Early Local Last Glacial Maximum in the Tropical Andes

See allHide authors and affiliations

Science  29 Apr 2005:
Vol. 308, Issue 5722, pp. 678-681
DOI: 10.1126/science.1107075


The local last glacial maximum in the tropical Andes was earlier and less extensive than previously thought, based on 106 cosmogenic ages (from beryllium-10 dating) from moraines in Peru and Bolivia. Glaciers reached their greatest extent in the last glacial cycle ∼34,000 years before the present and were retreating by ∼21,000 years before the present, implying that tropical controls on ice volumes were asynchronous with those in the Northern Hemisphere. Our estimates of snowline depression reflect about half the temperature change indicated by previous widely cited figures, which helps resolve the discrepancy between estimates of terrestrial and marine temperature depression during the last glacial cycle.

It is unclear whether the last glacial maximum (LGM) was a globally synchronous event or whether a local last glacial maximum (LLGM) occurred in the tropics at a different time (1). This issue has broad implications for our understanding of the forcings that underlie the massive global climate changes associated with glacial cycles, and it underpins all paleoclimate reconstructions for the tropics. An absolute chronology for the last glacial cycle and ice coverage in the tropics is unavailable, largely because of the limitations of radiocarbon dating and the scarcity of datable material in glacial environments above 4000 m above sea level (masl). Existing glacial chronologies for the tropics are based primarily on minimum-limiting radiocarbon dates and relative dating techniques (2). An absolute chronology would have important implications for paleoclimate reconstructions and the debates over the synchroneity of glaciation in the Northern and Southern Hemispheres (3), the synchroneity of continental and mountain glaciation (4), the magnitude of tropical climate change at the LGM (57), and the relation between sea surface temperatures (SSTs) and temperatures over continental regions (8). Snowline depression of ∼1000 m at a presumed LGM age of ∼21 thousand years before the present (ky B.P.) is often cited as evidence for terrestrial temperatures 5° to 6°C cooler than at present in the tropics (6). However, the climatic interpretation of snowline depression is complicated (9), because the timing of glacial coverage in the tropics has been poorly constrained. Here we present an absolute chronology of the LLGM based on cosmogenic dating (10Be) of erratics on moraines in the tropical Andes of Peru and Bolivia.

We selected field areas that allowed direct comparison of our terrestrial glacial chronology to existing lacustrine paleoclimate records from nearby lakes: Lake Junin (11°S, 76°W) in central Peru and Lake Titicaca (16°S, 69°W) on the Peruvian-Bolivian Altiplano (Fig. 1). Both lakes predate the last glacial-interglacial transition and have provided valuable paleoclimate proxy records in the form of sediment cores (1014). Lake Junin and Lake Titicaca lie on high plateaus (>3800 masl) between the eastern and western cordillera of the Andes and receive moisture predominantly from easterly winds crossing the Amazon Basin during the austral summer (15). We dated moraines in four valleys in the eastern cordillera bordering Lake Junin and in Milluni and Zongo Valleys in the Cordillera Real, approximately 35 km east of Lake Titicaca (fig. S1 and supporting online text). We used the concentration of cosmogenic 10Be to determine the exposure ages of boulders on moraines in the study areas (Table 1 and table S1) (16, 17). Analytical uncertainties on the entire set of exposure ages average 3.0 ± 1.5% (17). We assigned dated moraines to one of four moraine groups (A to D) based on an assessment of the position of the moraine within the valley, the geomorphic character of the moraine, and the exposure ages obtained from the moraine.

Fig. 1.

Site location maps of the study areas. (A) South America, with area covered by (B) indicated. (B) Map of Peru, Bolivia, and adjacent countries, showing the general location of Lake Junin and the Junin valleys, Lake Titicaca, and Milluni and Zongo Valleys in the Cordillera Real. The 3000-m contour line bounds the eastern and western cordillera of the Andes and the intervening high plateaus, including the Altiplano.

Table 1.

Exposure ages (10Be) for moraine groups A to C deposited during the last glacial cycle (17). Statistical data are shown for sample ages excluding outliers (I) and for all sample ages (II). As used here, the term “outliers” refers to exposure ages greater than 2 standard deviations (SD) from the mean age of the moraine group.

Moraine group (I) Mean age (excluding outliers) (years) SD of the ages (excluding outliers) (years) Age range (excluding outliers) (years) (II) Mean age (all samples) (years) SD of the ages (all samples) (years) Age range (all samples) (years) Number of samples (including outliers) Number of outliers (>2σ from mean) Relative SDView inline Analytical uncertaintyView inline
A 14,500 2,000 11,500-18,400 15,300 4,300 11,500-31,100 19 1 14% 4.1%
B 16,300 2,900 7,900-22,700 15,900 4,400 3,500-29,700 50 4 18% 4.2%
C 27,800 6,400 15,500-53,200 32,100 27,200 15,500-189,900 37 1 17% 3.2%
  • View inline* Relative SD = (SD of the ages, excluding outliers)/(mean 10Be age, excluding outliers).

  • View inline Analytical uncertainty is the blank-corrected uncertainty in the accelerator mass spectrometer measurements.

  • In the Junin valleys (Fig. 2A and fig. S1, A to D), 146 10Be exposure ages clearly distinguish moraines deposited late in the last glacial cycle from moraines that are considerably older (Fig. 3). Exposure ages of boulders on LLGM moraines are typically 34 to 22 ky B.P.; ages on older moraines generally predate the last glacial cycle (>65 ky B.P.) and in many cases are dated >200 ky B.P. (18). In the west-facing valleys, the LLGM moraine is the lower of two end moraines that are located midway downvalley; a younger (∼20 to 16 ky B.P.) moraine typically dams a lake (or paleolake) 1 to 2 km upvalley from the LLGM ice limit.

    Fig. 2.

    Exposure ages of moraines in the Junin valleys (A) and Milluni and Zongo Valleys (B) based on cosmogenic dating with 10Be. Thin gray lines are contour lines, with altitudes in masl; contour intervals are 200 m (A) and 500 m (B). Heavier solid lines indicate distinct moraines observed in the field and/or on aerial photographs; dashed lines indicate less distinct moraines or boulder trains. Ages shown were calculated without erosion or uplift (17). Analytical uncertainties on exposure ages average 3.0 ± 1.5% for all moraine groups combined (moraine groups A, B, C, and D). The LLGM in both areas was reached ∼34 ky B.P. (A) The Junin valleys include Alcacocha Valley (1), Antacocha Valley (2), Calcalcocha Valley (3), and Collpa Valley (4). In west-facing Junin valleys, LLGM glaciers (moraine group C) were small relative to glaciers that deposited larger moraines downvalley (moraine group D) before the last glacial cycle (≫65 ky B.P.). (B) In Milluni Valley (5) and Zongo Valley (6), all moraines that we have dated were deposited during the last glacial cycle (<65 ky B.P.). Contemporaneous moraines lie more than 1000 m lower in Zongo Valley than in Milluni Valley. Detailed sampling locations are illustrated in fig. S1.

    Fig. 3.

    Exposure ages and paleoclimate proxy records for 0 to 40 ky B.P. The top three panels are histograms and probability density (PD) curves (32) showing age distributions for cosmogenic 10Be exposure ages in moraine groups C, B, and A, respectively, from the top down; n refers to the number of samples included in the histogram and PD curves. Analytical uncertainties on exposure ages average 3.8 ± 1.4% for moraine groups A, B, and C combined (moraine group D is omitted). Two exposure ages >40 ky B.P. have been omitted from the histogram for moraine group C (see table S1). The bottom panel shows borehole gamma radiation readings [in counts per second (cps)] from the Salar de Uyuni, Bolivia [outer left y axis, black curve (27)], SSTs at the Cocos Ridge reconstructed from Mg/Ca ratios in foraminifer shells from core TR163-19 [inner left y axis, blue curve (23)], and the SPECMAP stacked benthic δ18O record [right y axis, red curve (29)]. Higher borehole gamma readings are interpreted as indicating wetter conditions (27), and higher (positive) benthic δ18O values are interpreted as indicating greater global ice volume (29). The vertical blue band spanning the four panels marks the LLGM in the tropical Andes.

    In the Cordillera Real (Fig. 2B and fig. S1, E and F), all 42 of the 10Be exposure ages fall within the last glacial cycle (table S1). In Milluni Valley, exposure ages on two sharp-crested parallel lateral moraines sampled at ∼4500 to 4600 masl lie between 34 and 23 ky B.P. (with two younger outliers). Exposure ages on a recessional moraine at ∼4650 masl are late glacial to early Holocene (18 to 9 ky B.P.). All exposure ages from Zongo Valley, even those from the lateral moraines at 3400 to 3500 masl, are late glacial (<20 ka) or younger (19).

    The pattern of LLGM moraine ages in the Junin region is consistent from valley to valley, whereas the pattern in the Cordillera Real is not. Exposure ages in the Junin region suggest that glaciers reached their maximum extent of the last glacial cycle as early as ∼34 ky B.P. Recession was under way by ∼21 ky B.P., followed by stillstand or readvance between ∼16 and 20 ky B.P. In Milluni Valley, the maximum extent of ice during the last glacial cycle occurred ∼34 to 23 ky B.P., with subsequent recession and several stillstands or readvances (20). In Zongo Valley, however, late-glacial ice extended downvalley more than 1 km lower in altitude than ice of similar age in Milluni Valley. We have not yet identified a moraine older than 20 ky B.P. in Zongo Valley. If the LLGM terminus descended past 3400 masl to ∼3000 masl, as seems likely from geomorphic evidence, Müller's (21) estimate of ∼4100 to 4200 masl for the LLGM limit was too high by at least 700 m and perhaps by more than 1 km.

    Our glacial chronology is consistent with the paleoclimate proxy records from Lake Junin and Lake Titicaca. Seltzer et al. (10, 11) used magnetic susceptibility (MS) and stratigraphic records to bracket the LLGM between ∼30 ky B.P. and ∼21 ky B.P. in a sediment core from Lake Junin with a basal age of ∼40 14C ky B.P. (radiocarbon years before the present). MS in Lake Junin sediments was elevated (relative to Holocene and late-glacial levels) even before 30 ky B.P., including sharp increases ∼37 to 35 and ∼33 ky B.P. (10, 11). Seltzer et al. (10) identified the glacial-interglacial transition with a dramatic drop in MS ∼21 ky B.P. and the onset of full deglaciation with the increase in organic carbon ∼16 ky B.P. Our record from the Junin valleys suggests maximum ice extent ∼34 to 22 ky B.P., ice retreat from the terminal moraine by 21 ky B.P., formation of a second end moraine upvalley between ∼20 and 16 ky B.P., and steady retreat after ∼15 ky B.P. Lake Titicaca sediment cores indicate a wet glacial period and the onset of deglaciation ∼24 to 19 ky B.P. The LGM and late-glacial period (until ∼15 ky B.P.) on the Altiplano were wet (12, 13) and lake levels were high (14). Temperature decreased by ∼5°C from 27.5 to 21 ky B.P. and warming began ∼21 ky B.P. (14). Seltzer et al. (11) identified the glacial-interglacial transition with decreases in MS and sedimentation rates ∼24 and ∼19.5 ky B.P., respectively, in two sediment cores from Lake Titicaca that show elevated MS before the transition. Our glacial chronology shows maximum ice extent in Milluni Valley ∼34 to 23 ky B.P. and construction of a recessional moraine (∼10 km upvalley and ∼50 m higher in altitude) by 18 ky B.P. Exposure ages suggest that ice remained in Zongo Valley almost into the Holocene.

    We estimate that equilibrium-line altitude (ELA) depression at the LLGM was 300 to 600 m in the Junin valleys and Milluni Valley and 800 to 1000 m in Zongo Valley (fig. S2). Even with uncertainty of ±100 m, our estimates of LLGM ELA depression for the Junin valleys and Milluni Valley are about half of published estimates of 900 to 1000 m for the tropical Andes (6, 7). We can use the position of LLGM moraines to constrain paleoclimatic change at the time of moraine deposition. If we assume that the change in ELA (ΔELA) was entirely a function of a change in temperature (ΔT), we estimate a ΔT of –2° to –4°C in the Junin valleys and Milluni Valley at the LLGM [using a moist adiabatic lapse rate of 6°C/km (7)]. Recent estimates put the LLGM temperature depression in the tropical Andes and Amazon Basin at ∼5°C (14, 22) and the change in SST in the eastern equatorial Pacific at –2.8° ± 0.7°C (23). If the Junin and Milluni estimates of ΔELA accurately reflect the ΔT at the LLGM, as these recent ΔT estimates suggest, then we can infer that the Zongo Valley glacier was anomalously long. The length of the Zongo Valley glacier can be explained by the effects of local factors such as shading by the sheer valley walls (24) and debris cover on the ice surface (25), compounded by orographically controlled precipitation resulting from moisture-bearing easterlies (26) and/or the positive feedback set up by ice advancing into a steep, positive precipitation gradient. Thus, our findings sound a note of caution for the use of generalized ELA estimates in paleoclimate reconstructions for areas of complex topography such as the Andes.

    We propose that a shifting balance between temperature and precipitation changes promoted glacier development and expansion early in the last glacial cycle and that warming initiated deglaciation ∼21 ky B.P. The SST record from the Cocos Ridge in the eastern equatorial Pacific (Fig. 3) shows overall lower SSTs from ∼70 to 12 ky B.P., with sharp decreases at ∼68 to 65 ky B.P., ∼38 ky B.P., ∼30 to 25 ky B.P., and ∼22 to 16 ky B.P. (23). Published moisture records suggest that the last glacial cycle was wetter than the present (27, 28). A moisture record for the Salar de Uyuni (∼20°S) on the southern Bolivian Altiplano (27) indicates a dry period from ∼67 to 61 ky B.P., followed by wetter conditions with periods of perennial lakes from ∼60 to 15 ky B.P., and finally dry conditions from ∼15 ky B.P. to the present. Ice may have begun to accumulate in the highest areas as soon as precipitation increased at ∼60 ky B.P. and then expanded after the sharp SST cooling at ∼38 ky B.P. (23). With increased precipitation and persistent cool temperatures, glaciers in the highest areas reached their maxima at ∼34 ky B.P., while ice began to accumulate and/or expand in the broader Junin valleys with lower headwalls. Glaciers retreated behind their end moraines at ∼22 to 21 ky B.P. in response to warming. Glaciers stabilized at their new positions for ∼5000 years, from ∼20 to 16 ky B.P.; the resulting recessional moraines dam lakes (or paleolakes) in the west-facing Junin valleys (fig. S1). The scarcity of younger moraines upvalley from these lakes indicates that the subsequent and final deglaciation was steady, leaving the basins relatively free of fluvioglacial deposits and till. Decreasing precipitation (driven by the precession cycle) and further warming after ∼15 ky B.P. made this last deglaciation a rapid one. However, in steep and deep valleys on the eastern side of the eastern cordillera, such as Zongo Valley, deglaciation proceeded differently. There, the combination of sheer valley walls that shaded the glacier; rockfall and erosional debris that insulated the ice; and orographically controlled precipitation that supplied the accumulation area maintained the glacier terminus at an altitude more than 1 km lower than those of contemporaneous termini of paleoglaciers on the western side (fig. S1).

    Our results indicate that the LLGM in the tropical Andes began ∼34 ky B.P., thus preceding by as much as ∼10 to 12 ky B.P. the maximum continental ice volume of the last glacial cycle as inferred from the marine oxygen isotope record (29). We estimate that ELA depression was typically about half the widely cited value of ∼900 to 1000 m, except where local factors enhanced glacier size. The close agreement between our 10Be-derived chronology and paleoclimate proxy records from nearby lakes shows that these tropical mountain glaciers are sensitive indicators of tropical climate that integrate a combination of climatic variables into a seemingly simple signal. The mountain glaciers of the tropical Andes apparently reached their maximum extent of the last glacial cycle before the continental ice sheets (Fig. 3) (29). Records of mountain glaciations from some other mid- and high-latitude regions have shown similar divergence from global ice volume records (2931), thus reflecting a complex mosaic of regional climatic patterns during the last glacial cycle.

    The asynchroneity in timing of the LLGM in the tropics and ice extent in the Northern Hemisphere implies that the climate-forcing mechanisms were decoupled at some level during the last glacial cycle. This, in turn, challenges conclusions about interhemispheric teleconnections that are based on the assumption that the LGM was globally synchronous (3) and that the magnitude of mountain snowline depression was uniform everywhere.

    Supporting Online Material

    Materials and Methods

    SOM Text

    Figs. S1 and S2

    Table S1


    References and Notes

    View Abstract

    Stay Connected to Science

    Navigate This Article