Tephra identification without pre‐separation in ashed peat

Cryptotephras in peat and sediment samples are traditionally separated for geochemical characterization using chemical or density floatation techniques following initial tephra identification and shard counting through analysis of ashed residue via light microscopy. However, these practices can be time consuming, subject to practitioner experience and material type, with a potential for sample loss. We present an alternative approach to identify cryptotephra in peat and sedimentary samples, where ashed material is mounted directly in epoxy resin and analysed through back‐scattered electron (BSE) imaging via scanning electron microscopy–energy dispersive X‐ray spectroscopy (SEM‐EDX). Semi‐quantitative, unsupervised chemical maps of epoxy mounts are created within 120 min using ‘Feature Analysis’ on AZtec software by Oxford Instruments. These maps locate grains of higher atomic mass and categorize phases based on geochemistry. We create a tephra identification method using an ombrotrophic peat sample doped with bimodal Vedde Ash, where recovery of the known tephra proportion in wet peat is 96%. We also propose an automated alternative to optical shard counting, whereby tephra counts identified via Feature Analysis can be ratioed to total grain counts acquired through ImageJ software and extrapolated to the inorganic fraction in wet peat. We apply the method to a minerotrophic peat from Brackloon Wood, Mayo, Ireland, where the Laki ad 1783–84 cryptotephra is successfully identified.


Introduction
Tephrochronology is the practice of dating and synchronizing sedimentary sequences via layers of volcanic ash (Lowe, 2011).It is a highly useful technique in geoscience as it establishes tie points between different palaeoenvironmental records (Lowe and Alloway, 2015) and can provide a record of past volcanism (e.g.Martin-Jones et al., 2017;Sun et al., 2017).'Cryptotephra' is tephra (volcanic glass) that is invisible to the naked eye and not readily apparent in palaeoenvironmental sequences due to small shard sizes (typically 25-120 µm) and/ or low concentration (Dugmore, 1989;Lowe and Hunt, 2001).Cryptotephra is particularly useful as it can travel thousands of kilometres from source before final deposition and incorporation into sedimentary archives and ice cores (Lowe and Hunt, 2001;Lowe, 2011).Traditionally, chemical or density separation techniques are used to isolate cryptotephra from sedimentary samples.In the case of organic-rich samples, tephra detection and counting are initially carried out by light microscopy analysis on ashed material (inorganic residue after carbon combustion) using the highly effective 'quick-burning' method (e.g.Hall and Pilcher, 2002).Sediment-rich ashed samples may also undergo sieving (e.g.Pilcher and Hall, 1992) and/or density separation (e.g.Turney, 1998;Blockley et al., 2005) during initial examinations for tephra.However, high temperature mobilizes sodium (Schaeffer et al., 1978;Nielsen and Sigurdsson, 1981) and therefore tephra separation for geochemical analysis is repeated at intervals where tephra has been located and counted (e.g.Dugmore et al., 1995;Turney, 1998;Swindles et al., 2010).The advent of chemical and later density separation techniques for tephra within sedimentary records led to a proliferation of tephrochronology studies and a step change in our understanding of the nuanced timing of palaeoclimatic processes.Holocene studies of environmental change and volcanic activity commonly employ these techniques for peat and lake sediment cores (e.g.Wastegård et al., 2001;Plunkett, 2006;Watson et al., 2017).However, these practices can be laborious and may increase the possibility of sample loss and contamination, while also potentially creating a bias towards felsic tephra recovery (Pollard et al., 2003;Blockley et al., 2005;Roland et al., 2015;Cooper et al., 2019).
There is ongoing debate over which of these methods are preferable (e.g.Roland et al., 2015;Cooper et al., 2019;Monteath et al., 2019).Most commonly, density separation is used on sediment-rich sequences while ombrotrophic peat samples are often subjected to chemical separation in preparation for geochemical characterization.The process of density separation requires pre-determined knowledge of expected tephra densities.It generally employs the use of sodium polytungstate as a heavy liquid, which is expensive and requires rounds of recycling and density adjustment between uses (Blockley et al., 2005).Buoyant, vesicular tephra grains can also become trapped in the lighter organic floatation step (Monteath et al., 2019).Sodium polytungstate needs to be washed completely from the final tephra float with distilled water before mounting for optical microscopy, as even the smallest remaining volume will crystallize upon drying and could obscure cryptotephra shards, therefore adding further length to the process.Additionally, where samples have high minerogenic fractions, density separation techniques are not always reliable for removing common minerals such as quartz.These minerals may easily become entrained in the sample through transfer between test tubes and could subsequently overwhelm the slide or may have similar density to tephra shards, making optical grain counting more difficult.When smaller sample sizes are used to evade this problem, several separations from the same sample may be necessary, again increasing preparation time.
Chemical separation is also a multi-stage process, whereby sulphuric (H 2 SO 4 ) and nitric (HNO 3 ) acid are commonly used to remove organic matter (e.g.Dugmore et al., 1992;Swindles et al., 2010).Samples with high contents of biogenic silica such as phytoliths and diatoms (i.e.lake sediment) may also be treated with 5% NaOH or KOH (Rose et al., 1996).While some have shown that chemical separation practices do not significantly alter tephra geochemistry at a range of tephra compositions (e.g.Monteath et al., 2019) some have argued that alteration of cryptotephra can be significant for basaltic and andesitic shards during chemical digestion of organic matter, using both acidic and basic techniques (Blockley et al., 2005;Cooper et al., 2019).As such, density separation was originally proposed not only as a way to better separate tephra from sediment-rich samples (Turney, 1998) but also as a solution to these potential issues during chemical digestion of ombrotrophic peat (Blockley et al., 2005).However, chemical separation is generally more efficient than density flotation (Roland et al., 2015), and therefore it is the preferred method for extracting tephra from organic-rich samples for geochemical analysis.
Notably, efficient optical shard counting via light microscopy, carried out either on ashed residue (e.g.ombrotrophic peat; Hall and Pilcher, 2002) and/or sieved and densityseparated ashed residue (e.g.mineral-rich samples; Turney, 1998;Blockley et al., 2005), is subject to user bias and experience.In particular, tephra may be more difficult to distinguish and count in mineral-rich samples, especially for mafic populations less readily identifiable through a lack of distinct vesicular morphologies.
Here, we present an intuitive, alternative method to identify and count cryptotephra in ashed peat samples using backscattered electron (BSE) imaging with scanning electron microscopy-energy dispersive X-ray spectroscopy (SEM-EDX).This method is possible due to advances in AZtec SEM-EDX software by Oxford Instruments and chemical-based discrimination using 'Feature Analysis'.Alternatively, the method can also be undertaken offline, where BSE images can be annotated on ImageJ software to define different greyscale populations before spot analysis via SEM-EDX (an incident electron beam that allows compositional determination of a sample through displacement of electrons, resulting in the release of characteristic X-rays).Under BSE, greyscale reflects bulk composition (Dilks and Graham, 1985).Heavier elements are brighter, and lighter elements darker, so different phases including tephra can be identified and characterized.The BSE approach offers the advantage that tephra is readily distinguished from certain types of organic matter or biogenic silica such as phytoliths, which can otherwise appear similar via light microscopy (Blockley et al., 2005).Additionally, tephra composition and compositional range can be distinguished from other inorganic grains (mineral phases) accounting for loss of sodium, where an ashing step has been employed.Finally, rather than shard counting optically, this approach offers a more automated advantage, as shard counting via Feature Analysis is unsupervised.Tephra counts can either be reported as absolute shard counts or can be expressed as a percentage of the total grain count and extrapolated to wet/dry peat as an estimate of tephra abundance within a full 1 cm 3 sample.This approach also provides information on the size distribution of identified tephra.Application of this method on peat/sediment samples minimizes the potential bias due to alteration or nonseparation of mafic shards.In addition, this method can be easily applied to ice core samples for efficient cryptotephra detection and characterization.It may also be applicable to tephra detection in carbonate-rich sediments following carbonate removal using dilute hydrochloric acid (HCl).We demonstrate this technique on an ombrotrophic peat sample spiked with the bimodal Vedde Ash, then apply it to a series of highly minerogenic samples from Brackloon Wood, Co. Mayo, Ireland, where basaltic shards from the Laki AD 1783-84 eruption have previously been identified (Reilly and Mitchell, 2015;Rosca et al., 2019).

Samples Method development samples: Dope1, VeddeM1 and VeddeF1
A peat sample was doped with a 50:50 mix of mafic and felsic tephra from the bimodal Vedde Ash, originally collected from its type locality in Norway (Mangerud et al., 1984).We used a peat sample from Liffey Head Bog (LHB), Co. Wicklow,.This sample has a low mineral content, with an average loss-on-ignition (LOI) of 97.6% and was found to be free of tephra during previous investigations (Rosca et al., 2018).Separate mafic and felsic portions of Vedde Ash were first homogenized using a pestle and mortar and then sieved to a grain size <300 µm.Using a microbalance, ~0.01 g (0.0101 and 0.0103 g) of mafic and felsic tephra respectively was then added to 25 g (~5 cm 3 ) of wet LHB sample in a ceramic crucible.The sample was then homogenized using a plastic spatula.This sample is referred to as Dope1 hereafter.To characterize the composition and heterogeneity of the mafic and felsic portions of the Vedde Ash sample and help determine expected tephra compositions in Dope1, we weighed up to 0.002 g of each portion separately.These samples are referred to as VeddeM1 and VeddeF1 hereafter respectively.

Application samples: Brackloon Wood
To develop the BSE imaging and Feature Analysis method, a 50 cm peat core was taken from Brackloon Wood (BW), Co. Mayo, Ireland.This site was previously found to contain the basaltic Laki AD 1783-84 cryptotephra between 12 and 16 cm depth (Reilly and Mitchell, 2015;Rosca et al., 2019).It is a small hollow, woodland peat deposit and samples were collected using a Russian corer in 2019.The site is highly minerogenic with LOI values ranging from 8 to 60%.Hence, this site acts as a suitable test of our technique in its ability to detect, count and characterize tephra amongst abundant mineral grains.The core was sub-sampled into 1 cm 3 intervals from 8 to 18 cm with a ceramic knife and each sample was transferred to a ceramic crucible.Sample labels are given as BW8-9 cm, BW9-10 cm continuing to BW17-18 cm.

Sample preparation
Dope1 and all BW samples were dried in an oven at 105 °C overnight then ashed in a muffle furnace at 500 °C for 4 h to remove organic matter and calculate ash%.At each stage, from wet to dry and then ash, the samples were weighed using an OHAUS TM 4 decimal balance.Samples were then transferred to test tubes with deionized water to avoid loss of the delicate material.BW samples were first sieved through 100 µm mesh to remove coarse mineral material, then through 25 µm mesh to remove fine clay particles.Therefore, the 100-25 µm fraction was retained.Dope1, VeddeM1 and VeddeF1 were sieved only through 25 µm mesh for ease of chemical characterization by BSE imaging.Samples were then centrifuged, the water decanted and residues were left to dry in an oven at 70 °C overnight.After evaporation to dryness, Dope1 and all BW samples were soaked in acetone for ~1-2 h to remove any remaining moisture to avoid reaction with epoxy resin when mounting.Dry samples, VeddeM1 and VeddeF1 were homogenized before mounting by gently shaking the test tubes and mixing with a micro-spatula.The residues were transferred onto double-sided adhesive on a glass plate (within the region bound by an epoxy mount holder) using a micro-spatula, ensuring sufficient coverage over the entire surface area.The spatula was tapped gently against the edge of the mount holder to capture any excess material and Struers epoxy resin was then poured on top of the sample.Mounts were subsequently dried at 40 °C overnight.Coarse grinding of the epoxy mounts was carried out by hand using Struers 1200 SiC grinding paper.Mounts were checked diligently with a binocular microscope using reflected light to ensure the maximum possible exposure of particle dimensions without loss of sample.Samples were then polished using 6 and 1 µm diamond suspension for 6 and 3 min respectively.Once a sufficient polished surface was achieved, with good grain exposure and minimal relief, samples were prepared for SEM-EDX analysis.Samples should be labelled clearly and given co-ordinates (i.e.N-E-S-W) along their side using a fine blade in order to record their position when mapping.All epoxy mounts were cleaned with ethanol and nitrogen gas (N 2 ) was gently blown across the surface to remove any surface particulates.PELCO Colloidal Graphite was painted around the top with a strip down the side of each mount for improved conductivity, before being introduced into a Cressington 208 Carbon Coater to be coated with 12 nm of carbon.

SEM-EDX and AZtec Feature Analysis
Instrument and software set-up We used a Tescan S8000 FEG-SEM (field emission gunscanning electon microscope) operating under high vacuum conditions at the iCRAG Laboratory, Trinity College Dublin.The S8000 is fitted with four 170-mm 2 Ultim Max EDX detectors and an X4 Pulse Processor, running Oxford Instruments AZtec microanalysis software (versions 5.1 and later 6.0).We used a voltage of 20 kV, a beam current of 300 pA, absorption current of ~180 pA and a working distance of 15 mm.With four EDX detectors operating, the selected beam current and voltage correspond to ~65 000 counts per second.X-ray energies for each peak were calibrated to concentrations using a range of minerals from the Astimex Standards Ltd standard mount MINM 25-53.
AZtec Feature Analysis collects a mosaic of BSE images of the sample mount and constructs polygons around areas of interest for analysis via EDX.The BSE brightness and contrast was set to highlight potential silicic material within the mounts, where greyscale thresholding was applied to construct polygons around features that have a brightness within the specified threshold (samples that comprise heavier elements, i.e. elements with higher atomic numbers, are brighter under BSE; Lloyd, 1987).Some phases such as quartz or other common minerals may be excluded to decrease mapping time.A Gaussian smoothing filter was applied to the polygons to account for irregularities and vesicles or holes within tephra or mineral grains.For this method, the field of view (FoV) was set to 750 µm and each BSE image had a resolution of 1024 × 778 pixels and a pixel dwell time of 1 µs.To avoid analysis of overly small features, a minimum feature size of 200 pixels was selected, which translates to an area of ~24 µm 2 , whereby the smallest grain diameters detected in this study were 5 µm.The features highlighted by the greyscale thresholding were then analysed to determine the chemistry.The number of features that needs to be analysed differs from sample to sample.Analysis via EDX was set at a process time of 2 (dimensionless) and an acquisition time of 0.1 s per feature.This short acquisition time provided sufficient counts for semiquantitative results and allowed for unsupervised Feature Analysis scans of the whole mount in a reasonable timeframe of 60-120 min.Classifications using an array of major elements in wt% were used to separate all features into different phases, with a maximum and minimum acceptable threshold to allow for semi-quantitative analysis of features that may be slightly outside the expected compositional range.For this study, the greyscale and geochemical thresholds were set using two or three snapshots from different regions on a sample at the fixed BSE brightness and contrast settings to ensure continuity of mapping across all sample mounts.Based on the Feature Analysis map, any potential tephra shards located were followed up with fully quantitative EDX spot analysis.
A flow diagram of the Feature Analysis set-up and parameters on AZtec is displayed in Fig. 1.For spot analysis, the FoV was set to 5 µm, where the beam rasters over the entire FoV and each analysis comprised 1 million counts.At ~65 000 counts s -1 , fully quantitative spot analysis takes ~15 s and delivers results in oxide wt%.Spot analysis at low beam current minimizes sodium migration and allows high precision and accuracy.Data quality for spot analysis was tested for Al 2 O, CaO, FeO, K 2 O, MgO, MnO, Na 2 O, SiO 2 and TiO 2 using Smithsonian NMNH 111240-52 basaltic glass (Jarosewich et al., 1980;Dixon and Clague, 2001) and obsidian from the Astimex MINM25-53 standard mount.Precision, measured by relative standard deviation (RSD), was better than 2% and accuracy (%bias) was typically better than 5% for all elements >1 wt% (Supporting Information Tables S1 and S2).Chlorine, SO 2 and P 2 O 5 were also analysed on the same standards to allow identification of salt, sulphides and apatite (Tables S1 and S2).
Feature Analysis also produces morphological information per feature, which can be exported offline if required.The equivalent circular diameter (ECD) is calculated in micrometres for every feature identified and this can be used as an estimate (2D) of grain size for tephra and/or mineral phases.Such a feature may be helpful if tephra layers are located in distal sedimentary records using this method and if the user wishes to create isopach maps for tephra distribution (e.g.Cashman and Rust, 2020).This information is also useful for determining sieve size if prior sieving is not carried out.

Feature Analysis development using Dope1, VeddeM1 and VeddeF1
Features (i.e.tephra and mineral phases) were categorized based on thresholds for their expected compositional range between one and nine elements (element wt%) (Table 1).Investigation of tephra categories and thresholds within the Feature Analysis recipe were established using VeddeM1 and VeddeF1 for mafic, intermediate and felsic tephra.As seen in Table 1, geochemical thresholds for tephra compositions were wide to ensure the best coverage.Other common mineral phases in the BW peat were then established as new categories using Dope1 so that Feature Analysis did not incorrectly identify them as tephra.These included carbonates, along with some apparent biogenic silica showing a combination of high calcium and lower silica content (<15 wt%) along with quartz, mica, plagioclase and K-feldspar.Due to the visibility of some mineralized plant matter, a category was established to identify this, defined by a high Mg content as confirmed by spot analysis.Using this approach, 86% of features were categorized by the Feature Analysis software.Categories can be changed depending on sample type and upon inspection of all feature data either on AZtec or offline.
The distinction between phase types for Dope1 in terms of greyscale is clear (Fig. 2a). Figure 2b demonstrates the successful categorization of each grain based on the geochemical thresholds.Any grains that lie outside of the specified thresholds are labelled as 'non-categorized' and their raw data can be exported and examined offline.A common reason that features were not categorized is that the pixel partially overlapped blank epoxy at the edge of the feature or within vesicles, leading to a lower analytical total.In such cases, the feature could be categorized after normalization to 100%.
Reduction of non-categorized data offline first involved separation of silicate and non-silicate data.Non-silicate data where Si was 0 wt% were reduced further to remove features where all other element values were either 0 or <1 wt%.These data were contaminants such as salts from handling (Na, K, Cl) and artefacts from the sample holder (Al) or from charging due to epoxy air bubbles (Cl).Therefore, they were discarded from the overall non-categorized count.Non-silicate features that represented the same chemistry as these artefacts or contaminants but at >1 wt% were also discarded.For 'noncategorized' silicates, elemental compositions (wt%) were converted to oxide wt% and normalized to 100% to correct for partial analysis of mineral phases or tephra.For each feature located by AZtec, a mean greyscale value, feature diameter (μm) and geochemical information are supplied.After normalization, the silicate data were reduced first by  greyscale.We used the known mean greyscale values of already identified tephra to isolate potential unidentified shards.For the features that fell within the range of tephra greyscale values, geochemical thresholds were then applied with the order of importance SiO For a single sample, this process can be easily carried out within 60 min but will vary depending on grain counts.

ImageJ grain counting
We propose a supervised but automated alternative to classic manual shard counting of cryptotephra.The proportion of tephra on the sub-sample mounts can be extrapolated to the original sample, since the ashed residue represents all inorganic matter (mineral phases and tephra) in the original sample.This requires that total grain counts on the mount are first obtained.ImageJ open access software (Schneider et al., 2012) provides an efficient method to achieve this.We used ImageJ version 1.53k.While Feature Analysis can also count total grains, ImageJ provides an alternative offline tool and will detect all grains within the full greyscale spectrum of the sample without incurring long processing times.A BSE image of the mount can be exported into ImageJ (Fig. 3a).Brightness and contrast can be changed to increase grain visibility.The method for grain counting followed that of Costa and Yang (2009) for the counting of pollen.Through the 'threshold' tab, grain populations can be highlighted to create a false colour image via greyscale threshold adjustment (Fig. 3b).Following this, the scale should be set.We chose a reference grain in each sample for Dope1 and BW samples that represented the smallest size desired for counting through cross-checking with the original BSE image (e.g.~3-5 µm).Under the 'analyse' tab, the user can then 'analyse particles' and set a pixel size (pixels μm -1 ) for the smallest grains on the mount along with the circularity index (0-1), which we found optimum at 0.9 (1 being a perfect circle).These parameters should be tested several times before final grain counting.The 'analyse particles' function produces a new image with count masks of each grain (Fig. 3c) along with a text file showing total counts.To account for small variations in image contrasting or thresholding, we carried out three replicate particle analysis runs for each sample (Dope1 and BW samples), where RSD ranged from 0.36 to 8%.The counts of tephra shards calculated via Feature Analysis can then be extrapolated to the original wet (or dry) sample via Equation (1) to calculate the amount of tephra in 1 cm 3 , whereby T total is the estimated total percentage of tephra in the original 1 cm 3 sample, t mount is the number of tephra shards on the mount, g mount is the total number of grains on the mount and I total is the inorganic percentage relative to the original wet (or dry) peat sample.This extrapolation may be most useful in sediment-rich samples, whereby the mounted proportion will represent a sub-sample of the original residue, whereas for highly organic peats with smaller inorganic fractions, the entire sample could be mounted.In this study, Equation (1) also acts as a simple method to test the recovery of tephra percentage in Dope1 by Feature Analysis after peat sample processing.

T t g I
(1) VeddeF1 and Dope1 were cross-checked for chemistry offline through oxide conversion and normalization to 100%.As shown in the kernel density plot in Fig. 4, VeddeM1 has a wide compositional range of mafic tephra at 43-53 wt% SiO 2 to felsic tephra at 63-75 wt% SiO 2 , with an intermediate phase between these endmembers.Respective proportions of these tephra categories are 42%, 53% and 5%.In contrast, VeddeF1 is composed of 100% felsic tephra (Fig. 4).Total mafic, intermediate and felsic counts between these samples, identified using Feature Analysis, were 301, 36 and 782 respectively.We used these relative proportions to calculate tephra populations expected within Dope1 based on the measured mass of each Vedde Ash sample added to LHB peat and their mass difference to VeddeM1 and VeddeF1 samples.This extrapolation assumes that all tephra populations have the same shard size distributions.At a 50:50 ratio of VeddeM1 to VeddeM2 within Dope1, we expect Feature Analysis to recover compositional proportions of 26.8%, 3.2% and 69.9% for mafic, intermediate and felsic shards respectively (Table 2).The spread of bulk tephra composition within Dope1 as identified by Feature Analysis broadly encapsulates this 50:50 ratio between the two Vedde Ash samples, but with marginally lower counts of mafic and intermediate tephra and higher counts of felsic tephra (Fig. 4).Mafic, intermediate and felsic tephra counts obtained through Feature Analysis for Dope1 are 399, 54 and 1770 respectively (Table 2).Bias between the proportions of tephra types to total tephra counts in Dope1 and the expected proportions derived from VeddeM1 and VeddeF1 are −33%, −24% and 14% respectively for mafic, intermediate and felsic populations.Mafic tephra in Dope1 is more strongly skewed towards larger clast sizes (>80 µm), compared to mafic tephra in VeddeM1 (Fig. 5).Meanwhile, felsic tephra in Dope1 has a skewed distribution towards comparatively smaller shard diameters (<30 µm) than felsic tephra in VeddeM1 and VeddeF1.Thus, the spiking of peat with Vedde Ash broadly showed the expected proportions of tephra composition using Feature Analysis, even after peat sample preparation (i.e.drying, ashing and transfer).However, clast sizes probably affected population estimations based on tephra sample mass between Dope1, VeddeM1 and VeddeF1.

Feature Analysis vs spot analysis via FEG-SEM-EDX
The accuracy of Feature Analysis in quantifying mafic and felsic tephra glass geochemistry was assessed via spot analysis using FEG-SEM-EDX on a selection of five tephra shards from each endmember population identified in Dope1.Note that accuracy is expected to be poorer for Feature Analysis than for spot analysis due to the fast acquisition time of 0.1 s.The purpose is not to produce quantitative data, rather to rapidly produce a large amount of semi-quantitative data for categorization.Feature Analysis recipe against the spot analysis results on the same shards.Results were converted to oxides and normalized to 100% to account for any shard vesicularity, as the minimum pixel size of 200 may result in partial analysis of resin and tephra, leading to poor totals.Despite this, Feature Analysis totals typically lie within 5% of spot analysis results for oxides present at >5 wt% (MgO, Al 2 O 3 , SiO 2 , TiO 2 and FeO total for mafic tephra, Na 2 O, Al 2 O 3 , SiO 2 and K 2 O for felsic tephra).Concentrations <2 wt% are typically below detection (TiO 2 and MgO in the felsic shards).Bias for CaO lies within 8% and 5% for the mafic and felsic members, respectively.Felsic shards, on average, yield lower totals than mafic shards due to their greater vesicularity, whereas mafic shards are more platy/blocky (Fig. 6).This accuracy is sufficient to classify material as tephra.However, when applied to unknown peat samples, all located shards should also be subject to fully quantitative EDX spot analysis for preliminary identification of tephra and eruption source.

Ashing and geochemical alteration
Ashing is essential for this method as mounting dried peat samples is impractical due to sample volume and the need to disaggregate the material.However, previous studies have highlighted the potential for high temperature ashing (>350 °C) to alter the geochemistry of tephra shards, particularly the mobile alkalis such as sodium and potassium (e.g.Dugmore et al., 1995).Increasing temperature allows greater diffusion of sodium, as the sodium atoms separate from their bridging oxygen atoms (Schaeffer et al., 1978;Nielsen and Sigurdsson, 1981).To check for alkali loss, %bias was calculated for each major element oxide between mean measured mafic and felsic shard compositions via FEG-SEM-EDX spot analysis (five shards from each type) and published Vedde Ash shard geochemistry derived from Tephrabase (Mangerud et al., 1984;Turney et al., 1997;Wastegård et al., 1998Wastegård et al., , 2000;;Björck and Wastegård, 1999;Zillén et al., 2002;Pilcher et al., 2005;Koren et al., 2008;Matthews et al., 2011) that display similar SiO 2 values to our shards.Results are shown in Table 4 and Supporting Information Fig. S1.Average concentrations of Na 2 O and K 2 O in the Vedde Ash are 2.68 ± 0.49 and 0.81 ± 0.11 wt% (mafic) and 4.55 ± 0.35 and 3.43 ± 0.14 wt% (felsic) respectively (errors are 1σ).The mean measured Na 2 O and K 2 O values for each endmember after ashing in this study are 1.54 ± 0.77 and 0.77 ± 0.21 wt% (mafic) and 5.33 ± 0.07 and 3.64 ± 0.11 wt% (felsic) respectively.Precision is better than 2% for analysis at >1 wt% (Tables S1 and S2).K 2 O for this study is well within error of published values, considering lower precision for low abundance K 2 O in mafic shards.Comparatively, Na 2 O appears depleted within the ashed mafic shards, showing a bias of −43% relative to the published mean (Table 4) and the measured mean lies outside of the published error at 1σ.The felsic shards show a positive bias of 17% for Na 2 O, and therefore no sodium loss is indicated in the ashed felsic shards.Measured Na 2 O concentration in both endmembers shows  overlap with published values when using 2σ error (Table 4).Differences in Na 2 O are unlikely to be accounted for by the analytical method itself, because at concentrations >1 wt% RSD and bias for EDX spot analysis of unashed mafic and felsic glass in this study are within 5% (Tables S1 and S2).Therefore, the apparent Na 2 O depletions in mafic tephra (at 1σ) are probably related to the high temperature loss of sodium during ashing.Nonetheless, geochemical identifications of tephra can still be made using the determined major element chemistry.Therefore, when ashing does lead to alkali loss, it is still possible to use data from ashed tephra to identify the tephra source through construction of bivariate plots.

Dope1 Feature Analysis: bulk tephra recovery in wet peat
The mean total grain count retrieved for Dope1 via ImageJ was 5087 (Table 5, Fig. S2).Total Feature Analysis counts within each category are given in Table 2.The biogenic silica count of 43 is noteworthy as these grains, usually phytoliths, can easily be confused for tephra in traditional techniques and require either harsh chemical digestion to remove them (potentially altering shard composition) or, alternatively, several density flotations.The total mass of doped wet peat was 25.02 g.The total ashed mass of Dope1, accounting for all inorganic matter (i.e.tephra and other mineral phases), was 0.045 g.Bulk tephra mass was 0.02 g.Using the tephra and wet peat mass, the known bulk tephra percentage in wet peat for Dope1 is therefore calculated as 0.081%.The total inorganic fraction in Dope1 is 0.179% based on total ashed and wet peat mass.As shown in Table 5, Feature Analysis successfully identified 2223 tephra shards within this doped peat sample.Therefore, we use Equation (1) to calculate the extrapolated bulk tephra abundance in the peat (assuming it is unknown) by using the ratio of tephra/total grains identified via Feature Analysis and ImageJ.Following this, tephra in Dope1 is estimated as 0.078%.Bulk tephra recovery by Feature Analysis is calculated by comparison of the extrapolated value to the known percentage of bulk tephra in the wet Dope1 sample.Hence, we achieve a recovery of 96% (Table 5).This is a highly promising result for this method, with its true potential better highlighted in our application to real 'unknown' peat samples such as Brackloon Wood (see below), where tephra counts are likely to be many orders of magnitude lower and mineral grains more abundant.
'Non-categorized' Feature Analysis data Despite the excellent bulk tephra recovery achieved for Dope1 we also examined the 'non-categorized' data offline to investigate if any tephra went unidentified.The total 'non-categorized' count after initial reduction (as specified in 'Feature Analysis development using Dope1, VeddeM1 and VeddeF1' above) was 728 (silicates and non-silicates), whereas the initial total exported from Feature Analysis was 4481.Nonsilicate mineral phases identified included 14 apparent iron oxide and 28 carbonate features.Following offline reduction and normalization of chemical data from silicate 'noncategorized' features, 302 additional tephra shards were identified, of which 5% were mafic, 6% intermediate and 89% felsic.Median diameters for each phase were 18, 25 and 14 µm respectively.Felsic and intermediate shards were highly vesicular, while mafic shards were platy/blocky in structure (Fig. 6).These populations were classified as 'non-categorized' Table 4. Bimodal Vedde Ash spot analysis (wt%) comparison to published major element geochemistry (Tephrabase; Björck and Wastegård, 1999;Koren et al., 2008;Mangerud et al., 1984;Matthews et al., 2011;Pilcher et al., 2005;Turney et al., 1997;Wastegård et al., 1998;Zillén et al., 2002).due to poor analytical totals, meaning that the analysed concentrations of a given element fell below the category threshold, while low-abundance elements were not detected.Such poor totals were a result of shard size ≤25 µm, high vesicularity or a combination of both, where analysis of the tephra was compromised by partial analysis of resin due to the minimum pixel size of 200.Notably, an additional 132 felsic features identified offline were counted by AZtec as separate features of already identified tephra.This double counting was apparent in vesicular tephra where portions of larger shards (50-120 µm) were only semi-exposed above the epoxy resin (Supporting Information Fig. S3) and thus despite Gaussian smoothing, they were counted as additional features with low analytical totals.Double counted tephra will be revealed to the user when inspecting identified tephra shards for spot analysis, as AZtec provides an ID for each feature it locates.

Shard type
The 'non-categorized' silicate populations that fell outside offline tephra greyscale and geochemical thresholding amounted to 252.These phases included quartz and feldspars which were also excluded due to diameters <25 µm, poor analytical totals, or impurities from Al, Fe, Ti, K or Na.Other phases included mafic minerals not defined in the Feature Analysis recipe (amphiboles and pyroxenes) and mineralized plant matter that had outlying chemistry to the recipe thresholds on Feature Analysis.Thus, regardless of already successful tephra recovery, we recommend that all 'noncategorized' silicate data are converted to oxides, normalized to 100% offline and categorized manually to double-check for tephra or gain insight into sample composition if required.

Feature Analysis
To test whether we can detect tephra in unprocessed natural peat samples using our new approach, Feature Analysis via BSE imaging was applied to highly minerogenic peat samples from Brackloon Wood.This site has previously revealed sparse findings of basaltic tephra from the Laki AD 1783-84 eruption at a depth range of 12-16 cm (Reilly and Mitchell, 2015;Rosca, 2018;Rosca et al., 2019).Feature Analysis maps from 1 cm 3 sub-samples between 8 and 18 cm revealed a total of 11 basaltic tephra shards between 10 and 16 cm.The greatest shard count per sample was five at 13-14 cm depth.Samples had a high abundance of quartz.Therefore, to reduce mapping times, we adjusted the greyscale thresholding to exclude all populations at a mean greyscale level of quartz or below, thus only targeting phases with atomic mass greater than quartz.Figure 7 shows results from a random subsample image of the BSE mount from BW13-14 cm, in which grain counting revealed a population of 92% quartz and one tephra shard.Additional phases identified above this greyscale threshold across all samples included numerous K-feldspar and plagioclase along with mica, zircon and amphibole (Table 6).As with Dope1, the non-categorized population was subjected to geochemical thresholding offline to investigate potentially unidentified tephra and did not reveal any additional shards.Feature Analysis maps for each BW sample took an average of 105 min.As Feature Analysis could be left to run unsupervised, the 10 samples were processed in 24 h (tephra identified, counted and geochemically characterized) including the time taken for calibration, standardization and Feature Analysis set-up.

Spot analysis via FEG-SEM-EDX
A selection of BSE images of the shards identified between 10 and 16 cm are shown in Fig. 8. Morphologies appear blocky to mildly vesicular, and diameters ranged from 11 to 30 µm with  an average of 21 µm.Small shard sizes suggest some tephra may have been lost during sieving.Therefore, we recommend removing this step for future studies of highly organic peat (e.g. in ombrotrophic bogs) or sieving at much smaller mesh sizes for mineral-rich peat (e.g. at 5 µm to complement the smallest diameters detectable by Feature Analysis at 200 pixels).All located shards were analysed via EDX spot analysis and gave clear geochemical signatures representing basaltic glass.The basaltic geochemistry is compared to proximal (Thordarson et al., 1996) and distal Laki tephra from peat bogs (Reilly and Mitchell, 2015;Rosca, 2018) and ice cores (Fiacco et al., 1994;Kekonen et al., 2005) in Table 7. Compositions are also compared with different proximal Icelandic basaltic tephra populations including other modern eruptions from the Grímsvötn system (AD 1432-57) from which Laki is derived, Veiðivötn of the Bárðarbunga volcanic system (AD 1477-1717) and of Katla (AD 1416(AD -1918) ) along with the published Laki AD 1783-84 compositions (Fig. 9).Major element results for FeO total vs. SiO 2 wt% show overlap with both Grímsvötn, Laki and Veiðivötn.However, TiO 2 vs. FeO total wt%, Al 2 O 3 vs.TiO 2 wt% and FeO total vs. CaO wt% clearly show that compositions of the basaltic glass are most consistent with the Grímsvötn Icelandic volcanic system and published Laki shard compositions.
As with the mafic Vedde Ash in Dope1, Na 2 O is depleted but K 2 O is unaltered (Table 7).Mean measured Na 2 O concentration was 0.71 ± 0.45 wt%, while the mean published concentration of Na 2 O was 2.58 ± 0.27 wt%, revealing a bias of 73% below published values for these Laki shards, with measured values also well outside of published error.As previously outlined, reliable, fully quantitative EDX analysis of unashed mafic glass in this study (Table S1) suggests that low sodium concentration in these shards is a result of high temperature ashing and not the analytical method.Comparatively, all other major element oxides are within 1σ of published values (Table 7).Therefore, tephra identification is still conclusive.Hence, Feature Analysis is an effective and efficient tool to identify cryptotephra and simultaneously get accurate, quantitative major element information from shards in an unknown sediment-rich sample.In highly organic peat (e.g.ombrotrophic samples) geochemical mapping may be quicker than the Brackloon samples (105 min) as mineral phases will be less abundant.

ImageJ and shard counts
BW13-14 cm displays the highest shard count (Table 6) and is therefore considered to represent the time of deposition for Laki AD 1783-84 tephra at Brackloon Wood.This is in accordance with results from Reilly and Mitchell (2015) and suggests that our method has captured a similar relative abundance of tephra per depth at this site.All samples from Brackloon Wood have a high minerogenic content, with grain counts in the thousands, as is apparent in the raw BSE sample mount for BW13-14 cm (Fig. 10).We subjected all BW samples to ImageJ to retrieve total grain counts and estimate the percentage of tephra in the original 1 cm 3 sample relative to other inorganic components.The total average grain count   (Fiacco et al., 1994;Thordarson et al., 1996;Kekonen et al., 2005;Reilly and Mitchell, 2015;Rosca, 2018), and modern eruptions from the Grímsvötn volcanic system (blue), Katla (orange) and Veiðivötn (white with black outline) (Tephrabase, Lawson et al., 2007;Streeter, 2011;Streeter and Dugmore, 2014).[Color figure can be viewed at wileyonlinelibrary.com] for BW13-14 cm was 44 420 (Fig. 10).The estimated tephra abundance in the original 1 cm 3 wet sample at 13-14 cm using Equation ( 1) is 0.005% (Table 6).Tephra percentage for BW10-11 cm, BW11-12 cm, BW12-13 cm, BW14-15 cm and BW15-16 cm was 0.003, 0.001, 0.002, 0.001 and 0.002% respectively.Such low tephra abundance is achieved here due to the exceptionally high quartz count in these samples and comparatively sparse cryptotephra shards.Notably, this technique in mineral-rich samples may be highly useful when the user wishes to acquire the amount of minerals vs. tephra, a process that is normally achieved by counting a minimum of 300 grains via light microscopy (van Harten, 1965;Swindles et al., 2010).Therefore, with Brackloon being highly mineralrich, optical shard counting would prove far less efficient.However, in terms of simply stating tephra abundance in samples with low tephra concentration such as Brackloon, it may be more practical to quote absolute shard counts per cm 3 , whereas samples with higher cryptotephra counts could be better represented by tephra percentages.

Discussion
This BSE imaging and geochemical mapping approach has proven successful in identifying cryptotephra in peat samples.However, while there are clear benefits to using this technique, it remains to be wholly comparable with conventional approaches and it does include a number of limitations.
The principal limitations of this approach relate to the time and cost of SEM analysis.The SEM-EDX sample holder used in this study can hold four epoxy mounts.Therefore, for the maximum mapping time of 120 min used in this study, eight samples (two batches of four) can be mapped in ~16 h (Table 8).As previously stated, Feature Analysis maps can be left unsupervised once the instrument has been set up as per Fig. 1.Hence, a minimum of four samples can be mapped during a working day, while the user may undertake other practical tasks (e.g.drying, ashing, sample mounting, polishing).Once samples are mapped, the user can also export raw BSE images and data from each map for cross-checking categories offline (e.g.identifying any potential tephra in the 'non-categorized' data).Therefore, after each map, or once a second batch has been set up to run overnight, the user can already begin to carry out offline reduction and/or calculate tephra abundance relative to total grain counts via ImageJ.As such, the main benefit of the Feature Analysis technique lies in the simultaneous nature of data collection and, further, the unsupervised collection of these data.On average, offline reduction (supervised) took ~1 h per sample, while ImageJ particle analysis (supervised and automated) can be conducted within 10 min per sample (Table 8).Thus, between Feature Analysis mapping (detecting, counting and compositionally characterizing tephra) and offline practices, eight samples can be processed within 2 days (Table 8).The user can therefore allocate a third day to carry out spot analysis via EDX on the located shards within each mount.As EDX spectra acquisition Table 8.An outline of the processing steps required for the BSE imaging and AZtec Feature Analysis method and the approximate time taken for each step.Time in hours is based on the processing of eight samples to complement the maximum amount of samples that can be analysed by Feature Analysis in 1 day at a mapping time of 120 min.
Step takes ~15 s, spot analysis can be carried out within 30-40 min for 100 shards, providing the positioning of the mounts are similar to that in which they were mapped, which is easy to achieve once mounts are labelled clearly.Therefore, tephra identification, counting and geochemical characterization for eight samples can be completed within 3 days (Table 8).
The traditional 'quick-burning' method employing ashing and a 10% HCl digestion for ombrotrophic peat (e.g.Pilcher and Hall, 1992;Hall and Pilcher, 2002) is straightforward and may take less than 5 min per sample to initially identify potential tephra presence using light microscopy (i.e. 1 day allocated for ashing, mounting and identifying tephra).After ashing again at finer resolution (e.g. 1 cm), shards are counted (shards cm -3 ) to pinpoint a tephra layer (Swindles et al., 2010).However, the time taken to count the number of shards (and sometimes mineral grains) is subjective to user experience and material type.For example, using the metric of eight samples, this may take up to 8 h (one hour per sample) if tephra is not readily identifiable.However, an experienced user may be able to make identifications and count tephra in a single sample within 10 min, especially for ombrotrophic peat.To evade issues of misidentification of tephra in a sediment sample or mineral-rich peat, sieving (Pilcher and Hall, 1992) and/or density separation (Blockley et al., 2005) is often used at this stage to remove mineral grains first before counting, adding several hours to the preparation time (subject to experience) before samples can be mounted on glass slides and examined for tephra.The preparation of fresh samples for the geochemical analysis of identified tephra shards using either chemical digestion or density separation may also incur an additional half day or full day.However, chemical techniques, especially for ombrotrophic peat samples, are generally much more efficient than density separation (Roland et al., 2015).
Based on the analysis of the minerogenic Brackloon Wood intervals in this study, 10 samples were processed from wet peat to mounted ashed samples with identified, counted and geochemically characterised tephra within 7 days (peat drying and ashing steps accounting for 2 days; Table 8).Notably, with the efficient and accurate counting of tephra at fine resolution by light microscopy being subjective to user experience, our technique offers both a more quantitative and hands-off approach.Additionally, since geochemical characterization of cryptotephra is dependent on an additional phase of sample preparation, our method performs what are normally separate tasks at the same time, while also using less sample from the original core.Furthermore, as basaltic tephra is often less obvious to a practitioner using light microscopy (e.g.lower vesicularity), making identifications of this tephra type will be easier using Feature Analysis.This may be especially the case in samples with large numbers of mineral grains, as mafic tephra identification may be easily obscured to the human eye.Therefore, our method has similar or indeed inferior efficiency to the traditional practices used on ombrotrophic peat samples (i.e. from wet peat to identification, counting and geochemical characterization of tephra), with the advantage that automation means that time is freed up for other tasks.However, time may be saved for more mineral-rich sequences, although these must be well homogenized to ensure the mineral to tephra shard ratio within the prepared sub-sample for Feature Analysis is relatively reflective of the original ashed residue.Notably, ashing as a key limitation in this study cannot be ignored due to its effect on Na 2 O concentration.Therefore, this approach may be best suited to the preliminarily classification of tephra geochemistry, and where geochemical correlation via other major oxides is not apparent, traditional practices for tephra separation will be required.As such, this method is presented simply as a more automated alternative to tephra detection, counting and characterization.
A single 1-L bottle of Struers epoxy resin costs approximately €250 (subject to fluctuation).Therefore, mounting is relatively economical at approximately €1 per sample (5 mL), whereby 200 samples can be mounted using a single 1-L bottle.However, for upscaling the Feature Analysis method to analysing hundreds of samples, as is often the case when an entire tephrostratigraphy is required, we acknowledge that SEM-EDX instrument use may incur significant expense to the user.Furthermore, generating Feature Analysis maps of intervals in an unknown core (i.e. 100 samples) at a maximum mapping time of 120 min will take more than a week (~200 h).Although improvements may be made with more efficient mounting (e.g.larger sample holders and/or using pre-drilled epoxy mounts that can take three separate samples each), it is important to highlight that our technique may be more beneficial when a sequence has first been scanned for tephra at coarse resolution using light microscopy after ashing (highly organic peat) or ashing, sieving and density flotation (mineralrich).As such, traditional methods probably remain the most efficient and economic for initially detecting tephra in a sequence.To further lower costs, BSE images of sample mounts acquired through SEM analysis could be exported and used offline on ImageJ where suspect populations can be identified and tephra presence can be narrowed down to a smaller batch of intervals for Feature Analysis or for moving straight to EDX spot analysis on potential shards.This may be most useful when BSE greyscale levels are known for certain mineral populations during SEM-EDX use (i.e. spot analysis, identifying greyscale and annotating a BSE image) so that they may be excluded on ImageJ.Acquiring a BSE image of a sample mount often takes no more than 15 min.Future applications of the Feature Analysis technique could also include tailoring of the geochemical thresholds to identify altered tephra, which may be particularly useful for the detection of mafic shards in the peat environment, as they may be more prone to depositional alteration by cation leaching, silica dissolution and hydration (e.g.Pollard et al., 2003).Finally, this technique would be highly useful for ice core samples, whereby the potential loss in sodium due to ashing can be avoided and thus geochemical identification of tephra will be more wholly reliable.

Conclusion
The advances that chemical and density tephra detection and separation practices have brought to the tephrochronology community are profound.However, detection, counting and geochemical characterization of tephra shards traditionally requires separate stages of sample preparation.We have shown that when mounting ashed peat samples in epoxy resin without pre-separation by chemical or density practices, tephra can be detected using SEM-EDX through BSE greyscale thresholding and semi-quantitative chemical maps using AZtec Feature Analysis within 120 min.This has proven successful for a doped ombrotrophic peat sample, in which a known proportion of Vedde Ash in peat was recovered by 96%.The Feature Analysis semi-quantitative method developed using the Vedde Ash doped peat sample is within 5% bias of fully quantitative EDX spot analysis for mafic and felsic tephra.Therefore, this method allows accurate, preliminary location and characterization of tephra in peat before acquiring more detailed spot analysis.Ashing at 500 °C for 4 h shows geochemical alteration of Na 2 O in mafic tephra; however, all other major oxides in both endmembers are intact and allow accurate tephra identification.Application of our technique to minerogenic samples from Brackloon Wood, in which total mineral counts are much higher than ombrotophic peat samples, has also proven effective in locating the Laki AD 1783-84 cryptotephra.For this method, especially where tephra counts are high, or relative mineral abundance to tephra is required, optical shard counting techniques can be replaced by a combined use of ImageJ and Feature Analysis, whereby tephra shard counts located by AZtec can be ratioed to the total grain count acquired by ImageJ and extrapolated to the original sample.
This method is highly user friendly.All feature mapped data can be exported offline where morphology and geochemistry can be further evaluated, allowing tailoring of the Feature Analysis recipe for the user's needs.Furthermore, ImageJ offers a more wholly offline approach for future applications, where greyscale thresholding could be evaluated before quantitative spot analysis is carried out on suspect grains via EDX.Overall, the benefit of this technique lies in the ability to detect, count and geochemically characterize tephra simultaneously in a more automated fashion.However, we acknowledge that this method may incur significant expense to the user through instrument usage when upscaling to hundreds of samples.Therefore, this method is best suited to sections where potential tephra shards have already been located at coarse resolution using the classic 'quick burning' approach and/or density flotation for mineral-rich sequences, or in sections where age range is known, whereby cryptotephras from known volcanic eruptions can be investigated.Finally, excluding the ashing step, our approach could be readily applied to cryptotephra studies of ice cores.

Figure 1 .
Figure 1.Flow diagram for Feature Analysis on AZtec software by Oxford Instruments, with FEG-SEM-EDX parameters included.

Dope 1 :
Figure 2. BSE image showing Feature Analysis classification using peat doped with Vedde ash (Dope1).(A) Visible greyscale difference between tephra and other phases.(B) Successful characterization of phases based on geochemistry.[Color figure can be viewed at wileyonlinelibrary.com]

Figure 4 .
Figure 4. Kernel density plot of the SiO 2 (wt%) compositional distribution of tephra within VeddeM1, VeddeF1, their total and within Dope1, as identified by Feature Analysis.[Color figure can be viewed at wileyonlinelibrary.com]

Figure 7 .
Figure 7. Random BSE sub-sample image from BW13-14 cm showing tephra greyscale difference to abundant quartz and feldspar grains (A) and the successful categorization of mafic (basaltic) tephra by Feature Analysis (B), where quartz grains are largely unmapped due to their abundance.[Color figure can be viewed at wileyonlinelibrary.com]

Figure 8 .
Figure8.A selection of BSE images of tephra shards recovered from Brackloon Wood.Tephra shards were found in the 1 cm 3 intervals between 10 and 16 cm.EDX spot analysis was conducted on all shards in each interval.

Figure 10 .
Figure10.Raw BSE image of the BW13-14 cm epoxy mount (A) and ImageJ (version 1.53k) count masks for each identified grain upon the resin, along with total grain counts between three particle analysis runs, their average and relative standard deviation (%) (B).
Table 3 shows the measured compositions for each of the major elements used to classify categories in the © 2024 The Authors.Journal of Quaternary Science Published by John Wiley & Sons Ltd.J. Quaternary Sci., Vol.39(5) 816-830 (2024)

Table 2 .
Feature Analysis and offline counts for Dope1 and percentage bias from expected composition based on VeddeM1 and VeddeF1.Shard size distribution of mafic and felsic tephra within Dope1, VeddeM1 and VeddeF1.Shard diameters are estimated in micrometres for each identified tephra shard through Feature Analysis.[Colorfigure can be viewed at wileyonlinelibrary.com]

Table 3 .
Comparisons of Feature Analysis tephra identification to spot analysis in oxide wt%.
'n/a'not applicable where Feature Analysis detected 0 wt% due to low concentration and fast acquisition time (0.1 s).*Total iron expressed as FeO.

Table 5 .
Standard deviation (σ) is indicated for published values and values within this study.*Total iron expressed as FeO.Bulk tephra recovery of Dope1 in wet peat, using known sampling masses, Feature Analysis tephra counts, ImageJ total grain counts and Equation (1).