The color systematics of volcanic ashfall samples in estimating eruption sequences: a case study of the 2017–2018 eruption at Shinmoe-dake, Kirishima volcano, Southwest Japan

The color of pyroclasts is fundamental, because it reflects various magma properties and eruption processes, including particle morphology, chemistry, and petrological characteristics. However, deriving the componentry ratio (CR) of pyroclasts for ongoing eruption monitoring remains challenging due to the lack of a robust quantitative standard. The derivation of the CR, as well as other petrological analyses, is too laborious and time-consuming to introduce as a sustainable monitoring method. To address this, we employed spectroscopic colorimetry to rapidly and quantitatively describe eruptive product colors, enabling CR derivation based on clear, objective standards for ash particle classification. Through color spectroscopy of bulk and sieved ash samples, we analyzed the major size fraction for time-series samples during the waxing stage of the 2017–2018 Shinmoe-dake eruption in Kirishima volcano, Southwest Japan. Our findings reveal that the color changes in bulk ash systematically changed with the evolution of componentry. This temporal color change was due to an increase in the amount of vesicular particles with clear glass against dark angular lava particles, as well as a grain size change, which we interpret as an indication of a transition from phreatic/phreatomagmatic to magmatic eruption. Subsequently, the color of the ash changed when the amount of different lava particles increased gradually, coinciding with a shift toward a more dominant effusion of lava. As the lava effusion continued, a slight reddening of the ash, indiscernible to the naked eye, was clearly detected by the spectrometer before the onset of intermittent Vulcanian eruptions. We interpreted this reddening as oxidation resulting from decreased ascent speed and caprock formation, which accumulates overpressure for Vul-canian explosions. These results highlight the potential of rapid, objective color value and componentry derivation for sustainable quasi-real-time monitoring of ongoing eruption materials.


Introduction
The most characteristic features of volcanic eruptions are the diversity of eruption styles and their complex temporal evolutions or eruption sequences.Our primary effort has been to elucidate why eruption styles change over time by combining various interdisciplinary approaches.Reconstructing the sequence of deposits formed by eruptions is the first step in constraining the sequence of eruptive phenomena over time.This first step is also exceedingly crucial, because sequential sampling of the products is conducted based on this stratigraphy for further analysis to elucidate the evolution of magmatic systems and fundamental magma ascent processes in conduits, including degassing, fragmentation, and crystallization, which control the eruption style and its transition.However, such sequential sampling has only been possible in the past for larger eruptions (VEI > 4), because the deposits from smaller eruptions are soon bioturbated, eroded, or entirely lost by rain or wind.This is not only true for eruptions in the past, but it is currently also substantially more difficult to obtain samples in the proximal area without risk during large eruptions to protect the lives of surveyors.In addition, to achieve the goal of forecasting devastating eruptions or paroxysms and mitigating the disaster as early as possible before they occur, sampling after the paroxysms is insufficient, and sampling before the paroxysms is more desirable to capture precursory signs or investigate processes that may evolve into paroxysms in the near future.To obtain constraints on the transition into such paroxysms, direct or in situ continuous monitoring in distal areas is fundamental for sustainably capturing the evolution from the beginning toward the paroxysms at safe sites.In this context, volcanic ash samples are ubiquitous in all types of eruptions.Several large explosive eruptions result in lapilli deposits during the climactic phases, but they also experience waxing stages with ash-emitting activity at the beginning phases.We have the capability to obtain ash samples from distal areas, even amidst paroxysms, when access to the proximal area, where lapilli-sized ejecta is actively deposited, becomes inaccessible.Thus, ash samples can be obtained more frequently than lapilli samples.These time-series ash samples may provide insights into the sequence of eruptions and the processes that control eruption phenomena, particularly during the waxing phases of a volcanic eruption.
The component ratio (CR) or componentry of volcanic ash particle samples has been used as an indicator to characterize explosive volcanic eruptions (Taddeucci et al. 2002;Ohba et al. 2007;Suzuki et al. 2013;Cassidy et al. 2015).The CRs of ash samples are expected to provide information on the fragmentation of magma, degassing, or the environment surrounding the vent systems, although this is not yet clearly understood.Thus far, the derivation of CR has been relatively laborious, and the standard for classification criteria depends strongly on the experience of the investigators, which would be arbitrarily based on features such as the shape, color, and/or transparency of the particles.Several studies have traditionally focused on the shape of volcanic ash (Heiken and Wohletz 1985;Dellino and La Volpe 1996), because the shape is directly manifested by vesiculation, fragmentation, aggregation, or abrasion during magma ascent in conduits and/or ash plumes, which are speculated to control eruptive phenomena.Recent studies have summarized methods for processing and analyzing ash samples to quantify shape parameters and have proposed a protocol to characterize them under the same standard for comparative studies (Liu et al. 2015;Ross et al. 2022;Comida et al. 2022).Shoji et al. (2018) introduced a convolutional neural network (CNN) as an objective method for classifying volcanic ash particles.However, there is no consensus on the standard that should be used for classification during componentry.
The color of volcanic ejecta has also been studied as an indicator of transitions in eruption styles or changes in the environment surrounding vent systems.This is because mineral phases have unique colors depending on their chemistry and molecular structure (Kennard and Howell 1941;Nassau 1978), and the processes during magma ascent toward the surface could change the mineral species and their composition through reactions with the melt under possibly changing fugacity conditions (Miyagi and Tomiya 2002;Yamanoi et al. 2008;Moriizumi et al. 2009;Miyagi et al. 2013).Despite pioneering studies, the colors of natural volcanic samples are often not described objectively, even with available standards like color charts (Geological Society of America 1964).Typically, they are described subjectively under less controlled conditions concerning the light source, sample size, and state.Miwa et al. (2015) introduced image analysis using three filters under an optical microscope, which would have been a primitive model of microspectroscopy introduced in this study.
In this study, we introduce color spectroscopy and visible microspectroscopy for ash samples and particles, respectively, to describe the color quantitatively and objectively.We describe time-series samples collected during the initial phases of the 2018 eruption at Shinmoe-dake, Kirishima volcano, Southwest Japan, which were successively sampled on-site during the eruption.During this time, the eruption style changed from phreatic to a magmatic eruption of lava effusion, accompanied by nearly continuous ash venting, followed by several intermittent Vulcanian explosions (Matsumoto and Geshi 2021;Maeno et al. 2023).
In this study, our approach involves introducing a quantitative color description for the sequential products of well-documented eruptions as follows: (1) we provide a rapid and objective sustainable method for monitoring eruption products; (2) we demonstrate a colorimeter as a tool to detect tiny and short phenomena that might be missed by other methods; and (3) we propose CR as a robust and comparable index for eruption sequences with quasi-real-time observations using visible microspectroscopy.

The 2017-2018 Shinmoe-dake eruption, Kirishima volcano
Shinmoe-dake is an active center of the Kirishima volcanic complex in Southwest Japan (Fig. 1).The Kirishima volcanic complex consists of several active and less active volcanic edifices that spread sporadically from northwest to southeast.Mt.Shinmoe-dake is located primarily in the central part of the complex and has a conical edifice approximately 3 km in diameter with a crater diameter of approximately 850 m.Before 2017, the crater was occupied by a circular lava lobe that effused after three sub-Plinian pulses of the 2011 eruption.The total amount of magma erupted in the 2011 eruption was estimated to be 2.1-2.7 × 10 7 m 3 dense rock equivalent (tephra of 0.7-1.2× 10 7 m 3 + lava of 1.5 × 10 7 m 3 ).This was a low-end VEI 3 eruption (Nakada et al. 2013).
After 6 years of dormancy, Shinmoe-dake erupted in mid-October 2017, primarily at the eastern rim of the 2011 lava lobe.Continuous to intermittent phreatic eruptions lasted for approximately 1 week (October 11 and October 15-16) and thinly dispersed white to light-gray fine ashfall around the volcano, reflecting low plume height and frequent changes in weak winds during the eruption.Most of the deposits were wet because of rain during this period.The total mass of the tephra fall of the 2017 eruption was estimated to be in the order of 10 7 -10 8 kg (Oikawa et al. 2018).
At the beginning of March 2018, a small ash eruption began 4 months after dormancy since the 2017 eruption (Matsumoto and Geshi 2021;Maeno et al. 2023).On March 1, 2018, an eruption started at approximately 11 a.m. from the east vent of the 2011 lava field inside the Shinmoe-dake summit crater.This ash venting activity (Phase 1) (Matsumoto and Geshi 2021) continued intermittently with increasing intensity until March 6, when a new lava flow effused from the same crater until March 9 (Phase 2).During Phase 2, a rapid tilt change and seismicity increase (LP) were observed in the morning (approximately 10:00) on March 6, which would have recorded lava extrusion or edifice deflation, accompanied by higher ash clouds than before (Yamada et al. 2019).The tilt change and high-level LP seismicity continued until March 9, and the total lava flow volume was estimated to be 1.4-1.5 × 10 7 m 3 (Chiba et al. 2018;NIED 2018).After March 8 (Maeno et al. 2023), Vulcanian eruptions occurred intermittently at several points in the lava field (Phase 3) until the end of June; however, their frequency gradually decreased.

Samples and analysis methods
Samples from the 2017 eruption were obtained during the ashfall from October 11 to 16, 2017.For the 2018 eruption, samples were collected from March 2 to 7, 2018, when the eruption intensified toward the climactic phase (Table 1; Fig. 2).Most of the samples from the 2017 eruption and during the waxing stage of the 2018 eruption were collected upon fallout on-site using flat stainless trays with good time and area precision (ashfall weight of 500 cm 2 per several min), as well as without loss of the fine fraction by wind immediately after deposition.The samples were collected from different sites, because the direction of the dominant wind changed gradually.During the October 2017 eruption, the distribution direction changed frequently during most rain events; however, we collected samples at sites 4-7 km from Shinmoe-dake (Fig. 1).In March 2018, we collected data from the east (before March 6 AM) to the south (evening of March 6) toward the northwest (March 7 AM) from the summit.Most sampling sites were in the distance range of 4-9 km from the summit of Shinmoe-dake; therefore, most samples were ash-sized.We also analyzed certain samples collected after the day of deposition, before March 5, and after March 8, as well as samples from the 2008 and 2011 eruptions, for comparison (Table 1).
The analytical procedure used in this study was as follows (Fig. 3): Dried bulk ash samples (Fig. 2a) were first analyzed by colorimetry and manually sieved into five to six size classes (> 1000, > 500, > 250, > 125, and > 0 μm; > 64 μm occasionally).The color of the sieved samples was analyzed using the same method as that of the bulk samples if sufficient amounts were available (> 0.5 g).Some samples, ranging between 500 and 250 μm, were then classified into several types under a microscope and manually counted for componentry by one investigator (TS).We then performed visible microspectroscopy on each particle in each sample.At least 100 grains were counted or analyzed using visible microspectroscopy for each sample.

Bulk colorimetry by colorimeter
Bulk colorimetry was performed using a handheld colorimeter (CM700-d, Konica Minolta, Inc., Tokyo, Japan) for dry ash samples in colorless transparent glass bottles (LABORAN Screw Tube Bottle; AS-One Corporation, Osaka, Japan) (Fig. 3).The Commission Internationale de l'Éclairage (CIE) recommended the CIE L*a*b* color system in 1976 as a color standard for objectively describing objects (Wyszecki and Stiles 1982).The CIE L*a*b* color system quantitatively and objectively describes colors, Fig. 1 Locality map of the Kirishima volcanic complex, Southwest Japan (modified from a GSI map).The sampling points are shown as solid boxes with numbers (Table 1) as follows: red and black numbers: 2018 ash samples; blue numbers: 2017 ash samples; black numbers with squares: 2011 ash samples (Suzuki et al. 2013).The broken circular lines represent the distance from the summit of Shinmoe-dake.The thick, broken lines indicate areas mentioned in the text  *Site numbers shown in Suzuki et al. (2013) and comparisons between samples are psychometrically consistent.The L* value is the psychometric lightness; black corresponds to L* = 0, and white corresponds to L* = 100.The a* and b* values represent psychometric chromaticism.A positive value of a* corresponds to red, and a negative value corresponds to green.A positive b* value corresponds to yellow, and a negative value corresponds to blue.The analysis conditions were maintained constant and were primarily the same as those described by Moriizumi et al. (2009).We adopted a 10° view angle and the CIE standard illuminant D65 (indirect sunlight at 6500 K) as the measurement conditions.After white calibration, color measurements of the samples were performed directly using a transparent glass container, as stated above.The effective analytical area of the spectrophotometer had a diameter of 8 mm.We measured each sample in triplicate from below the bottle's bottom side, and after each measurement, the sample was shaken to shuffle it, ensuring that the average value obtained reflected the randomly mixed ash.Colorimetry of the sieved samples was performed similarly for each size fraction if enough amount of sample was available (approximately > 0.5 g).We used two optical configurations of the spectrometer for measurement: specularly reflected light included (SCI) and excluded (SCE) conditions simultaneously.Typically, the differences between SCI and SCE conditions are affected by the surface condition.However, we only present the results from SCI, as there was no substantial difference in relative variation between the two datasets (Additional file 1).

Visible microspectroscopic colorimetry
We used a binocular reflectance microscope (Nikon SMZ800) with a visible light spectrometer at wavelengths of 400-700 nm (Fig. 3) (Glacier X, BWTECK Inc., Plainsboro, NJ, USA) (Yamanoi and Nakashima 2005).As noise signals are large in the 400-450 and > 700 nm ranges due to the less controlled light path in the microscope, we used signals of 450-700 nm for the analyses in this study.
We used a light-emitting diode light source with a dome shade (DM160; Wraymer Inc., Osaka, Japan) to minimize background noise.Each spectrum (reflectance % or R%) was obtained by subtracting the background from the raw intensity spectrum and dividing by the white standard (CNo.BWT110102, BWTECK Inc.).Subsequently, the normalized spectrum (normalized R %) was obtained by dividing the spectrum by its average intensity in the wavelength range of 450-700 nm to minimize the effect of grain size.The componentry of the samples was determined by minimizing the residual sum of squares (RSS) of the normalized R % between the unknown and several reference particles that were previously selected as characteristic types of particles in the studied samples.As the color values of the reference particles also varied, we analyzed 20 particles per type for the 11 reference types to obtain the average spectra along with their standard deviations.Further details about microspectroscopy are presented in the results and discussions chapter.

Grain size and color of bulk and sieved samples
The grain size distributions (GSDs) of the representative March 2018 ash samples are shown in Fig. 4. Most of the 2017 ash samples were fine-grained, with over 70 wt.% of samples being < 125 μm in diameter.Although the 2018 ash samples were not collected at the same locality, the median grain sizes were relatively in the narrow  2018 (2018c30).sco: black scoria, gry lava: gray lava, lgt gr lava: light gray lava, alt: altered lava, pl: plagioclase, px: pyroxene, gr pu: gray pumice.Black arrows indicate pyrites range of 1-3 phi (phi scale is defined by d = (1/2) φ = 2 −φ where d is the diameter of the grain in millimeters) and showed a gradual increase from March 2 (sampled at 6:17 on March 3) until the evening of March 6, 2018, and then decreased toward noon on March 7.The 2011 ash samples were different from each other, because they originated from different eruption styles (sub-Plinian or Vulcanian) and were collected at different localities at distances from approximately 3 km up to > 20 km from the summit.In particular, samples Pre10 and 11, collected at sites 10 and 11 during the sub-Plinian phase, were depleted of fine components (Table 1) (Suzuki et al. 2013).
The relationships between the color values of the bulk and sieved samples are shown in Figs. 5 and 6 in the CIE L*a*b* color system.The temporal changes in each color value are shown in Fig. 7.Although there was overlap among the samples from different years, the color values of the bulk ash samples showed systematically different trends between the 2017 and 2018 eruptions, and they also changed over time during each eruption (Fig. 5).Most of the 2017 samples (blue symbols in Fig. 5) had higher L* values and lower a* and b* values than those of the 2018 samples.The 2017 samples showed a slightly negative correlation of L* with a* and b* values, whereas the a* and b* values showed a positive correlation, although they were confined within an exceedingly narrow range.Over time, the L* value decreased, indicating a change in color from bright to dark.This is consistent with the observed increase in coarse grains, although not as abundant (Fig. 2b).The bulk ash samples at the beginning of the 2018 eruption (red solid triangle symbols enclosed by a green circle with "Mar6AM" in Fig. 5b) were also in the same color range as those of the 2017 samples until 15:00 on March 6 (Phase 1 to early Phase 2 of Matsumoto and Geshi 2021), and the L* value decreased from > 53 on March 2-4 to 52 on March 5 (Fig. 7b).However, the L* value began showing a relatively different positive correlation trend with the a* and b* values on the afternoon of March 6 (red arrow in Fig. 5b).Subsequently, on that evening (17:00-19:00), all three values decreased, exhibiting a similar trend.On the morning of March 7, however, the a* and b* values were higher than before, with a moderate L* value (orange arrow and circle with "Mar 7AM" in Fig. 5b).
Fig. 3 Schematic illustration of the procedure and the system of spectroscopic analyses in this study.There are three steps of analysis: 1 color spectroscopy of bulk ash samples; 2 color spectroscopy of sieved ash samples; 3 visible microspectroscopy of ash particles.Conventional detailed petrological analyses, which take time, follow as the next step (4).GSD: grain size distribution; SEM: scanning electron microscopy; EPMA: electron probe micro-analyzer.Please refer to the text for additional details in section "sample and analysis methods" From March 11 to the end of June (Phase 3 of Matsumoto and Geshi 2021), the color values exhibited two distinctive linear trends.One trend was similar to that of the evening of March 6 (March 11-20 and May 14).This is highly similar to the Vulcanian ash from the 2011 eruption (January 28, April 3, and August 31; area within Fig. 5b).The other showed substantially higher a* values of up to 3.0, with similar or slightly higher L* and b* values (March 13, 15, April 5, and June 27; area within Fig. 5a but outside Fig. 5b).They showed smaller slopes than those of the afternoon of March 6 in the a*-L* and a*-b* diagrams (red diamond symbols in Fig. 5b) and similar trends to those of sub-Plinian ash on January 26, 2011 (black circle symbols).
The sieved samples showed different values from those of the bulk samples and among samples of different grain sizes (Fig. 6a).In general, finer fractions (smaller symbols in Fig. 6a) showed similar color changes to those of the bulk samples but not precisely the same values.The ranges of values of the fine fraction were wider than those of the other coarser fractions but narrower than those of the whole bulk samples (Fig. 5a).The coarse fractions (larger solid symbols in Fig. 6a) showed relatively smaller values and widths of variation but showed a similar relative relationship to the finest fraction in terms of temporal change.The finer fraction in the morning of March 6 (small purple symbols in Fig. 6a) was included in a trend similar to that of the bulk samples, with a negative correlation between a* and L* values (green symbols and a dotted circle in Fig. 6b).Coarser fractions showed exceedingly low a* and b* values, with an L* of approximately 48-50 (purple symbols in Fig. 6a).Subsequently, in the afternoon of March 6, the finer fraction ["Mar6PM (fine)"] showed an exceedingly similar trend to that of the bulk sample (red arrows in Figs.5b  and 6b), while the coarser fractions showed small variations in color values within 6.0, 0.1, and 1.0 for the L*a*b* values, respectively ["Mar6PM (coarse)"].As the eruption proceeded, the samples from the evening of March 6 coarsened, and the color of the bulk ash (red circles in Fig. 6b) resembled that of the coarse fractions (Figs.6a and 7).Most of the coarse particles were colorless (i.e., small a* and b*) or black to dark on the evening of March 6 but exhibited a range of color values, with a positive correlation trend in the a*-L* and a*-b* relations (greenish symbols in Fig. 6a; enclosed as "Mar6evening").On March 7, particles of all size fractions showed a positive correlation with higher a* and b* values and moderate L* values (reddish symbols in Fig. 6a; "Mar7AM").This trend was primarily in the extrapolation, along with the trend of the coarse fraction in the evening of March 6.

Color spectra by visible microspectroscopy of ash particles
Sample 2018c02 consisted of eight types of particles that were also found in most other samples (Fig. 2c): black scoria (sco), gray lava (lgt gr lava), dark gray lava (gry lava), altered lava (alt), red lava, isolated plagioclase (pl), and isolated pyroxene (px) (Additional files 2 and 3).In certain samples collected in the afternoon on March 6, black scoriaceous to gray pumiceous particles were also found, which showed variation in vesicularity and color (Fig. 2df).The spectra of these typical particles that were manually selected under a microscope are shown in Fig. 8 (Additional file 4).Although the spectra of ash particles differ grain by grain, even for similar grain types, the characteristics of the spectra can be classified using the average intensity and spectrum forms of the normalized R% (Figs. 9 and 10, respectively).Scoria particles and isolated plagioclase noticeably exhibit low and high average intensities, respectively, which differentiate them from the others (Fig. 9).The isolated pyroxene particles have a unique spectral form with a right-ascending slope in relation to wavelength.
Other types of particles also have average intensities that are consistent with the qualitative observations: darkcolored lavas are lower in intensity than gray lavas, and light-colored pumiceous particles have a higher intensity than scoria.In addition, the normalized spectra, that is, the reflectance spectra divided by the average reflectance, were confined within a relatively narrow range for each type of particle (Fig. 10).

Reconstructing eruption sequence using the color of the ash sample and grain size
The bulk and sieved samples exhibited different color values.The samples of the finest fraction showed color values nearest to those of the bulk samples, except for the cases of fine-depleted samples.These facts indicate that the color of ash particles varies with grain size and that the color of the mixture of all size fractions, or bulk sample, should be the sum of the colors of each fraction.Furthermore, the relationship between the bulk and sieved samples indicates that the contribution to the bulk color is substantially stronger for finer fractions than for coarser fractions.This can be interpreted as follows: the coarser grains were coated with finer powdery grains, which diminished the effect of the coarser fractions.However, as observed for the fine-depleted samples, the contribution of the coarser fractions increased when the amount of coarser fractions increased.As samples that were collected at the nearest localities and during short intervals of time were settled under exceedingly similar conditions in terms of wind direction or transportation path of the ash cloud, the changes in grain size likely reflect changes in eruption conditions at the vent.Thus, the increase in coarser fractions can be correlated to intensified activity at the vent or vice versa.From this perspective, the color changes in the bulk and sieved samples with GSD observed at the beginning phase of the March 2018 eruption of Shinmoe-dake following the 2017 eruption can be interpreted as follows (Fig. 5c).
The 2017 eruption: The variation in color values of the 2017 ash samples indicates that they were primarily colorless (low a* and b*), with a gradual decrease in brightness (L*) and an increase in coarser dark lava particles (blue symbols in Fig. 5a, b or opposite direction to gray arrow in Fig. 5c).These dark lava particles were relatively altered and exhibited round shapes under the microscope (Fig. 2b), which could be correlated with the altered lava of the 2011 eruption ponded in the summit crater.The finer fractions were primarily comprised of the same constituents that fragmented in the shallow conduit.The increase in coarser dark lava particles thus indicates that the eruption became more violent and ejected coarser material, resulting in the enlargement of the vent.
Phase 1 of the 2018 eruption (Matsumoto and Geshi 2021): Similar results were obtained for the March 2-5 samples from the 2018 eruption.The ash samples showed exceedingly similar bulk color changes, indicating a decrease in L* with increasing a* values (red triangles in Fig. 5b).Therefore, we speculate that this also reflects vent opening phenomena through the 2011 summit lava lobe and that the coarsening of the ash sample (Fig. 2c) reflects intensified activity toward March 6 (Fig. 4).
Morning to afternoon of March 6: On March 6, 2018, continued ash emissions became increasingly intense over time due to the increase in the median size of the GSD (Figs. 4 and 7a).From the morning to the afternoon (until ca.15:00) of March 6, the variation of the bulk color values changed from a negative to positive correlation in a*-L* (green circle to red arrow in Fig. 5b).This indicates that the composition of the ash sample changed completely from a mixture of fine and coarse altered lava (green circle) to that of coarse fresh and altered lavas (red arrow) with fine vesicular color-rich particles (Fig. 2d-f ), which is entirely different from the previously observed trend.All bulk color values are plotted along a positive correlation trend with each other (along red arrows in Fig. 5a-c).During that time, the color of the finer fractions also changed from a trend similar to the bulk ash that started from 2017 to the morning of March 6 (green arrow in Fig. 6b 5a-c).This change was paralleled by an increase in the color values of the finer fractions that were maximized at approximately Fig. 8 Representative spectra by visible microspectroscopy of the 2018 ash samples of Shinmoe-dake.The legend is for reference particles in samples 2018c02, 2018c14, and 2018c25.sco, px, alt lava, part-alt lava, lgt gr lava, gry lava, and bl lava are the scoriaceous particle, isolated pyroxene, altered particle, partly altered lava, light-gray lava, gray lava, and black lava, respectively, erupted by 9:00 on March 6, 2018 (2018c02 in Table 1).ves pu is the vesicular pumice particle sampled from 17:05-17:20 on March 6 (2018c14).The gr pu and bl sco are gray pumiceous and black scoriaceous particles sampled from 20:52-21:17 on March 6, 2018 (2018c25) Fig. 9 Average intensity of reflectance (%) for the representative ash particle types.These particles are selected from samples 2018c02 and 2018c25 in Table 1, and the abbreviations are as in Fig. 8. sco and pl can readily be distinguished by the intensity of their reflectance Fig. 10 Normalized spectra derived from raw visible spectra in Fig. 8 divided by average intensities in Fig. 9.The forms of the normalized spectra are systematically different among different reference types.Normalization would have emphasized the difference in spectra by minimizing the difference in grain size; px can be discriminated by a spectrum form that has lower and higher values than others in the range < 550 nm and > 600 nm, respectively.Black particles have flat spectra, especially between 550 and 620 nm, whereas colored particles have steeper slopes in all the wavelength ranges of visible light.The symbols are the same as in Fig. 8 15:00 [Mar6PM (fine) in Fig. 6; orange lines in Fig. 7b-d], whereas coarser fractions remained at lower three-color values [Mar6PM (coarse) in Fig. 6a-1].The finer fraction consisted of newly appearing light-colored pumiceous grains [blue to green symbols enclosed by a circle with "Mar6PM (fine)" in Fig. 6a].These facts indicate that the continued intensified eruption resulted in eruptions with vesicular magma in the afternoon of March 6, even though erosion of the 2011 lava by the eruption continued.
Evening of March 6: After approximately 15:00 on March 6, the three bulk color values declined again (decreasing trend with a red arrow in Fig. 5b).This time, the linear trend was slightly different from that before 15:00, and the lower end of the trend had slightly higher L* and b* values and lower a* values in the evening ("Mar6evening" in Fig. 5b).During this time, the color change of the finer fractions was exceedingly similar to that of the bulk samples (Fig. 6b), whereas the variations in the a* and b* values of the coarser fractions were distinctive and exhibited a linear or slightly curved trend (symbols enclosed by "Mar6evening" in Fig. 6a) that could be extrapolated to that of the fine fraction observed before 15:00 [symbols enclosed by "Mar6PM (fine)"].These facts indicate that the pumiceous particles also began to increase in coarser fractions (Fig. 2f, g), which follows a linear trend with black-gray lava (Fig. 5a).However, the color of the dark lava had changed a little from that observed before, with slightly lower a* values and similar L* and b* values ("Mar6evening" in Fig. 5b).This change in color of the lava particles can be interpreted as follows: the erosion of lava continued as before, but the color of lava changed, likely from the 2011 lava to the newly effused 2018 lava with a darker color.
March 7: Then, on the morning of March 7, the bulk and sieved colors all changed to exhibit exceedingly high a* and b* values, with L* values in the same range of variation as before ("Mar7AM" in Fig. 6).Although the median grain size decreased compared to that in the evening of March 6 (Fig. 7a), the three-color values (especially a* and b*) of the coarser fractions also increased to the same extent as those of the finer fractions, which corresponded to an increase in pumiceous particles for all fractions (Figs.2h and 7b-d).In contrast, the trend of sieved sample colors on March 7 (red symbols enclosed by "Mar7AM" in Fig. 6a) indicated a higher a* value than that of the finer fractions on March 6 [blue-to-green symbols enclosed by "Mar6PM (fine)"].The higher a* values, relative to the same L* values, can be interpreted as results of oxidation, as indicated by Moriizumi et al. (2009), which shows a similar trend in color change.
After March 8: Although sieved data are not available, this tendency of increasing redness of the bulk ash samples became stronger after March 8 when Vulcanian explosions were the dominant eruption style (Fig. 7c).As Vulcanian eruptions are generally characterized by intermittent explosions with a certain time interval during which magma is stagnant at a shallow level below the crater, the tendency for redness is likely to imply that the magma ascent speed slowed and could be oxidized by interaction with the surrounding air immediately beneath the vent.Thus, the timing when the ash obtained higher color values (especially a*) might be interpreted as the signal for an imminent Vulcanian explosion, which is consistent with the fact that Vulcanian explosions occurred on March 8 for the first time and occurred intermittently thereafter (Fig. 7).

Color spectra of visible microspectroscopy and automatic derivation of componentry
The color spectra from visible microspectroscopy of the ash samples (250-500 μm in diameter) in the waxing phase of the 2018 eruption showed systematic variation, which might be useful for the objective derivation of the componentry.Therefore, we accumulated spectral data for several reference types of ash particles that were defined and collected by an investigator (TS) (Additional file 4).We then classified particles of unknown samples by introducing the RSS of the spectra between the references and the unknown and by constructing an Excel worksheet (Additional file 5).In addition, we conducted previous selection procedures, because they have unique characteristics in spectral form and average intensity, although they have large variations in the reference spectra (Fig. 8).Scoria particles had an exceedingly low average intensity of R% < 10, and plagioclase isolate particles had an exceedingly high average intensity of R% > 40 (Fig. 9).Isolated pyroxene particles had an exceedingly low normalized intensity in the wavelength range of 450-550 nm (Fig. 10) despite the relatively large intensity variation.Thus, by first selecting unknown particles that match these three types using the average intensity and spectral forms, we utilized the RSS of the spectra between the other reference particles and the remaining unknown particles to assign the particle type that resulted in the minimum RSS.The reproducibility of the RSS method for reference particles varied from less than 10% to nearly 90%, depending on the type (Additional file 6).However, certain reference types are not independent of others and are occasionally difficult to differentiate from each other even by manual collection (for example, 14ves_pu and 25 gr_pu, 02sco and 25bl_sco, and certain types of lavas).This type of misclassification or arbitrariness in selecting thresholds can often occur, because manual selection is based on the characteristics of the entire particle, whereas the spectroscopic data were collected from only one spot of a particle with heterogeneity within it, and different parts of the same particle may have different spectra to be classified differently.In particular, this can easily occur when analyzing pumiceous or lava particles on plagioclase and pyroxene phenocrysts.Thus, by automatically accepting the results as correct and excluding isolated crystals from the total sum (Fig. 11b), the reproducibility exceeds 70% for most reference types (Additional file 6).

Interpretation of the detailed sequence in the beginning phase of the 2018 eruption
In Fig. 11, the results of both manual and automatic selection by RSS are shown.Although there were certain discrepancies in the selection of particles, the results were consistent with each other, considering the precision of both selection methods.This enables us to interpret changes in the componentry as follows.
The time series of the componentry indicates that the small ash venting activity before noon on March 6 did not erupt pumiceous particles but rather gray lava fragments and an exceedingly small amount of black scoria.Pumiceous particles then appeared in the afternoon of March 6, with an amount of 20-70% for a grain size range of 250-500 μm.Matsumoto and Geshi (2021) and Maeno et al. (2023) showed componentry exceedingly similar to our results in this phase of the eruption sequence, although their samples had different particle sizes and were not in a constant size range.Our componentry data were all for 250-500 μm size particles, and we also monitored bulk and sieved color values as well as the GSD with a higher temporal resolution of sampling, as shown in Fig. 11.Thus, we confidently indicate that the waxing activity can be broadly parallel to the increase in grain size and pumiceous particles.Matsumoto and Geshi (2021) found two types of particles with different crystallinities in terms of nanolites for both the vesicular A and dense B types, which correspond to less crystalline (A1 and B1) and more crystalline (A2 and B2) particles, respectively.They discussed the shallow magma ascent mechanism based on their results, suggesting that the increase in non-crystallinetype particles indicates the beginning of the magmatic eruption.They also interpreted the difference between the two types to originate from different pathways of magma: crystalline magma from the slow ascending part at the periphery and less crystalline magma from the rapidly ascending part in the center of the magma conduit.Such changes in the crystallinity of the eruptive products were also found in our results, as black lava grains decreased with increasing gray lava grains in the morning and early afternoon of March 6, although the lava particles may have originated from both the 2011 and 2018 lava lobes, as discussed above.
In contrast, Maeno et al. (2023) found two types of lapilli on March 8-9 in the 2018 eruption: P1 and P2, based on the size distribution of microlites of lapilli samples.They discussed the magma ascent processes and the P2 magma, with a steep slope of crystal size distribution for small microlites, which was interpreted to have ascended rapidly from the deeper part of the conduit, whereas the P1 magma ascended relatively slowly but smoothly.Although we could not find such differences in ash particles as P1 and P2 lapilli, at least before noon on March 7, other than the increase in redness in bulk and sieved color, this intensified their interpretation of the occurrence of P2 magma only after March 8.In addition, the time series of the CR may indicate a more detailed change: the crystalline gray lava particles increased against darker lava particles in the afternoon of March 6, while light-colored pumiceous particles remained at approximately 20-40%.This may imply that the average magma ascent rate increased while the maximum rate remained constant, although it did not exceed the transition threshold for explosive eruptions, as indicated by Kozono et al. (2013) (Fig. 9).

Potentials for monitoring the color of bulk and sieved ash samples
As noted above, systematic correlations between changes in the color of ash samples and grain size can be linked to the eruption sequence (Fig. 5c).These changes in color values were consistent with actual observations under a microscope (Fig. 2), demonstrating that color value and grain size variations are valuable for monitoring eruptive sequences in terms of eruptive materials.The primary advantage of bulk sample color analysis is its rapidness in providing continuous data on material characteristics compared with other methods, such as petrological descriptions.In addition, more precise data on constituents can be obtained by incorporating a sieving process for bulk samples, although this may require additional time and labor, potentially complicating ongoing monitoring efforts.Therefore, bulk color monitoring is preferable for basic continuous monitoring procedures.If irregular changes are detected in bulk samples, or if sufficient time and resources are available for sieving, colorimetry of sieved samples could provide more detailed insights into eruptive conditions, as evidenced during the 2018 Shinmoe-dake eruption.
We propose that the relationship between the color and size of ash samples can be used to reconstruct eruption sequences (Fig. 5c).For example, if L* increases as a* and b* decrease, it might indicate an eruption involving altered materials (excluding oxidation).A darkening and decreasing color value may suggest an increase in coarse fresh lava fragments.If all L*a*b* values are positively correlated and exhibit a reddish trend, this suggests oxidation of the lava.Previous studies, such as Miyagi and Tomiya (2002), have individually discussed color changes caused by single processes.However, by compiling knowledge of color changes from various processes likely to occur in the magma plumbing system into a systematic color chart of L*a*b* before an eruption, we can effectively monitor eruption sequences (Fig. 5c).
However, this method may not be universally applicable to other volcanoes or even at different stages of activity at the same volcano, because the colors of eruptive materials are influenced not only by these processes but also by properties inherited from deeper sources, such as bulk composition, or by the conditions and locations where the active craters formed.Therefore, a relationship established at one volcano during a certain period may only be valid for that specific case, and data must be accumulated on a volcano-by-volcano basis, or its application may be limited to certain periods of activity.Nevertheless, there are likely some common characteristics in the trends of color change associated with each elementary process, such as crystallization, vesiculation, alteration, and oxidation.After creating color diagrams for several volcanoes, it might be possible to develop a standardized map applicable to these volcanoes and to identify deviations from this standard as the individual characteristics of each volcano.Specifically, if key factors such as magma composition or crater location are consistent over a certain period or among different volcanoes, the processes in the conduit (e.g., ascent rate, vesiculation, fragmentation, crystallization) that influence eruption styles may be limited to a few variables, such as grain size and redox state of the sample, thereby enhancing our understanding and prediction of transitions in volcanic activity.

Notes for continuous ash sample collection for monitoring
The samples used in this study were collected directly during the eruption before reaching the ground using trays or similar methods.However, some samples were retrieved from deposits on artificial structures.In such cases, the timing of the start of deposition could only be determined if the area had been cleaned prior to the onset of ashfall.Without this precaution, obtaining samples with precise timing and without contamination was challenging.In addition, finer particles are often blown away by strong winds at ground level, and fragile materials like pumiceous particles are prone to breaking into finer pieces due to rolling and abrasion.It is also observed that fine fractions of particles can infiltrate through coarser particles in the same deposit, creating a stratified structure that appears as if finer particles were deposited first.This information is crucial for interpreting sample characteristics and understanding the environmental context during collection.Time-series sampling is a fundamental and robust method for the detailed reconstruction of eruption sequences, as demonstrated by several studies.Although continuous and comprehensive sampling was not possible during the 2018 eruption, particularly from March 7-9, when significant eruption transitions occurred, this study underscores the importance of automatic continuous sampling of eruptive materials, as proposed by Shimano et al. (2013), for detailed monitoring of eruption sequences.If researchers are unable to remain at the sampling site, employing an automatic ash sampler would be advantageous for accurately timing ashfalls and minimizing sample dispersion or loss.
Unlike the 2018 eruption, most eruptions in 2017 occurred during rain, and deposits were often washed away by repeated rainfall.During phreatomagmatic eruptions, which are frequently associated with rainfall, the deposits can become very hard when dried, making it difficult to disaggregate them back into original particles.In such cases, it is advisable to sieve the samples before drying to safely obtain the coarser fraction.
Fortunately, this study successfully resolved differences in color values over time across various particle types.However, there may be cases where trends in color values are indistinguishable from those of other elementary processes due to overlap.This is particularly true for sediments that have been deposited over a long period, where color changes due to alteration or weathering are likely, as discussed by Moriizumi et al. (2009).Such variations, which may differ from one outcrop to another, should be cautiously approached when analyzing older samples.

Colorimetry and componentry of ash samples as monitoring methods of eruption sequences
Manual selection of ash particles lacks quantitative standards, and prolonged analysis periods often lead to fatigue, affecting the accuracy and consistency of particle selection and counting.Although this method is useful, it has a few limitations.For a more precise analysis of sample componentry, we should employ microscopic objective methods, as discussed in the previous section, because relying solely on color descriptions fails to capture crucial textural and compositional information.
For monitoring eruptive materials rapidly and sustainably, this type of throughput measurement and continuous derivation of the rapid componentry will be useful for broadly grasping the time sequence of an eruption.In particular, it is less costly to establish visible microspectroscopy than other petrologically conventional analytical machines, such as the electron probe micro-analyzer (EPMA).Unfortunately, it requires substantial time (for example, 1 min to select each ash particle, number it, perform spectral analysis under the correct position of the microscope, and store the spectral data).Therefore, it requires 2-3 h to analyze 100-200 particles to overcome the recommended statistical errors (Ross et al. 2022).The introduction of an automatic stage and automatic preset function would reduce the analysis time; however, this may be too costly.
Colorimetry of bulk ash samples is an exceedingly rapid and objective method for deriving material data from eruptions by simply drying the field-obtained samples.As noted in the previous section, bulk samples typically displayed color values similar to those of the finest fractions, which exhibited greater color variation than coarser fractions.Furthermore, the bulk color reflected that of the coarser fractions when the grain size distribution was coarser, suggesting that changes in the ratios of major particle types may be reflected in the bulk sample color values (Figs. 7 and 11).This indicates that rapid analysis of bulk ash color could be effective in identifying major components.However, while most ash samples contain fine fractions, the colorimetry of sieved samples can also provide significant insights into major componentry.It is important to recognize that estimating the proportion of ash particles from color values alone is generally challenging.Nevertheless, with a strong prior correlation between the componentry of major particle types and bulk (or sieved) color values (as shown in Additional file 7), it is possible to estimate the componentry of dominant particles through rapid spectrometric analysis of bulk or sieved samples.The relationship between sieved ash color measured by colorimetry and componentry ratio determined by microspectroscopy and manual selection suggests that componentry can be approximated by colorimetry of sieved samples (Additional file 7).Although a more detailed analysis is needed for precise estimations, this approach could be valuable for preliminary monitoring of eruptions and determining subsequent steps.

Comprehensive and sustainable classification methods for objective componentry
In this study, automatic derivation of components was performed based only on color spectra.However, conventional classification of ash particles also considers particle shape (Ross et al. 2022), and integrating shape parameters could enhance classification quality (Liu et al. 2015;Dürig et al. 2021).Miwa et al. (2015) incorporated shape parameters alongside RGB intensities as color data, and Shoji et al. (2018) included transparency intensity with shape data, successfully classifying ash particles using a CNN algorithm or machine learning techniques.
Visible microspectroscopy analysis in our study involved capturing images of all particles, with certain reference particles displaying distinct shape parameters like convexity, circularity, and solidity (Additional files 2 and 8); detailed discussions on such combined analyses will be discussed elsewhere.Yasuda and Hokanishi (2022) recently established another method for deriving the componentry of ash samples using backscattered field emission (FE)-EPMA images of thin sections.As FE-EPMA is a throughput machine, the images can be derived automatically by a preset function for several hours for thousands of particles; if we set a threshold in advance for the classification of particles, we can shortly derive componentry by image analysis on a personal computer.Thus, it may be more rapid than visible microspectroscopy to derive the componentry.In addition, as this method uses shape and textural information, the classification results can be more readily interpreted than color alone.However, it has certain limitations, notably in distinguishing between andesitic melt and plagioclase within samples, presenting challenges in thresholding due to their similar atomic numbers.We can differentiate them using the element mapping function of EPMA.However, this would increase the total analysis time, which would be comparable to that of microspectroscopy in this study.Thus, it would be better to prepare several methodologies to derive the componentry, and we may use the optimal method depending on sample conditions.
We did not introduce a machine learning method in this study to classify ash particles, because we focused on our primary objective of introducing spectroscopic methods for ash monitoring.In addition, the utilization of such statistical methods, often characterized as black boxes, might preclude the shared understanding and dissemination of techniques and results within our scientific community.However, as Shoji et al. (2018) were successful in classification using a CNN for integrating knowledge, it would be challenging to establish an integrated classification method by combining both shape and colorimetric data and introducing not only supervised but also unsupervised machine learning techniques that will lead us to the ultimate objective classification and monitoring of ash samples.

Conclusions
The color of the ash samples changed as the eruption proceeded.The relationship between grain size and color values indicates that the bulk ash color is strongly affected by the grain size distribution of the sample.Bulk color spectroscopy has the advantage of rapid and objective monitoring of eruptive materials that broadly reflect the componentry and eruption styles.
The combination of rapid spectroscopy for these samples was used to reconstruct the sequence of the 2018 eruption of Shinmoe-dake, which evolved in a few days as follows.From March 2 to 6, the color changes indicated coarsening and darkening of the ash samples, indicating an intensification of the explosion.From the morning to afternoon of March 6, the color change indicating the inclusion of pumiceous particles, first as fine and then coarse, provided a quantitative and graphical signal for magmatic eruption.On the evening of March 6, the color change of the darker end member indicated a change in the lava surrounding the vent (lava effused in 2011 compared to that in 2018).The increase in the redness of pumiceous particles on March 7 indicated oxidation as a consequence of the ascent rate decrease that resulted in Vulcanian explosions from March 8.
Visible microspectroscopy can quantify the color differences in ash particles, allowing for the objective derivation of componentry by pre-accumulating reference spectra.While this method is not yet comprehensive, integrating shape analysis with machine learning techniques could enhance its quality.Our findings hold potential for research into automatic and continuous sampling of eruptive materials, which would contribute to volcano monitoring.This approach enables multivariate time-series analysis of eruption mechanisms by integrating information from eruptive materials with geophysical data.

Fig. 4
Fig. 4 Grain size distribution of the representative samples from March 2 to 7 (cumulative size distribution from coarse to fine fractions) with those of some eruptions in 2011, 2017, and after March 11, 2018.The median size (size at cumulative fraction = 50%) changed temporally but broadly from fine (large phi) to coarse from noon to evening on March 6, and then it became finer in the morning of March 7

Fig. 5 a
Fig. 5 a Color values of bulk ash samples in the Commission Internationale de l'Éclairage (CIE) L*a*b* color system.Black symbol: 2011 and 2008; blue square: 2017; red: 2018 ash samples.b An enlarged view of the area with dotted lines in a. c Schematic illustration for interpreting color change of ash samples to estimate eruption styles or processes in the shallow conduit.Please refer to the text for additional details

Fig. 6
Fig. 6 Color values of sieved ash samples in the CIE L*a*b* color system.a) sieved samples, b) comparison between sieved and bulk samples.Small symbols: bulk samples (circled by dotted lines); thick solid triangle to the pentagon: sieved samples (legend indicates the diameter (μm) of the fraction.The color indicates the sampling timing from the morning of March 6 to the afternoon of March 7, 2018.The range of a* in this figure is narrower than in Fig. 5 left ) toward a trend with positive correlation [enclosed by a circle with "Mar6PM (fine)"], while coarser fractions changed from exceedingly low b* values (< 0) to those with negative correlations between a*-L* values [Mar6PM (coarse) in Fig. 6a], most of which are concentrated at a* = 0.0, b* = 1, and L* = [43-49].Although the change was not straightforward in order of time and included some fluctuations, the bulk ash color values (L*, a*, b*) systematically increased, following a positive

Fig. 7
Fig. 7 Temporal changes in (a) grain size.These changes are in the phi scale (phi 25, 50, and 75: size of 25, 50, and 75 cumulative weight % in descending order of size) and in (b-d) color of bulk and sieved samples at the initial waxing phase of the 2018 eruption in comparison with the bulk colors of the 2011 eruption, 2017 eruption, and later phase of the 2018 eruption

Fig. 11
Fig. 11 Componentry ratio (CR) deduced through the minimization of the residual sum of squares with results derived by hand-pick.a CR including isolated crystals (a-1: hand-picked, a-2: automatic), b CR excluding isolated crystals (b-1: hand-picked, b-2: automatic).Both results in terms of time-sequence by hand-picked and automatic procedure are consistent, although the standard of classification would have been different.Please refer to the text for additional details