Micro–X-ray fluorescence (µXRF) analysis of proximal impactites: High-resolution element mapping, digital image analysis, and quantifications

Impactites represent unique lithologies formed as the result of hypervelocity impact events on terrestrial planetary bodies (Stöffler and Grieve, 2007). The nature and emplacement of impactites are unequivocally related to the shock dynamics (e.g., size of the impactor), type of environment (oceanic vs. continental), and the target stratigraphy, which can be homogeneous or heterogeneous (Osinski et al., 2008). Impactites are subdivided into distal (at >2.5 crater diameter distance) and proximal impactites (Glass and Simonson, 2012), where the latter are found both within the crater as part of the continuous ejecta blanket at a distance of less than 2.5 times the crater diameter (Stöffler and Grieve, 2007). Proximal impactites are further subdivided into shocked target rocks, impact melt rocks, and impact breccias (Stöffler and Grieve, 2007). Shocked target rocks are nonbrecciated, melt-free rocks displaying unequivocal effects of shock metamorphism. Studies of their petrography and geochemical composition provide important constraints on the degree of shock and deformation that occurred within the crater (Melosh, 1989; French, 1998; Riller et al., 2018).

Impact melt rocks result from impact melting (Dressler and Reimold, 2001) and represent the solidified state of molten target rocks (Stöffler and Grieve, 2007). They vary from clast-rich to clast-free and mostly occur within impact structures in the form of coherent impact melt sheets emplaced on the crater floor (Grieve and Therriault, 2012). The geochemical and isotopic signatures of impact melt rocks are unique to every impact crater, as they depend on the nature and contribution of distinct target lithologies and the impactor (French, 1998). Based on both geochemical (e.g., Floran et al., 1978; Jahn et al., 1978; von Engelhardt and Graup, 1984) and theoretical (e.g., Simonds and Kieffer, 1993) data, impact melt rocks are considered to be geochemically homogeneous. However, this homogeneity applies essentially to major-element compositions, as trace-element concentrations (Reimold et al., 1990) and isotope ratios (e.g., Marchand and Crocket, 1977; Kettrup et al., 2003) are known to display considerable variability within impact melt rocks. This variability relates to target heterogeneity and/or dissimilar contributions of distinct target rocks incorporated at the local scale after turbulent motion of the melt (Kettrup et al., 2003; Thompson and Spray, 2017).

Impact breccias are classified according to the contribution of a single target lithology or multiple target lithologies (monomict vs. polymict breccias) and the absence or presence of impact melt particles within the matrix, with classification as a lithic breccia or suevite, respectively. Suevite is a polymict impact melt–bearing breccia with a particulate matrix containing fragments of shocked and unshocked mineral and lithic clasts (Stöffler, 1977; Stöffler and Grieve, 2007). In the case of suevite, the depositional mechanisms remain relatively poorly understood, and formational scenarios range from ground-hugging density currents (Siegert et al., 2017) to fallback following the collapse of the ejecta plume (Wittmann et al., 2007), ocean resurge within the crater (Ormö et al., 2007; Gulick et al., 2019), and phreatomagmatic (molten fuel coolant) interaction (Grieve et al., 2010; Artemieva et al., 2013; Osinski et al., 2020).

Due to the target lithology heterogeneities and their chaotic formational processes, proximal impactites are complex rocks that appear to be bound to crater-specific conditions. Their petrological and geochemical characterization sheds light on those parts of the target stratigraphy that were brecciated, melted, or shock-vaporized as a result of the impact. This knowledge can be used to reconstruct the pressure and temperature conditions under which proximal impactites formed (Melosh, 1989; French, 1998).

While macroscopic and microscopic observations aid in identifying the different mineralogic components within these impactites, such petrographic techniques lack insights into specific chemical variations both at the bulk level and between clasts and/or groundmass. Several analytical techniques generate accurate and precise concentrations of major and trace elements in geological samples. However, these are either destructive techniques based on solutions and powders (such as inductively coupled plasma–mass spectrometry [ICP-MS], inductively coupled plasma–optical emission spectrometry [ICP-OES], instrumental neutron activation analysis [INAA], and bulk glass bead X-ray fluorescence [XRF]) or semidestructive in situ techniques based on thin and/or thick sections (such as electron microprobe analysis [EMPA], laser ablation–inductively coupled plasma–mass spectrometry [LA-ICP-MS], and synchrotron-based XRF). All these techniques require time-consuming and/or laborious sample preparation, especially in the case of the in situ techniques. Micro–X-ray fluorescence (hereafter µXRF) is a powerful tool to overcome these problems because this technique causes very limited to no damage to a sample, and it can rapidly produce bulk elemental data sets. In addition, µXRF makes it possible to visualize and quantify chemical heterogeneity at a high spatial resolution (down to 20 µm) within a sample (de Winter and Claeys, 2017; Flude et al., 2017).

The usefulness of µXRF methodologies in the context of distal impactites has already been demonstrated, both in the case of Archean spherule layers (e.g., Goderis et al., 2013a; Hoehnel et al., 2018; Huber et al., 2019) and Cenozoic ejecta materials, including Australasian tektites (e.g., Goderis et al., 2017). In this study, we explored the potential of µXRF as a fast, inexpensive, and nondestructive technique particularly suitable to the study of proximal impactites. Additionally, µXRF constitutes an ideal approach to select optimal sample locations and prepare these phases for more detailed analysis such as LA-ICP-MS, especially in the case of rare, important, or highly heterogeneous samples, as impactites often are (e.g., scarce drill core intervals, impactites with meteoritic contributions, or fragile impact glass with possible clay alteration). We demonstrate the potential of high-resolution, qualitative to semiquantitative element mapping followed by quantitative digital image analysis based on these µXRF maps. Furthermore, we quantified a range of different impactite lithologies and compared the measured compositions with those determined by bulk powder–based techniques on the same samples. The methodology and case studies presented here highlight the advantages of µXRF to rapidly and efficiently document the composition of heterogeneous and proximal products from impact structures.

In total, 44 samples from the Chicxulub, Ries, and Popigai impact structures were selected to illustrate the three main types of proximal impactites, i.e., shocked target rocks, impact melt rocks, and impact breccias. These impact structures range in diameter from ~24 km to ~180 km and represent structures formed on distinctly different geological settings, with varying complexities of target lithologies that range from magmatic to metamorphic and/or sedimentary in origin.

The Chicxulub impact structure (Yucatán, Mexico) is ~180 km in diameter (Hildebrand et al., 1991). It has been dated at 66.052 ± 0.008/0.043 Ma and coincides with the Cretaceous-Paleogene boundary mass extinction (Sprain et al., 2018). The Chicxulub crater is the result of the impact of a 10 ± 4 km diameter (Alvarez et al., 1980) carbonaceous chondrite (Goderis et al., 2013b) on a Mesozoic carbonate and evaporite platform, which rests on crystalline target rocks ranging from Neoproterozoic age (ca. 546 Ma; Keppie et al., 2011; Renne et al., 2013) to Carboniferous age (326 ± 5 Ma; Zhao et al., 2020). The 40 Chicxulub samples used in this study originated from the International Ocean Discovery Program–International Continental Drilling Project (IODP-ICDP) Expedition 364 drill core: 9 crystalline target rocks, 21 impact melt rocks, and 10 impact breccias (Fig. 1A). This core was drilled offshore on the Chicxulub peak ring at Site M0077 (global positioning system [GPS] coordinates: 21.45°N, 89.95°W in World Geodetic System 1984 [WGS84] reference coordinate system) in 2016 and reached a depth of 1334.7 m below seafloor (mbsf; Morgan et al., 2016). This drilling extracted a near-continuous sequence of proximal impactites consisting of ~104 m of suevite, ~26 m of impact melt rocks, and ~610 m of granitoid basement material intruded by pre-impact mafic dikes and suevite/impact melt rock (e.g., Morgan et al., 2017; Kring et al., 2017). In the Expedition 364 drill core, the peak-ring lithologies all display pervasive hydrothermal alteration (Kring et al., 2020; Simpson et al., 2020).

The Ries impact structure (Germany) is an ~24-km-diameter impact structure (Pohl et al., 1977) with an age of 14.808 ± 0.021 Ma determined by ⁴⁰Ar-³⁹Ar (Schmieder et al., 2018; Schmieder and Kring, 2020). The target at the Ries structure is composed of a Hercynian crystalline basement covered by Mesozoic sedimentary rocks (Schmidt-Kaler, 1978). The basement consists of gneisses, amphibolites, and ultrabasic rocks with granitic intrusions. The sedimentary sequence is constituted by sandstone, siltstone, marl, and limestone (Pohl et al., 1977). Two samples were selected from the Ries impact structure: impact melt rock sample Pols-01 and suevite sample Aum-01 (Fig. 1A). Pols-01 was sampled in 2008 at Polsingen (GPS coordinates [WGS84]: 48.92°N, 10.71°E), which is the location of one of the few outcrops in the Ries impact structure displaying isolated small (<100 m) lumps of coherent impact melt rock (Pohl et al., 1977; Graup, 1999). This represents one of the two main forms of impact-melted materials in the Ries crater; the other type constitutes impact melt particles within suevites (Osinski, 2003; Stöffler et al., 2013). Suevite sample Aum-01 was sampled in 2008 in the Aumühle quarry, near Oettingen (GPS coordinates [WGS84]: 48.97°N, 10.63°E), which yield several well-exposed suevite sections (Osinski et al., 2004). In the Ries impact structure, impact melt rock and suevite are characterized by the same stratigraphic position as they both overlie the sedimentary Bunte Breccia following a sharp contact (Hörz et al., 1983; Stöffler et al., 2013).

The Popigai impact structure (northern Siberia, Russia) is ~100 km in diameter (Masaitis, 1994) and has been dated to 36.63 ± 0.92 Ma using the ⁴⁰Ar-³⁹Ar step-heating technique (Schmieder and Kring, 2020). Popigai is part of a late Eocene cluster of impact structures, and it likely formed through the impact of a type L or H ordinary chondrite (Tagle and Claeys, 2005; Kyte et al., 2011) on the Archean Anabar Shield. The target at the Popigai impact crater site is composed of Archean to early Proterozoic crystalline rocks that are covered by sedimentary sequences (Masaitis, 1994). The latter include Proterozoic to Cambrian conglomerates, quartzites, dolomites, and limestones, Permian terrestrial sandstones and argillites, and discontinuous thin beds of Cretaceous sandstones (Vishnevsky and Montanari, 1999). Popigai has been determined to be the source crater of the late Eocene clinopyroxene microkrystite spherule layer (Liu et al., 2009). Within the crater, the shocked target rocks are covered by an ~2-km-thick impactite sequence composed of three main units that vary in thickness within the crater (Naumov, 2002; Masaitis et al., 2005). The lower unit includes polymict breccia constituted of fragments from the different target lithologies; the intermediate unit contains alternating impact melt rock (tagamite in the Russian literature) and suevite sheets; and the upper unit presents suevite lithology (Masaitis et al., 2005). The suevites from the intermediate and upper units mainly differ in the contained impact melt clasts (vitroclasts), which are black and greenish-gray, respectively (Naumov, 2002). Two samples were selected from the Popigai impact structure: an impact melt rock Pop 10 (from Tagle and Claeys, 2005) and an impact breccia Pop C6 (Fig. 1A). Both samples were collected during the International Expedition in 1997. Pop 10 was sampled from a large outcrop of impactites along the Rossokhe River, west of the crater (GPS coordinates [WGS84]: 71.76°N, 110.25°E; Tagle and Claeys, 2005). Pop C6 was collected northwest of the crater along the Rossokhe River and was sampled from a pile of scattered and abandoned old cores that were drilled in Popigai during the intense exploration programs by the Union of Soviet Socialist Republics (USSR) in the 1970s and 1980s (Masaitis et al., 2005).

Because the µXRF measurements are performed on a horizontal plane, the measured surface needs to be as flat as possible. Naturally formed near-flat surfaces can directly be analyzed in the case of precious samples. However, preferably, rock samples are cut using a diamond board table saw and slightly polished to remove traces of contamination from the sawing blade (e.g., W, Co, Cu) and to acquire an even surface for easy operation of the instrument. A total sample thickness of at least 1 mm is required to account for the maximum attenuation depth of the X-rays, equivalent to ~800 µm (Beckhoff et al., 2006). The attenuation depth differs per energy level (i.e., per element) and depends strongly on the type of material analyzed. In the case of thinner samples such as polished thin sections (~30 µm thick), materials located below the sample can affect the analysis. Therefore, a plastic sample holder and stage are used, which are composed of very light elements that cannot be measured using the µXRF technique. Samples with irregular surfaces are loaded into an entirely plastic (e.g., polypropylene) holder filled with kinetic sand and carefully placed in the same horizontal plane prior to measurement (Fig. 1B). The plastic sample holder is placed on a motorized x-y-z stage composed of plexiglass (polymethylmethacrylate) material. In the case of the Bruker M4 Tornado (Fig. 1C), the dimensions (length × width × height) and weight of the sample holder cannot exceed 20 cm × 16 cm × 12 cm and 5 kg, respectively.

The selected proximal impactites were analyzed using a Bruker M4 Tornado benchtop micro–X-ray fluorescence instrument (Bruker Nano GmbH, Berlin, Germany) under near-vacuum conditions (20 mbar), hereafter referred to as the µXRF instrument (Fig. 1C). The mapping option, the stage precision, and the application of a vacuum represent the main benefits of this “desktop” instrument as opposed to the more common “handheld” portable-XRF (pXRF), allowing the detection of light elements up to Na (de Winter et al., 2017). The µXRF instrument uses a 30 W rhodium anode metal-ceramic X-ray tube and two 30 mm² silicon drift detectors (SSDs) with a resolution of 145 eV (Mn-Kα) at 300 kcps (kilocounts per second). The Rh X-ray source was operated under maximum energy settings (600 µA, 50 kV) and without the use of a filter. The X-ray beam was focused by a polycapillary lens on a spot with a diameter of 25 µm (Mo-Kα; de Winter and Claeys, 2017). A standard integration time of 1 ms per pixel and a spacing of 25 µm were employed for all µXRF mappings. In addition, µXRF spot analysis was carried out on a selection of homogenized powders, which were carefully flattened and mounted in a holder with plastic cups, using 10 points per powder and an integration time of 120 s for a defocused spot size of 200 µm.

Three different modes can be selected to perform the µXRF analysis with the Bruker M4 Tornado: spot, line scan, and map. The spot analysis obtains an XRF spectrum of a selected point with a spot size of 25 or 200 µm. For such a spectrum to be accurate and considered representative, it must be based on an analysis with an integration time between 60 s and 300 s (de Winter et al., 2017). For reproducibility purposes, it is also recommended to perform several random spot analyses (e.g., 10) on the same target to obtain a representative average of the measurements (de Winter et al., 2017). The line scan analysis consists of a series of—often densely spaced—spot analyses. In general, the µXRF spot and line scan analyses are useful if the user is interested in characterizing particular—often chemically homogeneous—phases or when micrometer- to millimeter-scale chemical changes throughout a phase or unit need to be accurately quantified. The relatively long measurement time on small surface areas makes the spot analysis and line scan modes unfavorable for geochemically characterizing large heterogeneous samples such as the proximal impactites in this study. A µXRF spot analysis and line scan campaign can still be useful for more in-depth geochemical studies, but only after a first characterization of the samples is carried out. Therefore, in this study, we focused on the µXRF mapping mode and do not discuss the spot and line scan analyses in more detail.

The µXRF mapping mode allows the user to cover areas that are significantly larger than those used in the spot and line scan modes, with a very short integration time per pixel (1 ms in this study). Hence, this mode is preferred as a first screening tool for (especially large) samples without characterized chemical composition. Each pixel on the acquired µXRF map has an XRF spectrum associated with that pixel. This implies that for a quantitative and representative spectrum study, an area of a complete sample or a masked-out region of interest needs to have at least 60,000 pixels to reach a total integration time of at least 60 s (de Winter et al., 2017).

Using the Bruker M4 Esprit software, XRF spectra were converted into two types of chemical maps. First, multi-element maps were produced, which show a color assigned to each element selected, where the color of every pixel is a tone combination of the colors selected (i.e., a ratio of the selected element abundances). Second, single-element heat maps were produced, which show the relative variations of a single element on a linear red-green-blue (RGB) color (0–255) heat scale in which the highest RGB value (255) corresponds to the pixel on the map with the largest amplitude of the main (often Kα or Lα) peak of that specific element. For all maps in this study, the RGB value of a pixel was calculated by the average RGB value of the pixel itself and the eight pixels directly surrounding that pixel (a square of 3 × 3 pixels or 5625 µm²). Spectra from bulk µXRF maps and regions of interest were quantified using the standardless fundamental parameters (SFP) method, resulting in volatile-free, normalized compositions in which the major elements are expressed as oxides, and the trace elements are expressed in elemental form. This standardless method uses the fundamental algorithm developed by Rousseau (Rousseau, 1984a, 1984b; Rousseau and Bouchard, 1986; Rousseau and Boivin, 1998), which is based on the theoretical Sherman equation (Sherman, 1955) that links XRF peak intensities with the respective elemental concentrations (de Winter et al., 2017; Vansteenberge et al., 2020).

To determine the accuracy of the µXRF SFP quantification technique, 40 slabs from the IODP-ICDP Expedition 364 core and the 40 associated thin sections were analyzed by µXRF. The quantified major- and trace-element data were compared with bulk powder–based ICP-OES and ICP-MS data from the same 40 intervals but not from the exact same subsamples presented in de Graaff et al. (2019) and Déhais et al. (2019) and acquired following the methodologies outlined in Poppe et al. (2020). Those 40 homogenized powders were also analyzed by µXRF defocused (200 µm) spot analysis to compare the ICP-OES and ICP-MS data with respect to the matrix. The comparisons made for the major elements in this study were between µXRF SFP and anhydrous normalized ICP-OES data, because it is not possible to obtain loss-on-ignition data with the µXRF instrument. This direct comparison is the reason why major elements are expressed as oxides and trace elements are expressed as elements, even if µXRF also allows all elements to be expressed in elemental configuration. In addition, the SFP quantification of the bulk µXRF map of impact melt rock Pop 10 was directly compared with bulk powder XRF data from the same sample. Details about this bulk XRF technique were presented in Tagle (2004).

The µXRF element maps can be studied in further detail by applying digital segmentation in image analysis software to quantify modal proportions of lithological components and to determine textural characteristics (including size, shape, orientation, and sorting) of these components. This independent, quantitative data set aids in the interpretation of complex depositional mechanisms for a range of lithologies, often sedimentary. Semi-automated digital image analysis methods are commonly used to characterize heterogeneous lithologies such as proximal impactites, but most techniques are based on scans of cut and polished hand specimens or microphotographs of thin sections (Chanou et al., 2014; Pittarello and Koeberl, 2013). Here, we present a novel approach based on the geochemical composition of rock samples.

The Image Analysis module of the Zeiss (Carl Zeiss GmbH, Jena, Germany) ZEN 3.1 Pro software was used for segmenting, classifying, and quantifying different components of the Popigai impactite samples. RGB-colored µXRF multi-element maps (Fe-Si-Ca) were imported into the software and scaled based on their resolution (25 μm per pixel). Then, a classification was made of the different clast types and the groundmass present in the sample. Subsequent segmentation took place with automatic selection of phases in a designated sample frame on the µXRF map. This segmentation was based on maximum and minimum thresholds of RGB values (0–255), with a minimum particle size threshold of 0.2 mm. We also applied the option “fill holes” to exclude secondary structures inside (mainly) impact melt particles. Interactive manual segmentation was only carried out when the contrast between clast and groundmass was not high enough or when clasts consisted of multiple compositionally distinct mineral phases.

After careful digital segmentation, an extensive clast-specific data set was produced. The image analysis software exports the following output measurements for each clast: (filled) area (mm²), convex area (mm²), length of major and minor axes of the best-fit ellipse (mm), aspect ratio (ratio between the major and minor ellipse axis length), perimeter (mm), roundness (a value from 0 to 1 based on [4π area]/[convex perimeter]²), sphericity (perfect circle equals 1), and convexity (convex perimeter/perimeter). It is also possible to select whether a particle at the edge of a sample touches the selected sample frame. This is useful to ensure that particles that are cut off by the outline of the sample are not incorporated in the shape parameters database because their morphologies are incomplete and hence inaccurate. However, since these particles are still part of the studied sample area, they are incorporated in the calculation of the modal abundances of classes and are also included in determining the degree of sorting using cumulative area and perimeter fractions (Chanou et al., 2014).

Subsequently, the SUM-classes export function in the ZEN 3.1 Pro software provides the total count of the clasts of one particular class, the cumulative area percentage, and average values for the particle parameters stated above. Based on these parameters, the modal abundances of the different components can be calculated together with their main particle characteristics.

The direct and qualitative result from µXRF mapping is a bulk map spectrum, which represents the average chemical composition of the complete sample surface. The sample composition, the measurement conditions, the spectrometers, and the X-ray source determine the general shape of the µXRF spectrum and the presence of the elemental fluorescence peaks. The latter are always located at their typical energy levels linked to the K, L, and M lines (Beckhoff et al., 2006). Figure 2A shows the bulk map spectrum of impact breccia sample Pop C6, with the main fluorescence peaks (Kα) highlighted in bold and the secondary (and smaller) peaks (Kβ) shown in regular font. In the bulk map spectrum, the 20 Kα peaks (Na, Mg, Al, Si, P, S, Rh [from the X-ray source], K, Ca, Ti, V, Cr, Mn, Fe, Ni, Cu, Zn, Rb, Sr, and Zr) and 7 Kβ peaks (Ca, Ti, Fe, Zn, Rb, Sr, and Zr) are positively identified. For example, the Fe-Kα peak occurs at 6.404 kilo-electron volt (keV), while the Fe-Kβ peak is located at 7.058 keV. Extra caution is required when identifying elements that display peak overlap, such as the Co-Kα peak (6.930 keV) and the Fe-Kβ peak. This problem can be solved by automatic or interactive peak deconvolution in the M4 Bruker software. Three common µXRF background spectrum characteristics are also shown in Figure 2A: the “Bremsstrahlung” and the Rayleigh and Compton peaks. The Bremsstrahlung is a wide plateau between 5 and 22 keV that is caused by the presence of backscatter without any excitation of the sample (Shackley, 2011). The Rayleigh and Compton peaks are four peaks generated due to the X-rays emitted by the Rh tube source. The Rayleigh peaks are caused by the elastic scattering of X-rays by the sample without any change in their energy (Potts and Webb, 1992) and are always located at 20.216 keV (Kα) and 22.724 keV (Kβ) for the Rh source (Fig. 2A; Vansteenberge et al., 2020). The Compton peaks are the result of inelastic scattering of X-rays, and their shift in energy compared to the Rayleigh peaks depends on the angle between the X-ray source and the spectrometer (lowest intensity at 90°). However, the scatter of the X-rays shifts the Compton peaks always toward slightly lower energies than the Rayleigh peaks (Potts and Webb, 1992).

From qualitative µXRF maps, it is possible to define regions of interest and to access the subspectra related to these areas. Figure 2B shows four subspectra of four different polygons within the Pop C6 sample, linked to chemically distinct lithological phases shown in the Fe-Si-Ca multi-element µXRF map. Note the difference in the total amount of counts for the bulk spectrum in Figure 2A and the regions of interest in Figure 2B. To allow the observation of the smallest peaks, the y-axis has been truncated at 100,000 counts in Figure 2B. Therefore, a comparison for Ca and Fe is not possible with the main Kα peaks, respectively at 3.692 keV and 6.404 keV, but it can be made using the secondary Kβ peaks of these two elements, respectively at 4.013 keV and 7.058 keV. The intensity comparison of the same elemental peak between several spectra provides semiquantitative information about the relative enrichment or depletion of a particular element in the regions of interest. The peak intensities show for instance that basement clast B2 is enriched in Al and Si relative to the other regions selected. Carbonate clast C1 is enriched in Mg, P, S, Ca, and Mn relative to the other regions. Impact melt clast M5 is enriched in K, Ti, and Fe.

Figure 3 displays both types of µXRF maps for the sample Pop C6 (spectra shown in Fig. 2). The µXRF multi-element map with Fe, Si, Ca, and K directly provides clear constraints on the assemblage of different lithological components present in the sample. This impact breccia contains a range of clast sizes, from millimeters to several centimeters. Most of the greenish-yellowish clasts in the impact breccia sample Pop C6 (Fig. 1A, #5) are enriched in Fe and are likely impact melt clasts. Some clasts are enriched in Ca, and most of these do not contain any significant Fe, Si, or K, especially two large clasts at the top right (C1) and bottom left (C2) corners (Fig. 3). A few areas display black pixels, corresponding to holes in the surface and to the edges of the samples, where the X-rays were diffracted and no longer reached the detectors. The single-element heat maps of Mg, Al, Si, K, P, S, Ca, Ti, Cr, Mn, Fe, Ni, Cu, Zn, Rb, Sr, and Zr provide semiquantitative information on the distribution of these elements in Pop C6 (Fig. 3), as their relative color scale is based on RGB values from 0 to 255. These heat maps are correlated with the spectra shown in Figure 2. For example, clast C1 displays a spectrum with low-intensity Al, Si, Ti, Cr, Fe, Cu, Zn, Rb, and Zr peaks compared to other clasts and the groundmass (Fig. 2B) and therefore low RGB values of these elements in the respective heat maps (Fig. 3). Clast B2 shows very high counts for Al, Si, and Sr peaks relative to the other parts of the sample (Figs. 2B and 3). The comparison of fluorescence peaks in an XRF spectrum can sometimes be difficult due to low intensities; this is the case, for example, with Cu and Rb (Fig. 2B). However, the heat maps can still help to visualize that Cu and Rb are slightly more abundant in the matrix, and only a few small clasts are enriched in these elements (Fig. 3).

The RGB scale of single-element heat maps is relative for each mapped area and depends on the lowest and highest Kα peak intensities. Therefore, it is impossible to directly compare several heat maps of the same element when samples are mapped separately. To overcome this problem, several samples can be mapped together (Figs. 1B and 1C) to obtain a relative scale that is equal for the complete map area. To illustrate this approach, we mapped three impact breccia samples and three impact melt rock samples from Popigai, Ries, and Chicxulub, both individually (Fig. 4A) and all together (Fig. 4B). Between the six samples, Fe is most abundant in Pop C6, whereas the three samples Aum-01, Pols-01, and 90_1_85_87 do not contain any Fe-rich areas when compared to the other samples (Fig. 4B). This is clearly different when observing the Fe heat map of each sample separately (Fig. 4A).

Besides a qualitative approach with multi-element maps and a semiquantitative approach with single-element heat maps, the µXRF mapping technique also makes it possible to perform a quantitative image analysis based on these element maps. Figure 5 shows the workflow and results based on digital segmentation of Fe-Si-Ca µXRF maps from the two Popigai impactite samples in the Zeiss ZEN 3.1 Pro image analysis software. Impact breccia sample Pop C6 (Fig. 1A, #5; Fig. 5A) consists of a relatively equal amount (both ~40 area %) of both impact melt particles and groundmass, supplemented with ~20% of other clasts consisting of (felsic) basement clasts and carbonate clasts. Due to the high abundance of both impact melt particles and a fine-grained groundmass that appears to be clastic, this sample is classified as a matrix-supported suevite. Pop 10 (Fig. 5B) consists of 81% melt matrix and is therefore classified as a clast-poor impact melt rock (<25% clasts). The clasts in Pop 10 are subdivided into basement clasts, carbonate clasts, and Fe-rich clasts. The latter are also associated with vesicles on the sample surface (Fig. 1A, #6).

Figure 6 exhibits the particle size and shape characteristics of 1282 segmented clasts in Pop C6, subdivided into the three clast types from Figure 5A and shown both with absolute and relative frequencies. The unimodal clast-size distribution of the impact melt particle population (Fig. 6A) is dominated by a fine size range of 0.2–0.4 mm with a median of 0.37 mm. The carbonate clasts display a finer median of 0.25 mm, but they also show an additional peak at 0.6–0.7 mm. The basement clasts show a coarser and more irregular distribution with a median clast size of 0.96 mm. The roundness curve (Fig. 6B) displays a difference between more angular impact melt clasts (median of 0.43) and more rounded carbonate and basement clasts (median of 0.54 and 0.60, respectively). The aspect ratio curve (Fig. 6C) does not exhibit a clear pattern and shows for all clast groups both nonelongated and elongated clasts, although the median of the impact melt clasts is slightly higher compared to the other groups (1.72 vs. 1.45–1.53). The difference in clast morphology is more pronounced in the sphericity plot (Fig. 6D). The impact melt clasts are less spherical (median of 0.44) than the carbonate and basement clasts (median of 0.79 and 0.75, respectively).

Figure 6E shows a sorting curve for sample Pop C6 based on the relationship between the cumulative perimeter and cumulative area fractions, using all segmented clasts (n = 1351). In addition, digital segmentation was carried out on a classic, visual sorting scale commonly used in geology, following the procedure of Chanou et al. (2014). The resulting five sorting curves correspond to schematic examples of five typical degrees of sorting, ranging from very well-sorted (characterized by a slope of ~1 and a continuous linear distribution) to well-sorted (slope of 1–1.2), moderately sorted (slope of 1.2–1.7), poorly sorted (slope of 1.7–2.1), and very poorly sorted (slope of 2.1–3 and a very discontinuous distribution) samples. Based on a comparison with these five general sorting curves, sample Pop C6 can be characterized as a poorly to moderately sorted sample, as it shows a rather discontinuous distribution and a slope of ~1.7.

To determine the accuracy of the µXRF standardless fundamental parameters quantification technique, we performed a comparison study between µXRF and ICP-OES and ICP-MS data based on a wide range of Chicxulub samples (homogenized powders, thick sections, and thin sections). Figure 7 shows the µXRF mapping results on three representative samples from the IODP-ICDP Expedition 364 core, both as thick and thin sections. These samples cover the three groups of proximal impactites, with a suevite (core 54_1, 64–66 cm, 659.58 mbsf), an impact melt rock (core 90_1, 85–87 cm, 729.54 mbsf), and a shocked granitoid target rock (core 95_2, 89.5–91.5 cm, 745.78 mbsf). The bulk map spectra of the thick and thin sections of 54_1_64_66 are similar up to the Mn peak, especially for the Na, Mg, Al, Si, P, and S peaks, which cannot be distinguished from one another (Fig. 7A). The heights of the amplitude of the Kα peak of Mn and Fe are larger in the thick section compared to the thin section at 1.7 and 2.6 times, respectively. With heavier elements than Fe, the intensity of the peaks within the thin section spectrum decreases even further, resulting in barely noticeable Kα peaks for, e.g., Ni, Cu, Zn, Rb, and Zr. Similar observations were made based on the spectra of the other two core samples (90_1_85_87 and 95_2_89.5_91.5). The intensity changes around the Mn and Fe Kα peaks observed from these bulk spectra are also visible in the heat maps of these samples (Fig. 7B). The six samples were mapped together to allow the comparison of the peak amplitude of the associated elements, which is linked to the concentrations. The relative colors are comparable for the thick and thin sections of all three samples for Mg, Al, Si, and Ca maps, but not for the Fe, Ni, Rb, and Sr maps.

Figure 8 shows binary diagrams based on 40 samples from the IODP-ICDP Expedition 364 core, displaying different proximal impactites and target lithologies. The horizontal axis shows element data based on bulk powder ICP-OES and ICP-MS, whereas the vertical axis shows data based on the µXRF SFP quantification method applied on three different subsamples (200 µm spot analysis of homogenized powders, mapping of thick sections, and mapping of thin sections, respectively). The analysis of each kind of subsample has a different purpose. The 200 µm defocused spot analysis of homogenized powders, which were also used for ICP-OES and ICP-MS analysis, shows the initial error of the µXRF SFP method, as the same matrix is used in both cases. The µXRF SFP quantification of thick section maps provides information on the supplementary error added by the heterogeneity of the samples. The µXRF SFP quantification of thin section maps gives information on the additional error introduced by an insufficient thickness of the samples.

The major elements, expressed as oxides Na₂O, MgO, Al₂O₃, SiO₂, K₂O, CaO, TiO₂, MnO, and Fe₂O₃, and the trace elements Ni and Rb are shown to illustrate the comparison between the different techniques (Fig. 8). The slope, intercept, and R² value of the linear regressions derived from the graphs shown in Figure 8 are reported in Table 1, and the µXRF SFP quantification data are reported in Table A1 (homogenized powders with 200 µm spot analysis), Table A2 (mapping of thick sections), and Table A3 (mapping of thin sections). For all major elements in the homogenized powder analysis, the slope of the linear regression is close to 1 (except for MgO and Fe₂O₃, which are 1.36 and 1.31, respectively) and associated with an excellent R² > 0.9 (except for Na₂O, which has R² = 0.88). This indicates that the µXRF SFP quantification on homogenized powders and the ICP-OES technique are in very good agreement for all the major elements.

For both thick and thin sections, from Na₂O to CaO (six major elements out of nine), the slope of the linear regression is close to 1 and associated with good R² > 0.77 (except for Na₂O and Al₂O₃ in thick sections, with R² = 0.59 and R² = 0.66, respectively). This indicates that the µXRF SFP quantification of thick and thin sections and the ICP-OES technique are in reasonable to good agreement for those elements. The slope and R² values for TiO₂ are similar for thin and thick sections but are not as good as for the other major elements described above. For MnO and Fe₂O₃, the thick sections show slope and R² values close to the relatively high values of Na₂O to CaO, in contrast to the thin sections, which yield lower values with slopes of 0.44–0.49 and R² values of 0.42–0.56. This change at MnO and Fe₂O₃ is also visible in Figure 8, in which the blue trend line of the thin sections starts to deviate from the other two trend lines. Both the slope and the R² values for the thin sections decrease drastically for all trace elements heavier than Fe₂O₃ (all slopes < 0.26, all R² < 0.42), as can be seen in the bivariate plot for Ni and Rb.

The case of Ni is particularly interesting, because at first the correlation between µXRF SFP quantification and ICP-MS data seems to be excellent for homogenized powders and thick sections (slope of 1.25 and 1.66, R² of 0.98 and 0.96, respectively). However, these values are only valid when all samples are taken into account (graph “Ni” in Fig. 8). Most of the samples fall within a low-concentration range (10–45 ppm). The only outliers are the two felsite samples (~100 ppm of Ni) and the two dolerite samples (230–300 ppm of Ni), which control the good correlation for the nickel. To visualize the ability of µXRF to quantify low Ni concentration, a second Ni graph without the data points of dolerite and felsite samples is shown in Figure 8 (graph “Ni*”), and the associated linear regression parameters are reported in Table 1. If only the two dolerite samples are removed, the slope and R² values are still in good agreement with ICP-MS data (slope of ~0.9–1.1 and R² of 0.75–0.87) for both homogenized powders and thick sections. If the two dolerite and the two felsite samples are removed, the linear regression parameters show a bad correlation with ICP-MS data (slope of ~0.35–0.61 and R² of 0.14–0.40). All these observations indicate that the µXRF SFP quantification is accurate for homogenized powders for all major elements and most trace elements and is relatively accurate for thick sections up to Fe₂O₃ and for thin sections up to CaO.

To visualize and further quantify the trends observed in Figures 7 and 8, the concentrations determined by the µXRF SFP technique were compared to those derived from bulk ICP-OES and ICP-MS data. Figure 9 displays a box-whisker plot showing the deviation of the µXRF SFP quantification from the bulk ICP-OES and ICP-MS results for the different major (as oxides) and trace elements. The median deviations and interquartile ranges visualized in Figure 9 are also summarized in Table 1. The trend is similar between homogenized powders (Fig. 9A) and thick sections (Fig. 9B), although the homogenized powders display for most elements a smaller interquartile range and whisker range. The method is relatively accurate from Na₂O to MnO. From Fe₂O₃ to Zr, the deviation from the bulk data increases (1) in the positive direction for homogenized powders (Fig. 9A) and for thick sections (Fig. 9B) and (2) in the negative direction for thin sections (Fig. 9C). Also, the precision decreases with heavier elements, visible as an increase in the size of the envelopes, both in the interquartile range and the whisker range. From Fe₂O₃ toward heavier elements, the technique tends to overestimate the concentrations in the homogenized powders and the thick sections, and to underestimate the concentrations in the thin sections.

The impact melt rock Pop 10 has been analyzed by bulk powder XRF (Tagle, 2004), and, therefore, the µXRF SFP quantification of this sample can be compared to this other bulk technique. The results are shown in Table 2. The concentrations obtained with the µXRF technique (on a homogenized powder and a thick section) are close to the values published by Tagle (2004). The thick sections show a deviation of less than 10% for the major elements (expressed as oxides) Na₂O, Al₂O₃, SiO₂, CaO, TiO₂, MnO, and Fe₂O₃. Larger deviations are observed for K₂O, MgO, P₂O₅, and SO₃. For the two first major elements, this difference is likely explained by comparison between different subsamples of the same sample (Tagle, 2004; this study). Therefore, even if Pop 10 had been determined to be a clast-poor impact melt rock (Fig. 5B), a variation in the presence of clasts from the target lithologies (Vishnevsky and Montanari, 1999) in the sample can explain the observed difference in MgO, K₂O, and SiO₂ values. In the case of P₂O₅ and SO₃, the concentrations are very low (<0.1 wt%), so the values are below the limit of quantification. As already shown with the analysis of the Expedition 364 samples (Fig. 9B) and for the bulk XRF comparison, the µXRF SFP quantification overestimates trace-element concentrations, except in the case of Y and Ba. However, the determined Ni, Cu, and Zr results exhibit a deviation of <15% compared to the bulk XRF values. Regarding the Ni quantification, it is important to notice that the Ni concentration of Pop 10 (>120 ppm) is higher than most samples from the Chicxulub IODP-ICDP Expedition 364 drill core used in this study, which indicates that µXRF SFP quantification of these high levels of Ni concentrations is meaningful. This confirms the potential of µXRF to track meteoritical components, which are frequently linked to Ni enrichment in impactites (e.g., Lambert, 1975; Grieve et al., 1981; Goderis et al., 2012). Due to peak deconvolution issues, the Co and V values cannot be verified, as they are biased by the overlapping Fe and Ti peaks, respectively.

The first major advantages of the µXRF mapping technique are the fast and easy sample preparation and the nondestructive nature of the analysis. These make µXRF a powerful tool to geochemically characterize proximal impactites, especially in the case of complex or precious samples such as impact breccias and drill core samples. When using an integration time of 1 ms per pixel with a spacing of 25 µm, the instrument takes approximately 1 h to map an area as wide as a standard geological thin section (~4 × 2 cm, ~8 cm²) and roughly 1 d for a map covering the maximum sample size (~16 × 20 cm, ~320 cm²). In case of time constraints for the completion of a map, the duration of the analysis can be shortened by increasing the spacing. This leads to a loss in spatial resolution, but in the case of large samples (>100 cm² surface area), smaller regions of interest can efficiently be identified and eventually rescanned later at a higher resolution. The µXRF technique is also competitively priced, as no consumables are involved, and the cost per measurement remains consequently relatively low.

However, a few issues must be kept in mind when interpreting µXRF maps. A commonly encountered problem is linked to focusing, as only one focus plane can be selected while performing µXRF mapping. When a sample is not completely flat or displays edges and holes, the excitation of primary X-rays and the subsequent detection of secondary X-rays are inhibited. This can lead to blurred areas on the µXRF maps, as can be seen for example in the bottom part of Pop 10, which is lower in elevation than the focal plane (Fig. 4, #6). Another problem is the presence of exogenous or preparatory materials such as tape, glue residue, or notes on the samples, which affect some element maps. This was observed for the label written at the bottom of the IODP-ICDP Expedition 364 thin sections, because the ink was enriched in S and Ti. Therefore, a careful selection of the mapping area is required to avoid any quantification problems for µXRF maps. It is also important to note that even if the µXRF is a surface technique, the X-rays penetrate the samples by several hundreds of micrometers (Beckhoff et al., 2006). This means that the analysis incorporates (potentially enriched) phases that are not exposed at the surface and observable by the naked eye.

Multi-element µXRF maps are useful to directly identify multiple mineral or lithological phases and to classify complex samples such as impact breccias, which often contain a wide range of different lithologies (Figs. 2 and 3). Theoretically, it is possible to create multi-element maps with as many elements as possible. However, a different color needs to be assigned to each element selected in the map. Due to the complementarity of colors, the most effective multi-element maps are those combining 2–4 elements.

In contrast, single-element heat maps visualize the elemental distribution of a single selected element (Figs. 3 and 4). Each map has its own linear RGB color scale (0–255) based on the lowest and highest peak amplitude for each element selected within the sample. These peak amplitudes are linked to concentrations, but other factors like matrix effects and element-specific characteristics such as fluorescence yield also play a significant role in peak intensity (Beckhoff et al., 2006). Because of the relative scaling, element distribution maps of individually scanned samples cannot be compared in terms of absolute concentrations. Heat maps can only be compared if all the samples are mapped together (Figs. 4B and 7B) or if the sample composition is well constrained by other techniques such as bulk XRF, ICP-OES and ICP-MS, EMPA, or LA-ICP-MS.

Digital image analysis of semi-automated segmentation based on µXRF maps of proximal impactites is successful in producing quantitative modal abundances (Fig. 5) as well as an extensive data set of particle parameters (Fig. 6). As the segmentation in this method is based on the geochemical composition, clast type misidentifications on sample surfaces due to visual aspects such as color—potentially influenced by postimpact alteration—are limited.

For instance, in Pop C6 (Fig. 1A, #5), a large gray clast is visible in the upper right (C1) while a medium-large brown clast is found in the lower left (C2). These clasts are visually distinct, but they have a similar chemical composition based on the µXRF maps (Fig. 3) and are therefore classified as clasts from a carbonate target lithology. Pop C6 is classified as a matrix-supported suevite (Fig. 5A) based on a groundmass that appears to be clastic, a clast/matrix ratio of 60% versus 40%, and a dominance of impact melt particles (~40%). The impact melt particles in Pop C6 are more angular and less spherical than the carbonate and basement clasts (Figs. 6A–6D). Pop C6 does not display clear bedding or a preferred orientation of clasts, and it is characterized as a poorly to moderately sorted suevite (Fig. 6E) based on the cumulative perimeter and area fractions of the entire clast population.

This quantitative information sheds light on the physico-mechanical processes of formation, transport, and emplacement of this Popigai suevite. The impact melt particles in Pop C6 are dominated by angular, green/gray porous clasts (Fig. 1A) that are rich in Fe and do not exhibit large numbers of other mineral or lithic clasts within them (Figs. 3 and 5A). These are interpreted as remnants of quenched melt, resulting in glassy or vitric melt particles, which are easily weathered out (hence leading to a porous texture). Another type of impact melt particle exists, illustrated by clasts M3 and M7 (Fig. 5A), as this type is characterized by higher SiO₂ and lower Fe₂O₃ contents compared to the vitric melt particles (based on μXRF SFP quantifications of these regions of interest, as shown in Table A4). The clasts from this second type of impact melt are also angular in shape and characterized by entrained mineral and lithic clasts from the felsic basement. These could represent brecciated parts from the central melt sheet transported into the suevite unit. Overall, Pop C6 fits the description of lithovitroclastic suevites with predominantly greenish melt particles and a small fraction of both crystalline and sedimentary clasts. These suevites are described from the lower part of the upper member of the Popigai crater fill sequence (Naumov, 2002; Raikhlin, 1996; Masaitis, 2003). This suevite is interpreted to have been emplaced by a combination of base surge and settling from the collapsed hot vertical part of the explosion cloud without significant water-melt interaction (Masaitis et al., 2018). Pop 10 consists of 19% clasts and is therefore classified as a clast-poor impact melt rock, which fits the literature description of the high-temperature impact melt rock or T2 melt rock observed in the Popigai impact structure (summarized by Whitehead et al., 2002).

In this study, we used Fe-Si-Ca multi-element maps of thick sections as a basis for the digital segmentation, but single-element heat maps of polished thin sections can also be used. Figure 7 indicates that the intensities of the RGB-colored heat maps do not show clear differences in elements ranging from Mg to Ca. The Fe heat maps display a decrease in intensities for the thin sections, but due to the large abundance of Fe in impactites, the main features (e.g., melt particles in suevites) can still be recognized and are therefore useful for the segmentation. The benefit of using µXRF maps on thin sections is that petrographic features such as shock metamorphism, in particular, mineral phases and the nature of a clast matrix or a groundmass (e.g., clastic, vitric, or microcrystalline microtextures), can directly be compared and studied. This allows for the segmentation of clasts not only based on chemical aspects, but also based on mineralogical features. Overall, the incorporation of geochemical micrometer-scale distribution maps as a basis for semi-automated digital segmentation unlocks a broad range of applications in geosciences. This technique produces quantitative textural data sets combined with geochemical information, in both a systematic and a time-efficient manner.

The µXRF SFP quantification on thick and thin sections exhibits a clear trend (Figs. 7–9). Our results indicate that the technique quantifies all elements identified in the spectra for the thick sections, thereby showing comparable results to the homogenized powders but with a lower precision in general, but especially for trace elements (visible as larger interquartile and box-whisker ranges in Fig. 9B compared to Fig. 9A). From Na to Mn, the SFP quantification for the thick sections displays good accuracy (slope ~1 and deviation <10% from Na₂O to CaO) compared to the bulk ICP-OES data (Table 1). For Fe and heavier elements, the µXRF method tends to consistently overestimate the concentrations, as indicated by the high positive deviation of several tens of percent (Figs. 8 and 9B; Table 1). As mentioned in the Results section, nickel is an important element to track meteoritical components in impactites (Goderis et al., 2012, 2013a). Our results for low-Ni-concentration (10–45 ppm) Chicxulub IODP-ICDP Expedition 364 samples exhibit the difficulty to properly quantify such low concentrations for this trace element (Fig. 8; Table 1). However, for homogenized powders and thick sections, higher Ni concentrations (50–300 ppm) are slightly overestimated (dolerite and felsite samples in Fig. 8; Pop 10 in Table 2). As many types of impactor are enriched in Ni (Tagle and Berlin, 2008) compared to most of the terrestrial target rocks, this implies that many impactites are Ni-rich. Therefore, the µXRF SFP quantification of Ni-rich samples (>50 ppm) is still relevant to provide first clues about the Ni concentration for further analysis. Regarding the thin section analysis, our results indicate that the technique also provides accurate data for elements ranging from Na to Ti (deviation <15%). However, for Mn and heavier elements, the SFP quantification underestimates the concentrations (deviation highly negative by several tens of percent) or is not consistent with the bulk ICP-OES and ICP-MS data (slopes and R² close to 0; Figs. 8 and 9C; Table 1).

As the thick and associated thin sections represent two different surfaces from the same core intervals, the main difference between the two sample types is their respective thickness, which is ~30 µm for thin sections and >1 mm for thick sections. X-rays penetrate the sample surfaces, and therefore the thickness of the sample and the composition of the nonvisible deeper phases influence the µXRF map results. Moreover, the attenuation depth of X-rays is different for each element and differs strongly depending on the type of geological material (Beckhoff et al., 2006). To explain the differences observed in Figures 7–9, the attenuation depths for a range of different geological materials were calculated (see Fig. A1 and Table A5 in the Appendix; Henke et al., 1993). The selection of these geological materials was based on the most important mineral groups within the IODP-ICDP Expedition 364 core determined by bulk powder X-ray diffraction (Morgan et al., 2017). In total, nine mineral groups were selected: calcite, quartz, muscovite, albite, anorthite, orthoclase, analcime, saponite, and clinochlore. Based on the raw attenuation depths and corresponding X-ray energies, the intercepts for each Kα peak of the elements from Na to Zr were calculated for the different mineral groups (see Table A5 and Fig. A1 in the Appendix). These intercepts correspond to the depth (in µm) that the X-rays need to reach in order for the sample to receive a full excitation and response. Considering a thin section thickness of 30 µm, the calculated intercepts indicate that for most of the selected materials, the µXRF detectors no longer receive a full response from the samples for the energies corresponding to the Cr-Kα and Mn-Kα peaks (5.4–5.9 keV). As the samples analyzed from the IODP-ICDP Expedition 364 core represent a mixture of the selected mineral groups, we conclude that SFP quantifications of elements from Na to Ti are applicable for µXRF maps of thin sections. Every element heavier than Mn or Fe cannot properly be analyzed because the intercepts of the Kα energies with the attenuation depths are larger than 30 µm. Detection of Fe based on µXRF mapping of thin sections remains possible, as shown in Figure 7B, despite an incomplete response due to a deeper calculated intercept. This observation is linked to the generally high abundance of Fe in impactite samples and the high fluorescence yield of Fe (Beckhoff et al., 2006). Therefore, Fe heat maps of thin sections can still be used to characterize proximal impactites, albeit not quantitatively. Another observation from these intercept calculations is that the analysis of thick sections should be carried out on samples with a thickness of at least ~1 mm to ensure correct µXRF quantifications of heavy elements such as Sr and Zr.

The advantage of µXRF mapping on thin sections is the possibility to directly compare the geochemical results with optical or scanning electron microscope images, or to use µXRF maps in combination with analytical techniques such as EMPA or LA-ICP-MS. However, a direct comparison of a thin section with a µXRF map of the corresponding counterpart slab often remains challenging due to the incomplete match between the two pieces as a result of the thin section preparation.

Figures 8 and 9 and Tables 1 and 2 demonstrate that the µXRF SFP quantification on thick sections leads to results that are comparable to both bulk ICP-OES/ICP-MS and bulk XRF for major elements (expressed as oxides, from Na₂O to Fe₂O₃). Therefore, we applied this quantification technique to characterize the two Popigai impactite samples, using spectra of both the bulk maps (Pop C6 and Pop 10) and the regions of interests for Pop C6. The latter are shown as annotated components (e.g., C1, B2, G3, M4, etc.) in Figure 5A. The geochemical results are compared to literature data in two ternary diagrams shown in Figure 10. Data for those ternary diagrams can be found in Table A4 in the Appendix. The µXRF bulk values of both samples fall within the range of concentrations determined by literature (Rabkin, 1959; Lutz, 1985; Vishnevsky and Montanari, 1999; Whitehead et al., 2002; Kettrup et al., 2003; Tagle, 2004; Tagle and Claeys, 2005) for their respective lithological unit (Fig. 10A). A similar observation is made for the impact melt particles M3 and M7 from Pop C6 (identified in Fig. 5A). All other impact melt particles analyzed in Pop C6 cluster together at slightly higher Fe₂O₃ and MgO concentrations relative to M3 and M7 (Fig. 10A). This difference between these two groups of impact melt particles can also be observed in the single-element heat maps (Fig. 3) and the photograph of Pop C6 (Fig. 1A). M3 and M7 are comparatively less porous and are characterized by entrained felsic basement clasts. Therefore, we interpret these clasts as microcrystalline impact melt particles that represent brecciated parts of clast-rich impact melt rock derived from melt bodies in the central part of the Popigai crater. The other melt clasts are interpreted as vitric impact melt particles. Their enrichment in Fe and Mg can be explained by the alteration of glass due to the formation of smectite, a common feature observed in many impact structures and also described in literature for Popigai (e.g., Masaitis et al., 2018). These vitric, altered greenish melt particles correspond to those found in the suevite described from the upper unit of the Popigai crater fill sequence (Naumov, 2002). The carbonate clast (C1, C2, and C3) compositions match well with the carbonate target rock (limestone) literature data (Fig. 10A; Kettrup et al., 2003). The groundmass area (G1, G2, and G3) concentrations fall within the range of the basement literature data but appear enriched in alkaline elements compared to the other clasts and bulk rock values (Fig. 10A). This could be due to the K-metasomatic hydrothermal overprint, which is known to affect large impact structures (e.g., Abramov and Kring, 2004) and has been observed at Popigai (Naumov, 2002). Figure 10B plots SiO₂ versus Fe₂O₃ + MgO + CaO and Ni, because Pop 10 and most of Pop C6 clasts display a Ni concentration above 50 ppm (even >100 ppm for most impact melt particles). The concentrations quantified by the µXRF SFP method for these oxides have been shown to be accurate (Figs. 8–9; Table 1), and for Ni, the concentrations are high enough to be meaningful (Fig. 8; Table 2). A good correspondence between the µXRF data and literature data is clear for the bulk Pop C6 and Pop 10, and for most of the impact melt particles of Pop C6, including the Ni values. The Ni concentrations of M3 and M7 are slightly lower than those of the other impact melt particles (Fig. 10B), which can be explained by larger felsic basement contributions to M3 and M7 (Fig. 3). The three groundmass areas analyzed are more enriched in SiO₂ than any other component within Pop C6 (Fig. 10B), which is also visible in the multi-element Fe-Si-Ca-K map and the Si heat map of the Popigai suevite sample (Fig. 3). This likely results from a significant incorporation and comminution of felsic lithologies in the groundmass, which are abundant within the different Popigai target rocks (Masaitis, 1994; Vishnevsky and Montanari, 1999).

The Chicxulub samples were used to demonstrate that the µXRF SFP quantification provides accurate and precise data for most major elements (Figs. 8–9; Table 1). This is confirmed by the good agreement between the data obtained for the Popigai samples and the literature data for the major elements and also for Ni (Fig. 10; Table 2). Therefore, we compared the data obtained by µXRF SFP quantification on the bulk maps of the six samples shown in Figures 1 and 4A (Pop C6, Pop 10, Aum-01, Pols-01, 54_1_64_66, and 90_1_85_87). The ternary diagrams used for this purpose (Fig. 11) are identical to those shown in Figure 10. The compositions of lithologies from the three impact structures result from the mixing between a mafic, a felsic, and a sedimentary carbonate component (Fig. 11A), which is linked to a distinct target stratigraphy (e.g., Popigai—Masaitis, 1994; Vishnevsky and Montanari, 1999; Ries—Pohl et al., 1977; Schmidt-Kaler, 1978; Kring, 2005; Chicxulub—Kring, 2005; Keppie et al., 2011). The µXRF measurements of all the six crater impactites fall within or in close proximity to fields of bulk data from literature values of Chicxulub (Kettrup and Deutsch, 2003), Popigai (Kettrup et al., 2003), and Ries impactites (Siegert and Hecht, 2018), respectively. The small deviation for the suevite sample from Popigai (Pop C6) can be explained by the incorporation of carbonate clasts in this sample. Relative to each other, the Ries impactites are enriched in a felsic component, the Popigai impactites are enriched in a mafic component, and the Chicxulub impactites are enriched in a carbonate component. For all three impact structures, the suevite samples are enriched in carbonate (as traced by elevated CaO concentrations) relative to the corresponding impact melt rocks (Fig. 11A). For these three impact structures, pre-impact sedimentary target rocks of different thicknesses (2.5–3 km for Chicxulub [Urrutia-Fucugauchi et al., 2011]; 1.5–1.7 km for Popigai [Vishnevsky and Montanari, 1999]; and 470–820 m for Ries [Schmidt-Kaler, 1978]) overlie crystalline basement lithologies. For all three craters, a carbonate component is part of the sedimentary cover, and for Chicxulub, the sedimentary target was composed of mostly carbonate and anhydrite. This explains the enrichment of the suevite samples in carbonate compared to the respective impact melt rocks, because the upper part of the impactite sequence incorporates more material from the upper part of the target stratigraphy. The main component of all these impactites is silicate melt, especially in the case of Ries (Fig. 11B). The Popigai impactites are more enriched in Ni than those at Chicxulub and Ries as described in the literature (Popigai—Tagle and Claeys, 2005; Chicxulub—Tagle et al., 2004; Ries—Siegert and Hecht, 2018). This observation can be explained by the presence of “up to several dozens of meters thick dolerite dykes and sills” (Kettrup et al., 2003, p. 735) within the Popigai target rocks, but it may also reflect a large meteoritic contribution (Tagle and Claeys, 2005). The two samples from Ries display similar Ni concentrations, which is in agreement with literature data (e.g., Siegert and Hecht, 2018). However, for Chicxulub and Popigai, the impact melt rocks are more enriched in Ni than the corresponding suevites. This is frequently observed for impact structures (e.g., Koeberl et al., 2007) and is generally interpreted to reflect a higher dilution of projectile components by (sedimentary) target rocks in the case of suevites (Goderis et al., 2012).

The µXRF SFP quantification method can accurately and precisely quantify major elements but tends to overestimate most of the trace-element concentrations in the case of thick samples. In the case of thin sections, the µXRF SFP quantification appears to be accurate and precise from Na to Ca, less reliable for Ti, Mn, and Fe, and not efficient in quantifying most trace elements. A comparison with other in situ semidestructive bulk techniques is presented in Table 3. For the purpose of this work, the limits of detection (LODs) for µXRF were estimated based on repeated measurements of the U.S. Geological Survey reference material basalt glass BCR-2G (Table 3). These estimates are conservative because the LODs strongly depend on the matrix and the homogeneity of the material studied (reference material or proximal impactites). Light elements (Na to P) have LODs ranging from ~7000 to 400 ppm, other major elements (K to Fe) show LODs ranging from ~70 to 20 ppm, and LODs for trace elements from Ni to Zr are all below 15 ppm (Table 3).

As a screening tool, µXRF outclasses scanning electron microscopy–energy-dispersive X-ray spectroscopy (SEM-EDS) for most of the characteristics, except the spatial resolution (Table 3). This study demonstrates that µXRF SFP quantification is a powerful technique, especially for the major elements. For a precise determination of trace-element concentrations, several in situ semidestructive techniques are recommended, such as LA-ICP-MS and synchrotron XRF (Table 3). However, the disadvantages of these techniques involve a smaller sample size, a longer sample preparation and measurement time, a necessity of standardization (not compulsory for synchrotron XRF), and a higher cost (Table 3). We therefore suggest to first perform µXRF mapping on geological samples as (1) a first geochemical characterization and (2) to be able to detect suitable areas for more in-depth research. Subsequently, these regions of interest on the sample surface can then be prepared for higher-resolution and more accurate measurements (including trace elements) by means of, for example, LA-ICP-MS.

The results presented in this study are based on the Bruker M4 µXRF instrument (installation in 2015 at the Vrije Universiteit Brussel; Fig. 1C), which uses intermediate-vacuum conditions (20 mbar) that make the mapping of lighter elements (from H to Na) difficult or impossible. However, later-generation µXRF instruments are now available that—due to better vacuum conditions—allow a more precise mapping of those elements (e.g., the Bruker M4 Tornado Plus with optional He-purge system to achieve <20 mbar vacuum conditions, which allow detection of C, N, O, F, and Na). Also, larger sample chambers exist, where, most of the time, the sample has a fixed position, and a single detector is moved across the sample surface (e.g., the Bruker M6 can acquire maps with an area of 80 cm × 60 cm at 100 µm resolution). Another difficulty associated with the XRF technique concerns the deconvolution of closely neighboring peaks. In the case of samples with, for example, high Fe contents and low Co peaks (or high Ti and low V concentrations), considerable data processing is required to obtain accurate Co results.

In general, the µXRF quantification can also be carried out by applying calibration curves, which require the incorporation of multiple homogeneous, matrix-matched reference materials. However, impactites are often heterogeneous in nature due to their complex emplacement mechanisms and the large variety of chemically distinct target lithologies incorporated in these samples. For example, the Pop C6 suevite consists of vitric impact melt particles intermediate to mafic in composition, granular to crystalline felsic basement clasts, and crystalline to mud-rich carbonate clasts. This wide range of clast matrices requires separate matrix-matched multiple-standard calibrations applied to the different groups of clasts. A multiple-standard calibration is not recommended for bulk quantification of impact breccia samples due to this large clast variability, and the µXRF SFP quantification can be considered to be sufficiently accurate and precise. The µXRF quantification of relatively homogeneous or simple impactites, such as clast-free to clast-poor impact melt rocks, could benefit from a fit-for-purpose matrix-matched calibration. Further research is needed in this field, as this endeavor would require the incorporation of a large set of international reference materials that can reflect the chemical variability encountered in target lithologies from preferably a large number of impact structures.

The µXRF technique presented in this study also has wider applications in other geoscience fields, such as the analysis of sedimentary or volcanic breccias, meteorites, and conglomerates. Such coarse-grained lithologies, often heterogeneous in nature, are particularly well suited for digital clast segmentation to quantify both bulk and clast-specific textural properties. In addition, the geochemical composition can be compared with these particle parameters such as clast size, shape, roundness, and degree of sorting. This implies that two extensive, independent data sets, derived from µXRF scanning and subsequent digital image analysis, can be obtained in a matter of hours.

In this study, we present an inexpensive, fast, high-resolution and nondestructive approach to geochemically characterize proximal impactites using µXRF. This technique allows rapid acquisition of the following data:

1. Qualitative multi-element maps: These maps make µXRF an efficient screening tool because simple multi-element maps (e.g., Fe-Si-Ca) are often sufficient to disentangle various lithological components.

2. Semiquantitative single-element heat maps: These maps allow the visualization of elemental distributions within the samples. Each heat map has a linear RGB-colored (0–255) scale, with the highest RGB value (255) determined by the pixel with the highest main fluorescence peak amplitude. When mapping multiple samples at the same time, this approach detects whether one sample is enriched in an element compared to another sample.

3. Bulk geochemical data and clast-specific compositional data based on a manual selection of regions of interest: As such, µXRF is unique as a bulk technique that also allows characterization of geochemically specific areas. The quantification can rely on the standardless fundamental parameters method, which is accurate for most major elements (with Na₂O–CaO being accurate within 10% compared to bulk powder techniques) but tends to overestimate most trace-element concentrations for samples thicker than 1 mm. Regarding thin sections, the quantification is also accurate from Na to Ca, but less reliable in the cases of Ti, Mn, and Fe, and not useable to quantify trace elements due to their higher attenuation depth of the X-rays. Alternatively, matrix-specific calibration curves can be constructed using international reference materials.

4. Quantitative image analysis, a semi-automatic tool of digital segmentation based on µXRF maps: Based on their geochemical composition, clasts can be segmented in an objective manner, unlike the use of visual characteristics (e.g., clast color). This technique results in an extensive quantitative data set of modal abundance, size, shape, and degree of sorting of the segmented clast population.

5. Bulk and clast-specific geochemical data superimposed on clast-specific textural data: When thin sections are mapped with this approach, the μXRF geochemical data set can also directly be integrated with mineralogical features. Such rapid, new workflows may also be adapted in geoscientific fields outside of impactite research to aid both in quantifying lithological components and in interpreting rock-forming processes.

In conclusion, we strongly recommend µXRF as both a visual screening tool and as a first quantitative geochemical tool for any new sample from known or potential impact structures.

We appreciate the valuable comments of reviewers Martin Schmieder and Lutz Hecht, which helped to improve this manuscript. The Analytical, Environmental & Geo-Chemistry Research Unit team is supported by Research Foundation Flanders (FWO grant G0A6517), Belgian Science Policy Office (BELSPO BRAINS), the Excellence of Science Program (ET-HoME ID 30442502), and Vrije Universiteit Brussel (VUB) Strategic Research Program. Pim Kaskes thanks FWO for the awarded Ph.D. Fellowship (projects 11E6619N, 11E6621N). Philippe Claeys thanks the FWO-Hercules foundation for financing the X-ray fluorescence analytical platform at the VUB. We thank Niels J. de Winter for valuable discussions about the technical capabilities of the µXRF instrument.

The material in the Appendix consists of μXRF major and trace element data sets that have been used in Figures 8 and 9 (Table A1–A3) and in Figures 10 and 11 (Table A4). In addition, Figure A1 and Table A5 show the attenuation depths of different minerals versus X-ray energies to provide additional explanation for the Discussion section “Differences between Thick and Thin Section Quantifications.”