Thermodynamic modelling of continental arc-adjacent magmatism: the Loicas Trough, N. Patagonia, Argentina

Abstract

Continental arcs are associated with volcanism concentrated into two main belts—the main arc and back arc, often separated by fold and thrust belts. The Loicas Trough, Argentina, is a post-orogenic extensional feature that obliquely cuts the fold and thrust belts. The trough hosts large Pliocene–Holocene volcanic centres, including Domuyo and Tromen, that lie between the main arc and back arc and thus provide a rare window into this setting. We present major and trace element data for the Loicas Trough, which we combine with geochemical modelling using the Magma Chamber Simulator (MCS) to explore the origin and evolution of the volcanism. The lavas display a wide continuous range from alkaline basalts to subalkaline rhyolites. Trace elements reveal variable extents of arc enrichment (2 < Nb/U < 28), which correlate with proximity to the trench and differentiation indices. Our results and MCS models indicate that the Loicas Trough parental magmas formed from compositionally zoned mantle. Best-fit models indicate that the differentiation occurs at middle and upper crustal levels, in sharp contrast to lower crustal hot zones beneath main arcs. Assimilation of partial crustal melts drives compositional evolution and obscures source signatures. Pure or high fraction end-member partial crustal melts are also identified at Domuyo based on their low Ba (~ 250 ppm) and moderate Sc contents (~ 8 ppm). We find evidence of similar lavas in transtensional settings adjacent to continental arcs worldwide, which do not adhere to the main versus back arc volcanism binary. We suggest the term arc-adjacent magmatism, where compositions are mainly controlled by extensive assimilation and reworking in the middle to upper crust.

Introduction

Continental arcs represent the Earth’s most compositionally diverse plate tectonic environment, are a key driver of planetary evolution, and have important societal consequences in the form of earthquakes and explosive volcanism. The subducting plate transports surface fluids and surface materials into the mantle, causing flux-induced melting. The plate motion induces mantle convection and locally drives shortening and extension in various parts of the overriding plate. This in turn results in the formation of wide chains of volcanic centres along the plate boundary, with a wide range of magma compositions resulting from this complex tectonic setting (e.g. Davidson et al. 2005; Plank 2014; Kelemen et al. 2014). The fluid-induced formation of continental arc lavas, and the thicker continental plate the lavas must transverse, leads to large long-lived magma plumbing systems and gas-rich felsic volcanoes, an explosive combination from a geohazards perspective (Hora et al. 2009; Oyarzún et al. 2021). The Andean arc is the textbook example of a continental arc, dominated by subduction of the Nazca Plate beneath the South American continental plate (Stern 2004). Quaternary volcanism in the Southern Volcanic Zone (SVZ) of the Andes primarily occurs in the main arc and back arc. The main arc volcanoes form dominantly in response to the main zone of slab dehydration, which induces melting of the sub arc mantle. Main arc lavas are generally subalkaline and inherit the enriched slab fluid trace element characteristics (Stern 2004; Moreno et al. 2007; Jacques et al. 2013; Hickey-Vargas et al. 2016). The back arc volcanoes form in response to subduction induced upwelling and decompression melting during slab-rollback, far from the fluid-enriched section of the sub-arc mantle. These back arc lavas are alkaline and have trace element characteristics akin to intraplate and plume volcanism (e.g. Kay 2001; Søager et al. 2013, 2015).

In this study, we investigate volcanism in the Loicas Trough extensional graben, in North Patagonia. The Loicas Trough intersects the fold and thrust belts of the southern Andes between the main and back arc regions of the Andean Arc (Miranda et al. 2006; Folguera et al. 2007b, 2008, 2012; Llambías et al. 2010). Therefore, Loicas Trough volcanism is not directly geographically linked with either the back arc or main arc volcanism, and represents a rare opportunity to investigate mantle and magmatic processes in this typically amagmatic zone. We present bulk rock major and trace element compositions of 70 lava samples from three volcanic centres in the Loicas Trough, and model their magmatic evolution using the Magma Chamber Simulator (MCS; Bohrson et al. 2014, 2020; Heinonen et al. 2020). Through this, we aim to determine the mantle sources and magmatic evolution in the Loicas Trough volcanic rocks. We demonstrate that the complex processes involved in their formation may be applicable to arc-adjacent volcanic rocks elsewhere as well as ancient arc systems, and represent a specific commonly unrecognized style of continental arc volcanism.

Geological setting

The Loicas Trough is a 300 km long NNW-SSE striking extensional feature situated in the fold and thrust belts of the Andes in the Southern Volcanic Zone (SVZ) from 35 to 38°S (Miranda et al. 2006; Folguera et al. 2007b, 2008, 2012; Llambías et al. 2010). The trough is located between the main arc volcanism of the Andean Cordillera to the west and the back arc volcanism of Payenia to the east (Fig. 1). The northern part of the Loicas Trough intercepts the main arc at ~ 110 km above the subducting Nazca slab, while the southern part is closer to the back arc domain, ~ 170 km above the Nazca slab (Hayes et al. 2012). The lithosphere is approximately 90 km thick and the continental crust is approximately 40 km thick in the region (Tassara and Echaurren 2012). The Loicas Trough is one of several large troughs that host volcanism situated in this region of the SVZ between the main and back arc (e.g. Varekamp et al. 2010; Rojas Vera et al. 2014).

Fig. 1

A Map showing the South American continent and plate boundary (Bird 2003). B Plate tectonic and structural map of the Loicas Trough, shown relative to the main arc, bounding fold and thrust belts, and the back arc Neuquén basin. The coloured fields indicate the studied areas, red = Domuyo, blue = Tromen, green = Palao (see Fig. 2), whereas the gray fields are back arc lavas. Modified after Folguera et al. (2008) with slab contours (blue dashed lines, SLAB 1.0) from Hayes et al. (2012), plate boundaries from Bird (2003) and the arc volcanoes are currently active stratovolcanoes from the Smithsonian Institute’s Global Volcanism Program database (https://volcano.si.edu/)

The formation of the Loicas and neighbouring troughs reflects local extension in response to stresses and movements between the Nazca and the South American plates since the Pliocene. The Nazca plate subducts at an oblique angle that results in an overall transpressional continental margin (Stern 2004). Additionally, the dip of the Nazca slab has been steepening since the Pliocene following a period of shallow-angle subduction (Ramos et al. 2014), which has caused thinning and upwelling underneath the South American plate as reflected in the back arc volcanism of the Payenia, Auca Mahuida, and Nevado volcanic fields (Fig. 1; Gudnason et al. 2012). This particular tectonic setting led to the formation of the troughs (Cobbold and Rossello 2003; Kay et al. 2006; Folguera et al. 2007a, 2015; Messager et al. 2010; Ramos et al. 2014; Rojas Vera et al. 2015).

The Loicas Trough hosts several larger volcanic complexes, including, from south to north: the Tromen Volcanic Complex, Cerro Wayle volcano, Cerro Palao volcano, the Domuyo Volcanic Complex, and the Varvarco Volcanic Field. There are also several domes and monogenetic cones such as Cerro Polco and Cerro Boliviano (Figs. 1 and 2) (Llambías et al. 1978a, 2010; Folguera et al. 2008). Volcanism has been occurring here from the Pliocene to the Holocene and temporally coincides with the latest episode of slab steepening (Kay et al. 2006; Miranda et al. 2006; Galland et al. 2007; Folguera et al. 2008; Pallares et al. 2019; Iannelli et al. 2023). Recent geophysical tomography studies have shown the presence of a hot, fluid or magma-rich zone beneath the trough at approximately 5 to 20 km depth, indicating still-ongoing magmatic activity in the trough (Pesce 2013; Chiodini et al. 2014; Tassi et al. 2016; Galetto et al. 2018; González-Vidal et al. 2018; Astort et al. 2019; Lundgren et al. 2020).

Fig. 2

Sample locations for all rocks and the large volcanic centres including Domuyo, Tromen and Palao-Polco volcanic complexes. Large volcanoes from the Smithsonian Institute’s Global Volcanism Program database (https://volcano.si.edu/), and topographic data from the NASA shuttle radar topography mission program

The Loicas Trough lithologies display a wide compositional variance from alkali basalts to high-K rhyolites (Llambías et al. 1978a, b), even within the same volcanic complex. The majority of samples from the Tromen Volcanic Complex display trace element signatures that are intermediate between a typical main arc and back arc compositions, reflecting progressively decreasing slab input into the mantle sources to the east (Søager et al. 2013; Jacques et al. 2013; Søager and Holm 2013; Holm et al. 2014; Pallares et al. 2019). Kay et al. (2006) concluded that the Tromen Volcanic Complex displays a stronger back arc signature than main arc affinity, similar to the back arc volcano Auca Mahuida located southeast of Tromen just beyond the Loicas Trough (Fig. 1). Although geochemical variations in the Tromen-Wayle area, in the southern part of the Loicas Trough, are well characterized (Kay et al. 2006; Llambías and Leanza 2011; Jacques et al. 2013; Pallares et al. 2019 and references therein), detailed geochemical studies are sparse for the central part of the Loicas Trough (Domuyo and Palao-Polco areas), since most studies in this region have emphasized the structural, geothermal and geochronological evolution (e.g., Zanettini et al. 2001; Miranda et al. 2006; Galland et al. 2007; Folguera et al. 2008; Llambías et al. 2010).

Samples and field observations

The investigated volcanic rocks are from the largest volcanic centres in the Loicas Trough, Domuyo, Tromen, Cerro Wayle, Cerro Palao, Cerro Polco, as well as monogenetic cones and lava flows surrounding these main centres. Samples include primitive to highly evolved lavas flows, hypabyssal shallow intrusions and pumice from ignimbrite deposits (Figs. 1 and 2). The volcanic rocks range from Pliocene (Grupo Domuyo Fm.) at ~ 4 Ma (Folguera et al. 2008 and references therein) to the Holocene (El Puente Fm.). Additionally, we analysed a few samples from the Late Cretaceous Pelan Fm. that represent the late Mesozoic igneous crust, though these samples are not discussed further in this study (Supplementary Material 1).

The younger (Pliocene to Holocene) volcanic samples are generally porphyritic and in the sections below the phenocryst phases are listed in general order of abundance. The groundmasses of Loicas Trough samples range from highly vesicular to massive and from glassy to fine grained. Detailed descriptions of eruptive styles, flow morphologies, and the compositional variations at each volcanic complex are given in the Supplementary Material 5. However, we note here a broad trend from generally more felsic lavas in the northernmost Domuyo Complex, to lavas with more variable compositions and more mafic compositions in the southern Tromen and Cerro Wayle volcanoes.

The mafic lava samples (Tromen-Wayle and Palao-Polco) range from aphyric to plagioclase-, olivine- and clinopyroxene-phyric. The total crystal modal abundance range from aphyric (< < 1%) to ~ 10% phenocrysts for the majority of the mafic samples. The intermediate samples from all complexes are generally feldspar-phyric, where plagioclase is the main and most abundant crystal phase, though clino- and orthopyroxene, amphibole or alkali feldspar are also present in some samples. Additionally, a few intermediate samples display glomerocrysts, and contain mafic enclaves. The intermediate lava samples have the most variable and overall highest modal crystal abundance, ranging from 6 to 38% with 15–20% being the norm. The felsic samples (from all complexes) are generally porphyritic and display the largest variation in crystal phases. The main phenocryst phase is feldspar, although the exact type varies between plagioclase, anorthoclase and sanidine. The dominant mafic phases are amphibole, clino- and orthopyroxene, biotite and oxides. Accessory phases include apatite and zircon. The felsic lavas are groundmass dominated and the total modal crystal abundance is moderate, ranging from 7 to 15%.

Methods

Major and trace elements

Weathered surfaces were removed using a rock saw, and pumice samples were rinsed in an ultrasonic water bath for 20 min and subsequently dried in an oven. The samples were coarsely crushed using a hydraulic steel press and a steel pestle and mortar, before milling to a fine powder in an agate disc mill or a tungsten carbide planetary ball mill.

Major element oxide contents were determined at AcmeLabs (Bureau Veritas Commodities Canada Ltd.) by lithium borate fusion dissolution and ICP-OES analysis (Code: LF300). Relative 95% confidence intervals on repeated measurements of the internal reference material (SO-19) analysed alongside the samples are within ± 5% for the major oxides, except for MnO and Cr₂O₃ (± 7 and 12%, respectively). Trace element concentrations of the evolved rocks likely to contain refractory phases such as zircon were also determined at AcmeLabs by lithium borate fusion dissolution coupled with ICP-MS analysis (Code: LF202). The 95% confidence intervals range from ± 2 to 12% on repeated measurements.

The trace element composition of primitive rocks devoid of refractory phases, two blanks and three rock reference materials (DISKO-1, BHVO-2, BCR-2) were determined by ICP-MS at the Geological Survey of Denmark and Greenland (GEUS) using a PerkinElmer 6100 DRC Quadrupole following the methods of Ottley et al. (2003). Repeated measurements of the reference materials are within ± 4% (2σ). The average concentrations overlap accepted values within 2σ (Jochum et al. 2016). Sample 130073, a basaltic andesite, was analysed at both GEUS and Acmelabs. Concentrations are in good agreement, with 5% difference or less between the replicate analyses for all elements. No correction is applied to the final trace element data.

Results

Major elements

The samples display a wide range of compositions, ranging from alkali basalt to high K rhyolites (Figs. 3 and 4) with SiO₂ and MgO contents ranging from 48 to 76 wt.% and 8 to 0 wt.%, respectively (Table 1 and Supplementary Material 1). The loss on ignition (LOI) is generally below 2 wt.%, with the exception of a few altered samples which are discussed below. The volcanic samples are silica-saturated to silica-undersaturated and follow an high-K calc-alkaline trend based on K₂O and SiO₂ (Fig. 4h). Additionally, the lavas evolve from metaluminous to peraluminous compositions, and at their most siliceous become peralkaline (Frost and Frost 2008). The volcanic rocks at Tromen-Wayle display the largest spread from basaltic to rhyolitic, whereas the Domuyo volcanic rocks are trachyandesitic to rhyolitic and the Palao-Polco rocks are trachybasaltic andesites to rhyolites (Fig. 3).

Fig. 3

TAS diagram (Le Maitre 2002) for the Loicas Trough volcanics. The concentrations are in wt.% and not normalized to anhydrous compositions due to the generally low LOI < 2 wt.%. Grey symbols are a GEOROC compilation (retrieved November 28, 2019, from https://georoc.eu/georoc/new-start.asp) of volcanic whole rock data from the Andes main and back arc for the central and transitional SVZ (Argentina/Chile). The back arc and main arc fields mark the 2D density distribution (lowest density contour, no. of bins = 10) of the main and back arc samples from the GEOROC compilation

Fig. 4

Harker diagrams (A–D, F–H) and major element ratios (E & I) against SiO₂. Fields on K₂O vs. SiO₂ (H) after Le Maitre (2002), and the units are in wt.%

Table 1 Major and trace element results for a selection of the most primitive and most evolved samples in our dataset

Primitive compositions exhibit considerable scatter for some major elements but the compositions generally converge as the rocks become more felsic (Figs. 3 and 4). The alkalis Na₂O and K₂O show scattered positive correlations with silica, and the entire magmatic series has high K₂O (Fig. 4h). The other major elements correlate negatively with silica. Silica and MgO display a hyperbolic relationship, whereas Al₂O₃ contents are initially constant and then decrease above 60 wt.% silica. The CaO/Al₂O₃ ratio decreases non-linearly with decreasing MgO, but decreases linearly with increasing silica. Bulk-rock Mg# are calculated from FeO_T contents, assuming Fe³⁺/Fe_tot = 0.3, as we lack a precise estimate of the oxygen fugacity (Fig. 4e). Iron initially increases with falling MgO, as seen by a rapid drop in Mg# from ~ 70 to 55 at ~ 48 to 51 wt.% SiO₂. This is followed by a steady decrease in Mg# with further differentiation. At high silica compositions (> 72 wt.%) Mg is completely depleted (Fig. 4a).

Trace elements

All lavas have enriched trace element patterns with elevated large ion lithophile element (LILE) contents compared to the primitive mantle (Fig. 5). Elements, such as U, Th and the LILEs correlate positively and strongly with SiO₂ (Fig. 6). The high field-strength elements (HFSE), such as Hf, Zr and Y, and the rare earth elements (REE) show increasing average concentrations with increasing silica content, but the trend is scattered and non-linear (e.g. Fig. 6c and f). On mantle normalized diagrams, the lavas display Nb–Ta troughs of variable magnitude, characteristic of arc-like enrichment (Fig. 6). A significantly diminished Nb–Ta trough is observed in the primitive rocks from Tromen and Palao-Polco, resulting in intraplate-like signatures similar to back arc volcanism, though Nb/U ratios (up to 20) in the Loicas Trough are not as high as pure back arc mantle melts (e.g. Nb/U up to 60 for the Rio Colorado basalts; Søager et al. 2013). The felsic rocks from all three volcanic complexes have stronger arc-like signatures seen as more pronounced Nb–Ta troughs, as Nb/U generally decreases with increasing silica contents (Fig. 6e). The Nb/U ratio also displays a hyperbolic relationship with Th/La ratios (Fig. 6i). While the different volcanic complexes generally display similar trace element patterns, ratios such as Th/Nb indicate location-unique signatures, with clear distinction of the Domuyo, Tromen and Palao-Polco volcanic regions (Fig. 6h). Additionally, we observe a group of samples that are consistently more enriched in elements such as the HFSE, Rb and the REEs. This enrichment is most prominently observed in Th concentrations, and is referred to as the high-Th group. The high Th group is present across all volcanic complexes (Fig. 6) and does not correspond to differences in analytical method, geography, formation, age, weathering or mineral assemblage/accumulation effects.

Fig. 5

Multi-element (A, C, E & G) and Rare Earth Elements (REE) diagrams (B, D, F & H) normalized to pyrolite/primitive mantle (PM) from Sun and McDonough (1989) and McDonough and Sun (1995), respectively. The REE concentrations in the primitive mantle model (PM, McDonough and Sun 1995) correspond to two times the concentrations of the chondrite model. Colours and symbols as in Fig. 3. Panels G and H show the most primitive lava from Tromen used as a parental composition in magma evolution models (130012), and the most evolved rocks from Domuyo (130058). TDM08 is a basalt from Varvarco, just north of Domuyo (Iannelli et al. 2023, under review). The primitive back arc and main arc samples are basalts from Auca Mahuida and Laguna del Maule, respectively (Kay 2001; Andersen et al. 2017), and are also used in MCS models as starting compositions (see discussion)

Fig. 6

Selected trace elements relative to wt.% SiO₂ (A–G) and trace element ratios (H, I). Trace element concentrations are in ppm, and SiO₂ in wt.%. The fields for the GEOROC compilation of the main and back arc are shown on the ratio plots, same as Fig. 3

The concentrations of Sr, Eu and Ti generally decrease with differentiation indices (i.e. SiO₂, MgO and Mg#), consistent with compatible partitioning of these elements into feldspars, amphibole and spinel forming pronounced negative anomalies in trace element patterns of the evolved samples (Fig. 5). Barium displays a different behaviour compared to Sr and Eu; it initially increases up to 56 wt.% SiO₂ and plateaus before decreasing with silica enrichment. A group of high silica rhyolites from Domuyo volcano display anomalously low Ba (c. 300 ppm), which is discussed below. Additionally, Sr/Y ratios are low (< 50) and thus non-adakitic, while Eu anomalies (Eu/Eu* = Eu_N/(Sm_N*Gd_N)^1/2) range from 1.16 to 0.24. Both Sr/Y and Eu/Eu* decrease linearly with silica (Fig. 6d). The rare earth element patterns (Fig. 5) are enriched and concave-up, with La_N/Yb_N ratios ranging from 3.8 to 15.3 (normalized to primitive mantle; McDonough and Sun 1995). The light REEs are steep with La_N/Dy_N ranging from 3.3 to 15.1 and the heavy REEs are flatter with Dy_N/Yb_N ranging from 0.9 to 1.6, a pattern typical of lavas that have experienced extensive amphibole and/or clinopyroxene fractionation (Fig. 6) (Davidson et al. 2013).

Discussion

Compositional variations in the Loicas Trough

The three volcanic complexes, Palao-Polco, Domuyo and Tromen, have overlapping trace element characteristics and compositions. Several traits from our results point to an Assimilation and Fractional Crystallization (AFC) dominated origin for the Loicas Trough magmas. For example, the sheer number of rhyolitic domes in the area suggests a prevalent driver for felsic melt generation. Petrographic observations from the Loicas Trough reveal that several samples included here contain xenoliths (e.g. sample 130064 from Domuyo and 130009 from Tromen), and sieved plagioclase crystal cores are abundant and can reflect an ante-/xeno-crystic origin. Orthopyroxene is a common phenocryst phase in the Loicas Trough samples, and Bohrson et al. (2020) have genetically linked orthopyroxene occurrence to the onset of assimilation in thermodynamic models in general.

The major elements span an extreme and continuous compositional range, and the rocks gradually transition from alkaline to subalkaline compositions, which is most easily achieved by assimilation of silica-rich material combined with fractional crystallization (DePaolo 1981). Fractional crystallization alone is less likely to cause the change in alkalinity of the Loicas Trough magma as extensively as observed. Furthermore, the hyperbolic relationship between several incompatible trace element ratios like Nb/U, Th/Nb and Th/La (Fig. 6i) indicate open system mixing is required to significantly alter these incompatible trace element ratios (e.g., Langmuir et al. 1978). The trace element patterns for the evolved rocks in the three complexes have also been extensively modified by differentiation, and incompatible trace element ratios generally correlate with differentiation indices. We consider AFC to be the most likely open system process and thus is an ideal starting hypothesis for modelling the Loicas Trough volcanism.

We use the magma chamber simulator (MCS) to gain insights into the above observed results and how they reflect the magmatic evolution (Bohrson et al. 2014, 2020; Heinonen et al. 2020). This allows us to evaluate which processes have the greatest impact on the simulated Loicas Trough magma composition by directly comparing models of fractional crystallization (FC) and assimilation fractional crystallization (AFC), and varying starting compositions and conditions.

Outliers and alteration

The impacts of alteration are evaluated and outliers identified prior to modelling the magmatic evolution. We use both the Chemical Index of Alteration (CIA) (Nesbitt and Young 1982) and the loss on ignition (LOI) to assess alteration of the Loicas Trough samples. These are well-established indicators of alteration, though neither are exclusively influenced by alteration, and have been successfully applied to quantify weathering of magmatic rocks (e.g., Waight et al. 2021). In the Loicas Trough volcanics, the CIA overlaps with the expected range for fresh magmatic rocks, ranging from 40 to 50%, and the LOI is generally below 2 wt.% (Fig. 7). The CIA and LOI increase with SiO₂ reflecting the natural melt evolution as indicated by the fitted curve, thus only values beyond the expected increase are indicative of alteration.

Fig. 7

Variation of Loss On Ignition (LOI) and Chemical Index of Alteration (CIA, Nesbitt and Young 1982) against silica (A–C). The fitted line (A–B) is a second-degree local polynomial fit (loess), and the grey interval indicates the 95% confidence interval

A handful of samples have LOI above 2 wt.% (Fig. 7b and c). These are dominantly pumice samples from ignimbrites and tuff deposits. One of these pumice specimens (130052 from Rio Atreuco, Domuyo) has a high CIA at 58% suggesting its composition has been significantly altered. Six additional rocks have intermediate LOI contents between 2 and 3 wt.% but not anomalous CIA (e.g. 130090), so whether the volatile contents are secondary or primary is difficult to determine. As generally few samples are affected by elevated CIA and LOI, and the high LOI is expected for the pyroclastic samples, they are not excluded. Finally, we observe distinctly low LOI for the low-Ba rhyolites (Figs. 6g and 7b), and suggest the low LOI for these rhyolites is most likely a signature of degassing.

MCS modelling

In the following section, we generate models for the magmatic melt evolution to assess the main processes responsible for the range of observed magma compositions. We use rhyolite-MELTS algorithms (Gualda et al. 2012; Ghiorso 2015 and references therein) and the Magma Chamber Simulator (Bohrson et al. 2014, 2020; Suikkanen 2020; Heinonen et al. 2020) to simulate different scenarios and compare the major and trace element trends of the models to the measured whole rock data. Using the MCS allows us to model thermodynamically constrained complex interactions between three subsystems: 1) The main magma chamber, 2) the wallrock and 3) the recharging/stoping material (bulk mixing subsystem). MCS models assume perfect fractional crystallization, where crystallized phases are chemically isolated. We follow the MCS terminology, where assimilation refers exclusively to assimilation of partial melt from the wallrock, as latent heat from the cooling magma melts the wallrock, whereas stoping refers to bulk rock assimilation (i.e. binary mixing).

Main MCS model

We tested several different model scenarios for the Loicas Trough volcanics including modelling pure fractional crystallization (FC), recharge events and AFC. Furthermore, we tested the effect on the modelled compositions with variable pressure conditions and initial primitive compositions using the MCS. Of these, we found an open system polybaric AFC model at two pressure stages, AFC at mid-level crust followed by pure FC at shallow levels, to be the best fit for the data. This model is referred to as the Loicas AFC model (Fig. 8a). Variations in magma chamber depth (pressure), initial magma composition and recharge of 10–50% of primitive melt are explored from this base model in order to best recreate the observed geochemical variations. For comparison, a pure FC scenario termed the Loicas FC model is also generated retaining the same initial system conditions as the Loicas AFC model (Fig. 8). The model input parameters are summarized in Table 2 and the MCS outputs can be found in Supplementary Material 4. We first describe the model conditions chosen to satisfy geological constraints while providing the best fit to the data, before comparing the results of the models to the Loicas Trough volcanic compositions.

Fig. 8

Diagrams comparing rock data and simulated MCS models. FC fractional crystallization, AFC assimilation fractional crystallization. A Ternary AFM diagram, alkali (A: Na₂O + SiO₂), iron (F: FeO_T) and magnesium (M: MgO) (plotted with ggtern, Hamilton and Ferry 2018). B–E Frost and Frost (2008) diagrams for granite classification showing the Loicas Trough data and MCS models. The Aluminium Saturation Index (ASI) is molar Al/(Ca + Na + K + 1.67). The Aluminium Index (AI) is molar Al₂O₃-(Na₂O + K₂O). Silica saturation is given by the FSSI (Feldspathoid Silica Saturation Index), where FSSI = (Qz—Lc—2Ne—2Kp)/100. Proportions of the normative minerals are estimated using the CIPW norm in GCDkit (Janoušek et al. 2016 and references therein). Mineral abbreviations: plagioclase–plg, orthopyroxene–opx, olivine–oli, alkali feldspar–alk, biotite–bio, spinel–sp, Fe-Ti oxides–Fe-Ti, amphibole–amf

Table 2 Initial input parameters for all MCS models and the wallrock

Model end-member compositions

The initial primitive melt composition for the Loicas AFC and FC models is assumed to be that of an alkali basalt from Tromen (sample 130012), the most primitive lava of our data based on MgO and SiO₂ contents (Table 1). We assess the effect of across arc source melt variation (e.g. Søager et al. 2013; Jacques et al. 2013; Pallares et al. 2019) by using a primitive back arc lava from Auca Mahuida volcano (sample RD17, Kay 2001) and a primitive main arc lava from Laguna del Maule volcano (sample LDM-12–34, Andersen et al. 2017) as potential initial compositions. These back- and main arc volcanoes align with the south- and northward continuation of Loicas Trough (Fig. 1).

The wallrock endmember used for assimilation is represented by the bulk Upper Continental Crust estimate (UCC) of Rudnick and Gao (2014). The initial magma to wallrock mass ratio is set to one (100 g magma to 100 g wallrock in MCS) which best fits the data, after testing a range of mass ratios (50—300 g wallrock). Models are initially simulated using rhyolite-MELTS 1.2.0, which uses a recently calibrated H₂O-CO₂ model for the primitive and intermediate compositional range, as recommended by Ghiorso (2015) for primitive and alkaline compositions. At more evolved compositions (~ 64 wt.% SiO₂ and above) rhyolite-MELTS 1.1.0 is used, which is optimized for the feldspar-quartz minimum (Gualda et al. 2012; Bohrson et al. 2014; Ghiorso 2015). The wallrock subsystem is simulated using rhyolite-MELTS 1.1.0 throughout. Wallrock melts are only assimilated when the partial melt fraction exceeds 15%, the upper recommended limit, to account for the high viscosity of felsic partial melts (Bohrson et al. 2014, 2020). The initial wallrock temperature is set to 550 °C, which is slightly elevated compared to a local geothermal gradient (375 °C at 15 km depth with a gradient of 25 °C/km; Rothstein and Manning 2003) accounting for the prolonged volcanic activity in the area.

Pressure considerations

Single pressure stage AFC models (tested from 10 to 1 kb) fail to replicate the compositions of the Loicas Trough volcanics. Single stage models at deep or intermediate pressures do not reproduce the high silica contents, while shallow single stage models critically underestimate the aluminium contents of the whole rock samples. Instead, at least two stages of AFC are required, first at mid crustal and then at upper crustal levels. The first stage of differentiation for all models occurs at intermediate crustal pressures (4 kb or ∼15 km depth) typical of the fractionation depths of other continental arc plumbing systems (e.g. Price et al. 2012; Andersen et al. 2017) and the presence of mid-crustal phenocryst phases such as amphibole (e.g. Davidson et al. 2007) in the Loicas samples. Model-iterations at pressures less than 4 kb in the Loicas AFC model severely under-predicted the Al₂O₃ contents compared to observed data (Marxer et al. 2021). The second stage of fractionation is modelled to occur at shallow crustal levels (1 kb or ∼4 km depth) and the pressure is changed from 4 to 1 kb when the magma chamber and wallrock reach thermal equilibrium (~ 950 °C). After assimilation ceases due to thermal equilibrium being reached at 4 kb, the modelled melt must evolve at shallower crustal levels as continued deeper differentiation in the absence of assimilation limits silica enrichment and thus the model would not reach silica contents above 73 wt.% as is observed in the Loicas lavas. The lower pressure second stage of magma differentiation is consistent with geophysical observations (e.g. seismic tomography) beneath the Loicas Trough and Domuyo, indicating the presence of shallow magma chambers (González-Vidal et al. 2018; Astort et al. 2019; Lundgren et al. 2020; Loucks 2021). The now cooler evolved melt is emplaced in the cooler shallower crust, and the heat budget is significantly diminished, thus assimilation of shallow wallrock partial melts is expected to be limited. Therefore, the second stage of differentiation is assumed to take place via pure FC.

Oxidation stage and water content

We assume the primitive initial melt of the Loicas Trough has QFM + 1 (quartz-fayalite-magnetite buffer), the back arc has QFM + 0, the main arc magma and wallrock (i.e. UCC) have QFM + 2, consistent with the elevated oxidation state of arc magmatism (Kelley and Cottrell 2009; Blatter et al. 2013). The oxygen fugacity is determined at the liquidus temperature for each melt, and used to estimate the initial FeO and Fe₂O₃ contents from FeO_T of the wallrock and the primitive magmas. The oxygen fugacity is allowed to evolve freely throughout the simulations (Gualda et al. 2012; Bohrson et al. 2014, 2020; Ghiorso 2015; Suikkanen 2020).

We tested different initial H₂O contents for the primitive melts and determined that MCS runs with initial water content at 2.5 wt.% or above were indistinguishable as water saturation was reached early on. Therefore, we assume initial water contents of 1 wt.% for the Loicas AFC and FC models and the back arc model, and 2 wt.% for the main arc and wallrock (e.g. Kelley and Cottrell 2009; Mandler et al. 2014). Additionally, we set the initial CO₂ content at 1 wt.% for the wallrock after Bohrson et al. (2020) but did not consider CO₂ for the primitive melt compositions.

Trace element model

The MCS models are expanded to include predicted trace element concentrations using the trace element engine in MCS (Heinonen et al. 2020). Trace element concentrations for each fractionation step are calculated using the initial concentrations of the primitive and wallrock endmembers, and partition coefficients for each of the predicted mineral phases. We use partition coefficients (K_d) from the Geochemical Earth Reference Model (GERM) database (https://kdd.earthref.org/KdD/) for the same rock types as the samples. The same set of partition coefficients is used in all models, and the trace element model input partition coefficients and references are given in Supplementary Material 3. Figure 10 shows select trace elements against silica for the Loicas AFC model including the partial wallrock melt and the Loicas FC model compared to the Loicas Trough lavas.

The bulk partition coefficients (D_bulk) for some elements, namely the REE (excl. Eu), Hf, Zr, U, Th, and Nb, in the model are underestimated relative to reality when based on the calculated GERM D_bulk. The concentration of these elements in the Loicas AFC model is consistently elevated compared to the data. As an example, consider Nb that is generally incompatible in the predicted mineral phases (calculated D_bulk ranging from ~ 0.001 to 0.4). Observed Nb concentrations remain constant with differentiation indices like SiO₂, which would require the actual D_bulk to be in the range of 0.6—0.9 (Fig. 10d). As discussed below on model limitations, MCS underestimates or fails to stabilize phases such as apatite, amphibole, and accessory phases like zircon and rutile. These minerals have high K_d for the above mentioned REE and HFSE, therefore the underestimation or absence of these minerals in the models has a high impact on the calculated bulk partition coefficients. For example, as the MCS model does not include zircon, we do not expect Zr and Hf concentrations in the melt model to fit the sample compositions. As an example for Hf, the average bulk partition coefficient, ${D}_{bulk}^{Hf}$, estimated by MCS using the GERM partition coefficients is ~ 0.1. The best fit ${D}_{bulk}^{Hf}$ to fit the data is ~ 0.3, and with a partition coefficient for Hf in zircon, ${K}_{d}^{Hf}$, of ~ 1000 in andesite (Fujimaki 1986), the fractionating crystal assemblage needs to include as little as 0.02% zircon to account for the best fit to the data. Therefore, to account for these missing phases for affected elements, we bypass the individual mineral partition coefficients or K_d. Rather than calculating the bulk partition coefficients using the GERM K_d, we determine the best fit bulk partition coefficient using iterations that results in the best fit of the model to the data for these specific elements (i.e. REE excl. Eu, Hf, Zr, U, Th, and Nb) (Supplementary Material 3).

Model limitations

The rhyolite-MELTS algorithms are limited to the underlying thermodynamic data and limited phase endmembers in the thermodynamic dataset (Gualda et al. 2012; Ghiorso 2015 and references therein). As noted above, rhyolite-MELTS is unable to predict the behaviour of important accessory phases such as zircon, which are not present in the thermodynamic database. However, even for phases that are included such as apatite, not all possible compositional end-members are present. We observe a poor correlation between the model and observed data for P₂O₅ that is a direct result of apatite fractionation being underestimated in the model (Fig. 9i).

Fig. 9

Harker diagrams and select major element ratios displaying the different models compared with the whole rock data (Fig. 4)

The biggest limitation of rhyolite-MELTS in our models is that it does not predict the crystallization of amphibole, which is observed as a phenocryst phase in several samples. The rhyolite-MELTS database includes some common amphibole endmembers, such as hornblende, but is extremely limited compared to the diversity of natural amphiboles. This limits amphibole stability in rhyolite-MELTS based models (Hirschmann et al. 2008; Gualda et al. 2012; Bohrson et al. 2014; Ghiorso 2015). Despite experimenting with pressure and initial water content, none of our models stabilized amphibole. The potential impact of amphibole as a fractionating phase is explored by manually fitting a three stage AFC mass balance model (Loicas manual AFC model) (DePaolo 1981). Mass balance fractionation models can provide very good fits to data, especially if several stages are involved. However, the solutions are non-unique, arbitrary and user-dependant, and thus do not yield genuine insights into the processes in the same way as a thermodynamically controlled model like MCS, which considers phase stabilities and heat budgets. Nevertheless, we include a mass balance fractionation model as a point of reference for the MCS model performance. A three stage model with three different fractionating assemblages is used to capture the two major kinks in the major element trends at ~ 5 wt.% MgO and ~ 62 wt.% SiO2 (Fig. 4). The three stages used in the mass balance model are: 1) fractional crystallization of a mixture of Fe-rich spinel and Fe-Ti oxide (oxides), plagioclase and olivine, 2) assimilation of rhyolitic melt and fractionation of a mixture of amphibole, plagioclase, clinopyroxene, oxides, apatite and olivine, and 3) assimilation of rhyolitic melt and fractionation of a mixture of amphibole, alkali feldspar, plagioclase, oxides, and apatite. The mineral compositions are based on the median mineral compositions of the phases present in the Loicas samples (from Traun 2023), and the full calculation can be found in Supplementary Material 2. We use the most evolved rhyolite from Domuyo (sample 130058, Fig. 5) as an assimilant, as felsic melts are commonly used as upper crustal assimilants in mass balance/mixing models to account for significant increases in silica contents, and this particular sample belongs to the low Ba-rhyolites which share several characteristics with modelled partial crustal melts (discussed below). The Loicas manual AFC model (pink model, Figs. 8 and 9) is generally consistent with the MCS Loicas AFC model (olive green model, Figs. 8 and 9). Therefore, even though amphibole is lacking in our models, the melt evolution trend appears unaffected, suggesting MELTS may ‘compensate’ for the lack of amphibole through crystallisation of other mafic phases like clinopyroxene, orthopyroxene, olivine and spinel for example. Given that all other phases observed in the rocks are predicted by the MCS models, we conclude that the lack of amphibole fractionation does not significantly affect the ability of the model to explain the major element evolution of the magma.

When modelling the trace element variations these limitations and underestimated phases can have a major impact on some trace elements (like Hf and Zr when zircon fractionation is absent). To solve this issue, we determined a best fit bulk partition coefficient for affected elements (REE excl. Eu, Hf, Zr, U, Th, and Nb). Consequently, by fitting bulk partition coefficients any information on how changes in the mineralogy of the model affect the concentration of these elements is lost. However, the modelled composition for these trace elements can still be used to illustrate the effect of changing initial compositions and different contaminants. Additionally, a well-fitted concentration prediction is critical when exploring combined trace element and isotope models.

Models of trace element concentrations that are dominated by the major phases like plagioclase, alkali feldspar, olivine and pyroxene (e.g. Sr, Eu, Sc, Ni and Ba), or elements that are consistently incompatible (e.g. Rb and Pb) are well predicted compared to the data. Therefore, MCS is suitable for modelling these trace element behaviours in the Loicas Trough case when combined the GERM partition coefficients (Fig. 10). The absence of minor and accessory phases for these elements therefore has an insignificant impact on the bulk partition coefficients of these elements.

Fig. 10

Trace element model predictions (A–F) for the Loicas AFC and Loicas FC models (Fig. 8). The red arrow indicates the evolution (pointing towards increasing degree of melting) of the compositions of the partial wallrock melt, and the yellow star is the hydrous bulk composition of the wallrock (UCC, Rudnick and Gao 2014). The trace element model partition coefficients can be found in Supplementary Material 3. Mineral abbreviations as in Fig. 8

Model results

A successful model should capture the broad major element trends, and replicate the main petrological and geochemical observations. We begin by comparing the predicted calc-alkalinity, alkalinity and silica saturation (Fig. 8). The granitic classification diagrams (Frost and Frost 2008) are useful to compare silica, aluminium and alkali saturation patterns of the models and our data. We observe a fundamental distinction between AFC and FC as predicted by MCS (Loicas AFC and FC models, Fig. 8). Both the Loicas AFC and FC models predict the transition from magnesian to ferroan compositions (Fig. 8a) and replicate a slight iron enrichment prior to FeO_T and MgO depletion as illustrated on the AFM diagram (Figs. 8b, 4e and 9e). However, in terms of alkalinity, peralkalinity and silica-saturation, the Loicas AFC model is more consistent with the data, as the Loicas FC model severely overestimates the alkalinity (Fig. 8c) and underestimates the silica saturation and aluminium saturation (Fig. 8d and e). Melt compositions predicted by the Loicas FC model remain within the alkaline silica-undersaturated to -saturated field, whereas the data and the Loicas AFC models cross the silica-saturation boundary and become increasingly silica-oversaturated. Crossing into the silica oversaturated field as indicated by the Feldspathoid Silica Saturation Index (FSSI) reflects assimilation of rhyolitic crustal partial melts in the model (Fig. 8e), which also affects the predicted fractionating mineral assemblage (Supplementary Material 4). Both the Loicas FC and AFC models predict the first phase to crystallize is spinel followed by olivine, clinopyroxene and then plagioclase. Our MCS model for the wallrock predicts that the main phases contributing to the partial wallrock melt are quartz, biotite and alkali feldspar, and the dominant residual wallrock phase is plagioclase (Supplementary Material 4). After assimilation is initiated the Loicas AFC model stabilizes orthopyroxene in response to the increase in silica oversaturation, which is not present in the Loicas FC model (Bohrson et al. 2020; Heinonen et al. 2020). Orthopyroxene is observed as a phenocryst phase in the rocks, supporting the Loicas AFC model to explain differentiation and mineralogy of the Loicas Trough magmas.

Assimilation and residual wallrock mineralogy

The AFC models are a better fit to the major and trace element data relative to the results of the FC models (Figs. 9 and 10). AFC models increase the melt silica-contents relative to the alkalis along a trend matching the data, whereas alkalis in the FC model become more enriched in the modelled melts relative to the measured sample compositions. Assimilation effects the predicted trace element concentrations of the models in two ways, both through mixing with the partial crustal melt and secondly via changes in the crystallizing assemblage (Fig. 10; Supplementary Material 4). For example, the Sc concentration in the rocks consistently decreases with differentiation (Fig. 11b). The decreasing Sc content is reproduced by the Loicas AFC model, where the addition of partial melts both drives the stabilization of orthopyroxene, in which Sc is compatible, and also acts to dilute Sc in the evolving magma. This Sc trend is not predicted by the Loicas FC model, where Sc contents increase continuously as orthopyroxene does not stabilize and Sc contents in the melts are not diluted by addition of low Sc crustal melts. The modelled Sc content of the Loicas AFC model fails to reproduce the magma Sc content at dacitic and rhyolitic compositions due to the model limitations regarding amphibole, as discussed above in Model Limitations.

Fig. 11

Trace element ratio models for the main arc and back arc AFC models (see text) compared to the Loicas AFC model, illustrating the across arc variation and slab fluid input, as approximated by the Nb-anomaly. The small blue diamonds are samples from Varvarco, provided by Iannelli et al. (pers. comm. 2020). A, B The red arrow indicates the composition of the partial wallrock melt and the yellow star indicates the bulk wallrock composition, as on Fig. 10C). The shortest great circle distance from each sample location to the Nazca trench is calculated using the plate boundary coordinate dataset by Bird (2003)

Residual phases in the melting wallrock impact the assimilating partial melt compositions, and therefore also affect the magma compositions. The role of residual mineralogy is well illustrated with Rb and Sc (Fig. 10b and e). Biotite and alkali feldspar are Rb-rich minerals and significantly contribute to the early partial melts, and as a result the Rb concentrations of the partial crustal melts are higher than bulk wallrock. However, once alkali feldspar and biotite are exhausted, Rb contents start decreasing as the melting assemblage changes to comprise plagioclase and quartz (Rb-poor minerals). The contribution of biotite to the partial melt is also seen in Sc concentrations, as Sc is compatible in biotite and other mafic phases. The Sc concentration of the wallrock partial melt is only slightly lower in the partial melt than the bulk wallrock compared to other compatible elements like Sr and Eu, which have very low concentrations in the partial melt.

The influence of residual plagioclase in the wallrock is observed in elements such as Na, Ba, Sr and Eu which are compatible in plagioclase (Blundy and Wood 1991). With differentiation the observed Na₂O content of the rocks gradually increases, where Sr and Eu decrease after feldspar saturation in the crystallizing assemblage. Ba content in the melt initially increases, then increases more gradually after feldspar saturation, and begins to decrease when the evolving melt reaches ~ 65 wt.% SiO₂. These trends are overall well reproduced by the Loicas AFC model, and less well captured by the Loicas FC model (Figs. 9g and Fig. 10a, c and f. In the Loicas AFC model, a large fraction of residual plagioclase in the wallrock retains Na, Ba, Sr and Eu²⁺, and the partial melt is thus extensively depleted in these elements. Low contributions of Na, Ba, Sr and Eu from wallrock melts caused by residual plagioclase in the wallrock dilutes the sodium contents in the magma (Bohrson et al. 2020; Heinonen et al. 2020). As heating and melting progress, plagioclase eventually starts contributing to the partial melt, leading to large increases in the partial melt concentrations of Ba, Sr and Eu. As Na is a major element, it is affected differently than the trace elements. The degree of Na₂O dilution in the AFC model is highly dependent on the assumed assimilant composition and the wallrock to magma ratio, which affects the degree of melting and thereby the residual mineral assemblage in the wallrock. A higher wallrock/magma ratio leads to a greater degree of Na₂O dilution (i.e. lower Na₂O contents in the magma) as there is a larger proportion of residual plagioclase at lower degrees of melting.

The critical melt fraction of the wallrock has a small effect on the modelled magma compositions. This is due to the limited change in the partial melt compositions at the range of critical melt fraction (typically within 5–15%, Bohrson et al. 2014, 2020). Varying the initial wallrock temperature affects the onset of assimilation and the effect is tested in additional models. A higher initial wallrock temperature leads to an earlier assimilation onset. This has a stronger diluting effect on the Na₂O content of the magma. Lower initial wallrock temperatures and later onsets of assimilation fail to reproduce the silica-oversaturated compositions of the data.

Recharge and replenishment

Recharge in the form of replenishment by more primitive magma is a common process during magma differentiation that changes the melt composition and affects the heat budget of the system (Davidson et al. 2001; Bohrson et al. 2014; Andersen et al. 2017). We explored the impact of recharge in our MCS models by adding different masses (20—50%) of primitive melt into the magma chamber during the early, mid or late stages of differentiation. We find that recharge does not significantly impact major element model predictions, rather it simply shifts compositions back along the same overall fractionation trend. However, recharge does add complexity to the trends, particularly if recharge occurs after the magma chamber has fractionated significant amounts of plagioclase and clinopyroxene. Recharge of a primitive melt and mixing across these major phase shifts could be a driver for some of the observed compositional scatter along these shifts (e.g. Al₂O₃ and TiO₂ scatter at 50% SiO₂ or 6% MgO, Figs. 4 and 9). A primitive melt recharge with a composition akin to the initial primitive melt also increases the overall heat budget of the system, which leads to greater crustal melting potential. However, because the mass of mafic melt also increases, there is no significant increase in the relative amount of partial crustal melt added. Given the limited observable impact of recharge, for simplicity, our final models do not include recharge events.

Aluminium saturation, pressure and accumulation effects

We find that assuming lower pressures and higher initial water contents in the parental melt composition (up to 2.5 wt.%) generally results in more silica-oversaturated compositions (not shown). However, low pressure conditions significantly affect Al₂O₃, as plagioclase stabilizes earlier at lower pressures, leading to low Al₂O₃ contents in the modelled magmas compared to the data (e.g. Marxer et al. 2021). At 4 kb pressure, the models predict a better fit to the observed ASI, though the ASI is still underestimated at higher degrees of differentiation (Figs. 8d and 9d). Models assuming even higher pressure fractionation (above 5–6 kb) have increased clinopyroxene fractionation and as a result silica contents decrease during early differentiation, inconsistent with the data.

Instead, we interpret the slightly underestimated ASI and aluminium contents, combined with the slightly elevated alkali contents in the model, partially reflect accumulation of phenocrysts phases in the samples (Figs. 8d, 9a, d and h). The whole rock data here represent bulk analyses, which may include pheno-/ante-/xeno-crysts and thus diverge from true melt compositions. Therefore the predicted evolution of melt compositions using MCS may not be directly comparable to the observed bulk rock compositions (e.g. Bohrson et al. 2014, 2020; Heinonen et al. 2020). Furthermore, phases with densities similar to the surrounding melt, such as feldspars, are more likely to remain suspended in the magma and cause bulk rock compositions to deviate from true melt compositions. Thus, the model assumption of perfect fractional crystallization would not apply. The presence of feldspar, observed in most samples, could indeed account for the slight discrepancy between the Loicas AFC model and the data for Al₂O₃ concentration (Figs. 8d and 9d, g and h, from 62 to 74 wt.% SiO₂). Based on a simple mass balance, it would require the presence of 6–17% plagioclase or alkali feldspar to account for the gap in Al₂O₃, consistent with the petrographic thin-section observations, where the modal abundance of feldspar phenocrysts is usually around 10%. Oxides such as iron spinels can have ASI as high as 58, due to their modest Al-concentrations but non-existent alkali content (Supplementary Material 2). However, their Al₂O₃ abundance is less than the melting models, and thus they alone cannot account for the difference in ASI. Additionally, as most samples above 65 wt.% have glassy groundmasses, it is possible that the evolved samples have experienced slight Na-loss due to devitrification as is relatively common in rhyolites (e.g. Ewart 1971) and could explain why the gap is more pronounced for specifically the most felsic samples.

The last phases to crystallize in the Loicas AFC model are alkali feldspar, quartz and titanite, which cause large fluctuations in the CaO contents of the Loicas AFC model at ∼73 wt.% SiO₂ (Fig. 9d and f). Rheological and density constraints mean that quartz and alkali feldspar are physically unable to fractionate, therefore this last fluctuation in the MCS model is not reflected in the bulk rock compositions. The Loicas FC model does not display the same fluctuation in CaO at the last stages of fractionation, as the silica-undersaturated composition does not stabilize quartz.

Low Ba rhyolites

The low-Ba rhyolite samples from Domuyo (Rhyolita Cerro Domo Fm.) have several geochemical features that are very similar to the crustal partial melts described above. These atypical rhyolites have low Ba contents (less than 300 ppm Ba) and somewhat elevated Sc (and FeO_T). This is the opposite of expectations from fractional crystallization, as Ba is generally incompatible and Sc is compatible (Blundy and Wood 1991). These anomalous Ba and Sc contents are coupled with very low Sr (< 5 ppm), CaO and LOI contents, low Eu/Eu* anomalies (< 0.4) and slightly elevated FeO_T. We interpret that these low-Ba rhyolites are distinct from the other rhyolites and not formed by AFC.

Extensive fractionation of alkali feldspar and sodic plagioclase can cause Ba concentrations to decrease rapidly (Blundy and Wood 1991). Our Loicas AFC model does predict that these minerals fractionate, but not at the extent required to account for the observed drop in Ba concentrations. Instead, we interpret these to represent local extrusions of early-stage partial melts from the wallrock with little mixing with the main magma chamber. The low-Ba rhyolite partial melts must have been generated prior to biotite exhaustion (Fig. 10e) and late stage plagioclase contribution to the melt (Fig. 10c), to explain their composition. Scandium contributions from biotite (or amphibole) to the partial melt could explain the elevated Sc (and FeO_T) contents, and abundant residual plagioclase in the wallrock could explain the low Ba, Sr and Eu/Eu* (Supplementary Material 3). However, it should be noted that the partial melts from the model are not as silica rich as the low-Ba rhyolites of the samples. This could reflect true wallrock compositions that are more evolved and thus have a higher proportion of felsic minerals like alkali-feldspar and quartz than the UCC wallrock composition assumed in our models (Rudnick and Gao 2014).

Across arc variations

The trace and major element results show that the primitive Loicas Trough samples have signatures intermediate between the main and back arc lavas, most strongly observed in the variable Nb-anomaly (Nb/U ranging from c. 7 to 27) for basaltic (i.e. SiO₂ < 52 wt.%) samples (Fig. 6). The primitive rocks from Tromen and Palao-Polco have OIB-like trace element signatures of elevated HFSEs, but additionally, display arc-like signatures with enriched LILEs, a small Nb–Ta trough and a positive Pb anomaly. These results are consistent with the observed across arc variation found when evaluating main and back arc compositions, and reflect a progressively decreasing slab input into the mantle source to the east (Søager et al. 2013; Jacques et al. 2013; Pallares et al. 2019). We interpret this across arc source variation is expressed as a compositional gradient between the subarc mantle and the sub continental mantle below the back arc continuously sampled by the Loicas Trough magmatic systems. The predicted Main arc and Back arc AFC models allow us to assess the expected outcome of across arc variation on the models, and act as bounds for the Loicas model (Kay 2001; Andersen et al. 2017). All conditional parameters and partition coefficients are kept identical to the Loicas AFC model, with the exception of the initial wallrock temperature for the Main arc model, which is increased from 550 °C to 700 °C. The initial main arc composition is less primitive than the back arc and Loicas initial melts, with lower MgO and higher SiO₂ content. It is assumed that the primitive main arc composition has already been fractionated and heated the wallrock, so the temperature of the wallrock is increased a corresponding amount (150 °C) to make the onset of assimilation of all three AFC models comparable (Table 2).

The Main arc AFC and Back arc AFC models predict broadly the same trend as the Loicas AFC model (Figs. 8 and 9), but several differences are important. Initially, the Main arc AFC model is less alkaline than the Back arc and Loicas AFC models and reaches silica-oversaturated compositions earlier in the model. The oxidized and wet starting composition of the Main arc AFC model also reproduces the characteristic calc-alkaline trend seen in the AFM diagram (Fig. 8a and 9e) due to a lack of initial iron enrichment (Hora et al. 2009; Kelley and Cottrell 2009; Blatter et al. 2013; Mandler et al. 2014). A closer look at the phases produced in the three MCS models reveals additional differences in the fractionating mineralogy that are not easily distinguished by the major elements alone (Supplementary Material 4). The first sequence of phases to stabilize in the Back arc AFC model are: Olivine, clinopyroxene, plagioclase and spinel. For the Loicas AFC model, the order of crystallization is: Spinel, olivine, clinopyroxene and plagioclase. For the Main arc AFC model, MCS predicts spinel, clinopyroxene and plagioclase. Thus, it is clear that for the silica-oversaturated hydrous and oxidized Main arc AFC model olivine fractionation is suppressed while spinel stabilizes earlier in the crystallization sequence when compared to the Back arc AFC model. This illustrates how MCS is able to reproduce the general calc-alkaline and tholeiitic trends as a proof of concept control test.

The impact of across-arc source variations on the slab fluid sensitive trace element ratios is assessed using the Main arc and Back arc AFC models (Fig. 8). The models predict a systematic change in the Nb/U and Nb/La as these ratios become more similar with increasing silica content driven by assimilation of partial melt (Fig. 11a and b). As the silica content increases, the source signatures are obscured, however, the Loicas Trough rocks consistently plot between the predicted main and back arc trends. Furthermore, the three volcanic centres capture the gradient in mantle compositions from the back arc to the main arc, as Tromen shows the largest variation in Nb/La and Nb/U, Palao-Polco displays intermediate variation and Domuyo has the smallest variation. These variations correlate systematically with perpendicular distance to the trench and melt differentiation (Fig. 11c), and are consistent with observations on the similarity between Tromen Volcanic Complex and the back arc when contrasted with the main arc (Kay et al. 2006; Llambías and Leanza 2011; Jacques et al. 2013; Pallares et al. 2019 and references therein). Our Domuyo samples only include intermediate to felsic compositions, therefore we have included samples from the immediately adjacent Varvarco Volcanic Field (Iannelli et al. 2023, under review) (Figs. 1 and 2) to represent primitive compositions near Domuyo. We infer that the three volcanic provinces have tapped into a compositionally zoned mantle at three different points on the sub-arc to back arc mantle gradient resulting in different source signatures, while still following a similar magma evolution trajectory in the crust. The Loicas Trough volcanic rocks, therefore, provide a snapshot of the compositional mantle gradient underneath the trough, consistent with other back and main arc studies of the Southern Volcanic Zone back and main arc volcanism (e.g. Søager et al. 2013; Jacques et al. 2013; Pallares et al. 2019).

Shallow AFC driven formation of Loicas Trough lavas

Combining our observations, major and trace element results and MCS models, we can construct a conceptual petrogenetic model for the Loicas Trough magmas. Figure 12 presents a schematic NW–SE cross section of the Loicas Trough, summarizing the main findings of this study. The defining attributes for the Loicas Trough volcanism are its off arc axis location within the faulted fold and thrust belts, east of and adjacent to the main arc with geochemical signatures inconsistent with purely slab fluid induced melting or purely intraplate decompression melting. Instead, melting is induced by oblique plate movement combined with slab roll-back which results in lithospheric extension through transtensional faulting and mantle decompression below the thrust and fold belts (Cobbold and Rossello 2003; Folguera et al. 2007a, 2015; Messager et al. 2010; Ramos et al. 2014; Rojas Vera et al. 2015). This setting is therefore also similar to those found in leaky transform faults and pull-apart basins at strike-slip plate boundaries and oblique subduction zones such as the Sumatra arc, Anatolian arc, and the Gulf of California (van Wijk et al. 2017; Sutrisno et al. 2019; Cosca et al. 2020).

Fig. 12

Conceptual cross section of the Loicas Trough, illustrating the three volcanic centres, the geological interpretation of the major and trace element results and the MCS models. In response to slab rollback and westward expansion of the decompression underneath the back arc, melting occurs underneath the Loicas Trough. The volcanic centres tap into a compositionally zoned mantle, which has a high slab-fluid induced mantle under the main arc and an intraplate back arc mantle to the west. The magmas evolved in two primary stages in the middle and upper crust, and induce anatexis, partial melt assimilation and locally partial melt extrusion (e.g. Rhyolita Cerro Domo Fm. at Domuyo)

The geochemical source signatures of the Loicas Trough magmas vary reflecting a heterogeneous and compositionally gradationally varied mantle that is intermediate between the back arc and main arc endmembers (Fig. 11). This results in distinctive trace element signatures in the parental melts (e.g., Nb/U) that correlate with perpendicular distance from the trench across the arc. The Domuyo Volcanic Complex is closest to the main arc and has a stronger arc trace element signature, whereas the Tromen Volcanic Complex is closer to the back arc and has a stronger back arc signature, though neither complex displays pure main arc or pure back arc signatures. The primitive magma compositions are also systematically more alkaline than the arc equivalents, indicating lower degrees of melting under the Loicas Trough compared to the main arc. In our MCS models we are able to well predict this signature by using intermediate oxidation states and water contents (Kelley and Cottrell 2009; Blatter et al. 2013).

The complete lava series displays the full range of primitive to highly siliceous lavas with increasing silica saturation, a trend driven by lithospheric processes of fractional crystallization, anatexis and assimilation. The latent heat of crystallization released from the magma chambers causes partial melting of the surrounding upper continental crust (Figs. 8, 9 and 10). Major element evidence for the dominant role of differentiation on the magma compositions is the transition from alkaline to sub alkaline compositions and the gradual increase in silica saturation (Frost and Frost 2008). Evidence for open system behaviour in trace elements is observed as changing incompatible trace element ratios such as Nb/La and Th/La with differentiation indices (Fig. 6). Evidence of crustal melting independent of assimilation into the magma is also indicated by the presence of low Ba rhyolites, whose low Ba concentrations are also associated with low Sr and Eu/Eu^*, and elevated Sc and FeO_T compared to the other rhyolites and dacites. These signatures of the low Ba rhyolites are reminiscent of the predicted crustal partial melt compositions from the MCS models, where biotite contributes to the crustal partial melt (elevated Sc and FeO_T), and plagioclase is a main residual phase in the wallrock (low Sr, Ba and Eu/Eu*).

We interpret that fractional crystallization and differentiation in the Loicas magmas are polybaric and progresses at middle to upper crustal levels. Geophysical surveys from the area indicate the presence of shallow magma chambers or fluids underneath the trough (González-Vidal et al. 2018; Astort et al. 2019). However, the MCS models indicate at least two stages of differentiation are needed to best reproduce the Al₂O₃ and SiO₂ trends of the Loicas rocks, the first stage at ~ 4 kb followed by a second stage at ~ 1 kb. These pressure levels coincide with studies of similar environments (Hammersley and DePaolo 2006; Andersen et al. 2017; Marxer et al. 2021; Oyarzún et al. 2021). Indications of shallow assimilation rather than deep assimilation in the Loicas Trough are also observed in non-adakitic Sr/Y ratios, which decrease with silica content directly contradicting typical deep crustal contamination signatures (Loucks 2021). These findings are also consistent with the transtensional tectonic setting, as extension favours shallow crustal ponding of magmas (Loucks 2021).

Global implications

The recognition of this unusual style of volcanism in the Loicas Trough begs the question of whether it is unique, or whether similar examples can be found elsewhere. Therefore, we searched for volcanic provinces, which share similar tectonic settings to the Loicas Trough volcanism. In other words, volcanoes located behind the main arc in oblique subduction zones but away from back arc and intraplate basins, and volcanoes near transform faults in active continental margins. We used Quaternary volcano locations from the Smithsonian Global Volcanism Program, and the plate boundary dataset from Bird (2003) to identify volcanoes in this setting around the Pacific Ring of Fire. Based on the location of these volcanoes, we compiled volcanic samples from the GEOROC database in a 25 km radius around each volcano, and compared these subregions to the Loicas Trough (Fig. 13, full compilation and references in Supplementary Material 6).

Fig. 13

Comparison of the geochemical compositions of volcanoes from 9 subregions to the Loicas Trough samples of this work and back arc and main arc compositions from the Andes as in Fig. 3 and Fig. 6. The data for the volcanoes were compiled from the Smithsonian: Global Volcanism Program (incl. all Quaternary volcanoes) and the GEOROC Compilation: Convergent Margins, https://doi.org/10.25625/PVFZCE (accessed on 15/10–2023). The joined dataset and specific references used for the Fig. 13 are given in Supplementary Material 6

The main characteristics of the Loicas Trough volcanism and its magma compositions are: Wide ranges in lava compositions from primitive to highly siliceous, by shallow AFC controls, relatively low degrees of melting, variable slab fluid influence in the source signatures, and proximity to continental arcs.

We found a non-exhaustive list of volcanic province examples that share these above-mentioned characteristics that includes: the neighbouring Loncopué and Bío Bío Alumine Troughs (Folguera et al. 2007a; Varekamp et al. 2010; Rojas Vera et al. 2014), the eastern part of Andean Central Volcanic Zone, the volcanoes in the paleo-Farallon subduction zone/modern San Andreas Fault in western USA, the eastern Cascades and Northern Cordilleran Volcanic Province in Canada (Edwards and Russell 2000), the northern Central Kamchatka Depression, the western North Island New Zealand (Price et al. 1992; Pittari et al. 2021), the north-eastern Central American arc by the Montagua Transform Fault, the northern Trans-Mexican volcanic belt (Verma and Nelson 1989; Ferrari et al. 2012), and north-eastern Sumatra in Indonesia (Sutrisno et al. 2019). Like the Loicas Trough, these localities are linked to transtension or lie on transform boundaries in an overarching convergent setting, and display a wide compositional range. The mantle sources for these localities are generally mixed, and display variable slab fluid input. Furthermore, large degrees of assimilation and interaction with the continental crust have been found to be critical in generation of felsic melts (Grove et al. 1988; Kelleher and Cameron 1990; Price et al. 1992, 1999; Cousens 1996; Schmitt et al. 2006; Hammersley and DePaolo 2006; Portnyagin et al. 2007; Carr et al. 2007; Donnelly-Nolan et al. 2011; Walker et al. 2011; Ferrari et al. 2012; Mandler et al. 2014; Bray et al. 2017; Carlson et al. 2018; González-Maurel et al. 2020; Wang et al. 2022). Figure 13 displays the slab-component, represented by Nb/U, and differentiation range, represented by SiO₂, of these localities compared to Loicas Trough samples from this work. There is generally a consistent first-order compositional trend, although we recognize each area will be affected by its local geological environment.

A detailed examination of examples of volcanic complexes from these localities reveals magmatic origin and evolution processes that are similar to the Loicas Trough. The Clear Lake volcano is part of the Coast Ranges of California. The source of volcanism is unclear but believed to originate from asthenospheric upwelling through a slab window. Hammersley and DePaolo (2006) modelled the magmatic evolution of these lavas using a polybaric two step AFC model, similar to our Loicas Trough model (Fig. 13), which was also consistent with middle and shallow crustal assimilation and differentiation rather than deep crustal differentiation. The Long Valley Caldera, Mono-Inyo Volcanic Chain, Coso Volcanic Field and Ubehebe Craters are situated at an old continental arc that has now become a transform boundary (Kelleher and Cameron 1990; Cousens 1996; Bray et al. 2017; Wang et al. 2022). The volcanism also ranges from basaltic to rhyolitic with glass domes similar to those at Domuyo (Fig. 13). The Newberry Volcanic Complex in Oregon, and Medicine Lake volcano and Lassen Volcanic Centre in California are situated in the Cascades east of the main arc axis in NW–SE striking rift zone similar to the Loicas Trough, and crustal assimilation is an important factor in differentiation (Grove et al. 1988; Bullen and Clynne 1990; Donnelly-Nolan et al. 2011; Mandler et al. 2014; Carlson et al. 2018).

The Altiplano-Puna Volcanic Complex is situated in the Central Volcanic Zone of the Andes, and like the Loicas Trough, located east of the main arc along extensional faults (Coira and Kay 1993; Stern 2004; González-Maurel et al. 2020). The rocks display both arc and back arc characteristics in their trace elements compositions (Coira and Kay 1993; Stern 2004), and oxygen isotopes show assimilation of continental crust at evolved compositions (González-Maurel et al. 2020).

The Kamchatka arc is a very complex subduction zone due to its double arc chain and intersection between the transform Aleutian Trench Plate Boundary (Ponomareva et al. 2007). In the northern part of the Central Kamchatka Depression closest to the transform fault, several large volcanoes (e.g. Shisheiski Complex), display mildly alkaline compositions and mixed source signatures with a smaller slab component (Portnyagin et al. 2007), similar to the Loicas Trough (Fig. 13).

The Central American Arc Volcanoes are intersected by the Motagua transform fault in northwestern part, and this intersection is correlated to several volcanoes (e.g. Ixtepeque, Ipala and Chiquimula Volcanic Field) east of the main arc in the Ipala Graben (Walker et al. 2011; Carr et al. 2014). Additionally, a handful of volcanoes (e.g. Conception and Cerro El Ciguatepe) are located in the Nicaragua Depression behind the main arc, which is connected the Hess escarpment fault to the south (Carr et al. 2007, 2014). These volcanoes also display variable alkalinity and source signatures from both diverse subducted sediments and intraplate mantle sources with significant overprinting crustal assimilation (Carr et al. 2007, 2014; Walker et al. 2011).

We suggest that the consistent characteristics of these volcanic provinces merit the recognition of a specific type of arc volcanism, which we term arc-adjacent volcanism. Our intention is not to give a one-fits-all interpretation of these volcanoes, as we acknowledge all volcanic regions are inevitable influenced by their local geology and physical environment. However, we draw comparisons to arc lavas, which despite being influenced by the nature of their slab (e.g. angle and thermal parameter) and variations in subducting material, and the thickness and nature of the overlying continental crust etc. (e.g. Carr et al. 2007; Plank 2014; Kelemen et al. 2014), global arc lavas still share a setting and first order compositional trend. By suggesting the term arc-adjacent volcanism, we aim to provide a first-order framework within which volcanoes and their lava compositions within transtensional and transform geological provinces adjacent to active continental margins could be understood as distinct from the main arc chain and back arc basins.

Arc-adjacent volcanic provinces would generally be found in transtensional settings adjacent to, but off-axis from, active or palaeo-arcs and back arcs, and feature wide ranges of lava compositions driven by extensive mid crustal to shallow AFC, with lower degrees of melting than beneath main arcs. We infer the arc-adjacent placement allows for mantle melting to occur with variable slab fluid influences, resulting in mixed mantle signatures between a slab fluid dominated sub-arc mantle and other mantle end-members. However, the lack of widespread extension arc-adjacent environments in comparison to back arc regions limits the rapid ascent compared to back arc settings, and provides optimal conditions for shallow magma stagnation, crustal heating and assimilation. We suggest that the recognition of this style of magmatism in the transform and transtensional settings by active margins may aid in the characterization of diverse continental magma compositions and identification of palaeo-arcs earlier in Earth’s history.

Conclusion

This study uses geological observations, combined with major and trace element results and MCS models to interpret the origin of the Loicas Trough magmas located in an eastern belt relative to the present-day main Andean arc (36–37°S). MCS models successfully replicate the observed major and trace element results and highlight the importance of AFC processes in generating these lavas (Fig. 12). Slab rollback and oblique subduction caused the formation of the Loicas Trough and decompression melting underneath the fold and thrust belts. The oblique orientation of the Loicas Trough relative to the main arc (Figs. 1 and 2), allowed volcanic centres to tap the compositional gradient between the main arc and back arc mantle sources (Fig. 11). Decompression melting of a mantle displaying an across arc gradient is reflected in geographical variations in both major and trace elements as primitive lavas become progressively more alkaline and intraplate-like towards the east. However, these mantle signatures are obscured early in their differentiation by the onset of assimilation of felsic partial melts of adjacent upper continental crust. This causes trace element patterns to develop a pseudo-arc like signature inherited from the wallrock, seen most strongly as an increase in the Nb–Ta anomaly with differentiation. Furthermore, assimilation of wallrock-derived partial melts causes the initially alkaline basalts to fully cross the silica-saturation boundary and become subalkaline high-K dacites and rhyolites.

Our models and petrological observations indicate that the majority of differentiation and assimilation takes place in the middle crust, although a late stage of shallow fractional crystallization is necessary to reach high silica contents significantly above 70 wt.% SiO₂. Although we find the gradual addition of partial crustal melts is instrumental in generating the wide range observed from alkali basalt to trachy-dacites and rhyolites, the composition of the most evolved low-Ba rhyolites from Domuyo is inconsistent with extreme AFC. These low-Ba rhyolites are interpreted as pure or almost pure crustal melts, further highlighting the importance of anatexis to explain the Loicas Trough magma compositions. The high Sc and Fe, and low Ba, Sr and Eu/Eu* of these rhyolites are consistent with an origin primarily from melting of hydrous mafic wallrock minerals, such as biotite, in the presence of residual plagioclase, and may be used to identify rhyolites formed as pure crustal melts elsewhere. The geochemical traits of the Loicas Trough volcanism can be identified in other volcanic centres with similar tectonic settings in continental arcs around the world. Thus, we propose ‘arc-adjacent’ volcanism as a newly recognised style of continental arc volcanism, defined by relatively low degrees of melting and extreme compositional variation driven by extensive upper crustal AFC, with trace element signatures intermediate between typical main arc and back arc compositions, typically found in transtensional settings adjacent to continental main arcs. The model presented here can aid in understanding the diverse ancient and modern day continental volcanism that occurs near continental margins with atypical source signatures. Future geochemical work in progress on the Loicas Trough volcanism will further evaluate the effects of open system processes, magma storage and plumbing systems through the use of radiogenic isotope proxies and in situ mineral analyses.