With particles and their locations with respect to the embryo preserved in Formvar, this method then allows for the application of microscopy techniques to provide information on the physical and chemical characteristics of non-soluble particles within hailstones. Using an OLYMPUS LEXT OLS4000 confocal laser scanning microscope (CLSM), a 2-D cross-section of each of the V-7 and NG-1 hail samples (Figs. 2 and 3, respectively) was created by identifying the embryo and using it as a reference point for scanning regions within and around it. For these two cases, the embryo is identified as the original nucleus through its uniform crystallographic structure under polarized light, contrasting with the radially aligned outer layers . The core's small, equiaxed ice grains share nearly identical c-axis orientations, indicating it formed from a single frozen droplet or graupel particle . This uniformity distinguishes it from the outer shells, where elongated grains align radially outward due to wet accretion water freezing onto the surface;. The radial alignment of c axes in these layers points to concentric growth around the embryo, while the embryo's central position and structural simplicity confirm it as the starting point . The transition from a single-oriented core to layered shells reflects the hailstone's formation sequence, where the embryo acts as the nucleation site for subsequent accretion . Additionally, in porous hailstones, the distribution and size of air bubbles differ significantly between the central zone and the external lobes, providing another distinguishing characteristic between the embryo and outer layers . In the case of the 4 cm V-7 sample, the 2-D cross-section (Fig. 2) covers the embryo (Sectors 4–5) and outer layers in both directions from the embryo to the outermost layers of the stone (Sectors 1–3, Sectors 6–8). Within this 2-D cross-section, individual sectors of equal size (i.e., labeled Sectors 1–8 in V-7, Fig. 2) were selected for higher magnification to identify individual particles quasi-randomly within each sector with respect to the embryo. A similar approach was implemented for NG-1 (Fig. 3); however, owing to the larger size of this stone (8 cm compared to the 4 cm size of V-7), a second CLSM sweep was required to examine the entire area encompassing the embryo. This additional sweep resulted in horizontally adjacent sectors (e.g., Sector 1, Sector 2) in the same vertical section that were then grouped under a similar layer number for analysis (e.g., in Fig. 3, Sector 1 and Sector 2 represented the same layer of the stone and thus were identified as “1” when analyzing particles in that area). These numbered layers cover the larger embryo (1–5) and a cross-section toward one end of the stone identified as the outer layers (6–10).

This approach provided particle size distribution and surface topographical information for 73 and 223 identified particles within the V-7 and NG-1 hailstones, respectively, including particles down to the technique-limited minimum size of 1 µm . This lower limit is achievable due to the CLSM optical sectioning capability, pinhole aperture design, and high-resolution imaging, which allow for clear visualization of structures approaching the diffraction limit of light without breaking it. It is worth noting that this resolution differs from the SEM, which can achieve resolutions down to 50 nm due to its use of electrons rather than light, offering even finer detail for nanoscale structures. software was used to calculate particle size, which is determined as the maximum length along the x and y axes. The increase in particle numbers analyzed between the samples is attributed to the increased efficiency in the analysis, allowing for more particles to be examined in a similar amount of time.

The sublimated hailstone samples were then coated with gold (V-7) and gold and palladium (NG-1), the difference owing to the availability of pure gold during lab analysis times and subsequently analyzed using a Zeiss Sigma field emission gun (FEG) SEM with an EDS X-Max 80 mm² detector. Secondary electron images were acquired at 15 kV and a working distance of approximately 8.5 mm, settings that dictated the clarity and detail of the imagery and allowed for the identification of heavy metals without disintegrating the non-soluble particles . The 2-D elemental cross-sectional maps (Figs. 2 and 3) were used to locate the same particles observed in the CLSM to investigate the elemental composition of the same identified individual particles. To minimize interference from the sample substrate, EDS spectra were acquired for 120 s from the center of each particle using a single-point analysis technique. Particle-free areas of the glass were also measured using EDS to obtain a control spectrum from the coated glass substrate. While this single-point approach was more labor-intensive, it increased our ability to reduce noise and minimize spectral contamination from the glass substrate. Monte Carlo simulations conducted at 15 kV accelerating voltage showed that for particles smaller than 1.5 µm, some x-rays are emitted from the glass substrate, though this impact is minimized for particles down to 1 µm. To further ensure reliable results, we classified particles based on element identification rather than quantifying with weight percentage values, which helps account for potential variations in signal intensity across different regions of particles. Additional details on this process and considerations when choosing coating material and EDS analysis techniques are available in .

A unique feature of this technique is its capacity to link the elemental composition along with the size of individual particles with their location and proximity to the hailstone embryo. Due to these particles' small size and complexity, EDS spectra were primarily used for element identification, focusing on the presence and relative abundance of key indicator elements to interpret particle identity. This approach was chosen instead of calculating quantitative elemental abundances from EDS measurements for several reasons.

First, the particles were covered by a ∼1 µm thick slightly varying layer of Formvar (a carbon-based material) and further coated with a 10 nm layer of either gold (Au) or a gold–palladium (Au–Pd) conductive coating. Additionally, the particles rested on a soda–lime glass substrate containing elements like silicon (Si), aluminum (Al), sodium (Na), calcium (Ca), potassium (K), and oxygen (O), which contributed to the X-ray signals. While we aimed for spectrum deconvolution to account for signatures from both the coating materials and the substrate , we recognize further complications introduced by slight spatial variations in Formvar thickness and thus relied on elemental presence and relative abundance rather than absolute quantitative abundances for our particle classification.

To further justify our classification approach, we note that analyzing particles with irregular geometries presents additional challenges in trying to quantify abundances . EDS is fundamentally designed for flat, homogeneous materials when applied to small, polyphase, or irregularly shaped particles. Complications arise due to assumptions in the matrix correction and irregularities in X-ray production, emission, and take-off angles to the detector that impact quantifying elemental abundances from the EDS spectrum. Additionally, the EDS spectrum represents all X-rays produced within an approximate 2–4 µm³ activation volume at the beam's impact location. Given that many aerosol dust particles (as an example noted in Sect. 1) exhibit heterogeneity at the submicron scale, numerous EDS spectra will reflect polyphase materials.

For these reasons, we did not place high confidence in the absolute values of elemental weight percentages. Instead, EDS spectra were interpreted based on the presence of characteristic elements corresponding to specific mineral phases. More specifically, particles with high concentrations of carbon (C) and oxygen (O) were interpreted as organics, while others with a significant presence of nitrogen (N) were identified as nitrates. Lithic fragments were identified as pieces of eroded rock rich in Si, Al, Na, Ca, Mg, and Fe. Agglomerated carbon–lithic or carbon–lithic–clay fragments showed signatures rich in C, Si, Al, Na, Ca, Mg, and Fe, representing potential organic mixtures with dust and other particles in the region (referred to as agglomerated mineral/organics). An example of this agglomerated mineral/organics classification is shown in Fig. 4a, characterized by strong Si, Al, Na, Ca, and Mg peaks in its EDS spectrum. The addition of CLSM, which provides information about particle color and general morphology, and SEM's high-resolution surface imaging with the EDS's elemental analysis allows for particle classification that considers both chemical composition and physical structure. Here (Fig. 4a), the CLSM image reveals a bright, irregularly shaped particle against the dark background. In contrast, the corresponding SEM image shows its complex three-dimensional structure with multiple crystalline components, supporting its classification as an agglomerate.

The middle panel (Fig. 4b) demonstrates a Fe-rich organic particle with dominant O and Fe signals. The CLSM image shows a reddish-brown coloration characteristic of Fe-rich particles, while the SEM reveals a smooth, folded surface with sharp-edge structures. This spectral and morphological evidence combination supports its classification as a Fe-rich organic particle. The right panel (Fig. 4c) reveals a salt particle with prominent Cl and Na peaks, characteristic of halite. The CLSM image shows a dark, well-defined particle. In contrast, the SEM image displays an irregular shape with uneven edges and lacks the symmetrical, straight-edge characteristics typically associated with cubic morphology, such as those found in halites. This elemental classification scheme also identified distinct minerals including quartz and calcite. We also detected particles containing unique indicator elements, such as groups of zinc chloride, copper zinc chloride, or copper chloride particles, with EDS spectra rich in Zn–Cl, Zn–Cu–Cl, or Cu–Cl, respectively.

The NOAA Air Resources Laboratory's Hybrid Single-Particle Lagrangian Integrated Trajectory model (HYSPLIT; ) was first used to generate a 24 h air mass back-trajectory using 0.28°×0.28° European Environment Agency Reanalysis datasets ERA5; with a vertical resolution of 37 pressure levels: 25 hPa intervals from 1025 to 750 hPa, 50 hPa intervals from 750 to 300 hPa, and 25 hPa intervals from 275 to 100 hPa. Trajectories were initiated at 17:00 UTC on 8 February and 22:00 UTC on 13–14 December, with hourly intervals starting at heights of 100, 500, 1000, and 1500 m a.g.l. (above ground level) from convective core coordinates on 8 February (64.75° W, 31.59° S, marked by a red star in Fig. 1) and on 13–14 December (64.45° W, 32.20° S; marked by a blue star in Fig. 1). These initiation coordinates were identified using channel 11 (8.4 µm) from the geostationary satellite GOES-16 .

These initiation height levels were chosen for the following reasons: 100 m is the lowest point to the surface and the first pressure level, 500 and 1000 m provide boundary layer variability, and 1500 m focuses on possible particle transport influences from the low-level jet . A trajectory matrix with a 7×5 grid and 0.30° spacing was also processed from the initial convective pixel coordinate at 17:00 and 22:00 UTC (on 8 February and 13–14 December, respectively) for 5 d to better understand which local and longer-range sources could have transported particles at the initiation location before hail-producing convection was observed. The area covered by all back-trajectories resulting from the matrix was divided into grid cells with dimensions of 0.28° longitude and 0.28° latitude (i.e., ERA-5 horizontal resolution). Each trajectory occurrence in each grid cell was then normalized based on the time spent over each grid cell and included trajectory endpoints for all the heights ERA5;. Residence-time coefficient pixels were overlaid on the C3S Land Cover classification gridded map from 2023, as shown in Fig. 1. This map provides a global description of the land surface divided into 22 classes, available through the C3S Climate Data Store and defined using the United Nations Food and Agriculture Organization's Land Cover Classification . This approach provides insight into the highest probability of specific land-use regions being possible source regions for the non-soluble particulates analyzed in this study.

This V-7 hailstone analyzed in this case included particles analyzed in both the embryo and the outer layers through a vertical cross-section of the 4 cm stone (Fig. 2). The CLSM analysis of V-7 revealed particle sizes ranging from 2 to 150 µm (Fig. 5), with an average particle size of 30 µm. This indicates that particle sizes vary, with many particles being much smaller or larger than the average particle size. The minimum size measured was 2 µm, close to the minimum observable particle size using this method 1 µm;. When focusing on particles within the embryo (22 particles), sizes ranged from 6 to 150 µm. In contrast, the particles we sampled in the outer layers (51 particles) were smaller over a narrower range, encompassing sizes from 2 to 122 µm. While the comparison between the embryo and outer layers in V-7 suggests that the embryo contained larger particles than the outer layers, a key point here is that both regions contained particles exceeding 100 µm.

With a large range of particle sizes identified in the hailstone, the elemental classification scheme described in Sect. 2.3 was applied to this hailstone sample to determine the compositions of these individual particles. Overall, the particles in V-7 were primarily characterized as agglomerated minerals/organics (∼41 %; Fig. 6a; Table 1) and distributed throughout the hailstone (Fig. 7). The particle sizes in this category ranged from 6 to 75 µm in the embryo (9 particles) and 2 to 55 µm in the outer layers (21 particles).

The second most prevalent group was agglomerated salts (∼16 %; Fig. 6a; Table 1), characterized by strong Cl and Na signals in EDS analysis and distributed throughout the hailstone (Fig. 7). The 10 particles in this category analyzed in the outer layers ranged from 5 to 122 µm in the outer layers. Of the 2 particles of this type sampled in the embryo, one was 13 µm, while the other was the largest particle recorded in this group and in the entire V-7 sample at 150 µm. Interestingly, even though this large particle was identified as a salt, it did not support a large single crystal but an agglomeration of smaller crystals (not shown). Instead, this particle exhibited a strong carbon peak, indicating a small salt particle atop a potential agglomerated mineral/organic particle.

At 8 % frequency, Zn–Cl-rich particles (Fig. 6a; Table 1) consisted of Zn–Cl content greater than 1 % by weight. A single particle of 33 µm was found in the embryo, while particles in the outer layers ranged from 7 to 25 µm (5 particles; Fig. 7). Organic particles (∼8 %; Fig. 6a; Table 1), containing more than 60 % weight of carbon and rich in oxygen, were found in both regions with sizes ranging from 22 to 33 µm in the embryo and 10 to 23 µm in the outer layers, with 3 particles in both regions (Fig. 7). Similarly, at 8 % frequency, clays (Fig. 6a; Table 1), characterized by high aluminum content and the presence of Si, K, Ca, Na, Mg, and Fe, were composed of 2 particles in the embryo of sizes 20 and 55 µm and 4 particles analyzed in the outer layers ranging from 4 to 47 µm.

Shifting to the larger-than-average particle sizes found in V-7 are quartz particles (∼7 %; Fig. 6a; Table 1), identified as silicon content greater than 30 % by weight, with 2 particles analyzed in the embryo of sizes 16 and 22 µm and 3 in the outer layers ranging 38 to 55 µm in size. Ammonia/nitrate particles (∼7 %; Fig. 6a; Table 1), which were nitrogen-rich with content greater than 7 % by weight, included 3 particles in the embryo ranging in size from 37 to 58 µm and 2 in the outer layers (31 and 45 µm). Two Fe-oxide particles (∼3 %; Fig. 6a; Table 1), dominated by iron and oxygen with no carbon or silicon, were located in the outer layers of the hailstone with sizes of 33 and 92 µm (Fig. 7). Although Fe oxides were only detected in the outer layers in this particular swath, a second swath conducted in a different direction across the hailstone (not shown) identified Fe-oxide particles in the embryo as well. This trend for all particle categories to be found throughout the whole sample extended to the final identified category, lithics (1 %; Fig. 6a; Table 1), characterized by a silicon content of 20 %–40 % weight, along with high concentrations of Na, Mg, and Ca; minor amounts of Al and K; and no detectable carbon from the Formvar. Only one particle, measuring 5 µm, was identified in this category and located in the hailstone sample's outer layer (Fig. 7); however, the second swath (not shown) revealed additional lithic particles within the embryo.

By analyzing the elemental composition of individual particles within the hailstone, the dominance of large, agglomerated particles (including those with salt) in the sample suggests a link to local land use, including hypothesized agricultural and biological sources. While particles of up to 100 µm are likely to remain suspended in the atmosphere for up to 2 d , the largest of the particles observed within the V-7 hailstone (i.e., 150 µm agglomerated salt in the embryo) were likely ingested into the storm from local sources near the SDC. The HYSPLIT back-trajectory analysis provides further insight into these links.

The analysis of the 24 h HYSPLIT back-trajectory (Fig. 8a) provides information on the possible origins of the particles in the V-7 hailstone sample with distinct patterns at different altitude levels. Levels 100 and 500 m a.g.l. show trajectories near the initiation point from the northwest but earlier in the 24 h period had curved over the northeastern part of the SDC, likely owing to the upslope flow of surface winds from the northeast . This upslope flow is supported by the terrain and height change of the trajectory seen in Fig. 8a (bottom figure). The 1000 and 1500 m a.g.l. trajectories came from the north and northwest near the initiation point but, moving back in time, showed a slight curvature north and northeast of the Cordoba Province at distances farther than the lower levels. Within this short time frame, the SDC will likely impact the particle sources found in the hailstone. However, longer-range transport of particles prior to 24 h may also have deposited particles in the area that could have been ingested into the 8 February supercell.

To investigate potential source areas further back in time, the residence-time coefficients calculated for the 5 d back-trajectory matrices (Fig. 8b) similarly highlight the regional influence indicated in the 24 h trajectories in that most trajectories lie within the Argentinian borders. More specifically, under the assumption that the particles arriving at the location where the hail-producing storm initiated are more likely emitted from regions where the air masses spent more time , the grid cells showing the high-residence-time coefficients (any grid that includes more than one HYSPLIT trajectory) are considered potentially principal sources for the particles found in the hailstone. The regions with the highest-residence-time coefficients for 5 d leading up to the V-7 storm initiation (orange, yellow, and red in Fig. 8b) are generally located over the SDC, city of Córdoba (white star in Figs. 1 and 8b); Argentina's largest natural salt lake (Laguna Mar Chiquita; 30.71° S, 62.56° W); and provinces such as Santiago del Estero, Chaco, Santa Fe, and Corrientes (see Fig. 1 as a geographical reference). These results reveal that sources within Argentina's geographical limits account for possible non-soluble particle sources in the analyzed hailstone sample. More specifically, in comparing the high-residence-time coefficient pixels from the 5 d back-trajectory with the C3S Land Cover map (Fig. 8b), we find the most predominant land uses (Fig. 8c) corresponding to shrublands, croplands, mixed vegetation, urban areas (mostly the city of Córdoba), and areas with a body of water (including the aforementioned salt lake), consistent with the imprints of soil dust, organics, salt, and minerals found throughout the V-7 hailstone, including the embryo, and emphasizing the importance of local land use on INP sources of hail formation.

For NG-1, particle size ranged from 1 to 256 µm (Fig. 9), with an average particle size of 35 µm. The largest particle in NG-1 was ∼100 µm larger than the largest in V-7, and the average particle size was also larger in NG-1 (35 µm) compared to V-7 (30 µm). This shows that both hailstones' particle sizes had large variations. When focusing on particles within the embryo (106 particles) and the outer layers (105 particles), large particles (i.e., >100 µm) were found throughout the hailstone, with large size variability observed in both regions. Like V-7, the elemental classification scheme applied to NG-1 showed that the particles were primarily composed of agglomerated minerals/organics (∼36 %; Fig. 6b; Table 1), with sizes ranging from 4 to 77 µm in the embryo (33 particles) and 9 to 109 µm in the outer layers (42 particles) and were similarly found throughout the NG-1 hailstone sample (Fig. 10).

While one relatively small lithic particle was identified in V-7 (Table 1, Fig. 6), lithics were the second most prevalent group in NG-1 (∼22 %; Fig. 6b; Table 1), with particle sizes ranging from 5 to 58 µm in the embryo (26 particles) and 8 to 181 µm in the outer layers (20 particles; Table 1), and were distributed throughout the hailstone, with 26 particles in the embryo and 20 in the outer layers (Fig. 10). Organic particles were also more prevalent in NG-1 (∼15 %; Fig. 6b; Table 1) than in V-7 (∼8 %; Fig. 6a; Table 1). This category contained the largest particle size found in NG-1, ranging from 9 to 256 µm in the embryo (22 particles) and 13 to 216 µm in the outer layers (10 particles), also distributed throughout the hailstone (Fig. 10). Similar to V-7, quartz particles were found throughout the hailstone (Fig. 10), with sizes ranging from 10 to 62 µm in the embryo (8 particles) and 3 to 69 µm in the outer layers (6 particles; Table 1). While less prevalent, ammonia/nitrate particles were also detected in NG-1, ranging from 23 to 36 µm, and found exclusively in the outer layers with 4 particles (Fig. 10; Table 1).

Despite the consistency in particle elemental categories found within both hailstones, some categories were unique to NG-1. Cu–Zn–Cl-rich particles (∼4 %; Fig. 6b; Table 1), with particle sizes ranging from 6 to 78 µm, were exclusively located in the outer layers with 8 particles (Fig. 10). A single Cu–Cl-rich particle (∼1 %; Fig. 6b; Table 1) measuring 32 µm was found in the outer layers (Fig. 10). Calcite particles (∼2 %; Fig. 6b; Table 1) showed size ranges of 12 and 24 µm in the embryo and 8 and 13 µm in the outer layers, with 2 particles in each region (Fig. 10). Finally, a single brass particle (∼1 %; Fig. 6b; Table 1) measuring 1 µm, characterized by strong Cu and Zn signals in EDS analysis, was found in the outer layers (Fig. 10).

Overall, while both hailstone samples showed similarities in certain categories such as quartz, nitrate particles, and Fe oxides (with NG-1 showing particles of 21 µm in the embryo and 61 µm in the outer layers), there were notable differences in the presence and distribution of lithics, organics, Cu–Zn–Cl-rich, Cu–Cl-rich particles, brass, and calcite, as well as the presence of salts in V-7 but their absence in NG-1. Were these different particle compositions linked to different trajectories associated with these different storms, including more non-local sources on 13–14 December? To explore this question, the HYSPLIT trajectories are again analyzed for the 24 h and 5 d time frames.

The 24 h HYSPLIT back-trajectory analysis was applied to the 13–14 December case (Fig. 11a) to determine potential source regions of the particles found within the NG-1 hailstone sample. Nearest the initiation point, the trajectories originate from the west and northwest at all analyzed levels. In particular, trajectories curved around the northern section of the SDC, moving upslope along the western side of the southern section of SDC instead of the eastern side as on 8 February. The 1500 m a.g.l. levels followed a similar path to that of the lower-level winds but started at regions farther north than the lower trajectories. This trajectory height and the direction are likely linked to the SALLJ and thus could have led to more remote sources of particles over a longer time period.

Therefore, residence-time coefficients for 5 d back-trajectories matrices were again calculated (Fig. 11b) to explore potential sources outside of Argentina's geographical limits. Unlike on 8 February, when the trajectories were primarily within Argentinian territories, on 13–14 December, air mass trajectories with high residence times also came from Paraguay and Brazil. Under the same aforementioned assumption that the particles arriving at the location where the hail-producing storm initiated are more likely emitted from regions where the air masses spent more time, the regions with the highest-residence-time coefficients (orange, yellow, red, and lighter shades of blue in Fig. 11b) were generally located northwest of the SDCs, in the boundary between Cordoba, the southwestern corner of La Rioja and San Luis provinces (Fig. 1), and in locations outside of Argentina. While the main land-use categories associated with the highest residence times are generally the same for both cases (i.e., Figs. 8c and 11c), the distribution shifts on 13–14 December from pixels associated with shrublands to more pixels associated with both cropland and mixed vegetation (Fig. 11b). The dominance of the cropland category again is consistent with the prevalence of agglomerated mineral/organics, but the trajectories from either the west side of the SDC or the regions north of Argentina could explain unique source regions to NG-1 resulting in particle elemental compositions not observed or more prevalent than in V-7 (e.g., lithics, Cu–Zn–Cl-rich, Cu–Cl-rich, brass, calcite). There were also low-residence-time pixels over the salt lake (Fig. 11b), potentially explaining why no salt was found in NG-1.