Regional tectonic setting

The Taranaki Volcanoes constitute the most westerly (albeit isolated) manifestation of volcanism associated with the Tonga-Kermadec subduction system in the New Zealand region. A well-defined Wadati-Benioff Zone dips westwards beneath the volcanoes (Adams & Ware, 1977; Reyners, 1983) and underlies them at a depth of 180 km; Egmont Volcano lies 180 km to the west of the Hikurangi Trench (Fig. 1). The Taranaki Volcanic lineament is located between the North and South Taranaki Basins, bounded by the NE-trending Cook-Turi Lineament to the north and the Taranaki Fault to the south (Fig. 1). The Tongaporutu High, to the east of the Taranaki Fault, is a major structural feature, which approximates the extension of the Median Tectonic Zone (MTZ) north from the Nelson region of the South Island. The MTZ marks the boundary between the Palaeozoic Takaka terrane to the west and the Mesozoic terranes to the east (Kimborough et al., 1993). Mortimer et al., (1997) concluded that this major terrane boundary occurs within the basement beneath Taranaki. Mesozoic greywackes of the Waipapa and Murihiku terranes occur to the east and are separated from Permian sequences (Brook Street terrane) and MTZ to the west, by the Taranaki fault. Egmont Volcano is located on basement dominated by granitic and dioritic rocks of the MTZ.

Geology of the Taranaki volcanic lineament

The volcanic centres that make up the Taranaki volcanic lineament (Fig. 1b) are a sequence of remnant edifices that progressively young to the south and show concomitant decreases in the degree of erosion with time. The oldest and most severely eroded centre is represented by a group of nearshore islands (the Sugar Loaf Islands) and onshore volcanic spires (including Paritutu on the outskirts of the city of New Plymouth) dated by the K/Ar method at 1·7 Ma (Neall, 1979). These are plagioclase-phyric, strongly porphyritic andesites, variously thought to be part of a ring fracture or feeders to now eroded vents (Grant-Taylor, 1964). The centre lies on the northeastern side of the lineament defined by the other three Taranaki centres and, on these grounds and limited geochemical data (see below), there must be some doubt about its affinity with the other centres.

The next youngest centre, Kaitake, which is K-Ar dated at 0·575 Ma [J. J. Stipp, personal communication, cited by Neall, (1979)], is a deeply eroded stratovolcano remnant, comprising a series of radial ridges about a central plateau at 684 m altitude. Dykes occasionally outcrop on the ridges but for the most part outcrop is poor. Lithologies are predominantly hornblende andesite with some diorite. A series of debris avalanche deposits are exposed at the north Taranaki coast (Maitahi Lahars) and these contain predominantly hornblende andesite clasts. The Maitahi lahar deposits are the only remnants of a previously extensive Kaitake ring plain.

The remnant cone of Pouakai lies 10 km SE of Kaitake and is K-Ar dated at 0·25 Ma [J. J. Stipp, personal communication, cited by Neall, (1979)]. The Pouakai Ranges rise to 1399 m and cover an area about half to two-thirds that covered by Egmont at its base. Pouakai lavas are also hornblende andesites. Much of the northern ring plain was protected by the Pouakai Ranges from inundation by Egmont lahars and up to 30 m of tephra from Pouakai is preserved across the north Taranaki landscape. To the south, the younger Egmont Volcano and its ring plain cover the Pouakai deposits.

The earliest identified activity at Egmont Volcano occurred at ~115-120 ka bp (Alloway et al., 1995) and the evidence of this activity is preserved as lahar deposits in the coastal cliffs of south Taranaki. The most recent volcanic activity was an eruption in ad 1755 (Druce, 1966; Topping, 1974). The present cone has been constructed over the last 7 kyr (Neall et al., 1986; Stewart et al., 1996).

Price et al., (1992) argued that magma generation along the Taranaki lineament is directly related to fluid loss from the slab, with fluid egress being structurally controlled by propagating tears or fractures in the subducting slab. Seismic data can be interpreted to indicate that the subducting slab beneath the North Island is segmented along a series of fractures that are orthogonal to the Hikurangi Trench (Reyners, 1993), and the Taranaki volcanic lineament may be a high-level expression of the development of one of these segmenting fractures.

Stratigraphic summary

A stratigraphic summary is provided in Table 1 and a geological map of the Egmont cone is presented in Fig. 2. Numerous episodes of cone collapse during Egmont's history have resulted in the construction of an extensive ring plain (Neall et al., 1986), which now contains the only evidence of earlier events at the Egmont centre. Most of the eruptive activity is therefore represented by clasts in debris avalanche deposits, which form the ring plain and range in age from ~120 ka to ~8 ka. Andesite clasts within these deposits have been extensively sampled and are regarded as representative of lavas that formed earlier edifices constructed during previous cone-building magmatic events. Although the ring plain deposits are the products of single events, they homogenized pre-existing lava flow sequences during cone collapse episodes and the clast samples from these deposits can only be crudely placed into a stratigraphic order. Consequently, only two very broad stratigraphic units are used in the discussion of the geochemistry of the ring plain deposits. `Old Ring Plain' deposits are those which were emplaced before 24 ka and `Young Ring Plain' deposits were emplaced between 24 and 8 ka.

Table 1. Summary of stratigraphic groupings for Egmont Volcano lavas (Stewart et al., 1996)

Figure 2. Geological map of young cone of Egmont Volcano (Mt Taranaki) showing distribution of major stratigraphic units. Geographic features mentioned in the text are also shown. `St' is the Staircase and `Sh' the Shark's Tooth.

Kahui Formation pyroclastic flows, which are radiocarbon dated at 8 ka bp (Neall, 1979), form a base tothe present cone. Their emplacement was followed by episodic construction of the present edifice by eruptions of pyroclastic material and lavas. The Kahui Formation separates lava flow sequences of the present cone from laharic deposits of the ring plain. Among the post-Kahui lava sequence, four groups of lavas of decreasing age have been recognized on the basis of field relationships and palaeomagnetic data (Downey et al., 1994; Stewart et al., 1996). In order of decreasing age of initial emplacement these stratigraphic units are: the Warwicks Castle group, the Fanthams Peak group, the Staircase sequence, and the Summit eruptives (Table 1).

Sampling

Field samples (500-1000 g) were broken from flow outcrops; outside surfaces were removed and samples immediately double-bagged in press-seal plastic bags to avoid contamination. A few samples were drilled from fresh pavements with a coring drill. The ring plain deposits were sampled by removing single large clasts (generally >500 g) and trimming away outside surfaces in the field before the samples were bagged. Overall sampling is strongly biased towards fresh flows of the young cone. Exposure in the Pouakai and Kaitake Ranges is poor and many outcrops are weathered or altered. Clast populations in laharic deposits generally represent a broad cross-section of the volcanic edifice existing at a particular time in the volcanic history and consequently one might expect considerably more variability than is observed in better constrained flow sequences of the young cone.

Major and trace elements

All samples were crushed using a WC shatter box and abundances of major and minor elements and selected trace elements (Table 2) were determined at La Trobe University by X-ray fluorescence (XRF) analysis. For samples analysed by spark source mass spectrometry (SSMS), a separate aliquot of sample was crushed in agate. Major and minor elements (Si, Ti, Al, Fe, Mn, Mg, Ca, Na, K, P, and S) were determined using methods similar to those described by Norrish & Hutton, (1969). In general, precision for each major or minor element is better than ±1% (1[sgr]) of the reported value. FeO abundances were measured by direct titration using a standardized CeSO₄ solution and H₂O and CO₂ by a gravimetric method. Selected trace elements were determined on pressed powder pellets using methods similar to those described by Norrish & Chappell, (1977). For most XRF trace elements, theoretical detection limits are of the order of 1-2 ppm and reproducibility is better than ±5% (1[sgr]).

Table 2. Whole-rock major and trace element data for Taranaki eruptives

For selected samples the rare earth elements (REE) and other trace elements were analysed by inductively coupled plasma mass spectrometry (ICPMS) at the VIEPS Trace Element Laboratory at Monash University using methods described by Price et al., (1997) or by SSMS at the Australian National University using methods described by Taylor, (1965, , 1971) and Taylor & Gorton, (1977). A few samples have been analysed by instrumental neutron activation (INAA) at the Australian National University using methods described by Chappell & Hergt, (1986). Precision for elements analysed by ICPMS, SSMS, or INAA is typically better than 5% with accuracy, based on analysis of BHVO-1, being, for most elements, better than 5% at the 95% confidence level (Table 2). Duplicate analyses of standard rocks and selected samples by ICPMS, SSMS, and INAA indicate that the three methods provide comparable data for the REE, Cs, Hf, and Th. Analyses of the same sample by the different techniques generally agree within 1-5% although for one or two elements in one or two repeated samples, differences of up to 10-12% were observed at low abundance levels. For Pb and U analyses the agreement between ICPMS and SSMS data is generally in the range 1-5%.

Lead, strontium, and neodymium isotopes

Powders were leached in hot 6 M HCl for 30 min, followed by sequential open beaker dissolution in HF-HNO₃ and HCl. Lead was purified using conventional HBr-HCl column chemistry. Unless a separate dissolution was made for Sr-Nd, the HBr eluate from the Pb column was collected, fumed with nitric acid and taken up in 1 M HCl for further processing on standard cation columns. Removal of Ba at this stage results in significant improvement in ion currents and running characteristics during Nd runs. Neodymium was further purified using reverse phase ion exchange chromatography on HDEHP-coated Kel-F columns.

Lead was loaded onto single Re filaments using silica gel-H₃PO₄. Procedural blanks vary between 30 and 50 pg and are negligible relative to the sample sizes used. Samples were run on a seven-collector Finnigan-MAT 262 in static mode at filament temperatures of 1250-1350°C, at ²⁰⁸Pb ion currents of (1-4) * 10^-11 A. Typically, three blocks of 10 * 8 s scans were collected, with in-run 2 SE of <= 0·05%. Mass fractionation is estimated to be 0·109% per mass unit, based on the SRM981 Pb standard; appropriate correction factors were applied to the data. Reproducibility for SRM981 (n = 78, 2[sgr]) is ±0·097% for ²⁰⁶Pb/²⁰⁴Pb, ±0·130% for ²⁰⁷Pb/²⁰⁴Pb, and ±0·175% for ²⁰⁸Pb/²⁰⁴Pb. These error ranges are consistent with repeat analyses for several young volcanic rocks. Mass fractionation was checked on selected samples using a ²⁰⁴Pb-²⁰⁷Pb double spike (Woodhead et al., 1995). Five runs gave an average fractionation factor of 0·091% (2[sgr] = ±0·034%), close to our empirical value of 0·109%.

Strontium samples were loaded in H₃PO₄ onto single Ta filaments. Mass fractionation was corrected by normalizing to ⁸⁶Sr/⁸⁸Sr = 0·1194. Typically, 5-7 blocks of 10 * 8 s integrations produced in-run precision (2[sgr]) of ±0·003%. ⁸⁷Sr/⁸⁶Sr (±2[sgr]) for SRM987 (n = 100) is 0·71023 ± 7 (0·01%), for BCR-1 (n = 6) 0·70500 ± 4, and for BHVO-1 (n = 19) 0·70348 ± 6.

Neodymium was loaded in H₃PO₄-doped 1 M HNO₃ onto the Ta side of a Ta-Re double filament assembly. Mass fractionation was corrected by normalizing to ¹⁴⁶Nd/¹⁴⁴Nd = 0·7219. Typically, 5-7 blocks of 10 * 8 s integrations produced in-run precisions (2[sgr]) of ±0·0025%. ¹⁴³Nd/¹⁴⁴Nd (±2[sgr]) for La Jolla (n = 100) is 0·511860 ± 16, for BCR-1 (n = 7) 0·512634 ± 18, and for BHVO-1 (n = 5) 0·512989 ± 13. Present-day CHUR was taken as 0·512631.

Petrography

Taranaki lavas range from vesicular red and black scoria through to non-vesicular to holocrystalline, porphyritic lavas. Inclusions are abundant in many andesite flows. Many of these are fragments of older andesitic eruptive material entrained in the host flow during eruption. However, inclusions of diorite and gabbro (inclusions up to 1 m in diameter), hornblendite, and Tertiary sediments are also common. Meta-amphibolite and schist xenoliths are relatively uncommon and inclusions of troctolite, dunite and pyroxenite are rare. Inclusions are particularly common and the inclusion population is most varied in andesite clasts of the laharic deposits of the ring plain. The ~23 kyr old Pungarehu Formation, which comprises extensive laharic, fluvial and debris avalanche deposits across the western ring plain of the volcano (Ui et al., 1986), is a major source for a wide variety of xenolith types.

In thin section, Egmont lavas are seen to range from holocrystalline to hypocrystalline. Some are seriate textured but most are porphyritic. Phenocryst proportions range from 25 to 55% [Table 3; see also Neall et al., (1986)]. Glomerocrysts are common and comprise clinopyroxene ± titanomagnetite ± plagioclase ± olivine with rare orthopyroxene and amphibole. Occasional very large (>10 cm) hornblende crystals are found. All lavas examined contained `phenocrysts' of (in order of abundance) plagioclase, clinopyroxene, titanomagnetite and hornblende. Olivine is present in small amounts in most lavas whereas orthopyroxene is rare, consistent with the high aH2O and low a_SiO2 inferred for these high-K rocks (Stewart et al., 1996). Groundmass assemblages include glass, plagioclase, titanomagnetite and clinopyroxene, with rare olivine, orthopyroxene, amphibole and ilmenite. Apatite and zircon are common accessories. Mineralogy and mineral chemistry are summarized in Table 3.

Table 3. Summary of mineralogy and mineral chemistry of Taranaki basaltic andesites and andesites (see Stewart et al., 1996)

MAJOR AND TRACE ELEMENT CHEMISTRY

Our database of chemical analyses from the Taranaki volcanoes comprises 141 major and trace element analyses representing all phases of eruptive activity in the region. The representation is, however, uneven. Early periods of activity are only poorly represented by samples from the Sugar Loaf centre and the Kaitake and Pouakai Ranges because outcrop in these areas is limited and alteration common. Samples of early Egmont eruptives are clasts from debris avalanche deposits, and age relationships are only generally known. In contrast, samples of the present lava cone (<8 ka bp) provide a detailed picture of younger magmatic evolution. The chemical analyses reported here are mostly from lava flow sequences and consequently pyroclastics and tephras, which generally have higher SiO₂ contents, are poorly represented. Only from the most recent phase of activity (the Summit sequence) have we analysed samples of pyroclastic flow deposits and these are the samples showing the highest SiO₂ abundances.

Analyses of representative whole-rock samples from Egmont Volcano are presented in Table 2 along with an average andesite composition for Ruapehu volcano in the TVC. For all the variation diagrams, major element data have been recalculated on an anhydrous basis and assuming an Fe₂O₃/FeO ratio of 0·2. Egmont eruptives range in composition from medium-K and high-K basalts and basaltic andesites through to high-K, high-Si andesites (Fig. 3). Using Gill's, (1981) definitions, the majority of samples are high-K, low-Si andesites, and young Taranaki eruptives are generally significantly more potassic than their equivalents in the TVC [Fig. 3; see also Price et al., (1992)]. Consequently, they tend to be relatively more alkalic; basalts and basaltic andesites tend to be weakly hypersthene or even nepheline normative (Table 2).

Figure 3. K₂O vs SiO₂ variation diagrams for Taranaki eruptives and comparisons with Ruapehu andesites and dacites. (a) Details of variation among Taranaki eruptives. (b) Comparison with Ruapehu and classification of Gill, (1981).

Major elements

Major element variation is illustrated in Fig. 4 using silica variation diagrams. TiO₂, FeO_total, MgO, and CaO abundances decrease systematically and linearly with increasing SiO₂ abundance, whereas Na₂O and K₂O contents increase (Figs 3 and 4). Al₂O₃ variation is complex; the stratigraphic groups are to some extent distinguished in the Al₂O₃ vs SiO₂ diagram. Pouakai and Kaitake data are moderately scattered, with one, probably altered, sample having a very high Al₂O₃ content (>20%) and other samples having Al₂O₃ contents similar to those observed in young Egmont cone lavas. Old Ring Plain material forms a scattered linear array within which SiO₂ and Al₂O₃ abundances are correlated. Some of the Young Ring Plain samples have compositions overlapping with data for Fanthams Peak samples, which cluster along a well-defined linear array at higher Al₂O₃ abundance for a given SiO₂ value. The lavas of the Warwicks Castle group are distributed into two distinct sub-groupings; one shows relatively high and constant Al₂O₃ abundances and the other, which includes the Turehu Hill samples, has lower Al₂O₃ and SiO₂ contents. Data for some of the Young Ring Plain eruptives overlap in the Al₂O₃ vs SiO₂ diagram with the higher Al₂O₃ Warwicks Castle group. Staircase group samples form a tight cluster of data points with relatively low Al₂O₃ abundances, and the Summit group defines a scattered linear distribution with relatively low and constant Al₂O₃ abundances.

Figure 4. Major element variation as a function of SiO₂ abundance for Taranaki eruptives.

Some of these distinctions are also evident in other variation diagrams. Fanthams Peak data form a distinctive linear array in the MgO vs SiO₂ diagram and some of the Young Ring Plain data overlap this field. In the same diagram, the two Warwicks Castle groups are separated. The Old Ring Plain data form a crude linearly distributed field at relatively higher MgO for a given SiO₂ value. The Pouakai and Kaitake data overlap with the field of Egmont cone and Young Ring Plain data.

K₂O vs SiO₂

The data presented in Fig. 3 confirm earlier observations (Dickinson & Hatherton, 1967; Neall et al., 1986; Price et al., 1992) that young Egmont lavas are distinctly more potassic than equivalent andesitic eruptives in the TVC, but our new, more comprehensive data set for Egmont allows a closer examination of this feature. In detail (Fig. 3b), K₂O abundance varies with time in Egmont Volcano and within the Taranaki volcanoes in general (Fig. 5). The highest K₂O contents are observed in the Summit group and progressively older stratigraphic groupings show a systematic shift to lower K₂O contents at a given SiO₂ value. In the K₂O vs SiO₂ diagram, Summit, Fanthams, Staircase, and Warwicks Castle groups each define separate linear arrays showing positive correlation between K₂O and SiO₂ abundances. Within the Summit sequence the most potassic and siliceous eruptives are also the youngest.

Figure 5. Temporal variations in K abundance (a) and ⁸⁷Sr/⁸⁶Sr isotopic ratios (b) for Taranaki eruptives. In (a), the K content at 55% SiO₂ is indicated for each group. Su, Summit group; St, Staircase group; Fa, Fanthams Peak; Wa, Warwicks Castle group; YP, Young Ring Plain; OP, Old Ring Plain. Po, Pouakai data; K, Kaitake data. Boxes show total ranges of age, isotopic ratio and K for each group.

Some of the Old Ring Plain samples are distinctly less potassic than other Egmont samples and have K₂O contents that are very similar to those of some Ruapehu andesites with similar SiO₂ contents. Samples from the Pouakai and Kaitake Ranges have K₂O contents scattering between the elevated levels seen in Warwicks Castle group samples and the Ruapehu medium-K values.

Trace elements

Trace element abundances for representative Taranaki samples are listed in Table 2 and variation for some elements is illustrated in Fig. 6 using SiO₂ variation diagrams. Rubidium, Ba, and Zr abundances increase systematically with increasing SiO₂ content and the variation in these elements is similar to that shown by K₂O (Fig. 3). Plots of K content vs abundance of Rb and Ba (Fig. 7) further illustrate this point, with abundances of these elements being strongly correlated with K abundance. Zr abundances are elevated in some of the Old Ring Plain samples, but for most data there is a correlation between K and Zr contents.

Figure 6. Trace element variation as a function of SiO₂ abundance for Taranaki eruptives. Ruapehu andesites and dacites (shown as diagonal crosses) are plotted for comparison in the Ni and Cr diagrams (g and h) to illustrate the relatively more fractionated nature of Taranaki andesites and mixed or hybrid character of Ruapehu lavas.

Figure 7. Trace element variation as a function of K abundance for Taranaki eruptives and comparisons with variation for Ruapehu andesites and dacites. Ruapehu data from Graham & Hackett, (1987) and R. C. Price (unpublished data, 1997).

Strontium abundances show much less coherent variation than is observed for K, Ba, Rb, and Zr. The Sr vs SiO₂ variation diagram shows some of the distinctions between different groups that have been recognized in the Al₂O₃-SiO₂ variation diagram. Fanthams Peak samples form a coherent group with Sr concentration showing a positive linear correlation with SiO₂ content, in contrast to the Summit group, within which there appears to be a slight decrease in Sr abundance with increasing SiO₂ content. The Warwicks Castle group is distributed into two overlapping sub-groups (high and low Sr) and, within each, Sr and SiO₂ abundances are not correlated. Samples from the Old Ring Plain sequence show a spread in Sr abundances to lower values and this is also the case for samples from the Pouakai and Kaitake centres.

Abundance variation for Ni, Cr, V, and Sc is similar to that shown by MgO and FeO; abundances decrease systematically as SiO₂ contents rise (Table 2). This is illustrated in Fig. 6 using V and Sc as examples of this type of behaviour. There is a suggestion that some of the structure identified in the Al₂O₃ vs SiO₂ diagram (see above) can be distinguished in the Sc vs SiO₂ diagram; the Fanthams Peak and Summit groups can be distinguished, as can two sub-groups in the Warwicks Castle group. Vanadium abundances are negatively correlated with SiO₂ contents.

Abundances of Ni and Cr are generally low (Table 2). Nickel abundance is <10 ppm for most samples from the cone sequence (Warwicks Castle to Summit groups), with only five samples having Ni contents >12 ppm. Among the samples from the Old Ring Plain sequence four have Ni contents >20 ppm, ranging up to 49 ppm. Most samples of the cone flows have Cr contents <25 ppm and Cr variation is similar to that observed for Ni.

Normalized trace element plots (Fig. 8) are characterized by features considered to be distinctive of subduction-related magmas from elsewhere (e.g. Pearce, 1982; McCulloch & Gamble, 1991). Relative to normal mid-ocean ridge basalts (N-MORB), Taranaki lavas show: enrichment in strongly incompatible large ion lithophile elements (LILE) such as Rb, Ba, and K; strong depletion in Nb relative to K and Th; and enrichment in Pb over Ce. The light rare earth elements (LREE; La and Ce) are enriched [(La/Yb)_n = 7·0-16·1] over heavy rare earth elements (HREE) and Y, which show abundances similar to those for N-MORB. One basaltic clast from the Old Ring Plain sequence (T90/42A) has a normalized pattern showing a significantly more subdued arc signature.

Figure 8. MORB-normalized plots of trace element data for Taranaki eruptives. Data for (a) representative samples from the young cone of Egmont Volcano, (b) the ring plain, and (c) the older Taranaki centres. (d) An average abundance pattern for Ruapehu andesites. Normalizing values are from Sun & McDonough, (1989).

Comparisons with Ruapehu (TVC andesite) trace element behaviour

Data for andesites from Ruapehu Volcano in the TVC are compared with Taranaki data in Figs 6, 7, 8 and 9. As is the case for Egmont data, Ruapehu analyses define strong positive correlations between K content and Ba, Rb and Zr abundances (Fig. 7), but the two volcanoes show distinctly different trends in the diagrams; Ruapehu eruptives have mean K/Ba = 33·2, K/Rb = 286, and K/Zr = 110·6 whereas Egmont lavas (excluding Kaitake and Pouakai data) have mean ratios of 21·4, 347, and 162·9, respectively. A few Pouakai and Kaitake samples tend to have Ba contents above those observed in other Taranaki eruptives and this may indicate that the samples in question are altered.

Figure 9. Ruapehu andesite-normalized (see Table 2 for normalizing data) plots of trace element data for Taranaki eruptives.

In the Sr vs K diagram (Fig. 7), data for the two volcanoes define two very different fields. Egmont eruptives have higher Sr contents with Old Ring Plain data scattering down to lower Sr values approaching those of Ruapehu lavas. Ruapehu eruptives have relatively constant Sr contents across a range in K abundance. The Pouakai and Kaitake centres show Sr contents similar to the range observed in the Old Ring Plain eruptives and distinctly higher than those observed at Ruapehu.

The comparison between Taranaki and Ruapehu trace element patterns has been examined further using normalized trace element plots in which an average Ruapehu andesite has been used as the normalizing composition (Fig. 9). Ruapehu eruptives show the same distinctive arc signature as Egmont eruptives, but when Taranaki data are normalized to a Ruapehu average, distinct differences emerge. Younger Egmont eruptives (from the young cone and Young Ring Plain) have Rb, Th, U, Nb, Zr, Ti, and Y abundances that are similar to those of the average Ruapehu andesite, but Ba, K, Sr, and P abundances are distinctly higher in the Egmont eruptives. The LREE show higher abundances in the younger Egmont eruptives but the middle rare earth element (MREE) and HREE abundances are comparable. Younger Egmont eruptives show significantly higher K/Nb (or La/Nb) and K/Ta ratios, and generally have higher Nb/Ta and Ce/Pb ratios.

Among the Old Ring Plain samples the trace element patterns are variable. A basaltic clast from the Inaha lahars (T90/42A) shows a pattern that is distinctly different from others. Relative to the Ruapehu reference average, it is depleted in Rb, Th, U and Pb, but enriched in Nb and Ti. Like other Egmont samples it is enriched in Sr and P relative to the average Ruapehu andesite. The other Old Ring Plain samples show patterns similar to those of other Egmont samples.

Samples from the Kaitake and Pouakai centres have normalized trace element characteristics similar to those observed in other Taranaki eruptives (Fig. 9f), but the pattern for a single sample from Paritutu is different, with abundances of LILE and high field strength elements (HFSE) being significantly higher than those observed in other Taranaki samples. Relative to the Ruapehu reference, the Paritutu sample does, however, show the same relative depletions (Nb and Pb) and enrichments (Sr and P) as are observed in other Taranaki samples. The higher LILE abundances in the Paritutu sample relative to other Taranaki samples could indicate that the former is more fractionated; this is consistent with the fact that this sample has the highest SiO₂ content (60·34%) of any Taranaki sample.

Oxygen isotope compositions

A detailed oxygen isotope study of Egmont carried out by R. B. Stewart et al. (unpublished data, 1997) showed that [delta]¹⁸O values range from 4·8 to 6·3%° for whole rocks (Table 4). This is within the range exhibited by MORB and island arcs rather than extending to higher values found in continental arcs where interaction has occurred between ascending magmas and continental crust. The lowest [delta]¹⁸O value observed (+4·8 in sample T90/10; see Table 4) in the suite of analysed samples can be taken to indicate interaction of the magma represented by this sample with a strongly depleted fluid, perhaps as a consequence of hydrothermal circulation in the middle to upper crust (e.g. Muehlenbachs, 1986).

Oxygen isotope disequilibrium clearly existed between some minerals and this is attributed to: (1) interactions with fluids in the mantle in the case of olivine; (2) re-equilibration (for titanomagnetite); and (3) high-level interactions between magma and water which modified groundmass [delta]¹⁸O composition (R. B. Stewart et al., unpublished data, 1997).

Mineral pair [delta]¹⁸O geothermometry indicates crystallization temperatures of between 1000°C and 1200°C, compared with 860-965°C derived from oxide geothermometry on groundmass magnetite-ilmenite pairs (Stewart et al., 1996). This is consistent with the crystallization and cooling sequence inferred for the magma, with the phases giving higher temperatures crystallizing at depth and the oxide temperatures representing closure in the oxides during or immediately after lava extrusion.

Petrological variation in eruptive sequences of Egmont Volcano

The young Egmont cone consists of clearly defined sequences of lava flows, which we have grouped stratigraphically and geochemically into packages (Warwicks, Fanthams, Staircase and Summit). Available data providing geochronological control on the probable ages of each group suggest that magmas have been emplaced during eruptive cycles with a duration of the order of 1-2 kyr (Downey et al., 1994).

Contrasting geochemical variation between the different stratigraphic groups is exemplified by the K₂O vs SiO₂ plot (Fig. 3). Within each group SiO₂ content and incompatible element abundances increase and compatible element abundances decrease. Each group occupies a distinct compositional space. A general observation is that within groups SiO₂ abundance increases and between groups K₂O increases.

Modelling of the geochemical variation in the Egmont eruptives is problematic because of the complexity of the processes involved and the nature of the material sampled. Phenocrysts record very complex histories suggesting either that rocks could represent mixed magmas (e.g. Eichelberger, 1975; Tsuchiyama, 1985) or that phenocryst fractionation could have been counteracted in some cases by resorption of phenocrysts during decompression as the magmas ascended (e.g. Nelson & Montana, 1992). Most of the samples from Egmont Volcano are from porphyritic flows in which phenocryst abundances vary between 30 and 60 vol. %, and consequently samples do not necessarily represent melts on simple liquid lines of descent. The design of quantitative treatments that adjust lava compositions for phenocryst accumulation is virtually impossible because of all the variables involved and the difficulty of constraining these. For example, much of the olivine is xenocrystal (Stewart et al., 1996) and at least some of the phenocrysts of other phases could be as well. Modal abundance of plagioclase does not correlate with Al₂O₃ abundance nor is modal clinopyroxene content correlated with Sc concentration, and it is therefore not possible to identify samples in which phenocryst accumulation has taken place.

For all these reasons, quantitative modelling can provide only a very broad approximation to the processes actually involved, even for individual stratigraphic flow groups. Consequently, the following discussion aims to use data and quantitative models in a more general way to evaluate the various processes that may have influenced the evolution of Egmont magmas.

Crystal fractionation in Egmont lavas

Within the whole sample suite and within individual stratigraphic groups, trends in major element variation are consistent with a broad control by crystal fractionation processes (e.g. Bowen, 1928; Gill, 1981; Gamble et al., 1990; Graham et al., 1995); Mg, Fe, and Ca abundances decrease systematically as Si content increases, and K concentration increases. The common occurrence of cumulate-textured mafic and ultramafic inclusions in the lavas is also evidence that crystal fractionation has been an important process controlling major element variation. Inclusions of this type range in composition from dunite, through wehrlite, to gabbro and diorite; cumulate mineralogy is dominated by clinopyroxene, olivine, plagioclase, orthopyroxene, titanomagnetite, and amphibole (Stewart et al., 1996). All these minerals also occur as phenocrysts, xenocrysts, or in glomerocrysts in the lavas.

The problems involved in quantitative modellingof major and trace element data for Egmont andesitescan be illustrated with an example. Table 5 showsa least-squares mixing model (Wright & Doherty,1970) for the major element chemistry of one of theevolved summit series lavas (T89/16 from thesummit dome); one of the less evolved summit serieslavas (T89/10) has been assumed to be parental.Models involving either plagioclase-clinopyroxene- titanomagnetite-amphibole or plagioclase-clinopyroxene-titanomagnetite-orthopyroxene give reasonable fits for the major element chemistry for most components, although there is a high residual for K₂O. If, however, the least-squares solution for the major elements is used as the basis for trace element modelling, the fit between model and actual abundances is not particularly good (Table 5). The explanation for the poor fit between major and trace element modelling is in part explained by a closer consideration of plagioclase fractionation.

Table 5. Major and trace element models for Summit eruptives of Egmont

The major element model calculates a bulk composition for the model parent composition by combining the supposed daughter composition with the compositions of the fractionating phases. The outcome does not depend upon the path by which the daughter evolved; all that matters is the bulk extract. Let us suppose, however, that plagioclase has been added back into fractionated melts; this could be related to sidewall accumulation in magma conduits or a magma chamber, or simply arise through inefficient segregation (flotation) of plagioclase within derivative magmas. The major element modelling will still provide a reasonable approximation to the bulk process, but the direct translation of the major element least-squares model to a trace element fractionation model is not necessarily appropriate.

A check on whether or not plagioclase has been removed by fractionation or added back into fractionated magmas can be provided by plotting SiO₂ (as a fractionation index) against Eu anomaly expressed as Eu/Eu*, where Eu is the measured chondrite-normalized Eu abundance and Eu* is the normalized abundance obtained by extrapolation between normalized Sm and Gd abundances. Because plagioclase is a dominant component in the fractionation process, it is expected that, provided f O₂ remains reasonably constant (Drake & Weill, 1975), simple fractionation will result in a negative correlation between Eu/Eu* and SiO₂; greater fractionation causes more plagioclase to be removed and a negative Eu anomaly should become more pronounced. When data for Summit series lavas are plotted in an Eu/Eu* vs SiO₂ diagram (Fig. 12), they define a crude positive trend indicating that something more complex than simple crystal fractionation has taken place. At least some Summit series eruptives represent zones of plagioclase accumulation in crustal magma chambers and this could explain the poor fit between major and trace element models for these eruptives.

Figure 12. Europium anomaly expressed as Eu/Eu* vs SiO₂ abundance for Summit (triangles), Staircase (diamonds), and Fanthams Peak (crosses) groups. Eu is the measured Eu abundance normalized to chondritic abundance and Eu* is the normalized abundance obtained by linear extrapolation between Sm and Gd.

Although the example illustrates the difficulty of quantitatively modelling the complex processes by which andesite magmas evolve, it does provide a crude demonstration that crystal fractionation processes were important in controlling geochemical variation within Egmont magma batches.

Contrasts between Egmont and Ruapehu volcanoes

Egmont and Ruapehu volcanoes share a convergent margin setting and their products originate in the same subduction system fluxing the same mantle wedge. Differences between them are therefore most likely to be due to: (1) progressive, cross-arc changes in the nature of the mantle wedge and/or the geochemistry of slab input; (2) variation in the thickness and/or composition of the mantle wedge involved in each case; (3) contrasts in the thickness and composition of the crust involved beneath each volcano; and/or (4) different degrees of partial melting taking place in the mantle wedge to produce different primary magmas.

Isotopic comparisons

Lead isotopic data (Fig. 10) are not consistent with extensive involvement of old crust in the genesis of Taranaki magmas, and differences in crustal composition are unlikely to provide an explanation for the contrasts between Ruapehu and Taranaki eruptives. Data for the two volcanoes have a common trend in the Pb isotope diagrams, with Taranaki Pb isotopic ratios extending to less radiogenic compositions. Basement beneath Taranaki is probably Waipapa terrane and intrusive rocks of the MTZ, and these have Pb isotopic compositions that overlap with those of Torlesse terrane greywackes presumed to form the basement beneath Ruapehu (Fig. 10). The Pb isotope data are consistent with the conclusion that crustal contamination has played a significant role in the evolution of Ruapehu lavas, but has been of lesser importance at Taranaki.

Graham & Hackett, (1987) pointed out that ⁸⁷Sr/⁸⁶Sr isotopic ratios in Ruapehu eruptives show a broad positive correlation with SiO₂ abundance; a feature they attributed to assimilation of a crustal component during crystal fractionation. The Taranaki data show a very limited range in ⁸⁷Sr/⁸⁶Sr isotopic ratios and there is no apparent correlation between SiO₂ abundance and isotopic composition (Fig. 13), consistent with the conclusion that assimilation of older crust has not played a major role in the generation of Taranaki magmas. Oxygen isotope data (R. B. Stewart et al., unpublished data, 1997) for Egmont Volcano fall within the range of data for oceanic island arcs and MORB and show no evidence for extensive interaction with material rich in ¹⁸O.

Figure 13. ⁸⁷Sr/⁸⁶Sr isotopic ratio vs SiO₂ abundance for Taranaki eruptives and comparison with Ruapehu andesites and dacites. Ruapehu data from Graham & Hackett, (1987) and R. C. Price (unpublished data, 1997).

Mineralogical and petrographic considerations

Amphibole is rare in Ruapehu eruptives (Cole et al., 1986; Graham & Hackett, 1987) but is common in Taranaki lavas, indicating that magmas parental to many Egmont lavas were relatively hydrous. Stewart et al., (1996) also demonstrated that Taranaki magmas were relatively oxidized.

An apparent paradox is immediately obvious. The subducted slab beneath Egmont Volcano has presumably already lost fluid beneath the TVZ to the east and would be relatively dehydrated; intuitively one might expect magmas generated further from the trench to be less hydrous. An explanation may be that lower degrees of melting are involved in the mantle wedge beneath Taranaki, so that, even though a lower fluid flux is involved, the water contents of parental magmas are higher. Such an interpretation would also be consistent with the relatively undersaturated nature of Taranaki eruptives. An additional consideration is that fluid may be lost progressively and continuously from the descending slab, with dehydration reactions involving phases such as phengitic mica generating hydrous fluid at different depths deep into the mantle and well below amphibole breakdown (Sorensen et al., 1997).

Xenoliths of crustal metamorphic rock are common in many Ruapehu lavas and they include meta-quartzites, pyroxene hornfels, pyroxene granulites and schists (Graham & Hackett, 1987). Such xenoliths are relatively less common in Taranaki eruptives, and cumulate-textured diorite, gabbro and hornblendite inclusions are much more common and reach large sizes. The xenolith population at Taranaki provides evidence for an extensive suite of cumulate rocks at depth, confirms the importance of amphibole in the evolution of Taranaki magmas, and supports the conclusion drawn from the isotopic information that, in contrast to Ruapehu, older crust has not been extensively assimilated.

Trace and minor element considerations

It has been recognized for some time (Dickinson & Hatherton, 1967; Neall et al., 1986) that young Egmont lavas are more potassic than Ruapehu lavas. The higher K is associated with higher abundances of other LILE (Ba and LREE) but the ratios of these elements to K (or to each other) are not the same in the two volcanoes. Early hypotheses proposed to explain potassium variation across subduction-related magmatic systems included pressure-dependent variation in partition coefficients for K (Dickinson, 1968), pressure dependence of K concentrations in eclogitic (slab) melts (Marsh & Carmichael, 1974) or variation in degree of slab melting with depth (Jakes & White, 1972), and enrichment of K in fluids or melts by wall rock reaction in the mantle wedge (Best, 1975). Recent models have concentrated on the influence of crustal thickness (e.g. Meen, 1987), the thickness of the melting mantle column in the wedge (Plank & Langmuir, 1988), and variation in degree of melting within the mantle wedge (e.g. Stern et al., 1993), although some models for generating high-K magmas in subduction-related arcs involve a complex interplay between crust and magmas derived from lithospheric and asthenospheric mantle sources (Edwards et al., 1991). The generally higher abundances of LILE and HFSE in Taranaki lavas may, in part, reflect greater degrees of crystal fractionation, but they may also indicate intrinsically higher abundances in parental magmas, which would be consistent with the proposal that lower degrees of melting were involved, at source, in the mantle wedge. The isotopic data appear to preclude more complex models involving crust or magmas derived from lithospheric mantle.

Although both Taranaki and Ruapehu andesites have normalized trace element patterns characterized by the `arc' signature (e.g. low Ce/Pb ratios and high Ba/Nb ratios relative to N-MORB), there are contrasts in the extent to which the arc signature is developed. Ba/Nb ratios in Taranaki eruptives are generally higher than in Ruapehu equivalents although some Older Ring Plain samples and one Pouakai sample have Ba/Nb ratios similar to or lower than those observed in Ruapehu eruptives (Fig. 14). This would suggest that slab fluid influence is higher in later Taranaki eruptives than at Ruapehu.

Figure 14. Trace element ratios plotted against Zr abundance for Taranaki eruptives and comparisons with Ruapehu andesites and dacites. MORB composition is from Sun & McDonough, (1989). Ruapehu data from Graham & Hackett, (1987) and R. C. Price (unpublished data, 1997).

Ce/Pb ratios are commonly interpreted to indicate sediment influence in the slab fluid component. A comparison of Taranaki and Ruapehu Ce/Pb ratios shows a clear contrast in behaviour with the Ba/Nb variation (Fig. 14). Ce/Pb ratios are similar for most Taranaki and Ruapehu samples but are elevated in some Fanthams Peak, Warwicks Castle, and ring plain samples. If the Ba/Nb ratios can be used to suggest a more significant slab fluid influence in Taranaki magmas, then the Ce/Pb data could be taken to mean that the fluids involved in magma genesis beneath Taranaki are compositionally different from those entering the wedge beneath Ruapehu. In particular, the influence of sediment on slab-derived fluids declines at deeper levels within the subduction system.

Woodhead et al., (1993) argued that Ti/Zr ratios of arc lavas can be used to draw inferences about the degree of depletion in mantle sources from which magmas parental to the lavas were derived. Taranaki lavas are clearly fractionated and the Ti/Zr ratios have been influenced by processes as well as source composition, but a comparison of trends for Ruapehu and Taranaki volcanoes provides an indication of the nature of their respective sources. In a Ti/Zr ratio vs Zr abundance plot (Fig. 14), Ruapehu and Taranaki samples define two distinctly separate but converging trends. Least evolved Taranaki samples have significantly higher Ti/Zr ratios at low Zr abundance and this might imply that magmas parental to the Taranaki suite eruptives have derived from a mantle wedge source that is distinctly more depleted than is the case beneath Ruapehu. The convergence of the trends is probably related to more significant magnetite fractionation influencing evolved compositions at Taranaki. K/Rb and Cs/Rb (Fig. 14) ratios are relatively elevated in most Taranaki eruptives, which is also consistent with derivation of parental magmas from a more depleted mantle source. There may be other factors affecting K/Rb ratios, and one possibility, interaction between parental magmas and the lower crust, is discussed in a later section.

The Nb/Ta ratios for Taranaki eruptives and particularly Ruapehu lavas are low relative to MORB and, because samples analysed by ICPMS methods were crushed in WC, the possibility that they have been contaminated with Ta, with consequent artificially lowered Nb/Ta ratios, cannot be precluded. Comparisons of Nb/Ta data with those for other subduction-related volcanics are limited, because not many analytical programs routinely analyse for Ta. Feeley & Davidson, (1994) reported Nb/Ta ratios ranging from 6 to 16 for Volcán Ollagüe calc-alkaline andesites (average of 11·4), but in their study, the Nb abundances were determined by relatively imprecise X-ray fluorescence analysis and the authors did not report the methods used for crushing samples. Peate et al., (1997) presented ICPMS data for Vanuatu arc lavas and volcanogenic sediments that were apparently crushed in agate, and they obtained Nb/Ta ratios ranging from 9·8 to 23·4 with an average value of 16·7. These values overlap with the range we have obtained for Taranaki samples.

Nb/Ta ratios reported in other work on New Zealand-Kermadec subduction-related volcanics are similar to those we have obtained. For example, Gamble et al., (1993) obtained Nb/Ta ratios ranging from 2·8 to 16 (average 11·6) for Taupo Volcanic Zone and Havre Trough basalts, and Briggs & McDonough, (1990) obtained ratios of 6·7-20 (average 13·4) for samples from the Alexandra Volcanics. Unfortunately, in both these studies, all samples were crushed in WC and Nb analyses were carried out by X-ray fluorescence spectrometry.

Regardless of the possible effects of contamination on the absolute values for Nb/Ta ratios, it should be valid to make a relative comparison of Ruapehu and Taranaki data presented here; samples from both volcanoes cover a similar range in major element composition and all have been prepared for analysis in an identical manner. On this basis, it could be concluded that Ruapehu eruptives show relatively lower Nb/Ta ratios than their Taranaki counterparts.

Clearly, Egmont lavas lack the characteristics of primitive arc magmas (e.g. Tatsumi & Eggins, 1995), with even the lowest Si compositions being evolved. Among the few samples showing more elevated Mg, Ni, and Cr there is evidence for contamination by xenocrystic olivine (Stewart et al., 1996). In contrast, although low-Si (basaltic) compositions are not common at Ruapehu, there is a higher proportion of comparatively magnesian basaltic andesites and low-Si andesites, and Ruapehu andesites show a spread in composition to higher Ni and Cr abundances (Fig. 6 and Table 2). In general terms, Ruapehu andesites are less evolved than those at Taranaki and this may reflect differences in crustal structure beneath the two volcanoes. The crust beneath Egmont Volcano is of the order of 25 km thick whereas the crustal thickness beneath the TVZ is estimated to be 15 km (Stern & Davey, 1987). Heat flow is much higher in the TVZ (Hochstein et al., 1993), which is considered to be a zone of pervasive extension and normal faulting (e.g. Stern, 1987; Cole, 1990). Consequently, magmas ascending beneath Taranaki are more likely to become trapped and fractionated beneath the crust than magmas generated beneath the TVZ.

Strontium is enriched in Taranaki relative to Ruapehu samples. The abundance of Sr remains relatively constant with increasing K and Si in the Ruapehu suite whereas Sr contents are elevated in older Taranaki eruptives. They become progressively higher in the cone sequences of Egmont Volcano, reaching maximum values in some of the Summit flows before decreasing slightly in the most evolved Summit group eruptives. One of the effects of higher water contents in andesitic magmas is contraction of the liquidus field of plagioclase (e.g. Gaetani et al., 1993). Later crystallization of plagioclase in some Egmont andesites may partially account for the difference in Sr behaviour between Ruapehu and Egmont suites.

Trends in K₂O abundances with time at Taranaki

The early magmatic history of Egmont Volcano, pre-dating the construction of the present-day cone, is recorded in the andesite clast assemblages of laharic deposits within the ring plain of the volcano. Data from this succession show that the strongest K enrichments are developed in the youngest Egmont eruptives.

Some of the andesites from the oldest Egmont laharic deposits have distinctive geochemical features that are atypical of the younger eruptives. Sample T90/42A (Table 2) is an example of this unusual rock type. It has relatively low SiO₂ abundance (~49%), and, although still fractionated, relatively high MgO (6·75%), Ni (119) and Cr (113) abundances. Normative Hy is low (2·42%) so that the rock is relatively undersaturated. The sample still manifests the arc trace element signature but, relative to other Taranaki samples, this is somewhat more subdued, with lower La/Nb ratios and higher Ce/Pb ratios; Nb abundances are 2-3 times higher than normal (15 ppm). Isotopically this sample is distinctive. It has the lowest ⁸⁷Sr/⁸⁶Sr isotopic ratio and a relatively unradiogenic Pb isotopic composition. These clasts appear to represent relatively low-degree melts from a depleted mantle source.

Data available for Pouakai Volcano, to the northwest of Egmont along the Taranaki lineament, indicate that most of the eroded remnant of this large volcano is similar to the older (relatively lower K) eruptives of Egmont, although higher-K material similar to the younger Egmont eruptives has also been recovered from laharic deposits associated with Pouakai Volcano.

The data can be interpreted to indicate that the earliest eruptives of the Taranaki volcanoes included andesitic magmas not dissimilar to those that characterize TVZ andesitic volcanoes to the east. The potassic character is manifested most strongly in the later stages of construction of the Taranaki volcanoes.

The origin of Taranaki magmas

Magma evolution in the Taranaki volcanoes

The arguments presented earlier [see also Stewart et al., (1996)] lead to the conclusion that magmas parental to Taranaki eruptives were relatively hydrous, oxidized, high-Mg basalts generated in the mantle wedge by fluxing of fluid from the subducting slab; none of these primary magmas are represented in the sample set. Hydrous primitive magma ponded at the crust-mantle boundary, evolving to high-Al basalt by olivine-pyroxene-spinel fractionation. P-T conditions at the crust-mantle boundary were such that amphibole began to crystallize from the magmas and within anhydrous lower-crustal and mantle wall rocks, buffering melt compositions to basaltic andesite (Foden & Green, 1992) and underplating the crust with amphibole, olivine, and pyroxene cumulates. Not all high-Al basalt magma was ponded and trapped at the crust-mantle boundary. Some rose to the surface, fractionating olivine and clinopyroxene and eventually plagioclase, and evolving to fractionated basaltic andesites and low-Si andesites. Much of the earlier formed amphibole was partially or completely resorbed as these magmas ascended higher in the crust, moving, in P-T space, away from the amphibole stability field. Xenoliths and xenocrysts of olivine, pyroxene, amphibole, and plagioclase from wall rocks and earlier cumulates were entrained by the ascending magmas.

Eruptions are probably fed from small, high-level magma chambers immediately beneath the volcano. These are sites in which there is a complex interplay between magma egress and recharge, crystal fractionation, magma mingling and mixing, and plagioclase accumulation (Smith et al., 1996). Segregation of felsic magma and crystal cumulates resulted in the formation of relatively hydrous magmas, which cooled until amphibole was again stable. The most evolved samples we have analysed from Taranaki are pumices from the Burrell Ash containing pale green amphibole and abundant complexly zoned plagioclase phenocrysts. Many clasts and bombs from this unit are complexly layered, mixed pumices.

The temporal geochemical changes within Taranaki eruptives could be simply a consequence of progressive variation in the degree of partial melting in the mantle source. This could also be associated with changes in the composition of the mantle sources being progressively melted and progressive changes in the nature of slab-derived fluids. With time, declining levels of partial melting at source could lead to progressively rising K contents, a more significant fluid trace element signature, and a dilution of depleted mantle wedge signature. The temporal changes could wholly reflect changes in the nature of the primary magmas.

An alternative explanation, based on a model proposed by Foden & Green, (1992), has been offered by Stewart et al., (1996), who suggested that progressive underplating and intrusion of magmas into the lower crust could raise the geothermal gradient so that eventually P-T conditions at the crust-mantle boundary rise above the amphibole stability field. Amphibolites generated by the passage of earlier more hydrous magmas begin to melt incongruently and produce small amounts of relatively K-rich melt. The model provides a possible explanation for the increase in K with time and is also consistent with the petrography and mineral chemistry of lava samples, xenocrysts, glomerocrysts and xenoliths. The xenoliths are cumulates and fragments of amphibolitized upper mantle and lower crust incorporated as the magmas were tapped from the crust-mantle boundary. Strontium, Nd, Pb, and O isotopic data indicate that crustal contamination was limited and this is also consistent with the model; the only contamination is from small quantities of lower-crustal melt at the more advanced stage of development of the system. One of the implications of the model is that magmatism associated with subduction systems should become more potassic with time because progressive underplating of the arc thickens the crust and raises geothermal gradients. Amphibolitization of the lower crust means that with time, a thick hydrous underplate develops and accumulation of magma and heat from melt crystallization eventually raises the geothermal gradient to a point where the underplate begins to melt incongruently and contaminate magmas with potassic melt.

Some of the effects of amphibole melt contamination on trace element behaviour have been tested using simple numerical models. In the models the minor and trace element composition of an amphibolitic lower-crustal component has been assumed to be similar to that of amphibolite xenoliths contained in some of the lavas. An example of one of these compositions is shown in Table 6. This particular xenolith sample (T95/10X1), which was originally 25 cm in diameter, is composed almost exclusively (>95%) of large (>10 cm) bladed hornblende crystals with minor plagioclase and clinopyroxene (<5%). In the models it has been assumed that amphibole with a trace element composition similar to that of xenolith T95/10X1 (Table 6), has melted by the simple reaction Hornblende -> Melt + Clinopyroxene (Holloway & Burnham, 1972); the models are not affected very strongly if spinel is assumed to be an additional phase produced in the melting reaction.

Table 6. Analysis of amphibolite xenolith from Taranaki andesite

Table 7 shows calculations of bulk distribution coefficients and melt compositions for 25, 50, and 77% melting, assuming the melting stoichiometry of Holloway & Burnham, (1972); at 77% melting, amphibole is no longer a residual phase and the melts are in equilibrium with clinopyroxene. Figure 15 shows the effect on K/Rb, K/Sr, Ba/La, and Ba/K ratios and K abundance, when model melt compositions are mixed with a relatively low-K basaltic andesite from Ruapehu.

Table 7. Trace element models for melting of amphibolite

Figure 15. Variations in K/Rb, K/Sr, Ba/La, and K/Ba ratios with K abundance for Ruapehu and Taranaki eruptives and comparison with trends produced by mixing between a low-K Ruapehu basaltic andesite and model melts produced by breakdown of amphibolite. The models are explained in the text and in Table 7. Models 2 and 3 involve both residual clinopyroxene and amphibole during amphibolite melting. Model 1 represents more extreme melting where the only residual phase is clinopyroxene. Numbers alongside symbols for melting model in (a) indicate per cent mixing between the basaltic andesite and the model melt. Vectors labelled C and M indicate general trends produced by crystal fractionation, or crystal fractionation coupled with crustal assimilation (C), and mixing of basaltic andesite and melts produced by partial melting of amphibolite (M).

Clearly, the models are poorly constrained; the calculations involve very broad assumptions about the trace element composition of the amphibolite component that has melted, partition coefficients, and the nature of the melting reaction. None the less, the calculated variations illustrated in Fig. 15 can be used to argue that progressive contamination of andesitic magmas by melts derived from underplated amphibolitic crust, particularly at low degrees of partial melting, could produce geochemical variation broadly similar to the temporal geochemical variation observed at Taranaki, and this mechanism could also explain some of the geochemical differences between Ruapehu and Taranaki eruptives.

The models can be used to explain differences in K/Rb and Ba/La ratios and K abundance data between Taranaki and Ruapehu, but this explanation is less convincing for Ba/K and K/Sr data, which may indicate that other factors such as plagioclase fractionation and/or differences in source component contribution may also be influencing the overall variation.