changed in different ways during these past arid intervals. This suggests that past aridification had different effects on these taxa. Thus, I argue that hypotheses for past extinction that involve a synergy among climate drying and forest clearance, hunting and biological invasion must consider taxon-specific responses to past aridity.

Keywords: aridification; extinction; palaeoclimate; palaeoecology; radiocarbon; stable isotope.

Figure 1.

Locations of the study site in southwest Madagascar (Tampolove) and sites with proxy records used to infer past changes in climate and other palaeoenvironmental variables (vegetation, erosion, fire and larger herbivore abundance) (A), and direct dates from different layers of each record used to establish chronologies in age-uncertain records sensitive to climate (B). Geographical groups of sites in A (defined by close proximity and comparable elevation and mean annual precipitation taken from WorldClim 2.1) yield the colour-coded records in B. Names of records discussed in the text are given in bold, consisting of site name and either record collection ID or lead author name and publication year. Note that only age-uncertain records with greater than one direct date during the Holocene are shown in B and that some of these records also include proxy data that are sensitive to vegetation.

Figure 2.

Locations of sites in southwest Madagascar that include previously published δ¹⁵N data from modern plant tissues (n = 1366) (A), and plant tissue δ¹⁵N data expressed relative to atmospheric nitrogen (AIR) separated according to photosynthetic pathway and as a function of mean annual precipitation (B). MAP estimates come from WorldClim 2.1, and dashed lines for plant groups with n > 100 give linear fits within each group. Abbreviated site names include Beza Mahafaly (BM), Ambohimahavelona (AMB), Andrevo (AND), Ranobe (RAN), Tsimanampesotse (TSI) and Cap Sainte Marie (CSM).

Figure 3.

Tampolove animal collagen δ¹⁵N and δ¹³C value distributions and changes through time (A,C) plotted against past change in relative sea level (B). The δ¹⁵N and δ¹³C values from directly ¹⁴C-dated tortoise and hippo collagen are marked according to mean calibrated years before present, and broad trends are marked by a linear model to the tortoise data (green) and a second-order polynomial to the hippo data (blue), with the 95% confidence intervals marked with grey shading. Box plots illustrating δ¹⁵N and δ¹³C data from the colour-coded taxa have widths scaled to sample size, boxes that illustrate interquartile ranges, and whiskers that extend to minimum/maximum points that fall within 1.5 times the interquartile ranges. The δ¹³C data are expressed relative to Vienna Peedee belemnite (VPDB).

Figure 4.

Tampolove tortoise and hippo collagen δ¹⁵N values plotted against time and relative to regional palaeoclimate records from northwest Madagascar (brown) and southwest Madagascar (red). Chronological uncertainty is propagated using GeoChronR, and grey ribbons mark quantiles around proxy estimates given by black lines in each case aside from the tortoise and hippo δ¹⁵N records (green and blue, respectively). Age–depth models are based on U/Th dates (Anjohibe/Anjohikely records), ¹⁴C dates (Tampolove and Ihotry records) and a combination of U/Th and ¹⁴C data (Asafora record). Solid distributions associated with AK1 and ANJ94−2 mark the modelled distributions for intervals that include hiatuses in these records. Red horizontal shading marks intervals with maxima in at least hippo collagen δ¹⁵N values that are discussed in the text.