There is some utility in having an estimate of the concentrations of elements in the continental crust as opposed to the sub-oceanic crust, mantle and core, partly for the development of theory of crustal evolution, and as an addition to our general store of knowledge. Estimates made in the past were based on a minority of data and a good deal of intuition, it is difficult enough to arrive at the percentages of sandstones, carbonates, granite etc at the Earth's surface without considering the changes at depth.

However we do know with reasonable assurance that all the main body of oceanic crustal basalt is derived from a process of partial melting of the underlying mantle. Continental crustal rocks have been derived from oceanic lithosphere by a further process of partial melting along subduction zones, a process which has been in operation without change from the Early Archaean times. We can assume that all continental rock was derived originally from basalts created mainly at ridge spreading centres. Continental rocks of all kinds share the same signature as seen in andesites of currently active arc segments.

Weathering and chemical changes to primary rocks cannot create or destroy elements, merely shift them about. The thin layer of sediment found on the sea beds is mainly of continental origin, and may be considered as a displaced part of it. The true original unmodified composition of the continental crust therefore may be taken as being an average of all andesites and their related rocks erupted since the earliest Archaean time.

Averages of active arcs

There are about twenty active island and continental marginal arcs operating in this era and there is now available data from some 1000 to 3000 samples per arc. Unfortunately a great deal of this is partial data, to quote one example, for the Mexican Arc sector, there are 792 samples analysed for La but only 308 for Sm and 10 for Ho. The figures quoted in the attached table are the number analysed for silica, other major elements are somewhat less, and trace elements range from 1 - 80% of that figure. For a few elements, eg W, Be, Bo, As, Sn,, Bi, Se, Te etc at most a dozen or so determinations are available and for some arcs, none at all. These elements are included as a general indication only. Figures for the number of determinations for each element may be added later.

From all arcs, rocks which are there by accident have been excluded, eg the Horoman Peridotite of Hokkaido appears to be fairly unmodified mantle, even if it does occur in an arc sector. Nephelinites of Honshu have also been excluded and lherzolites from the Durango region of Mexico have been excluded from the Mexican Arc as though quite tiny in volume, they affect the average Ni, Cr etc. However most arcs include alkaline basalts or atypical highly potassic members, such as those of Muriah, Tambora and Batu Tara in Indonesia. These have been retained as they have an arc signature and are part of the sequence, though an end member. Many alkali basalts are not of the oceanic type and on variation diagrams lack the linear Al2O3 distribution seen in Madeira for example. At least some of these alkali basalts have the continental Nb-Ta depleted signatures, eg the Durango basanites of Luhr etal (1989, JGR 94) and so are accepted as an integral part of the arc. Some such basalts do not have enough data to be classified but have been left in because of their close association with the usually older andesitic rocks, eg the Mango basalts of Fiji. The main effect is seen in the Na, K, Rb, Ti, Cr, Ni and REE levels but the numbers are small.

Should ignimbrites be included? Their huge volume and the lack of fractionated members has usually been taken to mean that ignimbrites originate from the partial melting of the continental crustal keel. In deep Himalayan valleys mobilised sediments of rhyolitic composition can be seen breaking through the overlying more basic layered metasediment towards the surface. While they appear at the surface as rhyolitic ignimbrites and flows, a basic residue is left at the continental base so the average composition of the crust has not been greatly changed if at all.

The 3000 samples of the Andean Arc include a hundred or more ignimbrites. With some dubiety these have been left in as at least some rhyolites may be formed by the partial melting of subducted oceanic basalt. As a result it may be seen that the Andean average is higher in LILE than other arcs.

The Aeolian Arc in the Mediterranean Sea ranges from normal calc-alkaline rocks in the islands of Salina, Alicudi etc to highly potassic leucitic members in both Vulcano and Stromboli. This is likely to indicate that some crustal refusion has been involved and so the AeolianArc has been left out for now. The entire Roman province, though with invariably a crustal signature, is also leucitic with greatly elevated LILE, especially K, Rb, Ba, Cs, U, Th and light REE. We may give an average later but the Roman province will not be included in the crustal average at this point.

The Significance of Arc Averages

It may well be asked, "Do arithmetic means of a diverse body of rock, often showing a curved relationships between elements have any real meaning?" Global averages do however allow us to think in terms of a vastly simplified model of crustal evolution, which involved processes too complex for us to grasp.

	Variation diagram for all arc averages.
	Webpage of data for the above graph. The means of 19 active andesitic arcs, for a total of some 16,000 analysed rocks.
	Fingerprints for the same arc averages as above. Note the fan-like divergence from the more basaltic juvenile arcs to the more andesitic mature arcs.
	An EXCEL workbook (spreadsheet) of the above
	Cumulative REE frequency diagram for averaged andesitic arcs. Arcs 1,2,3 are Tonga, IBM and Scotia "primitive" arcs, showing a progression in maturity toward the Andes (No 15.) The most REE enriched are Mexico (16,) and the Indonesian Arc (17) which owe this to the presence of ultra-potassic and alkaline rocks. No 18 is the average, assumed to be average continental crust.

The more "Primitive" arcs including the IBM (Izu-Bonin, Marianas), the Tonga Kermadec Arc and the Scotia Arc, display a much wider range of K than expected though at lower average levels than in the continental marginal, more mature arcs of the Andes, Mexico, Central America and the Honshu Arc. The evolution of the crust appears to parallel the evolution of the andesites, from a basalt or basaltic-andesite average composition in the barely emergent proto arcs towards andesite in the continental arcs. The Cascades are, due to the large volume of associated basalts, not as mature as might have been expected. In all, the averages are more basic than we might have assumed and are in fact rather more basic than earlier estimates of average continental crust made 40 -60 years ago.

Average Composition of Sub-oceanic Crust

This is a difficult proposition as the majority of oceanic crustal samples collected are now aphyric glasses due to their lack of alteration. This means that while all cumulative rocks are omitted, their higher fractionates may be present, so any average will be less magnesian than the true average. If we use only non-glass whole rocks then some at least will have a changed composition due to alteration. As the range of partial melts is between 6 - 10.4% MgO we could restrict ourselves to this range, but even then a rocks of 6% MgO may be an unmodified low degree melt or may be a higher fractionate of a higher degree melt, in which case the Zr/Nb will be higher, and the La/Lu lower. We also have the problem that most oceanic sampling is done on the surface, and high magnesia cumulate intrusive gabbros are seldom included.
The best thing may be to look at some fresh glasses, but bearing in mind they will somewhat less magnesian than the true average.

	Frequency plot of whole rock MORBs from the PETROS file. Note a small number of cumulate rocks of > 10% MgO
	Frequency plot of silica for the MBF glass file of the Smithsonian Institute. No cumulates, but a few fractionated rocks of up to 75% SiO2. Main body of data shows very low spread.
	The MBF file of MORB glasses showing MgO distribution with higher fractionates of low MgO
	Average of trace elements in oceanic basalts from the East Pacific Rise. Just a single average with three lines.
	Averaged Oceanic basalt glasses for the main ridge sectors. Note that the Smithsonia Magma Batch file is not included in the overall average as much of it has been included in the PETDB files. Melson & O'Hearn began this whole project of analysing only alteration-resistent glass by EMP back in about 1975, so that while they have had many emulators, to them belongs the credit. Considering the wide range of compositional changes with both degree of melt, and (in a minority of samples) with fractionation, the homogeneity is amazing. Exclusion of crystal cumulate whole rocks means the true average oceanic crust may be 1-2% higher in Mg and correspondingly lower in FeO)).