Glacial weathering, sulfide oxidation, and globalcarbon cycle feedbacksMark A. Torresa,b,c,1, Nils Moosdorfa,d,e,1, Jens Hartmannd, Jess F. Adkinsb, and A. Joshua Westa,2

aDepartment of Earth Sciences, University of Southern California, Los Angeles, CA 90089; bDivision of Geological and Planetary Sciences, California Instituteof Technology, Pasadena, CA 91125; cDepartment of Earth, Environmental, and Planetary Sciences, Rice University, Houston, TX 77005; dInstitute forGeology, Center for Earth System Research and Sustainability (CEN), Universität Hamburg, 20146 Hamburg, Germany; and eLeibniz Center for TropicalMarine Research, 28359 Bremen, Germany

Edited by Thure E. Cerling, University of Utah, Salt Lake City, UT, and approved July 5, 2017 (received for review February 21, 2017)

Connections between glaciation, chemical weathering, and the globalcarbon cycle could steer the evolution of global climate over geologictime, but even the directionality of feedbacks in this system remain tobe resolved. Here, we assemble a compilation of hydrochemical datafrom glacierized catchments, use this data to evaluate the dominantchemical reactions associated with glacial weathering, and explore theimplications for long-term geochemical cycles. Weathering yields fromcatchments in our compilation are higher than the global average,which results, in part, from higher runoff in glaciated catchments.Our analysis supports the theory that glacial weathering is character-ized predominantly by weathering of trace sulfide and carbonate min-erals. To evaluate the effects of glacial weathering on atmosphericpCO2, we use a solute mixing model to predict the ratio of alkalinityto dissolved inorganic carbon (DIC) generated by weathering reactions.Compared with nonglacial weathering, glacial weathering is morelikely to yield alkalinity/DIC ratios less than 1, suggesting that enhancedsulfide oxidation as a result of glaciation may act as a source of CO2 tothe atmosphere. Back-of-the-envelope calculations indicate that oxida-tive fluxes could change ocean–atmosphere CO2 equilibrium by 25 ppmor more over 10 ky. Over longer timescales, CO2 release could act as anegative feedback, limiting progress of glaciation, dependent on lithol-ogy and the concentration of atmospheric O2. Future work on glacia-tion–weathering–carbon cycle feedbacks should consider weatheringof trace sulfide minerals in addition to silicate minerals.

weathering | carbon cycle | glaciation | oxidation | sulfide

Episodes of glaciation represent some of the most dramaticchanges in the history of Earth’s climate. Basic questions

remain unanswered about why the planet has occasionally but notalways supported glaciers, about why some glaciations have beenmore intense than others, and about how glaciers have shapedEarth’s landscapes, chemical cycles, and climate. Before the Phan-erozoic (i.e., before ∼540 Mya), several glaciations are thought tohave approached complete or near-complete planetary ice cover(“Snowball Earth” events; ref. 1). In contrast, more recent glacia-tions, including the Quaternary glaciations of the last million years,have been characterized by oscillating climate with ice extent lim-ited to high latitudes even during glacial maxima (2). The formationof ice cover has long been known to fundamentally alter Earth’senvironment, changing albedo (3), influencing atmospheric andocean circulation (4), and reshaping landscapes (5). These diverseeffects may contribute to planetary-scale feedbacks that influencethe transition to and the evolution of glaciation.Chemical weathering influences the composition of the atmo-

sphere by releasing ionic species from minerals that alter the al-kalinity and redox condition of the ocean–atmosphere system.Weathering of silicate minerals produces alkalinity, removingcarbon from the atmosphere in the process (6). If glaciers reduceweathering fluxes from land area that was previously undergoingsilicate weathering, glaciation may decrease global alkalinity pro-duction, allow atmospheric CO2 levels to rise, and thus generate anegative feedback that ends glaciation (7). On the other hand,some evidence suggests that glaciation increases rates of physical

erosion (5, 8, 9), which is thought to promote more rapid silicateweathering (10, 11). If glaciation increases erosion rates, and iferosion enhances silicate weathering and associated production ofalkalinity, then glacial weathering could act as a positive feedbackserving to promote further glaciation (5, 12).Glaciation may also affect rates of weathering of nonsilicate

minerals. Studies of glacial hydrochemistry have pointed to the im-portance of sulfide oxidation and carbonate dissolution as majorsolute sources (13–15). Because the residence time of sulfate in theoceans is long relative to Ca, sulfide–carbonate weathering can act torelease CO2 to the ocean–atmosphere system over timescalesof <100 My (16, 17). Sulfide oxidation also influences global budgetsof redox-sensitive elements including sulfur, iron, and oxygen (16, 18,19). Although sulfide and carbonate minerals are usually present atlow abundance in most rocks, their rates of dissolution are typicallyorders of magnitude faster than for silicate minerals. For example,log rates from laboratory experiments are −6 to −10 for pyrite (20),vs. log rates of −10 to −16 for silicate minerals (21). Thus, even asmall proportion of sulfide or carbonate present in rocks can dom-inate solute production (22, 23). Indeed, weathering of these tracephases is predominantly “supply limited” in the present day, so fluxesincrease with erosion that supplies more minerals for reaction (17,24). In contrast, silicate weathering is not always limited by supply ofweatherable minerals (25). As a result, if glaciation enhances phys-ical erosion rates, the increased exposure of fresh rock materialcontaining unweathered minerals should drive increased sulfide andcarbonate weathering fluxes, but may not drive increased silicateweathering fluxes in tandem. Thus, the net effect of glaciation couldbe to supply CO2 to the atmosphere, rather than to remove it.

Significance

We compile data showing that, as hypothesized previously, watersdraining glaciers have solute chemistry that is distinct from non-glacial rivers and reflects different proportions of mineral weath-ering reactions. Elevated pyrite oxidation during glacial weatheringcould generate acidity, releasing carbon to the atmosphere. Weshow that this effect could contribute to changes in CO2 duringglacial cycles of the pastmillion years. Over the longer, multimillion-year timescales that Earth transitions into and out of glaciatedstates, sustained addition of pyrite-derived sulfate to the oceanscould shift the balance of the global carbon cycle toward increasingCO2 in the ocean–atmosphere, thus providing a negative-feedbackmechanism preventing runaway glaciation. This mechanism de-pends on oxidation and thus sufficient O2.

Author contributions: N.M., J.H., and A.J.W. designed research; M.A.T., N.M., and J.H.performed research; M.A.T. and J.F.A. contributed new reagents/analytic tools; M.A.T.,N.M., J.H., J.F.A., and A.J.W. analyzed data; and M.A.T. and A.J.W. wrote the paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.1M.A.T. and N.M. contributed equally to this work.2To whom correspondence should be addressed. Email: joshwest@usc.edu.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1702953114/-/DCSupplemental.

8716–8721 | PNAS | August 15, 2017 | vol. 114 | no. 33 www.pnas.org/cgi/doi/10.1073/pnas.1702953114

The directionality of long-term glaciation–weathering feed-backs depends on the extent to which glacial weathering is a netsource or net sink of CO2 to the atmosphere, which in turn de-pends on the balance of sulfide vs. silicate mineral weathering. Agrowing number of studies have explored various aspects ofglacial weathering by analyzing the chemical composition ofglacial runoff, often focusing on specific case studies, and insome cases synthesizing global datasets (13, 26, 27). Here, weassemble an expanded compilation of glacial weathering dataand use this compilation to identify characteristic features ofpresent-day glacial weathering, including the relative roles ofsulfide-carbonate vs. silicate mineral weathering compared withrivers not dominated by glacial processes. We then consider thepossible implications for global-scale fluxes in the Quaternarycarbon cycle and for planetary feedbacks over longer timescales.

Materials and MethodsWe compiled a database containing stream chemistry data from glaciatedcatchments, together with spatial data describing the contributing catch-ments (SI Appendix, Fig. S1, and Datasets S1 and S2). The dataset is based onsolute concentrations for 7,700 samples, from 95 glaciated catchments(glacial cover proportion >0.4%, and in most cases much higher; Fig. 1) withreported concentrations of Ca, Mg, Na, and K for at least one sample.Measured discharge was available for 52 of these catchments. The contri-bution from chemical weathering to the measured stream chemistry wasdetermined following correction for precipitation inputs, as detailed in SIAppendix. To provide a basis for comparing the glacial weathering data, werefer to the chemical composition of the world’s large rivers (28), whichprovide a spatially integrated view of weathering processes over continentalareas and reflect the composition of solutes delivered to the oceans. Forreference, we also consider data from the Global River Chemistry (GloRiCh)database of solute chemistry from 1.27 million water samples from∼15,500 catchments around the world (29) (Dataset S3).

ResultsGeneral Characteristics of Glacial Weathering. The mean of the sum ofcation concentrations (Ca plus Mg plus Na plus K) in the 95 glaci-erized catchments is 530 μmol·L−1, decreasing to 453 μmol·L−1 aftercorrection for atmospheric inputs. The weathering-derived solutesare dominated by Ca (58.6 mol %, SI Appendix, Table S2). Mg andNa contribute 17.5% and 17.4%, respectively, followed by K(6.5%). For the 54 glacial catchments with sufficient data to esti-mate dissolved pCO2 using PHREEQC (Dataset S1), 45 are su-persaturated relative to the atmosphere (median pCO2 for theglacial catchments, 692 ppm; mean, 4,029 ppm).For the 52 catchments with available runoff data, mean runoff is

1,730 L·m−2·y−1, which is 5.8 times above the global average riverrunoff (30). The mean cation weathering yield from the glacialcatchments is 21.5 t·km−2·y−1. For comparison, the mean cation yieldfrom the world’s large rivers is 18 t·km−2·y−1, with a correspondingmean cation concentration of 2,433 μmol·L−1 (31). Thus, relative toglobal average river water, glacial catchments are characterized bybelow average cation concentrations but above average cation yields,due to high runoff (26).Within the glacial dataset, precipitation-corrected cation con-

centrations are positively correlated with the proportion ofcarbonate-rich sediment in the catchments (r = 0.26, SI Appendix,Table S3) and negatively with coverage of metamorphic rocks(r = −0.24) and glacial cover (r = −0.27). The cation yield of theobserved catchments is correlated with runoff (r = 0.28), but thiscorrelation is weaker than reported for nonglaciated catchments(32, 33). Correlations between lithology classes and cation yield aregenerally insignificant, although some significant correlations can beseen for individual elements (SI Appendix, Table S3). Additionalweathering of glacially derived sediment is thought to take place inproglacial environments (34). Consistent with this hypothesis, cationyields in our compilation show considerable scatter but indicate aslight increase with the proportion of glacial cover (Fig. 1). Of allanalyzed variables, cation yield is most highly correlated with the

proportion of glacial cover and the sampling distance from theglacial snout (SI Appendix, Table S3). Subglacial weathering fluxesalso depend on hydrology, for example, differing for warm- vs. cold-based ice (35, 36) and along different flowpaths (37), and thesedifferences may explain some of the variability within the dataset.

Distinct Stoichiometry of Glacial Weathering.Our compilation providesthe opportunity to test the hypothesis that the solute stoichiometryof glaciated rivers is distinct from nonglaciated rivers, because thelarge number of data points allows for statistical comparison be-tween distributions. Although we do not directly evaluate temporalvariability, dissolved major element concentrations in river watersvary only slightly over time (38) relative to differences betweencatchments. Dissolved K:Na ratios, which reflect the stoichiometryof silicate mineral dissolution, are shifted to higher values in theglacial dataset compared with either the world’s large rivers or theGloRiCh dataset (distributions are significantly different at the P <0.05 level via a Kolmogorov–Smirnov test), consistent with results ofprevious studies and suggesting changes in the stoichiometry ofweathering-derived solutes driven by glaciation (15, 39).Ca:Na and SO4:Na ratios (Fig. 2) are also statistically different (at

P < 0.05 level) when comparing our glacial database vs. eithernonglacial dataset. In contrast, when comparing between the twononglacial datasets (GloRiCh vs. the world’s large rivers), Ca:Naand SO4:Na ratios are not statistically different. Glacial solutegeochemistry is significantly distinct even after subsampling theGloRiCh database for rivers with the same range of catchmentareas as the glacial database (SI Appendix, Fig. S3 A and B). GlacialSO4:Na ratios are weakly correlated with the distance from theglacier snout (SI Appendix, Fig. S3C) and do not depend oncatchment size (SI Appendix, Fig. S3A) or the proportion of glacialcover (SI Appendix, Fig. S3D). So, although additional weatheringmay occur in the proglacial environment (Fig. 1), it does not appearto be a major control on elemental ratios (SI Appendix, Fig. S3).Idealized versions of the reactions expected to generate dis-

solved Ca, Na, SO4 are summarized as follows:

Carbonate  dissolution:  2H+ +CaCO3 ↔Ca2+ +H2CO3, [1]

Silicate  dissolution:  4H+ +Ca2SiO3 ↔ 2Ca2+ +H4SiO4, [2]

Carbonic  acid: CO2ðdissÞ +H2O↔H2CO3 ↔ 2H+ +CO32−, [3]

Sulfuric  acid  from  sulfide  oxidation:  FeS2 + 15O2 + 14H2O

↔ 16H+ + 8SO42− + 4FeðOHÞ3.

[4]

Evaporites may also contribute Ca and SO4. Assuming glacia-tion is not associated with preferential evaporite weathering (an

0 25 50 75 100Glacial cover (%)

0

0.5

1.0

1.5

2.0

Cor

rect

ed c

atio

n co

ncen

trat

ion

(mm

ol/L

)

0

25

50

75

100

Wea

ther

ing

rate

(t k

m-2

yr-

1 ) Dominant lithology(>70% catchment area) Siliclastic sediment

Metamorphic rock Volcanic rock Plutonic rock No dominant lithology

Carbonate-rich sediment Trendline:r = -0.31; p < 0.01

Trendline:r = -0.33; p = 0.02

0 25 50 75 100Glacial cover (%)

A B

Fig. 1. Scatter plots of (A) cation concentration corrected for precipitationinput (*), and (B) total cation weathering flux, both plotted against glacialcover. Fluxes do not decrease to zero as glacial cover approaches completecover, indicating significant weathering under ice.

Torres et al. PNAS | August 15, 2017 | vol. 114 | no. 33 | 8717

EART

H,A

TMOSP

HER

IC,

ANDPL

ANET

ARY

SCIENCE

S

assumption we test below), elevated SO4:Na ratios during glacialweathering suggest relatively more sulfide mineral oxidation com-pared with silicate weathering in glacial catchments, and elevatedCa:Na ratios suggest more carbonate weathering than silicateweathering. These observations are consistent with previous studiesemphasizing the role of sulfide and carbonate mineral weathering inglacial catchments, even when these phases are minor constituentsof the bedrock (e.g., refs. 13–15 and 26).

Glacial Weathering as Net CO2 Source.Higher dissolved ion yields andhigher SO4:Na and Ca:Na ratios together suggest that sulfide andcarbonate dissolution rates are enhanced in association with glacialweathering. To first order, glaciation thus appears to act as a CO2source, by maintaining total solute fluxes while shifting weatheringtoward reactions that release rather than consume CO2.In principle, glacially enhanced Ca and SO4 fluxes could de-

rive from preferential dissolution of sulfate-bearing evaporitesor weathering of Ca-bearing silicates by sulfuric acid, reactions

that would not act as CO2 sources. We have no a priori reason toexpect that such effects dominate globally. Nonetheless, fullyunderstanding the net effect of glaciation on the carbon cycledepends on quantifying (i) the proportion of the dissolved sulfatederived from sulfide mineral oxidation, and (ii) the extent towhich this oxidative flux of protons (Eq. 4) is balanced by alka-linity generation from silicate and carbonate weathering (Eqs. 1and 2). To approach these questions quantitatively, we use amixing model to estimate the relative contribution of differentweathering reactions to the solute load for each catchment in ourcompilation. The model uses Na-normalized elemental ratios(Cl:Na, Ca:Na, and Mg:Na) to partition the dissolved load be-tween silicate, carbonate, evaporite, and precipitation sources(e.g., ref. 40 and details in SI Appendix). We apply our model toall samples from the world’s large river database (28) and a subsetof samples from the glacial (n = 84) database due to the datarequirements. We do not consider the GloRiCh data because thenonuniform sampling distribution in this database (e.g., multiplesamples from many small catchments in one region) compli-cates interpretation of results from our probabilistic approach,described below.We do not precisely know the end-member ratios appropriate

for each catchment within the large river and glacial catchmentdatasets, and these ratios may be modified by secondary pro-cesses. Thus, we cannot calculate a single model result for eachcatchment. Instead, we randomly sample across a uniform dis-tribution of possible end-member values (e.g., Fig. 2B), repeatingfor 10,000 combinations. For each combination, we use the frac-tional contribution from each solute source to calculate a ratio ofalkalinity to dissolved inorganic carbon equivalents (Alk:DIC)produced by weathering reactions.The Alk:DIC ratio allows us to evaluate the net effect of

weathering on atmospheric pCO2 over different timescales (17).The pCO2 of the atmosphere and ocean are coupled by gas ex-change and, through carbonate equilibria, are set by the ratio ofdissolved alkalinity to DIC concentrations in the oceans (Fig. 3).The ocean ratio is, in turn, determined by the supply of alkalinityand DIC from weathering reactions and their removal throughthe burial of carbonate minerals and organic carbon. Carbonateremoval is fixed at an Alk:DIC of 1:0.5 (Eq. 1; 1:0.5 line in Fig.3). The Alk:DIC of weathering varies, from 0 to infinity. Over thetimescales characteristic of carbonate burial (greater than ∼104 y),weathering will change the Alk:DIC ratio of the oceans (and thuspCO2) if it supplies alkalinity and DIC in a ratio that deviatesfrom the 1:0.5 associated with carbonate burial (Fig. 3). Thus,silicate weathering by carbonic acid (supplying Alk:DIC at a1:0 ratio, higher than 1:0.5) will consume CO2, whereas car-bonate weathering by carbonic acid (1:0.5) will have no net effect(6, 41). Carbonate weathering by sulfuric acid (0:0.5 ratio) willrelease CO2 over tens of million years until sulfate is reduced inmarine sediments, returning alkalinity (reverse of Eq. 4; ref. 16).On timescales shorter than required for the output flux via car-bonate precipitation to respond to changes in input fluxes (103 to104 y), pCO2 will increase if weathering inputs have an Alk:DIClower than the ocean (Fig. 3), which is presently approximatelyequal to 1, and vice versa. We thus evaluate model results withrespect to Alk:DIC thresholds of 1 (characteristic of shorttimescales) and 2 (long timescales). CO2 is released if weatheringoccurs with a ratio below these thresholds; CO2 is consumedabove these thresholds.We determine the Alk:DIC of weathering in each catchment on

the basis of the equivalents of cations released for combinations ofreactions, coupling acid-generating reactions (sulfide and CO2dissolution) with acid-consuming reactions (silicate and carbonatedissolution) (17) (Fig. 3A). We use the proportions of solutesfrom carbonate, silicate, and sulfide weathering, as inferred fromthe end-member mixing analysis, to tabulate the equivalents ofeach reaction required to produce the solute chemistry. Protons

-3

-2

-1

0

1

2

log

10(S

O4 /

Na)

-2 -1 0 1 2log10(Ca/Na)

-2

-1

0

1

2

log

10(M

g/N

a)

SO / NaMean / Median

0.9 / 0.43.7 / 1.00.5 / 0.4

Ca / NaMean / Median

2.9 / 1.19.3 / 4.30.8 / 0.7

GloRiChGlacialWorld

Silicate

Carbonate

Evaporite

A

B

Mg / NaMean / Median

1.2 / 0.52.6 / 0.8 0.8 / 0.7

1:1

Fig. 2. Stoichiometry of dissolved ions in the glacial data compilation presentedin this study (blue points, and blue dash-dot curves in histograms), theworld’s largerivers (red points, and red dashed curves; data from ref. 28), and the GloRiChdatabase of 17,000 nonglaciated catchments (gray points, and solid gray lines; ref.29). (A) SO4/Na vs. Ca/Na molar ratios, with population histograms (normalized tonumber of catchments) of both ratios shown along each axis (reported mean andmedian values are not log-transformed). (B) Ca/Na vs. Mg/Na ratios, with the rangeof each lithological end-member illustrated on the figure and the values listed in SIAppendix, Table S4. The GloRiCh data illustrate that our choices of end-membersbracket most river compositions.

8718 | www.pnas.org/cgi/doi/10.1073/pnas.1702953114 Torres et al.

not derived from sulfide oxidation (Eq. 4) are assumed to befrom carbonic acid (Eq. 3). For each catchment, based on theend-member mixing results across 10,000 different end-memberratios, we determine the proportion of the model simulations,F(x), that yield Alk:DIC < 1 and the proportion that yield Alk:DIC < 2. A higher value of F(x) indicates greater likelihood ofCO2 release over either short or long timescales, respectively, withF(x) > 0.9 reflecting that 90% of the model solutions for a givenriver predict CO2 release rather than consumption.To compare between the glacial and nonglacial datasets, we plot

the cumulative distribution of the F(x) values for the catchments ineach data compilation (Fig. 4). Calculated proportions of eachweathering reaction are sensitive to the Ca:Na ratio of evaporiteinputs, which is not well known and which may vary from site to site.To illustrate the sensitivity to this parameter, we plot multiplecurves for each dataset, assuming different maximum values of (Ca:Na)evaporite [the F(x) curves are not sensitive to choosing differentminimum values]. Independent of the assumed value for (Ca:Na)evaporite, more catchments from the glacial database have higherF(x) than the signature of average continental weathering reflectedin large rivers (Fig. 4). We thus infer that glacial catchments aremore likely than nonglaciated catchments to yield low Alk:DICratios, as expected for glacial catchments characterized by moresulfide oxidation relative to silicate weathering. For example, for∼20% of the catchments in the glacial database, more than 90% ofmodel results yield Alk:DIC < 2, whereas none of the world riversyield model results where more than 90% have Alk:DIC < 2. Thisresult is not strongly biased by lithological differences in end-member composition; as shown in SI Appendix, Fig. S5, glacierizedcatchments spanning the range of mapped lithology types yield highF(x), that is, high likelihood of Alk:DIC < 1 or 2. Our probabilisticapproach, necessitated by weak constraints on catchment-specificend-member values, can at most be interpreted to reflect a greaterlikelihood of glacial weathering to act as a CO2 source, relative tononglacial weathering. Some model solutions include seeminglyunlikely but not altogether impossible combinations, for example,the majority of cations sourced from precipitation and evaporitessimultaneous with the majority of SO4 sourced from sulfides.Further work partitioning solute sources rigorously by individualcatchment, for example, including constraints from S isotopes (17,24), will be important to more thoroughly test the probabilisticinferences made here. Nonetheless, our data suggest that present-day glacial systems are characterized by more sulfide oxidationand carbonate weathering than nonglacial systems, consistent withprevailing interpretations (13).

DiscussionWhy Increased Oxidation Fluxes from Glacial Weathering. The dis-tinct geochemistry of glacial weathering might be coincidental;glaciers in the present day are typically found in steep, rapidlyeroding mountainous regions (8), where fluxes from sulfide oxida-tion are relatively high independent of glaciation (16). Although ourglobal analysis cannot conclusively distinguish causality, we think itis more likely that high sulfide weathering fluxes are the result ofglaciation, driven by glacial enhancement of erosion. Several studieshave proposed that glaciation increases erosion rates (5). Becausesulfide weathering is typically supply limited, increases in erosionrate as expected from glaciation should increase oxidation fluxes(17, 24). It is also possible that sediment production and commi-nution during glacial retreat (8, 9) exposes fresh sulfides and facil-itates their oxidation, even if longer-term rates of erosion (and thusof sulfide oxidation) are not elevated as a result of glaciation. In thislatter case, high sulfide oxidation rates in the present day may re-flect a transient postglacial increase but would not be sustained overtimescales >105 y. High present-day sediment fluxes from glaci-erized catchments (8) and the oscillations in the Os isotope com-position of seawater over Quaternary glacial–interglacial cycles (42)both support a causative rather than coincidental relationship be-tween glaciers, erosion, and sulfide weathering fluxes. However, itremains unclear whether this is a transient effect of deglaciation or along-term sustained increase in erosion and oxidation.Low temperatures associated with glacial catchments may also

contribute to the high sulfide relative to silicate weatheringfluxes, and resulting low Alk:DIC ratios. If low temperaturesinhibit silicate mineral weathering (11), whereas high rates ofsulfide weathering are sustained by microbially mediated reac-tions (43), we would expect that colder, glaciated catchmentswould have higher ratios of solutes derived from sulfide vs. sili-cate mineral sources. Additional weathering of glacially derivedsilicate material during fluvial transport may shift the overallweathering balance toward CO2 consumption (17). However,significant amounts of glacial debris are deposited without suchreworking, and we do not observe systematic trends betweenNa:SO4 and catchment area (SI Appendix, Fig. S3), suggesting thatadditional weathering of glacial detritus does not necessarily ne-gate enhanced sulfide oxidation.

Oxidative CO2 Release over Quaternary Glacial Cycles. The modelresults indicate that glacierized catchments are more likely thannonglacierized catchments to generate weathering fluxes withAlk:DIC < 1 (Fig. 4), the threshold for causing pCO2 changesover millennial timescales. Could the magnitude of changes inglacial weathering be significant as a CO2 source over timescales

2 2.1 2.2 2.3 2.4 2.5Alkalinity (mEq kg-1)

2

2.1

2.2

2.3

2.4

2.5

DIC

(m

mol

kg-1

)

Contours = log( pCO ) ( atm) 2 1:1

i

iiiiv

v

B3.80

3.553.30

3.05

2.80

2.55

2.31

A

Reaction (Alk:DIC)i. Silicate + Carbonic (1:0) ii. Silicate + Sulfuric (0:0) iii. Carbonate + Carbonic (1:0.5) iv. Carbonate + Sulfuric (0:0.5) v. Carbonate Burial (-1:-0.5)

Fig. 3. Alk:DIC framework for evaluating effect of weathering on pCO2.(A) Tabulation of Alk:DIC associated with different reaction combinations,after normalization by the charge equivalents of cations released. (B) Con-tours of oceanic pCO2 (equivalent to atmospheric pCO2 over timescales ofgas exchange) as a function of alkalinity and DIC, calculated using CO2sys(67). Dashed black line shows 1:1 addition of Alk:DIC (stoichiometry ofconstant pCO2 over timescales shorter than the response time of the oceancarbonate system). Vectors correspond to combinations of reactions in A.

0 0.2 0.4 0.6 0.8 10

0.2

0.4

0.6

0.8

1

Cum

ulat

ive

prop

ortio

n of

dat

aset

≤ F

(X)

0 0.2 0.4 0.6 0.8 1F(X) for Alk/DIC ≤ 2

Glacial Rivers

A B

Color shading = upperbound on Ca:Na of

evaporite end-member(light = 0.2, dark = 2)

F(X) or fraction of model results for anindividual river with Alk/DIC ≤ 1

Worlds' Largest Rivers

Fig. 4. Effect of glacial vs. nonglacial weathering on the carbon cycle.(A) Proportion of model simulations, F(x), that yield Alk:DIC < 1 for glacialcatchments (blue lines) and the world’s large rivers (28), red lines. (B) As in A, butproportion of model simulations that yield Alk:DIC < 2. The Monte Carlo simu-lations used to calculate F(x) propagate uncertainties on solute chemistry andend-member compositions (Fig. 2 and SI Appendix). Regardless of assumedmaximum Ca:Na ratio of evaporite weathering, more glacial catchments arelikely to be characterized by CO2 release compared with the world’s large rivers.

Torres et al. PNAS | August 15, 2017 | vol. 114 | no. 33 | 8719

EART

H,A

TMOSP

HER

IC,

ANDPL

ANET

ARY

SCIENCE

S

of Quaternary glacial–interglacial cycles? Simplistically taking thedifference in median glacial vs. nonglacial SO4:Na ratios at facevalue (Fig. 2), glacial weathering might increase sulfide oxidationyields on average by ∼2.5× relative to nonglacial weathering. As-suming present-day global sulfide oxidation fluxes of∼1.5 × 1012 mol/yS (44), Last Glacial Maximum (LGM) glacial weathering over10% of global land area (14.5 × 106 km2; ref. 2) could generate anadditional ∼2.25 × 1011 mol/y S, releasing up to 4.5 × 1011 molCO2/y if sulfide oxidation produces CO2:SO4 in a 2:1 ratio (themaximum possible based on reaction stoichiometry; ref. 16). Over10 ky, for example during glacial onset or deglaciation, theoxidation-related flux would then be on the order of ∼4.5 × 1015

mol CO2, equivalent to a ∼25 ppm increase in atmospheric con-centration (for an atmosphere of 1.8 × 1020 mol). In the moreextreme case of doubling the entire global sulfide oxidation flux, assuggested from modeling of Os isotope variations (42), the asso-ciated increase in SO4 flux of 7.5 × 1011 mol/y could potentiallyamount to a much larger CO2 release of 1.5 × 1016 mol, equivalentto a ∼80 ppm increase in CO2.An increased oxidation flux of 2.5–7.5 × 1011 mol S/y associated

with glaciation would require glacially facilitated erosion of 16–48 ×1014 g sediment/y, assuming average pyrite S content in erodedrocks of 0.5 wt% S, as is characteristic of sedimentary rocks (45). Toproduce this material, sediment yield from an LGM glaciated areaof 14.5 × 106 km2 would have to be ∼110–330 t·km−2·y−1, well withinthe range of rates of glacially enhanced erosion rates observedduring the present deglaciation (8). Resulting changes in concen-trations of sulfate in the oceans or O2 in the atmosphere would bevery small and likely difficult to detect relative to the much largerreservoirs. Driving all oxidation by atmospheric O2 would consume5–14 × 1011 mol O2/y, or 5–14 × 1015 mol O2 over 10 ky, less than0.01–0.04% of the atmospheric reservoir of ∼4 × 1019 mol O2 andwithin the range of observed declines in pO2 over the past 1 My(46). Similarly, proxies or models of total weathering fluxes to theoceans (36, 47, 48) may not capture modest increases in thesesulfate fluxes because oxidative weathering is a small portion of thetotal global weathering flux.The scenarios explored here are highly simplistic, based on ap-

proximate estimates of 20–100% changes in global sulfide oxida-tion fluxes resulting from glaciation, and release of CO2 in a2:1 proportion to SO4. Moreover, the timing of enhanced oxidativeweathering during glacial–interglacial cycles is not well known. Inaddition, these calculations do not include other biogeochemicalprocesses that affect the carbon cycle, some of which may also beinfluenced by glacial erosion, such as organic matter cycling (e.g.,refs. 49 and 50) and sulfur redox transformation (e.g., refs. 43 and51). Nonetheless, our calculations serve to illustrate that oscillationof weathering between more and less sulfide oxidation over Qua-ternary timescales, expected based on the chemistry of present-dayglacial weathering, could play a quantitatively meaningful role inthe carbon cycle over glacial–interglacial cycles.

Glacial Weathering and Stable pCO2 for the Last 1 Myr. Our analysissuggests that present-day glacial weathering causes net CO2 re-lease, but measurements from ice cores show no long-term changein mean atmospheric pCO2 averaged over multiple glacial cyclesduring the past 1 My (52, 53). Highly responsive global silicateweathering feedbacks could provide one mechanism explainingconstant pCO2 (54). Another explanation, consistent with ourobservation of glacially enhanced CO2 release in the present day,is that the extent of glaciation has adjusted over time such that theamount of CO2 released via glacial weathering compensates forthe external driver that initially forced decreases in pCO2 andglobal cooling before 1 Ma. This forcing may have been via de-creased fluxes from solid Earth degassing (6) or greater weath-erability of continental silicate minerals (55–57). As pCO2dropped and climate cooled between 3 and 1 Ma (58), glacialadvance would have stimulated greater release of CO2 until

glaciation began producing enough CO2 during each glacial cycleto prevent further declines in atmospheric concentrations—even-tually reaching a stable steady state as seen for the past 1 My. Thishypothesis could potentially be tested with high-resolution pCO2records during Pleistocene cooling between 3 and 1 Ma.Climatic stability also followed the rapid cooling and onset of

Antarctic glaciation at the Eocene–Oligocene transition, 34–33 Ma. We hypothesize that CO2 release from oxidative glacialweathering could have contributed to the transition to stableOligocene climate, in addition to possible silicate weatheringfeedbacks (59). Nd and Pb isotope records from the southernOcean suggest a brief pulse of increased Antarctic weathering,∼33.9–33.5 Ma (60, 61). Longer-lived changes in 206Pb/204Pb areproposed to reflect increased carbonate weathering (61), andincreases in 187Os/188Os, a potential proxy for weathering ororganic- and sulfide-rich rocks (42), also continue after 33 Ma(62). Although aridity limits present-day Antarctic denudationand solute fluxes (63), sulfide oxidation associated with glacia-tion should at least be considered in evaluating the early Oli-gocene carbon cycle.

Glacial Weathering and Carbon Cycle Feedbacks over Earth’s History.Over geologic timescales, instead of glacial erosion driving a positivefeedback via silicate weathering and CO2 drawdown (5, 12), our datasuggest that C release dominates in glacial settings, such that glacialweathering could act as a negative feedback preventing the Earthsystem from descending into colder conditions. A feedback via glacialsulfide oxidation depends on elevated oxidation rates being sustainedover millions of years. Persistent increases in glacial erosion overthese timescales, as suggested for the Plio-Pleistocene (5), couldmaintain increased oxidative weathering fluxes, providing a mecha-nism for weathering to limit the magnitude of global cooling. How-ever, the evidence for glacially enhanced erosion rates continuingover >1 My remains controversial (e.g., refs. 8 and 64). Alternativelyor in addition, low temperatures could nudge global weathering awayfrom silicate weathering and thus toward lower Alk:DIC ratios.Either way, our analysis points to sulfide oxidation as a plausible

mechanism by which glaciation might be self-limiting, shifting thebalance of the global carbon cycle by increasing CO2 supply to theatmosphere–ocean system (Fig. 5). The O2 concentration in theatmosphere places a further constraint on the extent to which such amechanism may be effective (19). The O2-rich atmosphere of theCenozoic holds a sufficiently large reservoir to accommodatechanges in sulfide oxidation without changing atmospheric O2concentrations outside of known bounds (16). At times of very lowor effectively no atmospheric O2 concentration such as in the Ar-chaean and Paleoproterozoic (65), glacially driven CO2 release byoxidative weathering may have been less active (e.g., ref. 66). Thiscontrast raises the intriguing possibility of whether glaciations earlyin Earth’s history, when atmospheric O2 levels were lower, wereprone to being more globally pervasive compared with glaciations inthe Phanerozoic. Similarly, the effectiveness of an oxidativeweathering feedback depends on the sulfide content of rocks un-dergoing weathering, hinting at the possibility that growth of asulfide reservoir in Earth’s crust over time could have helped tostabilize climate.

Glaciation+ Sulfide

Oxidation CO2

Negative feedback via oxidative weatheringpreventing runaway glaciation?

+Glaciation

+ SilicateWeathering CO2

In absence of oxidative weathering, positive silicate weathering feedback?

–

– –

Fig. 5. Earth system feedbacks via glacial weathering. Sulfide oxidationseen in present-day glacial weathering may inhibit runaway glaciation. Sucha stabilizing feedback may not operate at lower pO2, for example, early inEarth’s history or on other planets.

8720 | www.pnas.org/cgi/doi/10.1073/pnas.1702953114 Torres et al.

ACKNOWLEDGMENTS. We thank three reviewers for constructive com-ments. M.A.T. was supported by a University of Southern CaliforniaCollege Merit Fellowship and a Caltech Texaco Postdoctoral Fellowship,N.M. by a Deutscher Akademischer Austauschdienst exchange fellow-ship, J.H. by the German Science Foundation (DFG-project HA4472/6-1

and the Cluster of Excellence “CliSAP,” EXC177, Universität Hamburg) andBundesministerium für Bildung und Forschung Project PALMOD (Ref01LP1506C), and A.J.W. by National Science Foundation Grant EAR-1455352.Requests for access to the GloRiCh database should be addressed to J. Hartmannat geo@hattes.de.

1. Hoffman PF, Kaufman AJ, Halverson GP, Schrag DP (1998) A neoproterozoic snowballearth. Science 281:1342–1346.

2. Ehlers J, Gibbard PL, Hughes PD (2011) Quaternary Glaciations—Extent andChronology : A Closer Look (Elsevier, Amsterdam).

3. Budyko MI (1969) The effect of solar radiation variations on the climate of the Earth.Tellus 21:611–619.

4. Kennett JP (1977) Cenozoic evolution of Antarctic glaciation, the circum-AntarcticOcean, and their impact on global paleoceanography. J Geophys Res 82:3843–3860.

5. Herman F, et al. (2013) Worldwide acceleration of mountain erosion under a coolingclimate. Nature 504:423–426.

6. Berner RA (2004) The Phanerozoic Carbon Cycle: CO2 and O2 (Oxford Univ Press,Oxford).

7. Kump LR, et al. (1999) A weathering hypothesis for glaciation at high atmospheric pCO2

during the Late Ordovician. Palaeogeogr Palaeoclimatol Palaeoecol 152:173–187.8. Koppes MN, Montgomery DR (2009) The relative efficacy of fluvial and glacial erosion

over modern to orogenic timescales. Nat Geosci 2:644–647.9. Hallet B, Hunter L, Bogen J (1996) Rates of erosion and sediment evacuation by glaciers:

A review of field data and their implications. Glob Planet Change 12:213–235.10. Riebe CS, Kirchner JW, Finkel RC (2004) Erosional and climatic effects on long-term

chemical weathering rates in granitic landscapes spanning diverse climate regimes.Earth Planet Sci Lett 224:547–562.

11. West AJ, Galy A, Bickle M (2005) Tectonic and climatic controls on silicate weathering.Earth Planet Sci Lett 235:211–228.

12. Foster GL, Vance D (2006) Negligible glacial-interglacial variation in continentalchemical weathering rates. Nature 444:918–921.

13. Tranter M (2003) Geochemical weathering in glacial and proglacial environments.Treatise of Geochemistry, Surface and Ground Water, Weathering, and Soils, edsHolland HD, Turekian KK (Elsevier, Oxford), pp 189–205.

14. Raiswell R (1984) Chemical models of solute acquisition in glacial melt waters.J Glaciol 30:49–57.

15. Tranter M, Brown G, Raiswell R, Sharp M, Gurnell A (1993) A conceptual model ofsolute acquisition by Alpine glacial meltwaters. J Glaciol 39:573–581.

16. Torres MA, West AJ, Li G (2014) Sulphide oxidation and carbonate dissolution as asource of CO2 over geological timescales. Nature 507:346–349.

17. Torres MA, et al. (2016) The acid and alkalinity budgets of weathering in the Andes–Amazon system: Insights into the erosional control of global biogeochemical cycles.Earth Planet Sci Lett 450:381–391.

18. Berner RA, Canfield DE (1989) A newmodel for atmospheric oxygen over Phanerozoictime. Am J Sci 289:333–361.

19. Bachan A, Kump LR (2015) The rise of oxygen and siderite oxidation during the Lo-magundi Event. Proc Natl Acad Sci USA 112:6562–6567.

20. Williamson MA, Rimstidt JD (1994) The kinetics and electrochemical rate-determiningstep of aqueous pyrite oxidation. Geochim Cosmochim Acta 58:5443–5454.

21. White A, Brantley SL (2003) The effect of time on the weathering of silicate minerals:Why do weathering rates differ in the laboratory and field? Chem Geol 202:479–506.

22. White AF, Bullen TD, Vivit DV, Schulz MS, Clow DW (1999) The role of disseminated calcitein the chemical weathering of granitoid rocks. Geochim Cosmochim Acta 63:1939–1953.

23. Hercod DJ, Brady PV, Gregory RT (1998) Catchment-scale coupling between pyriteoxidation and calcite weathering. Chem Geol 151:259–276.

24. Calmels D, Gaillardet J, Brenot A, France-Lanord C (2007) Sustained sulfide oxidationby physical erosion processes in the Mackenzie River basin: Climatic perspectives.Geology 35:1003–1006.

25. West AJ (2012) Thickness of the chemical weathering zone and implications for erosionaland climatic drivers of weathering and for carbon-cycle feedbacks. Geology 40:811–814.

26. Anderson SP, Drever JI, Humphrey NF (1997) Chemical weathering in glacial envi-ronments. Geology 25:399–402.

27. Anderson SP (2005) Glaciers show direct linkage between erosion rate and chemicalweathering fluxes. Geomorphology 67:147–157.

28. Gaillardet J, Dupré B, Louvat P, Allègre C (1999) Global silicate weathering and CO2

consumption rates deduced from the chemistry of large rivers. Chem Geol 159:3–30.29. Hartmann J, Lauerwald R, Moosdorf N (2014) A brief overview of the GLObal RIver

Chemistry database, GLORICH. Procedia Earth Planet Sci 10:23–27.30. Fekete BM, Vorosmarty CJ, Grabs W (2002) High-resolution fields of global runoff

combining observed river discharge and simulated water balances. Global BiogeochemCycles 16:15-1–15-10.

31. Moon S, Chamberlain CP, Hilley GE (2014) New estimates of silicate weathering ratesand their uncertainties in global rivers. Geochim Cosmochim Acta 134:257–274.

32. Bluth GJS, Kump LR (1994) Lithologic and climatologic controls of river chemistry.Geochim Cosmochim Acta 58:2341–2359.

33. Maher K (2010) The dependence of chemical weathering rates on fluid residencetime. Earth Planet Sci Lett 294:101–110.

34. Anderson SP, Drever JI, Frost CD, Holden P (2000) Chemical weathering in the fore-land of a retreating glacier. Geochim Cosmochim Acta 64:1173–1189.

35. Hodgkins R, Tranter M, Dowdeswell JA (1997) Solute provenance, transport and de-nudation in a High Arctic glacierized catchment. Hydrol Process 11:1813–1832.

36. Tranter M, et al. (2002) Direct effect of ice sheets on terrestrial bicarbonate, sulphateand base cation fluxes during the last glacial cycle: Minimal impact on atmosphericCO2 concentrations. Chem Geol 190:33–44.

37. Hindshaw RS, et al. (2011) Hydrological control of stream water chemistry in a glacialcatchment (Damma Glacier, Switzerland). Chem Geol 285:215–230.

38. Godsey SE, Kirchner JW, Clow DW (2009) Concentration–discharge relationships re-flect chemostatic characteristics of US catchments. Hydrol Process 23:1844–1864.

39. Anderson SP (2007) Biogeochemistry of glacial landscape systems. Annu Rev EarthPlanet Sci 35:375–399.

40. Négrel P, Allègre CJ, Dupré B, Lewin E (1993) Erosion sources determined by inversionof major and trace element ratios and strontium isotopic ratios in river water: TheCongo Basin case. Earth Planet Sci Lett 120:59–76.

41. Urey HC (1952) On the early chemical history of the Earth and the origin of life. ProcNatl Acad Sci USA 38:351–363.

42. Georg RB, West AJ, Vance D, Newman K, Halliday AN (2013) Is the marine osmium isotoperecord a probe for CO2 release from sedimentary rocks? Earth Planet Sci Lett 367:28–38.

43. Boyd ES, Hamilton TL, Havig JR, Skidmore ML, Shock EL (2014) Chemolithotrophicprimary production in a subglacial ecosystem. Appl Environ Microbiol 80:6146–6153.

44. Berner EK, Berner RA (1996) Global Environment: Water, Air, and Geochemical Cycles(Prentice-Hall, Upper Saddle River, NJ).

45. Nielsen H, et al. (1991) Lithospheric sources of sulfur. SCOPE 43: Stable Isotopes:Natural and Anthropogenic Sulphur in the Environment, SCOPE (Wiley, New York).

46. Stolper DA, Bender ML, Dreyfus GB, Yan Y, Higgins JA (2016) A Pleistocene ice corerecord of atmospheric O2 concentrations. Science 353:1427–1430.

47. Jones IW, Munhoven G, Tranter M, Huybrechts P, Sharp MJ (2002) Modelled glacialand non-glacial HCO3

−, Si and Ge fluxes since the LGM: Little potential for impact onatmospheric CO2 concentrations and a potential proxy of continental chemical ero-sion, the marine Ge/Si ratio. Global Planet Change 33:139–153.

48. von Blanckenburg F, Bouchez J, Ibarra DE, Maher K (2015) Stable runoff and weatheringfluxes into the oceans over Quaternary climate cycles. Nat Geosci 8:538–542.

49. Cartapanis O, Bianchi D, Jaccard SL, Galbraith ED (2016) Global pulses of organiccarbon burial in deep-sea sediments during glacial maxima. Nat Commun 7:10796.

50. Horan K, Hilton RG, Burton KW, Selby DS, Ottley CJ (2015) Enhanced oxidativeweathering in glaciated mountain catchments: A stabilising feedback on atmosphericcarbon dioxide? Proceedings of the American Geophysical Union, Fall Meeting 2015(American Geophysical Union, Washington, DC), Abstract EP21E-02.

51. Pasquier V, et al. (2017) Pyrite sulfur isotopes reveal glacial-interglacial environmentalchanges. Proc Natl Acad Sci USA 114:5941–5945.

52. Petit JR, et al. (1999) Climate and atmospheric history of the past 420,000 years fromthe Vostok ice core, Antarctica. Nature 399:429–436.

53. Higgins JA, et al. (2015) Atmospheric composition 1 million years ago from blue ice inthe Allan Hills, Antarctica. Proc Natl Acad Sci USA 112:6887–6891.

54. Zeebe RE, Caldeira K (2008) Close mass balance of long-term carbon fluxes from ice-core CO2 and ocean chemistry records. Nat Geosci 1:312–315.

55. Raymo ME, Ruddiman WF, Froelich PN (1988) Influence of late Cenozoic mountainbuilding on ocean geochemical cycles. Geology 16:649–653.

56. Kump LR, Arthur MA (1997) Global chemical erosion during the Cenozoic: Weath-erability balances the budgets. Tectonic Uplift and Climate Change, ed Ruddiman WF(Springer, Boston), pp 399–426.

57. Caves JK, Jost AB, Lau KV, Maher K (2016) Cenozoic carbon cycle imbalances and avariable weathering feedback. Earth Planet Sci Lett 450:152–163.

58. Zachos J, Pagani M, Sloan L, Thomas E, Billups K (2001) Trends, rhythms, and aber-rations in global climate 65 Ma to present. Science 292:686–693.

59. Zachos J, Kump L (2005) Carbon cycle feedbacks and the initiation of Antarctic gla-ciation in the earliest Oligocene. Global Planet Change 47:51–66.

60. Scher HD, Bohaty SM, Zachos JC, Delaney ML (2011) Two-stepping into the icehouse:East Antarctic weathering during progressive ice-sheet expansion at the Eocene-Oligocene transition. Geology 39:383–386.

61. Basak C, Martin EE (2013) Antarctic weathering and carbonate compensation at theEocene-Oligocene transition. Nat Geosci 6:121–124.

62. Ravizza G, Peucker-Ehrenbrink B (2003) The marine 187Os/188Os record of the Eocene–Oligo-cene transition: The interplay of weathering and glaciation. Earth Planet Sci Lett 210:151–165.

63. Nezat CA, Lyons WB, Welch KA (2001) Chemical weathering in streams of a polardesert (Taylor Valley, Antarctica). Geol Soc Am Bull 113:1401–1408.

64. Willenbring JK, Jerolmack DJ (2016) The null hypothesis: Globally steady rates oferosion, weathering fluxes and shelf sediment accumulation during Late Cenozoicmountain uplift and glaciation. Terra Nova 28:11–18.

65. Lyons TW, Reinhard CT, Planavsky NJ (2014) The rise of oxygen in Earth’s early oceanand atmosphere. Nature 506:307–315.

66. Holland HD (2002) Volcanic gases, black smokers, and the great oxidation event.Geochim Cosmochim Acta 66:3811–3826.

67. van Heuven S, Pierrot D, Rae JWB, Lewis E, Wallace DWR (2011) MATLAB Program De-veloped for CO2 System Calculations, ORNLCDIAC-105b (Carbon Dioxide Information Anal-ysis Center, Oak Ridge National Laboratory, US Department of Energy, Oak Ridge, TN).

Torres et al. PNAS | August 15, 2017 | vol. 114 | no. 33 | 8721

EART

H,A

TMOSP

HER

IC,

ANDPL

ANET

ARY

SCIENCE

S