LETTERSPUBLISHED ONLINE: 29 AUGUST 2016 | DOI: 10.1038/NGEO2788

Parental arc magma compositions dominantlycontrolled by mantle-wedge thermal structureStephen J. Turner1*, Charles H. Langmuir2, Richard F. Katz1, Michael A. Dungan3 and Stéphane Escrig4

The processes that lead to the fourfold variation in arc-averaged compositions ofmafic arc lavas remain controversial.Control by themantle-wedge thermal structure is supported bychemical correlations with the thickness of the underlying arccrust1–3, which a�ects the thermal state of the wedge. Controlby down-going slab temperature is supported by correlationswith the slab thermal parameter3–7. The Chilean SouthernVolcanic Zoneprovides a test of these hypotheses.Hereweusechemical data to demonstrate that the Southern Volcanic Zoneand global arc averages define the same chemical trends, bothamong elements and between elements and crustal thickness.But in contrast to the global arc system, the Southern VolcanicZone is built on crust of variable thickness with a constant slabthermal parameter. This natural experiment, along with a setof numerical simulations, shows that global arc compositionalvariability is dominated by di�erent extents of melting thatare controlled by the thermal structure of the mantle wedge.Slab temperatures play a subordinate role. Variations in thesubducting slab’s fluid flux and sediment compositions, aswell as mantle-wedge heterogeneities, produce second-ordere�ects that are manifested as distinctive trace element andisotopic signatures; these can be more clearly elucidated oncethe importance of wedge thermal structure is recognized.

The chemical compositions of arc-front stratovolcanoes,averaged for individual arcs and normalized for di�erentiation,display coherent systematics1,2. By comparing these globalsystematics with the regional chemical variability of theChilean Southern Volcanic Zone (SVZ; Fig. 1), new constraintscan be developed that are applicable on both the global andregional scales. Globally, incompatible elements correlate wellwith one another over fourfold concentration ranges (Fig. 2b–f).More compatible elements, such as Ca (Fig. 2a) and Sc, correlatenegatively with incompatible elements2,8, whereas heavy rare-earthelements vary little. Correlations among incompatible elementstranscend standard geochemical groupings associated with slabprocesses. Elements believed to be immobile (for example, high-field-strength Nb, Zr), or concentrated in subducting sediments9,10(for example, Th, La), or ‘fluid-mobile’ (for example, K, Pb, U, Sr),are all mutually correlated. Magma compositions also correlate withthe sub-arc Moho depth (Fig. 2g–i), and a proxy for the thermalstructure of the down-going slab3–6, the slab ‘thermal parameter’[� = slab age⇥ convergence rate⇥ sin(dip angle)]11, but not withthe depth of the slab beneath the arc or calculated sub-arc slab-surface temperatures12. Successful models of arc volcanism mustaccount for these first-order relationships.

These relationships are not dependent on the index used torepresent arc compositions. Here we use values normalized to

6% MgO (‘6-values’), but the chemical relationships persist toMg#s (Mg#=atomic Mg/(Mg+Fe)) in equilibrium with the mantle(Supplementary Fig. 1a), and are independent of filtering choices2.The Ce/H2O ratio in olivine-hosted melt inclusions4–6 has also beenused to represent compositional variability among volcanic arcs.Averaged melt inclusion H2O concentrations, however, are similarin all arcs13. Hence, arc-averaged Ce/H2O is largely a function of Ceabundance, and Ce/H2O and Ce6.0 correlate well3 (SupplementaryFig. 1b). The recently proposed ‘Continental Index’7 is alsowell represented by Ce6.0 (Supplementary Fig. 1c), or any otherincompatible element 6-value.

Because the global systematics persist across a wide range offractionation and are present even at highMg#’s, they cannot be pro-duced by intra-crustal processes2. Although assimilation, magmamixing, and high-pressure crystal fractionation impact di�erenti-ated magma compositions in all arcs, particularly continental arcs,these e�ects are minimized by filtering, normalizing and averagingmafic magma chemistry. Models accounting for the global first-order compositional variability of primitive arc volcanics must lookdeeper, to the mantle wedge or the slab.

Plank and Langmuir1 attributed correlations between majorelement 6-values and Moho depth to variations in the extentof mantle-wedge melting (F), hypothesizing that F varies withdistance between the Moho and the sub-arc slab depth (the ‘columnheight’). More precise determinations of slab depths14,15 reveal that6-values do not correlate well with the column height3, despitegood correlations with Moho values. More recently, global-scalecorrelations between � and arc-averaged melt inclusion Ce/H2Omotivated the proposal that this compositional diversity is instead afunction of slab temperature4–6. Slab melting also has been invokedto account for variability of the Continental Index7.

Building on this work, Turner and Langmuir3 demonstrated thatthe global arc chemical systematics are consistent with either oftwo general endmember models. In the first model, the flux ofslab melts to the mantle wedge is constant, while wedge thermalstructure varies as a function of the overriding plate thickness.If sub-arc lithospheric thickness increases with crustal thickness(an assumption that is justified if the entire sub-arc lithosphere isproportionally thickened or thinned by the same tectonic forces),then the mantle wedge beneath a thick-crusted arc resides athigher pressure and lower temperature, leading to lower F anddi�erent residual mineralogy, and higher incompatible elementconcentrations in melts. In the second, the wedge thermal structureis constant, while the slab flux varies as a function of the slabthermal structure. Hotter slabs provide a larger flux to the mantlewedge16, leading to higher concentrations of slab-derived elementsin mantle-wedge melts. Notably, both models require melting of the

1Department of Earth Sciences, University of Oxford, Oxford, Oxfordshire OX1 3AN, UK. 2Department of Earth and Planetary Science, Harvard University,Cambridge, Massachusetts 02138, USA. 3Department of Geological Sciences, University of Oregon, Eugene, Oregon 97403, USA. 4École PolytechniqueFédérale de Lausanne, CH-1015 Lausanne, Switzerland. *e-mail: stephent@earth.ox.ac.uk

NATURE GEOSCIENCE | ADVANCE ONLINE PUBLICATION | www.nature.com/naturegeoscience 1

© Macmillan Publishers Limited . All rights reserved

LETTERS NATURE GEOSCIENCE DOI: 10.1038/NGEO2788

/100

34 35 36 37 38 39 40 41Latitude (° S)

34 35 36 37 38 39 40 41Latitude (° S)

15

20

25

30

35

40

45

50

Dep

th to

Moh

o (k

m)

42

40

38

36

34

a b

c

100 km

200 km

Don CasimiroPlanchón−PeteroaDescabezado GrandeTatara−San PedroLongavíAntucoCallaquiTolhuacaLlaimaVillarricaMocho−ChoshuencoPuyehueAntillancaOsorno

Global range

Global range

74 72 70 68Longitude (° W)

Latit

ude

(° S

)

25

50

75

100

Φ

Figure 1 | Variation of Moho depth and the slab ‘thermal parameter’ (�) along the SVZ between 33� and 43� S. a, Coloured symbols denote arc-frontvolcanoes, black symbols denote other volcanoes. b, Moho depth18 beneath each volcano. The bracket indicates the global range of arc Moho depths(excluding the Central Andean Arc). SVZ Moho depths span the upper half of the global range. c, Slab thermal parameter15 (�) calculated o�shore eachvolcano, with the bracketed range representing the global range of �-values. In contrast to Moho depth, the range of �-values along the study area doesnot vary significantly.

subducting ocean crust in all volcanic arcs to su�ciently mobilizeSr and the light to middle rare-earth elements2,16.

A third possible model is that the slab thermal structurecontrols the H2O flux to the wedge, and that variable wedgeH2O concentrations regulate F . This model is not considered herebecause current estimates of H2O fluxes beneath global arcs17 haveno apparent relationship with magma compositions3.

These hypotheses lack quantitative sophistication because manyvariables in the subduction system are poorly constrained, andphysical models of magma/mantle flow in subduction zones are atan early stage. This leads to a dearth ofmodel predictions that wouldfacilitate critical tests of the models at the global scale.

The Chilean SVZ (Fig. 1), however, provides a naturalexperiment to distinguish the two models because the arc crustalthickness varies from 30 to 50 km (ref. 18) while � varies negligibly(Fig. 1b,c). If � is the primary control on volcanic compositions,there should be little chemical variability along the SVZ. If crustalthickness and the resulting wedge thermal structure are important,then the volcanoes should have chemical variations that areconsistent with the global systematics and occupy a significant partof the global range.

Combining published SVZ data19–22 with new data collectedas part of this study, permits this test. Volcano averages for theSVZ were calculated following previously established methods1,2,normalized to 6wt% MgO. Though the northern SVZ volcanoesdi�erentiate along more calc-alkaline paths than the southern SVZvolcanoes, this divergence occurs below ⇠6wt% MgO, and thusdoes not a�ect the normalized values (see Methods).

SVZ volcano averages closely correspond with the globalarc averages (Fig. 2a–f). Incompatible trace element 6-valuescorrelate with major elements, and with each other, transcendingcustomary element groupings. Furthermore, averaged SVZvolcano compositions overlap with global arc segments for crustal

thicknesses of 30–50 km (Fig. 2g–i). The SVZ thus represents globalvariations on a regional scale.

These results are inconsistent with slab thermal structure beingthe dominant control on magma composition, because the largechemical variations in the SVZ occur despite constant � inthis region (additional discussion in Supplementary Methods).Instead, the data are consistent with the mantle-wedge thermalstructuremodel, because both the internal chemical systematics andthe relationships with crustal thickness mimic the global trends.Leading-order, global chemical variations are therefore most likelythe consequence of a relatively constant slab flux operating onmantle wedges with varying physical conditions related to upperplate thickness.

The proposed relationship between crustal thickness and thepressure and temperature of melting3 can be more rigorously testedby using physical models to explore the systematics of wedgethermal structure. The models must take into account that, fromsouth to north, the sub-arc slab depth increases from 90 to 120 kmand the slab dip angle decreases from 37� to 27�. The depth of thelithosphere–asthenosphere Boundary (LAB) is poorly constrainedalong the arc, but behind the arc the LAB is estimated to increasenorthwards from approximately 70 to 100 km (ref. 18). If LAB andslab depth both increase by 30 km, then even as the upper platethickness varies, the wedge column height does not, which rulesout models that call on di�erent column heights. To quantitativelyexplore the e�ects of these parameters, we present numericalmodels23,24 that have been run for the purpose of constraining thethermal budget of the SVZ mantle wedge and testing the wedgethermal structure hypothesis.

The results of two model runs are presented (Fig. 3a–d) usinginput parameters representative of northern (⇠34� S) and southern(⇠41� S) cross-sections of our study area. F is calculated assuminga single mantle H2O concentration (0.6 wt%), providing a direct

2

© Macmillan Publishers Limited . All rights reserved

NATURE GEOSCIENCE | ADVANCE ONLINE PUBLICATION | www.nature.com/naturegeoscience

NATURE GEOSCIENCE DOI: 10.1038/NGEO2788 LETTERS

1.51.61.71.81.9

2.0

Dy/

Yb

1

2

3

4

La/S

m

50

100

Zr6.

0

7

9

a

b

c

d

e

f

g

h

i

111

13Ca

6.0

200

400

600

Sr6.

0

5

10

15

Nd 6.

0

Na6.0 Nd6.0

1.52.02.53.03.54.0

Na 6.

0

200

400

600

Sr6.

0

Depth to Moho (km)

5

10

15

Nd 6.

0

Chemical model results (ambient mantle = DMM)Chemical model results (ambient mantle = DMM + 2% EMORB)

1.5 2.0 2.5 3.0 3.5 4.0 5 10 15 20 30 40 50

Figure 2 | Chemical systematics of SVZ volcano and global arc averages. a–i, Coloured symbols as in Fig. 1. Open circles are arc-averaged 6-valuecompositions2. All incompatible element 6-values and various element ratios correlate with one another and with Moho depth. SVZ volcano averages spanthe upper half of the global range of arc-averaged incompatible element concentrations. The chemical variability of the SVZ is consistent with the results ofthe quantitative chemical modelling from previous work3 (red and black dotted lines on b–i). The 6-values for Don Casimiro and Planchón–Peteroa aremissing from d because Dy data is not available.

Distance from trench (km)

0 400 800 1,200

T (°C)

F (0.6% H2O)

T (°C)

Path

c

T, path aF, path a

F, path bT, path b

T, path cF, path c Max

F b

enea

th A

rc (0

.6%

H2O

)

30 50 70 90 110Lithosphere thickness (km)

Slab dip = 37°; age = 21 MyrSlab dip = 27°; age = 34 Myr

Nor

ther

n SV

Z m

odel

dept

h (k

m)

Sout

hern

SV

Z m

odel

de

pth

(km

)

Southern SVZ

Northern SVZ

0a b

e

c d

50

100

150

0

50

100

150

0 0.06 0.12 0 50 100 150 200 250 300 350 400

0

500

1,500

1,000

0

0.05

0.10

0.15

0.20

0.25

Path

b

Path

a

Figure 3 | Results of 2D numerical simulation. a–d, Temperature (black lines) and extent of melting (F) (blue lines) (a,c), along paths through the modelsdepicted in b and d, respectively. Although path a corresponds to the actual southern SVZ slab depth (90 km), while path b intersects the slab at 120 km,both reach the same maximum F. The maximum F for the southern model (F= 11.3%) is nearly twice that of the northern model (F=6.6%). e, Modelresults using slab dip and age values observed in the SVZ indicate that these parameters have a minor influence on F compared to lithospheric thickness.See Methods for model description.

comparison of the thermal energy available for melting. The‘southern SVZ’ model predicts F that is almost double that ofthe ‘northern SVZ’ model, corresponding well with the regionalincompatible element variations. The di�erence in lithosphericthickness is the primary control on F , with minor additionalvariability from dip angle and slab age (Fig. 3e). The rangeof F values expected globally (5–20%) from the wedge thermalstructure chemical model corresponds to lithosphere between 100

and 40 km thick. The range of F values calculated for the SVZchemical model (6–12%) is then consistent with both the regionalcrustal thickness values and the global trends.

The numerical simulations also demonstrate why there areno global correlations with slab depth, and why mantle columnheight does not relate directly to F . Two di�erent depth versusF profiles from the same model run are shown on Fig. 3a. Thesolid lines corresponds to a vertical cross-section intersecting the

NATURE GEOSCIENCE | ADVANCE ONLINE PUBLICATION | www.nature.com/naturegeoscience

© Macmillan Publishers Limited . All rights reserved

3

LETTERS NATURE GEOSCIENCE DOI: 10.1038/NGEO2788

slab at 90 km depth, while the dashed lines intersect the slab at120 km. Because of the competing changes in temperature andpressure, the maximum F value reached in both cross-sections isthe same, despite significantly di�erent column heights. The criticalparameter controlling F is thus the LAB depth.

The results from the SVZ suggest the overriding plate exerts aprimary control on global and SVZ chemical variations throughits influence on the wedge thermal structure. If variations inslab thermal structure do not produce the global-scale chemicalvariability of volcanic arcs, then why do arc compositions correlateglobally3–6 with �?

It is likely that additional factors indirectly relate� to themantle-wedge conditions. Across-arc compression in the overriding plateincreases as the down-going slab age and dip angle decrease25,26.Because slab age and dip angle contribute most of the varianceto � , arcs with lower �-values are most commonly compressional.A compressional strain regime results in lithospheric (includingcrustal) thickening, which may explain why greater Moho depthscorrespond to low �-values. The correlations between � andmagma chemistry may be further strengthened by the fact thatconvergence rate, the third parameter in � , directly a�ects thetemperature of the mantle wedge24. A more rapid convergence rateleads to higher wedge temperatures and higher F , and also tohigher � , thus providing an additional link between � and magmachemistry that is not causal.

There is evidence that slab temperatures vary beneath arcfronts26, so why don’t slab temperatures typically control arc-frontmagma chemistry? A surprising aspect of the successful models isthat they use a constant average slab flux, with a slab that must meltat every subduction zone. This suggests that, with the exceptionof a few unusual cases27, the total slab flux to the mantle sourceof arc-front volcanism is relatively insensitive to slab temperature.This outcome is likely if slab melts are transferred to the wedgeacross a large range of depths and temperatures, and then eventuallymigrate to the source region for arc-front volcanism28. To morefully constrain such mantle-wedge processes, thermal models thatincorporatemelting andmelt transport of heat need to be developed.

We have emphasized the aspects of arc petrogenesis accountedfor by the thermal structure of the SVZmantle wedge. These resultsdo not preclude a secondary role for the slab in determining localvariations in chemical compositions. Variations among volcanoesin arcs that are constructed on plates of nearly constant thickness,such as the Aleutians27,29 and Marianas9, seem to require a diversityof slab components. There is also an important role for heterogeneityin the ambient mantle wedge, which appears to contribute most ofthe isotopic variability to the SVZ volcanic rocks21,22. The diversityof sediment compositions worldwide also has an important influ-ence30. The results of the present work suggest that disentanglementof these factors by comprehensive investigations of individual arcsmay be aided by a recognition of a primary role of variable extentsof melting produced by wedge thermal structure.

MethodsMethods, including statements of data availability and anyassociated accession codes and references, are available in theonline version of this paper.

Received 4 February 2016; accepted 19 July 2016;published online 29 August 2016

References1. Plank, T. & Langmuir, C. H. An evaluation of the global variations in the major

element chemistry of arc basalts. Earth Planet. Sci. Lett. 90, 349–370 (1988).2. Turner, S. J. & Langmuir, C. H. The global chemical systematics of arc front

stratovolcanoes: evaluating the role of crustal processes. Earth Planet. Sci. Lett.422, 182–193 (2015).

3. Turner, S. J. & Langmuir, C. H. What processes control the chemicalcompositions of arc front stratovolcanoes? Geochem. Geophys. Geosyst. 16,1865–1893 (2015).

4. Cooper, L. B. et al . Global variations in H2O/Ce: 1. Slab surface temperaturesbeneath volcanic arcs. Geochem. Geophys. Geosyst. 13, Q03024 (2012).

5. Plank, T., Cooper, L. B. & Manning, C. E. Emerging geothermometers forestimating slab surface temperatures. Nat. Geosci. 2, 611–615 (2009).

6. Ruscitto, D. M., Wallace, P. J., Cooper, L. B. & Plank, T. Global variations inH2O/Ce: 2. Relationships to arc magma geochemistry and volatile fluxes.Geochem. Geophys. Geosyst. 13, Q03025 (2012).

7. Gazel, E. et al . Continental crust generated in oceanic arcs. Nat. Geosci. 8,321–327 (2015).

8. Plank, T. & Langmuir, C. H. Tracing trace elements from sediment input tovolcanic output at subduction zones. Nature 362, 739–743 (1993).

9. Elliott, T., Plank, T., Zindler, A., White, W. & Bourdon, B. Element transportfrom slab to volcanic front at the Mariana arc. J. Geophys. Res. 102,14991–15019 (1997).

10. McCulloch, M. T. & Gamble, J. Geochemical and geodynamical constraints onsubduction zone magmatism. Earth Planet. Sci. Lett. 102, 358–374 (1991).

11. Kirby, S. H., Durham, W. B. & Stern, L. A. Mantle phase changes anddeep-earthquake faulting in subducting lithosphere. Science 252,216–225 (1991).

12. Syracuse, E. M., van Keken, P. E. & Abers, G. A. The global range of subductionzone thermal models. Phys. Earth Planet. Inter. 183, 73–90 (2010).

13. Plank, T., Kelley, K. A., Zimmer, M. M., Hauri, E. H. &Wallace, P. J. Why domafic arc magmas contain ⇠4wt% water on average? Earth Planet. Sci. Lett.364, 168–179 (2013).

14. England, P., Engdahl, R. & Thatcher, W. Systematic variation in the depths ofslabs beneath arc volcanoes. Geophys. J. Int. 156, 377–408 (2004).

15. Syracuse, E. M. & Abers, G. A. Global compilation of variations in slab depthbeneath arc volcanoes and implications. Geochem. Geophys. Geosyst. 7,Q05017 (2006).

16. Hermann, J. & Rubatto, D. Accessory phase control on the trace elementsignature of sediment melts in subduction zones. Chem. Geol. 265,512–526 (2009).

17. van Keken, P. E., Hacker, B. R., Syracuse, E. M. & Abers, G. A. Subductionfactory: 4. Depth-dependent flux of H2O from subducting slabs worldwide.J. Geophys. Res. 116, B01401 (2011).

18. Tassara, A. & Echaurren, A. Anatomy of the Andean subduction zone:three-dimensional density model upgraded and compared against global-scalemodels. Geophys. J. Int. 189, 161–168 (2012).

19. Hildreth, W. & Moorbath, S. Crustal contributions to arc magmatism in theAndes of central Chile. Contrib. Mineral. Petrol. 98, 455–489 (1988).

20. Tormey, D. R., Hickey-Vargas, R., Frey, F. A. & López-Escobar, L. Recent lavasfrom the Andean volcanic front (33 to 42� S); interpretations of along-arccompositional variations. Geol. Soc. Am. Spec. Pap. 265, 57–78 (1991).

21. Hickey, R. L., Frey, F. A., Gerlach, D. C. & Lopez-Escobar, L. Multiple sourcesfor basaltic arc rocks from the southern volcanic zone of the Andes (34–41� S):trace element and isotopic evidence for contributions from subducted oceaniccrust, mantle, and continental crust. J. Geophys. Res. 91, 5963–5983 (1986).

22. Jacques, G. et al . Geochemical variations in the Central Southern VolcanicZone, Chile (38–43� S): the role of fluids in generating arc magmas. Chem.Geol. 371, 27–45 (2014).

23. van Keken, P. E. et al . A community benchmark for subduction zone modeling.Phys. Earth Planet. Inter. 171, 187–197 (2008).

24. England, P. C. & Katz, R. F. Melting above the anhydrous solidus controls thelocation of volcanic arcs. Nature 467, 700–703 (2010).

25. Lallemand, S., Heuret, A. & Boutelier, D. On the relationships between slab dip,back-arc stress, upper plate absolute motion, and crustal nature in subductionzones. Geochem. Geophys. Geosyst. 6, Q09005 (2005).

26. Jarrard, R. D. Relations among subduction parameters. Rev. Geophys. 24,217–284 (1986).

27. Yogodzinski, G. M. et al . The role of subducted basalt in the source of island arcmagmas: evidence from seafloor lavas of the western Aleutians. J. Petrol. 56,441–492 (2015).

28. Wilson, C. R., Spiegelman, M., van Keken, P. E. & Hacker, B. R. Fluid flow insubduction zones: the role of solid rheology and compaction pressure. EarthPlanet. Sci. Lett. 401, 261–274 (2014).

29. Class, C., Miller, D. M., Goldstein, S. L. & Langmuir, C. H. Distinguishing meltand fluid subduction components in Umnak Volcanics, Aleutian Arc. Geochem.Geophys. Geosyst. 1, 1004 (2000).

30. Plank, T. & Langmuir, C. The chemical composition of subducting sedimentand its consequences for the crust and mantle. Chem. Geol. 145,325–394 (1998).

AcknowledgementsWe wish to acknowledge the invaluable assistance and guidance of D. Sellés and ServicioNacional de Geología y Minería of Chile for assisting with our field campaigns.

4

© Macmillan Publishers Limited . All rights reserved

NATURE GEOSCIENCE | ADVANCE ONLINE PUBLICATION | www.nature.com/naturegeoscience

NATURE GEOSCIENCE DOI: 10.1038/NGEO2788 LETTERSR. Hickey-Vargas provided us with many of the samples used in this study as well.Z. Chen carried out some of the ICPMS analyses and provided analytical assistance forthe rest. L. Cooper provided discussions and scientific input during our joint samplingexpeditions and in the years following, and also took on a large amount of the samplepreparation. A helpful review by G. Yogodzinski significantly improved this manuscript.This work was supported by NSF grant EAR-0948511, NERC grant NE/M000427/1, andERC grant 279925.

Author contributionsC.H.L. conceived the SVZ as a natural experiment and obtained the funding. S.J.T., S.E.,C.H.L. and M.A.D. collected samples. S.J.T. and S.E. acquired the data. S.J.T., C.H.L. andM.A.D. interpreted the chemical data. S.J.T. carried out the geochemical modelling.

R.F.K. wrote the numerical code and guided S.J.T. in its application. S.J.T. wrote themanuscript with assistance from C.H.L.; M.A.D. and R.F.K. helped develop and revisethe manuscript.

Additional informationSupplementary information is available in the online version of the paper. Reprints andpermissions information is available online at www.nature.com/reprints.Correspondence and requests for materials should be addressed to S.J.T.

Competing financial interestsThe authors declare no competing financial interests.

NATURE GEOSCIENCE | ADVANCE ONLINE PUBLICATION | www.nature.com/naturegeoscience

© Macmillan Publishers Limited . All rights reserved

5

LETTERS NATURE GEOSCIENCE DOI: 10.1038/NGEO2788

MethodsNumerical models.Models use the methods of England & Katz24, and vary basedon convergence rate, slab dip, and the LAB depths behind the arc. For the southernmodel at ⇠42� S, LAB depth is 70 km, slab dip is 37�, and slab age is 21Myr. For thenorthern model at ⇠34� S, LAB depth is 100 km, slab dip is 27�, and slab age is34Myr (ref. 18). For the calculations, the overriding plate was aged until the1,100 �C isotherm on the right boundary was equal to the LAB depth. The stagnantlid and coupling depth were set at 10 km above the LAB.

Major and trace element analysis.Whole-rock major and trace elementconcentrations were measured on a newly collected set of samples from the ChileanSouthern Volcanic Zone (SVZ) stratovolcanoes, and new trace element analyseswere performed on a set of samples previously collected and partially analysed byRosemary Hickey-Vargas and colleagues. Whole-rock samples were crushed to agrain size of ⇠1mm and then powdered using either an agate mill or a shatterboxwith an alumina-ceramic crucible until a flour-like texture was produced. Majorelements were measured by X-ray fluorescence (XRF) at the University of Lausannefollowing Pfeifer and colleagues31. New trace element analyses were performed atHarvard University by solution nebulized inductively coupled plasma massspectrometry (SN-ICP-MS). Detailed analytical methods and uncertainties ofSN-ICP-MS measurements can be found in Turner and colleagues32.

Data filtration. The goal of this study is to better constrain models for primarymagma diversity. To this end, the data have been filtered to minimize secondarye�ects (for example, crystal fractionation, magma mixing, excess crystalaccumulation, and crustal assimilation) which are known to produce much of thechemical variability within individual SVZ volcanoes33–37. These crustal processesgenerally decrease the Mg content of a magma, leading to greater crustaloverprinting at lower Mg abundances. The resultant increasing variance ofelemental abundances with decreasing wt% MgO, creates characteristic ‘wedgeshaped’ arrays on plots of incompatible elements versus MgO (see SupplementaryInformation for an example of this e�ect). These wedge shapes typically come to a‘point’ once MgO exceeds 5wt%, indicating that samples with >5wt% MgO arecomparatively free from the more extreme crustal di�erentiation e�ects.

Some SVZ magmas with >5wt%MgO are still a�ected by secondary processes,and in particular mixing between higher-MgO magmas and evolved, low-MgOmagmas is commonly observed38,39. Evolved magmas are typically highly enrichedin incompatible elements, so mixed magmas do not represent primitivecompositions. Because evolved magmas have typically undergone extensiveplagioclase fractionation, leading to large negative Eu anomalies40, mixed magmasinherit a negative Eu anomaly. To try to eliminate such mixed magmas from theaverages we include only samples with Eu anomalies >0.9.

Magma modification by crustal assimilation has been identified along theentire SVZ33–35. Crustal assimilation typically results in preferential enrichment ofcertain elements—most notably Rb—that are particularly abundant in crustallithologies34,35. Rubidium in each volcano was compared to K concentrations tofilter out samples with excessive Rb enrichments. Because the ratios of Rb to K canalso vary among primary magmas, this filtration can only be applied relative toother samples from the same volcano. See Supplementary Information for the exactK/Rb cuto�s used for each volcano.

Calculation of 6-values.Magmas that have not been a�ected by open-systemprocesses evolve via crystal fractionation. Early-stage crystal fractionation does nota�ect incompatible element ratios, but does lead to variable elemental abundances.Therefore, this study considers variability of elemental abundances only amongmagmas that have 5.5 < wt%MgO < 6.5, which corresponds to a narrow range of

extents of crystal fractionation. Volcano ‘6-values’ are calculated by averaging thesesamples from each volcano, following the methodology of Turner and Langmuir2.For volcanoes with su�cient data, only trace element data on dissolved rocksolutions by ICPMS from Harvard University were included in averages, tominimize inter-laboratory biases. Ultimately, as shown by Turner and Langmuir2,the results of this study are not dependent on the use of 6-values as the chemicalindex of choice, as it can be shown that 6-values are strongly correlated with severalother indices for representing the average compositions of volcanic arcs. A figuredemonstrating this point is available in the Supplementary Information. Additionalfiltration steps were necessary in a few specific instances, which are also outlined indetail in the Supplementary Information.

Code availability. Code is available by e-mail request to the author.

Data availability. The data from this study is available in the article’sSupplementary Information, and will also be made available from the GEOROConline database (http://georoc.mpch-mainz.gwdg.de/georoc).

References31. Pfeifer, H.-R., Lavanchy, J.-C. & Serneels, V. Bulk chemical analysis of

geological and industrial materials by X-ray fluorescence, recent developmentsand application to materials rich in iron oxide. J. Trace Microprobe Tech. 9,127–147 (1991).

32. Turner, S. J., Izbekov, P. & Langmuir, C. The magma plumbing system ofBezymianny Volcano: insights from a 54 year time series of trace elementwhole-rock geochemistry and amphibole compositions. J. Volcanol. Geotherm.Res. 263, 108–121 (2013).

33. Reubi, O. et al . Assimilation of the plutonic roots of the Andean arc controlsvariations in U-series disequilibria at Volcan Llaima, Chile. Earth Planet. Sci.Lett. 303, 37–47 (2011).

34. Dungan, M. A., Wul�, A. & Thompson, R. Eruptive stratigraphy of theTatara–San Pedro complex, 36 S, Southern Volcanic Zone, Chilean Andes:reconstruction method and implications for magma evolution at long-lived arcvolcanic centers. J. Petrol. 42, 555–626 (2001).

35. Jicha, B. R. et al . Rapid magma ascent and generation of 230Th excesses in thelower crust at Puyehue–Cordón Caulle, Southern Volcanic Zone, Chile. EarthPlanet. Sci. Lett. 255, 229–242 (2007).

36. Tagiri, M., Moreno, H., López-Escobar, L. & Notsu, K. Two magma types of thehigh-alumina basalt series of Osorno Volcano, Southern Andes(41� 060 S)-Plagioclase dilution e�ect, Ganko. J. Miner. Petrol. Econ. Geol. 88,359–371 (1993).

37. Tormey, D. R., Frey, F. A. & Lopez-Escobar, L. Geochemistry of the activeAzufre–Planchon–Peteroa Volcanic Complex, Chile (35� 150 S): evidence formultiple sources and processes in a Cordilleran arc magmatic system. J. Petrol.36, 265–298 (1995).

38. Gerlach, D. C., Frey, F. A., Moreno-Roa, H. & Lopez-Escobar, L. Recentvolcanism in the Puyehue—Cordon Caulle Region, Southern Andes, Chile(40 · 5� S): petrogenesis of evolved lavas. J. Petrol. 29, 333–382 (1988).

39. Rodriguez, C., Sellé, D., Dungan, M., Langmuir, C. & Leeman, W. Adakiticdacites formed by intracrustal crystal fractionation of water-rich parentmagmas at Nevado de Longavi Volcano (36.2� S; Andean Southern VolcanicZone, Central Chile). J. Petrol. 48, 2033–2061 (2007).

40. Weill, D. F. & Drake, M. J. Europium anomaly in plagioclase feldspar:experimental results and semiquantitative model. Science 180,1059–1060 (1973).

© Macmillan Publishers Limited . All rights reserved

NATURE GEOSCIENCE | www.nature.com/naturegeoscience