Geochemical insight during archaeological geophysical exploration through in situ X‐ray fluorescence spectrometry

Geophysical techniques are widely applied in archaeological exploration, providing rapid and non‐invasive site appraisal. Geochemical analyses contribute significantly in archaeometry, but conventional laboratory apparatus requires that samples are removed from their in situ context. Recent advances in field‐portable apparatus facilitate in situ geochemical analysis, and this apparatus is deployed in this paper alongside conventional geophysical analysis to characterize the archaeological prospectivity of a site. The target is subsurface debris at the crash site of a World War II Mosquito aircraft.

Here, in situ XRF spectrometry is applied as part of a conventional deployment of magnetic and electromagnetic (EM) methods to characterize a potential archaeological site, specifically the crashsite of a World War II aircraft. The additional geochemical insight reduces the ambiguity in the interpretation of the geophysical data: geophysical anomalies are co-located with enriched concentrations of copper and zinc ions, associated with brass (copper-zinc) alloy in the aircraft's ammunition. The in situ data compare favourably to XRF and mass spectrometry applied under laboratory conditions, but the same survey locations show variability given the changing supply of chemical elements to the ground surface. In situ XRF spectrometry can offer a valuable complement to a campaign of exploratory field geophysics, but only under certain site conditions as considered in the discussion.
2 | X-RAY FLUORESCENCE (XRF) SPECTROSCOPY -FUNDAMENTAL THEORY XRF spectroscopy determines the elemental composition of a sample material using high-energy, short-wavelength (X-ray) radiation (note: spectroscopy and spectrometry are distinct; the former is a technique, whereas the latter is the quantitative analysis of data). When bombarded with X-ray radiation, different elements can be identified by the characteristic 'fluorescent' energy that they emit (Weltje & Tjallingil, 2008).
Although challenging to define, bespoke calibrations can be made (Quye-Sawyer et al., 2015;Scott et al., 2016) and allow the XRF data to be used as an absolute rather than relative indicator of composition (Środoń, Drits, McCarty, Hsieh, & Eberi, 2001). Laboratory XRF practice mitigates the effects of surface morphology by (destructively) grinding samples into a fine powder. Equivalent sample preparation is impractical for in situ XRF spectrometry hence field-portable XRF instruments have faced scepticism in the geochemical community (Frahm, 2013). However, recent research (e.g. Schneider et al., 2016) has reported similar accuracy and precision between field-and laboratory-based observations. The instrument deployed here is a hand-held Bruker Tracer IV-SD spectrometer (Figure 1), an energy-dispersive instrument with a rhodium target. The detection of elements lighter than calcium can be challenging since these have a low 'fluorescence yield' (i.e. their energy emissions are weak; Krause, 1979;Berlin, 2011), but this is overcome here with the use of a silicon drift detector (Speakman, Little, Creel, Miller, & Inanez, 2011). Sensitivity is further improved by including a Bruker 3 V Vacuum Pump (Figure 1) to inhibit the attenuation of fluorescent energy by air in the spectrometer's analysis chamber. The presence of water also impedes XRF analysis, since water scatters the X-ray radiation; therefore, in situ XRF surveys may always be vulnerable to the presence of groundwater (e.g. Tjallingil, Röhl, Kölling, & Bickert, 2007), especially for low-yield elements.
The sample area (spot size) of an XRF measurement is typically 1 cm in diameter. However, the depth penetration of XRF energy in soil is on the millimetre-to-centimetre scale, hence in situ XRF measures only the surface chemistry of host soil. While it may be detectable with geophysical methods, a target would therefore be invisible to XRF sampling unless the ground surface is enriched in relevant marker elements via some source-to-surface transport mechanism (e.g. ploughing, groundwater circulation; Hedges & Millard, 1995;Campana, 2009). Even then, such transport may not only be in a FIGURE 1 A Bruker Tracer IV-SD hand-held XRF spectrometer, deployed at Nuthampstead airfield (August 2015). Here, the Bruker spectrometer is held in the operator's right hand, and the 3 V Vacuum Pump in their left [Colour figure can be viewed at wileyonlinelibrary.com] vertical direction hence the strongest concentrations of ions may not be observed directly above the source. As such, in situ XRF prospection will probably always benefit from the constraint provided by conventional geophysical survey. Mosquito crashed in the grounds of Nuthampstead shortly after its take-off from RAF Hunsdon (also in Hertfordshire). Records suggest that the port engine detached from the aircraft, causing it to invert and impact the ground at a near-vertical angle. The crash caused an intense fire, and claimed the lives of the two crewmen (members of 487 Squadron Royal New Zealand Air Force). Their bodies were recovered from the site, along with some wreckage, but it is doubtful that all debris was cleared from the site and some components (including armaments and the starboard engine) may remain present today.

| FIELD SURVEY
The airfield has been extensively ploughed, but runways still remain and evidence of military infrastructure are present as cropmarks. The likely crash site has been identified by Nuthampstead Airfield Museum using contemporary photographs of the impact (e.g. Before describing these surveys in more detail, the detectability of the Mosquito aircraft is considered; first by geophysical survey, then through geochemical analysis.

| Geophysical detection of the target
The wingspan of a Mark VI Mosquito is 16.5 m, and it is 12.5 m noseto-tail. In horizontal flight, the tip of its fin and rudder is 3.8 m above the base of its belly ( Figure 4). The speed and steep angle of impact into soft clay soil suggests that any remaining components of the Mosquito could be buried several metres beneath the surface, although evidence for the potential depth is very sparsely reported.
Most surveys for aircraft wreckage can exploit the presence of aluminium and/or steel in the ground (i.e. relying on contrasts in electrical and/or magnetic properties; e.g. Osgood, 2014), but the Mosquito was one of the few World War II aircraft to be made chiefly of wood.
Aluminium is only used in the rudder and elevator and, at this site, the steel engine and armaments may not be present. Therefore, in addition to any remaining aircraft components, it was assumed that magnetic

| Geochemical detection of the target
With little precedent for similar XRF practice, it was initially unclear which elements could diagnose the crash site. While aluminium enrichment might ordinarily be consistent with buried aircraft wreckage, this is unlikely to be significant for the wooden Mosquito. Additionally, any small aluminium anomaly may be masked by the high background aluminium content in Nuthampstead's clay soil and, furthermore, attenuated by groundwater. To identify alternative geochemical targets, the XRF characteristics of surface debris from the putative crash site were considered, including: 1. brass ammunition cartridges: British cartridge brass from the World War II period, used in 0.303 ammunition, is an alloy of 70% copper and 30% zinc, occasionally containing small quantities of lead (Pb). Cartridges may also have jacket of cupronickel alloy.
None of the cartridges recovered show signs of melting (the melting point of most brass alloys exceeds 900°C), but all had exploded.
2. cannon rounds: this ammunition is made principally from steel, possibly alloyed with a nickel-chromium-molybdenum (Ni-Cr-Mo) blend. British aircraft carried several variants: armour-piercing ammunition may be tipped with a tungsten (W) carbide alloy, whereas explosive and incendiary variants haveTNT and phosphorous (P) cores, respectively.
3. burnt wood: although dominated by light elements (e.g. carbon, oxygen), traces of heavier elements, such as lead, could be present in any paint residue.
In addition to these fragments, a sample of burnt soil was tested to monitor any chemical alteration caused by the impact fire. Figure 5 shows the concentrations of elements in the debris fragments, expressed in parts per million (on a log scale due to the variability between elements). All XRF analyses use a 'trace mudrock' calibration for which the spectrometer operates at 40 kV. This manufacturer-defined setting was the most appropriate for Nuthampstead's clay rich soil, though this implies that the measured concentrations are relative rather than absolute indicators. Elements lighter than calcium and those too scarce to be detected (e.g. molybdenum, tin, antimony), are absent from this plot. Each concentration is compared to a background value (orange bars, Figure 5 The brass sample (green, Figure 5) is dominated by copper, with concentration exceeding 10 5 ppm. A high zinc fraction is also recorded (~80,000 ppm), with arsenic (As) and nickel also increased in abundance. The steel sample is iron-enriched, although with a surprisingly low concentration of~250,000 ppm. The low value could again indicate a calibration issue, or non-ideal conditions of the sample surface caused by corrosion (Dungworth, 1997;Scott et al., 2016). Lead is somewhat enriched in both metallic samples, but in very low concentrations which may approach the limit of instrumental sensitivity. The burnt wood sample is generally depleted in metallic elements although no element is obviously enriched against the background trend. The burnt soil samples show little significant alteration with respect to background.
Despite the vulnerability to calibration effects, any geochemical anomaly presented by the Mosquito would likely be in elements associated with brass, specifically copper and zinc. In addition to ammunition, the Mosquito was held together with~50,000 brass screws, therefore brass may be highly abundant in the ground. While iron could also have been an attractive target, the concentrations of copper and zinc are more significant above the background geochemistry, and its associated variability, in our observations at Nuthampstead. Soil samples were also taken from each XRF survey position for laboratory validation. Laboratory XRF analysis was conducted with the Bruker spectrometer on soil samples that were kiln-dried for several days, at 60°C, then ground with a pestle and mortar. Selected samples (17 in total) were also analysed by inductively coupled plasma mass spectrometry (ICP-MS). ICP-MS is regarded as a more precise means of quantitatively measuring elemental composition than XRF (Pye & Croft, 2007), being less vulnerable to calibration issues, but requires more extensive preparation of samples. Aliquots of 100 mg of dried-and-ground soil were dissolved in 5 ml of hot Aqua Regia (37% hydrochloric acid and 68% nitric acid, in a molar ratio of 3:1) at 140°C for one hour. A dilution series of 1:100 was made in 2% nitric acid and analysed for elemental concentrations on an Agilent ICP-MS instrument. Quartz minerals can be resistant to dissolution in Aqua Regia hence differences can exist between compositions evaluated through ICP-MS and XRF analysis of dissolved and undissolved samples. However, the samples in this experiment appeared to be completely dissolved in the Aqua Regia, therefore measurements with the two systems should be comparable. Additionally, for the elements considered in this study, comparisons were made of reported XRF versus Aqua Regia digestion ICP-MS measurements for standard soil samples: no significant differences between the two methods was observed for any element.     classified using Spearman's rank correlation coefficient (r s ). Figure 8 shows the correlation between different elements, with symbols coloured according to their distance along the transect. The frames in each plot are coloured according to the strength of correlation: green defines a strong correlation (r s > 0.65), red a moderate correlation (0.45 < r s < 0.65) and black a weak correlation (r s < 0.45) as no correlation. For clarity, only correlations between copper, zinc and lead are shown (others are included in Supporting Information Figure S1).  Concentrations determined through ICP-MS analysis (Figure 9b) are of the same order of magnitude as the equivalent XRF data, but differences in base-levels (evident for nickel, copper, zinc and arsenic) are evident. These are attributed to the inappropriate calibration of the XRF survey, implying that these in situ surveys should be considered relative rather than absolute indicators of concentration.

| Laboratory XRF and ICP-MS spectrometry
Nonetheless, anomalies in copper and zinc remain well-defined, 60 m along the transect, but trends in arsenic and lead are inconsistent. A lead anomaly is distinct in the ICP-MS record, approaching 20 ppm above background. The XRF energies for arsenic and lead are very similar: 10.543 keV for Kα for arsenic and 10.551 keV for Lα for lead. Therefore, the spectral interference between these elements makes it challenging for XRF to distinguish between arsenic and lead, particularly at low concentrations. As such, the XRF anomaly in arsenic is likely a false positive. Lead is feasibly associated with the crash, since World War II aircraft were balanced using lead weights.
It is worth noting that ICP-MS gives evidence of an aluminium anomaly. While the variability of the observed concentrations impedes its definition, aluminium concentrations appear consistently high 60-70 m along the transect, approaching 10,000 ppm (~5%) above background.  (Figures 6 and 7) are observed at the study site, which appear consistent with an aircraft crash at this location. Specifically, these are a widespread magnetic anomaly and enriched concentrations of elements associated with brass alloy.
The low-amplitude magnetic anomalies observed in both the Grad601 grid and the G-858 transect are interpreted as the response to the thermoremanence in burnt clay. Assuming a near-vertical impact, the area of this response is not inconsistent with the footprint of the Mosquito (~16 m × 4 m), which would have been affected by the impact fire. Additionally, the power spectrum of the G-858 response indicates that the magnetic source is located within 1 m of the ground surface, based on modelling the burnt layer as a thin layer with random magnetization. Spector and Grant (1970) show that for a verticallyextended random magnetic layer, the slope of linear sections of a power spectrum of log-power versus wavenumber (= 1/wavelength) is a factor of 4π times the source depth. Figure  The higher amplitude magnetic anomalies (> ±100 nT/m) observed in the Grad601 grid could be responses from larger fragments of ferrous wreckage, but a further survey would be required to evaluate the size and/or depth of these potential targets.
This interpretation is greatly strengthened by the XRF spectrometry. Co-located with the magnetic anomalies are local geochemical anomalies, particularly evident for elements (copper and zinc) FIGURE 9 Laboratory validation of in-field XRF spectrometry data. (a) laboratory analysis of handheld XRF following grinding of dried soil samples, again including a three-point median trend. b) concentrations as measured in ICP-MS analysis (including for aluminium, absent in previous XRF analysis). The dashed black line in these plots is the median average value for each element; error bars in ICP-MS analysis are smaller than the symbol [Colour figure can be viewed at wileyonlinelibrary.com] FIGURE 10 Power spectrum of magnetic field strength, recorded by the upper sensor of the G-858 gradiometer. Linear section i (fit to blue data) expresses a gradient of ˗24.7 m, corresponding to a depth of 0.8 m for the associated causative body. Linear sections ii (fit to red data) and iii (fit to grey data) are assumed, respectively, to correspond to elevation variations of the sensor and ambient noise [Colour figure can be viewed at wileyonlinelibrary.com] associated with brass. Besides iron and aluminium, brass is the most significant metallic component of the fully-armed Mosquito aircraft.
The geochemical evidence is particularly compelling since, in the absence of other information, the air-crash is the most plausible means of introducing these elements into the ground at this location; by contrast, the burnt layer alone could be more simply explained by (for example) disposal at some point in the recent history of the site. The full suite of geophysical and geochemical observations is therefore consistent with an air crash at the site identified within Nuthampstead Airfield.

| Efficacy of in situ XRF surveying
To use in situ XRF surveying as an archaeological exploration tool, some mechanism must exist to transport 'exotic' (i.e. absent in the background) geochemical elements from their buried source to the ground surface. No metallic fragments were observed in the laboratory-powdered soil samples, suggesting that elements at the site are transported in groundwater rather being present in shards of metallic debris.
At Nuthampstead, ploughing appears to be an effective transport mechanism, and the time since ploughing appears to be a key control on the clarity of the XRF anomalies. The survey in November 2014 was conducted soon after a period of ploughing, potentially supplying the ground surface with a 'fresh charge' of metal-rich groundwater.
Anomalies and their correlation coefficients were both reduced in the August 2015 dataset (e.g. Figure 8) compared to November 2014. Ordinarily, it might be expected that the drier ground conditions in summer would yield higher geochemical concentrations (e.g. Schneider et al., 2016) but, at the time of this acquisition, the ground had been undisturbed for several months. Metal ions could therefore have been flushed from the site by (for example) rainfall, or transported back into the subsurface. However, some ions must also remain adsorbed onto soil grains, otherwise, XRF analysis of dry soil (including in the laboratory analyses) would have detected no geochemical anomaly at all. Given that the sample size of the XRF instrument is~1 cm 2 , it is unlikely that analyses are conducted at precisely the same location between different time periods; however, the changes in the XRF responses are not a shift in the position of the geochemical anomalies, but in the scatter and the correlation of geochemical concentrations. Separate to instrumental effects (e.g. calibration and sensitivity), the measured concentrations are therefore a function of: a. the abundance of a given element in the source material, b. the groundwater solubility and adsorption potential of that given element, c. the efficiency of any source-to-surface transport mechanism.
Calibration issues are often unavoidable in archaeological XRF surveying (e.g. Scott et al., 2016). A non-specialist should therefore consider XRF spectrometry as a qualitative tool for 'anomaly spotting', rather than interpreting the absolute values of the recorded concentrations. Bespoke calibrations are recommended if absolute concentrations are required (for example) for comparative archaeometric purposes (Scott et al., 2016) or where forensic analysis may lead to litigation (Bergslien, 2013;Ruffell & Wiltshire, 2004;Sbarato & Sánchez, 2001). Validation with laboratory analysis is also advocated since XRF scattering effects are minimized in powdered samples; furthermore, such samples represent a homogenized volume of material, therefore the measurement is less susceptible to 'skin' anomalies.
With respect to the efficiency of acquisition, in situ XRF spectrometry compares favourably with established geophysical methods. Not only is the cost of equipment similar to many geophysical systems, the rate of data return (40 samples/hour, here distributed across a 100 m transect) is comparable to (for example) surveying with electrical resistivity tomography. While XRF spectrometry would probably be impractical as an initial reconnaissance tool, it can contribute valuable insight to the understanding of a target once that target has been identified.

| CONCLUSIONS
In situ XRF spectrometry provided a valuable geochemical complement to a suite of geophysical field acquisitions. Localized increases in the concentration of diagnostic metallic elements improved the detectability of the crash site of a World War II aircraft, adding confidence to the interpretation of a suite of geophysical data.
Specifically, increases in the local abundance of copper and zinc were identifiable as originating with brass ammunition cartridges among the aircraft wreckage. The applicability of in situ XRF at a given site requires not only that anomalous elements are present in detectable abundance, but that some source-to-surface transport mechanism (e.g. ploughing) is active. While in situ XRF responses should be validated under laboratory conditions, the portable XRF spectrometer offers a useful complement to a programme of field geophysical survey.