Glaciers are among the most sensitive responders to climate change, and their well-documented retreat over the past century is a clear signal of global warming.1,2,3 Long records of glacier fluctuations provide an important context for assessing the role of natural variability on this recent glacier retreat, helping to isolate an anthropogenic signal.4 Research over the past century established the current paradigm that warmth during the early-to-middle Holocene caused widespread retreat of high-altitude cirque glaciers in the western United States, followed by a resurgence of alpine glaciation, or Neoglaciation,5 beginning after 6ka (Fig. 2b), with the most recent advance during the Little Ice Age (LIA).6,7 The ages of these cirque moraines, which exist just distal to the LIA moraines, were largely determined using relative dating techniques.8,9,10,11 The moraines were then correlated with similar moraines that represented the type sites of Neoglacial events across the American West, such as Temple Lake in Wyoming,12 TL in Colorado,13 and Recess Peak in California.14 The numerical ages of these type moraines, however, were also not well constrained, and in the few instances where numerical dating was subsequently applied,15,16 some of these moraines have instead been attributed to the late-Pleistocene Younger Dryas cold event (12.9–11.7 ka).17,18,19,20 The ages of nearly all of these particular cirque moraines thus remain highly uncertain,21 which has important implications for our general understanding of Holocene climate,22 glacier change,23 proposed rapid climate change events during the Holocene,24 and silicate weathering estimates.25,26 Moreover, the presence of Neoglacial moraines beyond the LIA extent would imply a cooler pre-LIA climate, in contrast to regional and hemispheric temperature reconstructions22 that show a cooling trend through the middle-to-late Holocene culminating with maximum cooling during the LIA.

Here we address these issues by using cosmogenic 10Be to date 130 boulders from 19 moraines deposited by 14 cirque glaciers across the western United States (U.S.) (Fig. 1) that were previously interpreted as Neoglacial or early Holocene in age (Fig. 2) (see Supplementary Discussion). Our 10Be chronology demonstrates that each of the moraines originally interpreted as Neoglacial was deposited during the latest Pleistocene to earliest Holocene (between ~15 and 9 ka), indicating that, with the exception of some isolated locations,27,28 cirque glaciers in the western U.S. did not extend beyond their LIA limits during much, if not all, of the Holocene.

Fig. 1
figure 1

Shaded relief map of the western United States, sample locations, and moraine boulder ages. Boxes indicate individual boulder ages and uncertainties (regular font) and the arithmetic mean moraine age and standard deviation (bold) from this study. Italicized and underlined ages are outliers as defined in the text. Moraine arithmetic mean ages and uncertainties (standard deviation) from prior surface exposure age dating studies are also provided (blue font)

Fig. 2
figure 2

Shaded relief map of the western United States with locations of records discussed, and moraines ages from this study. a Black lines with labels denote field locations for this study, red circles show six other locations where previous cosmogenic-based cirque moraine chronologies exist and have been recalibrated, 19,31,58,59,60 and blue box denotes location of a sea surface temperature record.61 From south to north, LK—Lake Katherine; BB—Baboon Lakes; WP—Wheeler Peak; CL—Chicago Lakes; SP—Satanta Peak; TL—Triple Lakes; FJ—Fourth of July; BL—Blue Lake; DH—Deadhorse Lake; MB—Medicine Bows; TP—Temple Peak; SC—Stough Creek; HM—Horse Mountain; EC—Enchantment Lakes Basin. b Arithmetic mean (blue squares) and standard deviation (1σ) of boulder ages from moraine crests and distinct glacial events (black squares) based on moraine ages and stratigraphic positions from this study (see text). The lower curves show a probability density function of all 10Be ages from this study (blue curve and fill) and combined with recalibrated ages from six other locations (black curve and fill), and an approximate density curve based on previously interpreted ages of moraines from this study (pink dotted curve and fill)


Moraine chronologies

Moraines sampled for 10Be dating are 15–30 km up-valley from moraines marking the Last Glacial Maximum (LGM) and <2 km from cirque headwalls. The moraines in 10 cirques occur just distal (within several hundred meters) to presumed LIA moraines or associated rock glaciers, while no LIA-associated moraine or rock glacier occurs up-valley in the other four cirques. We interpret the 10Be ages to represent the time of onset of glacier retreat from the moraines. The age of each moraine and its uncertainty is reported as the arithmetic mean and standard deviation (1σ) of the 10Be ages, and we add the production rate uncertainty (~1–5%) in quadrature when comparing our ages to records based on other dating methods. See Methods section for details on sample preparation and age calculations.

Based on geomorphic relations and the precision of our ages, we can distinguish at least six discrete episodes of moraine-building events (Fig. 2b), with three latest Pleistocene cirque-glacier events occurring prior to the Younger Dryas cold interval, one within the interval, and two after. To constrain the event ages, we only consider analytical uncertainty in the 10Be ages because of equivalent production rate uncertainty. Moraines representing the discrete events are identified by either being from stratigraphically distinct moraines from the same cirque or by differing in age at 1σ from the other events (Table S1). The moraines from the Medicine Bow Mountains (MB) in the Snowy Range of southeastern Wyoming alone identify three distinct events, with their ages confirming a millennial-scale signal (14.5 ± 0.3, 11.5 ± 0.3, and 10.5 ± 0.3 ka). The oldest moraine ages from the Lake Katherine (LK) moraine in the Sangre de Cristo Mountains, New Mexico (15.1 ± 0.4 ka), the Upper Chicago Lake (CL) moraine in the Colorado Front Range (15.2 ± 0.7 ka), and the Blue Lake (BL) moraine in the Uinta Mountains (14.4 ± 0.3 ka) overlap with the oldest moraine age from the Snowy Range (14.5 ± 0.3 ka) and may thus be part of the same event. The age of the Deadhorse (DH) moraine in the Uinta Mountains of Utah (13.8 ± 0.3 ka) falls between the ages of the two older moraines from the Snowy Range and records a second event. Multiple other sites record moraine ages from 13 to 12 ka that overlap with each other at 1σ (Fig. 2b), but only the Horse Mountain (HM) moraine in the Tobacco Root Mountains, Montana (12.9 ± 0.4 ka) demonstrates a distinct third event. The age of the Fourth of July (FJ) moraine in the Colorado Front Range (11.6 ± 0.1 ka) is included with the intermediate Snowy Range moraine (11.5 ± 0.3 ka) to record a fourth event. The youngest DH moraine in the Uinta Mountains (10.4 ± 0.4 ka) is included with the youngest Snowy Range moraine (10.5 ± 0.3 ka) as a fifth event. Finally, the youngest distinct event is recorded by the inner Triple Lakes (TL) moraine from the Colorado Front Range (9.1 ± 0.9 ka).

Glacier length changes

In order to quantify changes in glacier length relative to their LGM positions, we scaled glacier positions as marked by moraines from each valley to normalized unit lengths, from 1 at their LGM positions to 0 at cirque headwalls (Fig. 1f). Although not necessarily a simple function of climate, previous work1,29,30 demonstrates that records of change in glacier length over large geographic areas capture first-order patterns of glacier response to climate. We combined our data with cosmogenic exposure ages from other moraines in the western U.S., with all ages recalibrated to be consistent with our ages (Figs. 1 and 2).

This compilation shows that most glaciers began retreating from their LGM moraines between 22 and 18 ka (ref. 30). The LK moraine from the Sangre de Cristo Mountains, New Mexico, together with moraines of similar age from the Colorado Front Range, Snowy Range, and the Yellowstone region of Wyoming, reveal 75–90% of retreat from LGM limits by ~15 ka, when simulations with a global climate model suggest that deglacial warming for this region had only reached ~50% of interglacial values (Fig. 3f). Such a nonlinear response may be due to hypsometric factors30 or, in the case of the Yellowstone Ice Cap, dynamic controls.31 By 14–13 ka, glaciers across most of the western U.S. had retreated by at least 90% to reach new high-elevation positions and deposit moraines 1–2 km from cirque headwalls (Fig. 3f). We attribute this regional-scale glacier recession primarily to warming, since its spatial coherency and magnitude are unlikely to be explained by changes in precipitation.32 Transient climate modeling shows that the combination of increasing boreal summer insolation33 (Fig. 3a) and atmospheric CO2 (ref. 34) (Fig. 3b) caused the warming that led to this widespread deglaciation of western U.S. glaciers (Fig. 3f).

Fig. 3
figure 3

Normalized moraine distance from cirque headwalls through the last deglaciation across the western United States compared to reconstructed and simulated climate forcing. a Summer insolation at 45°N (JJA) and 45°S (DJF) (ref. 62). b Atmospheric carbon dioxide concentrations from the WDC and EPICA ice cores.34,63,64 c Northern hemisphere multiproxy temperature reconstruction for 30–60°N (refs. 22,48). d, e Sea surface temperatures off the northern California (ODP1019)61 and southern Alaskan (EW0408-85JC)38 coasts. f, g Normalized distances between LGM moraines and cirque headwalls in the western U.S. (f—this study and calculated results from ref. 30) and from New Zealand and Australia (g—calculated results from ref. 30), and simulated local summer temperatures from transient modeling.47,65 Orange (this study) and black (from ref. 30) circles are the mean of boulder surface exposure ages on each moraine and the bars represent the standard deviation (1σ) of the boulder ages and the production rate uncertainty, added in quadrature. The simulated temperatures are for the ALL (blue) and single-forcing orbital (green) and greenhouse gas (red) simulations. The blue vertical bar represents the YD interval, the end of which corresponds to the Pleistocene/Holocene boundary. EPICA European Project for Ice Coring in Antarctica, WDC West Antarctic Ice Sheet Divide ice core, LGM Last Glacial Maximum, YD Younger Dryas cold period


Glacial events from ~15 to 13 ka occurred when simulated interglacial summer temperatures in the western U.S. were largely established (Fig. 3f) and atmospheric carbon dioxide levels were relatively stable (Fig. 3b). Such glacier variability may thus reflect forcing by natural modes of multidecadal climate variability that influences the western U.S. glaciers.35 Alternatively, modeling has shown that even in a stable climate, stochastic interannual variability can lead to kilometer-scale changes in glacier length on centennial to millennial timescales.36,37 Our data also identify cirque-glacier fluctuations that fall within the Younger Dryas cold interval (Fig. 2b), suggesting a forced response to this event that has been identified in other western U.S. paleoclimate proxy records.38,39,40,41

Of the 19 moraines dated in this study, six have mean ages that indicate glaciers in some parts of the western U.S. persisted until the early Holocene (Fig. 2b). Model simulations indicate that maximum Holocene summer temperatures occurred between 11 and 9 ka in response to summer insolation (Fig. 3f), which is generally consistent with sea surface temperatures off the Oregon and Alaskan coasts (Fig. 3d, e), while our data suggest that some glaciers persisted during this warmest interval. Similar to the latest Pleistocene moraines, glacier variability during the earliest Holocene suggested by the moraine ages may have been forced by centennial to decadal climate variability and/or stochastic interannual climate variability. Following this final glacier phase, remaining cirque glaciers retreated further and may have completely disappeared in the western U.S. before onset of Neoglaciation some 7–6kyr later that led to renewed glacier growth, culminating in the LIA.17,21,23,42 We thus conclude that the LIA-associated moraines in western North America represent the most extensive glaciation of the last ~11–10ka, likely in response to a decrease in boreal summer insolation33 (Fig. 3a) and regional (Fig. 3f) and hemispheric (Fig. 3b) cooling through the mid-to-late Holocene,22 resulting in any record of earlier Neoglaciation being overridden by the LIA advance.

The glacial records from the western U.S. are in good agreement with well-dated records from the European Alps, which show that after retreating from their LGM positions to high-elevation cirques by ~15–13 ka, glaciers remained within their LIA limits throughout the Holocene.43,44 Glaciers in the Southern Alps of New Zealand also retreated far up-valley from their LGM positions by 15ka (Fig. 3g),45,46 consistent with a dominant greenhouse gas forcing that synchronized deglaciation in the two hemispheres.30 In contrast to the western U.S. and European Alps, however, glaciers in the Southern Alps of New Zealand then re-advanced during the Antarctic Cold Reversal,45 retreated during the Younger Dryas,46 and then continued to slowly retreat throughout the Holocene to their LIA positions45 (Fig. 3g). Transient climate modeling suggests that this anti-phased hemispheric glacier behavior during the Holocene (Northern Hemisphere regrowth and advance, Southern Hemisphere retreat) occurred in response to differing combinations of orbital and greenhouse gas forcing.30 In the Northern Hemisphere, the large decrease in summer insolation following its peak at ~8 ka dominated the temperature response, leading to cooling throughout the Holocene (Fig. 3f). In contrast, Southern Hemisphere glaciers experienced rising local summer insolation throughout the Holocene, leading to relatively cooler early Holocene temperatures that subsequently warmed (Fig. 3g). Thus, while glacial retreat in the two hemispheres was largely in phase during the last deglaciation, the hemispheric response became out of phase during the Holocene (Fig. 3f, g). This change in glacier response was likely caused by a switch from greenhouse gas-dominated forcing from ~18 to 11 ka during the last deglaciation47,48 to largely local summer insolation forcing through the Holocene. The general advance of glaciers during the LIA in both the Northern and Southern hemispheres remains an enigma, but may be related to complex global climate feedbacks driven by solar minima and more frequent explosive volcanism.49,50 The global retreat of alpine glaciers over the past century, however, indicates that the system has switched again to one dominated by greenhouse gas forcing,4 but now associated with anthropogenic carbon emissions.


Sample collection

Samples for surface exposure dating were collected from the tops of boulders resting on moraine crests at all of the field sites (Fig. S1) using hammers and chisels. Special care was taken to select boulders that met the following criteria: ~0.5 m or taller to limit the potential for excessive snow cover, or prior burial; resting on moraine crests to minimize chances of overturning or toppling during moraine degradation; little-to-no surface weathering features (e.g., pitting, exfoliation); and flat tops and/or simple surface geometries to limit uncertainties made for shielding calculations. Approximately 1–3 kg samples were collected from the top 2–4 cm of each boulder for cosmogenic nuclide chemistry and analyses. Shielding measurements were made in the field using a Suunto clinometer and boulder surface geometries were measured using a Brunton compass. Latitude, longitude, and elevation for each boulder were collected using a handheld Global Positioning System and crosschecked with topographic maps.

Laboratory Techniques

We prepared samples for 10Be analysis at Oregon State University and the University of Wisconsin-Madison following the methods of Licciardi51 and Rinterknecht,52 with modifications by Goehring.53 The top 1–2 cm of each sample was crushed in a Bico disk mill and sieved to the 250–710 µm size fraction. The magnetic fraction of each sample was removed using a Frantz Isodynamic Mineral Separator. All samples were first leached for 24–48 h in 5–10% HNO3 and then for an additional 24–72 h in a 1% HNO3 + 1% hydrogen fluoride (HF) solution in warm ultrasonic baths to remove all non-quartz minerals and meteoric 10Be from the quartz surfaces. Small subsamples were then analyzed for Al, Ca, Fe, K, Mg, Na, and Ti by inductively coupled plasma-optical spectroscopy following digestion in HF to determine quartz purity; samples were leached further if purity criteria were not met. After sufficient purity was obtained, 20–50 g of sample together with ~0.25–0.75 µg of Be carrier were added to Teflon beakers and dissolved in concentrated HF. The HF was then evaporated and several HClO4 and HCl evaporations were performed to convert the samples to chloride form. Samples were then passed through cation and anion exchange columns to isolate Be. After completing the chromatography, the pH of each sample was adjusted to 5 with NH4OH to precipitate any Ti remaining in the sample as Ti(OH)4, which was removed by centrifugation. Following the Ti-removal, the pH of each sample was adjusted to 8 to precipitate Be(OH)2 and then washed three times in ultrapure water adjusted to pH 8 to remove boron. The samples were transferred to acid-cleaned, low-Boron quartz crucibles and evaporated at 80–90 °C in a high efficiency particulate air filter laminar flow bench. The Be(OH)2 was then converted to BeO by heating the covered crucibles in a rapid mineralizer for 60 min at 999 °C. Finally, samples were packed into targets in a laminar flow hood for accelerator mass spectrometry (AMS) analysis at the Purdue Rare Isotope Measurement (PRIME) Laboratory.

Each batch of samples was processed with one or two system blanks to account for background 10Be introduced during processing or AMS measurement. Additionally, we made repeated measurements (n = 7) from a single sample (DPI-MB-01) throughout the duration of the project, as well as a limited number of replicate measurements (n = 3) for another sample (LL-MB-07). The standard error of the replicate measurements for DPI-MB-01 (350 yr) is equal to the internal uncertainty of the sample (see Supplementary Data file).

Exposure age calculations

After correcting for the system blank (<1%), 10Be concentrations in quartz were calculated from the 10Be/9Be ratio and carrier and sample amounts. Exposure ages were calculated for each boulder following the methods and online program of the CRONUS Online Calculator v.3 (ref. 54). All raw 10Be/9Be measurements were calibrated against the Revised ICN Standard (07KNSTD) following the protocol outlined by PRIME Lab. Cosmogenic nuclide exposure ages were calculated using the scaling model LSDn and the regional production rate of 3.92 ± 0.31 atom g−1 yr1 from Promontory Point in Utah.55

To calculate moraine ages and their uncertainties, we report the arithmetic mean of the boulder exposure ages and the standard deviation (1σ), but include alternate mean-age and uncertainty calculations in the Supplementary Data file. Outliers in a set of boulder ages from a single moraine are initially defined as being >5 standard deviations from the mean value of the other boulder exposure ages following a similar method as Putnam et al.56 This method of identifying outliers accounts for 4 of the 14 identified in this dataset. We then used Chauvenet’s criterion to identify ten additional outliers (see Supplementary Data file).

Because the majority of the boulders sampled had pristine surfaces that showed no clear post-depositional weathering features and were generally glacially polished and/or striated, we do not apply an erosion correction to our surface exposure ages. We also do not apply a snow correction to our surface exposure ages, as no correlation between boulder height and surface exposure age is distinguishable from our dataset, and we consider our boulder sampling sites, which are from the exposed moraine ridges, to be generally wind swept throughout the year. However, assuming our boulder ages were covered by 0.5 m of snow (ρ = 0.3 g cm−3) for 4 months of the year, a correction of ~3% would be necessary,57 which is well within the age uncertainties from the scaling and production calculations.