Sample description and preparation

Crustal xenoliths that can reach 20 cm in size are embedded in volcanic debris that occupy the western flank of the crater of the volcano (Fig 2). The two samples under study (laboratory codes BT1611 and BT1612) were collected on the western side of the maar crater near the pyroclastic dam which frequently overflows during the rainy season (March-October). The sampling points (N6°26’46.43”/ E10°17’41.41”; ~1100 m a.s.l.) are separated by ~200 m distance; the sample closer to the Nyos dam (BT1612) is shown in Fig 3. Collected samples contain a mixture of granite fragmented to various grain sizes embedded in a dark and fine matrix of volcanogenic material. The porphyroid pink granite itself is composed of orthoclase, microcline, quartz, plagioclases and biotite. The target material for luminescence dating were the sand-sized quartz crystals in the dark matrix. Because of their small grain size these are expected to have had the best chance for complete luminescence signal resetting during the maar explosion, either by heat, light, pressure, or a combination of any of these factors.

The two samples were prepared under subdued red-light conditions (640 ± 20 nm) after marking the outer surface with yellow paint to avoid incorporation of light-influenced material into the luminescence sample. The outer layer of at least 2 cm of the brittle samples was removed with a low-speed water-cooled diamond saw; the interior part was then gently crushed in a steel mortar to obtain grain sizes in the range 90‒200 μm. Subsequent chemical preparation included treatment with 10% HCl and 10% H₂O₂ to dissolve carbonates and oxidise organic matter, respectively. Heavy minerals and feldspars were removed from the polymineral samples by applying density separation using sodium-polytungstate (2.62 g cm^-3 < ρ < 2.75 g cm^-3). Etching the remnant grains in 48% HF for 45 min served to remove the outer layer of the grains influenced by external α-radiation and to dissolve any remaining feldspars. Subsequently, samples were washed in 10% HCl to remove fluorides that could have precipitated during etching.

The absence of feldspars in the sample was proven by an IRSL test measurement (n > 3 for each sample) that yielded signals at instrumental background level for regeneration doses in the range of the expected equivalent dose (D_e). Regarding the large size of the samples (several 100 g), comparatively few pure quartz grains (~50 mg) were left after sample preparation. Future sampling should take this into account.

Instrumentation

Luminescence measurements were carried out with a Risø TL/OSL DA-15 reader attached to a DA-20 controller (TL in the UV spectral range) [24] and with a Freiberg Instruments lexsyg standard reader (optically stimulated luminescence, OSL) [25].

The Risø system is equipped with an in-built ⁹⁰Sr/⁹⁰Y β-source delivering a dose rate of 1.97 Gy min^-1 to coarse grain quartz samples. UVTL signals were detected with an EMI 9235QB UV-sensitive photomultiplier tube after passing through a 7.5 mm Hoya U340 glass filter.

The lexsyg system features an in-built ⁹⁰Sr/⁹⁰Y β-source yielding a quartz coarse grain dose rate of 3.34 Gy min^-1 and a Hamamatsu H7360-02 photomultiplier tube allowing detection of wavelengths in the range 300‒650 nm. Detection filters further constrained the spectrum to ~330‒380 nm (Delta-BP 365/50 EX plus Schott BG3) for recording LM-OSL. Optical stimulation was achieved by green LEDs emitting at 525 ± 25 nm with a maximum power density of ~36 mW cm^-2 at sample position.

All TL measurements were conducted in N₂ atmosphere and using a heating rate of 5 K s^-1. The background was recorded in a second run on the same aliquot immediately after signal readout and subtracted channel-wise to obtain net signals. Prepared quartz grains were attached to aluminium (Risø) or stainless steel cups (lexsyg) using silicone oil and a 5 mm spray mask to obtain an acceptable signal-to-noise ratio in luminescence measurements.

Initial tests and measurement protocols

To assess the general suitability of luminescence methods to date the phreatomagmatic eruption having formed the Nyos maar, a range of initial measurements was conducted, specifically aimed at testing

1. whether the eruption mechanism was successful in completely erasing the geological (inherited) luminescence signal of the fragmented granitic bedrock and
2. which luminescence emissions of the quartz samples provide thermally stable and non-fading signals.

To address the first issue, a heating plateau test was carried out for the two samples. A plateau in the ratio of natural TL (NTL) to TL after administering an additive β-dose on top of the natural signal (NTL+ β) demonstrates sufficient luminescence signal resetting during the eruption in this temperature range of the TL glow curve. The heating plateau further indicates the range of a thermally stable luminescence signal for determining the D_e [26]. The measurements were performed on two aliquots for NTL and two aliquots for NTL+ β for each sample (due to scarcity of sample material). TL glow curves were normalised to the integrated TL signal from 350‒400°C glow curve temperature of a 50 Gy test dose administered after the NTL or NTL+ β signals were measured (second-glow normalisation). This normalisation integral was considered because the heating plateau itself was determined with these measurements; for subsequent analyses the heating plateau range of each sample was adopted.

The optically stimulated luminescence (OSL) signal of quartz can be advantageous over the TL signal because of less severe heat treatments necessary in the course of repeated measurements for D_e determination, with potentially more reproducible signals. However, a requirement for accurate optical dating is the presence of a so-called fast component in the bulk OSL signal [27]. Therefore, the composition and thermal stability of the OSL signal of both samples was checked by measuring linearly modulated OSL (LM-OSL) [28]. Here, the stimulation power density was ramped from 0 to 36 mW cm^-2 during a period of 2 h (at 125°C) after bleaching the sample for 60 s in the lexsyg reader using green stimulation (36 mW cm^-2; 125°C), administering a regenerative β-dose of 500 Gy and preheating at 200°C for 10 s. The resulting LM-OSL curve allows quick visual assessment of presence or absence of the OSL fast component. Furthermore, LM-OSL curve deconvolution provides insights into the relative proportion of fast, medium and slow components [27].

Scarcity of prepared sample material for BT1611 led us to apply a UVTL single-aliquot regenerative dose (SAR) protocol for D_e determination. The measurement sequence includes four regenerative dose points (giving signal L_x) from which the lowest dose point is repeated to check the accuracy of sensitivity correction (‘recycling ratio’). This was done by means of a constant test dose (giving signal T_x) in between the regeneration cycles. The sensitivity-corrected dose points (L_x/T_x) were fitted with a linear function to deduce a functional relationship between TL and dose, applying the functions ‘calc_TLLxTxRatio’ and ‘plot_GrowthCurve’ contained in the R package ‘Luminescence’ (version 0.6.4) [29–31].

With this protocol, however, a significant sensitivity change of the sample during the first TL readout cannot be detected or corrected for by test dose monitoring. In this case, the SAR protocol would produce systematically inaccurate results. For sample BT1612, we therefore additionally applied a multiple-aliquot additive dose (MAAD) [26] and a single-aliquot regeneration and additive dose (SARA) protocol [32]. While the MAAD protocol is not affected by sensitivity changes in the course of the first measurement, the SARA sequence automatically takes into account these changes. Consistent SAR, MAAD and SARA ages would thus confirm that the SAR protocol yields accurate dose measurements for the investigated samples.

Specifically, for the MAAD measurements the sample material was divided into five portions which were further split into three aliquots. Each portion received different additive β-doses of 0, 20, 40, 60 and 80 Gy prior to TL readout. The integrated TL signal was then plotted against the additive β-dose and the data points fitted with a single saturating exponential function, whose intercept with the negative part of the dose axis gives the absolute value of the D_e. The UVTL instrumental background was recorded separately and its integrated signal subtracted from the bulk integrated signal.

In the course of the SARA protocol, four portions of sample material (three aliquots each) were β-irradiated with increasing additive doses of 0, 20, 40 and 60 Gy. Subsequently, the D_e was measured by regenerating UVTL signals such that these bracket the signal induced by the natural plus the additive dose. In other words, a SAR protocol without test dose correction was applied. When plotting the obtained (not sensitivity-corrected) D_e values against the known additive β-dose, the dose axis intercept of the linear fit to these data points returns the ‘true’ D_e (sensitivity-corrected). The slope of this fit further corresponds to the dose recovery ratio in the SARA experiment, i.e. a slope of 1 suggests that a known added dose was successfully recovered in the measured dose [33].

Finally, a test for anomalous fading was carried out, consisting of repeated cycles of β-irradiation with a dose similar to the D_e and UVTL measurement (L_x), while a pause of varying duration (up to ~15 h) was inserted before measurement. Test dose cycles in between regeneration cycles provided proxy data for sensitivity changes, and the L_x/T_x ratio was plotted against the normalised delay to assess signal loss over time (following the approach of Auclair et al. [34]).

Dose rate assessment

Due to removal of the α-influenced rim of the quartz grains, only external β-, γ- and cosmic radiation contribute significantly to the environmental dose rate. Beta- and γ-dose rates were calculated from K, Th and U concentrations determined from representative bulk maar tuff material (Fig 3) by thick-source α-counting (Th, U) [26,35] and ICP-OES (K). Radioelement concentrations were then translated to dose rates with conversion factors published by Guérin et al. [36] as implemented in the program ADELE [37], which was also used for age calculation.

The size range of quartz grains (90‒200 μm) used for luminescence measurements does certainly not reflect the ‘original’ size spectrum after bedrock disintegration in the course of the maar eruption, which has implications for the amount of β-dose absorption within the grains. The original size distribution of quartz grains was assessed by gently disaggregating representative maar tuff without cracking individual quartz grains followed by density separation to extract quartz only (see section sample preparation). The diameters of several hundred quartz grains were then assigned under the microscope to the size fractions 200‒600 μm, 600‒1500 μm and 1500‒2200 μm, which were weighed subsequently (the number of grains <200 μm and > 2200 μm was negligible). Within uncertainties of our analyses, all three size fractions amounted to almost identical masses, so that we adopted a grain size range of 400‒2000 μm for age calculation. This range certainly covers >90% of the grains studied and should thus adequately account for the effects of internal β-attenuation (see also S4 Fig).

The matrix of pyroclastic deposits contains both fragmented granitic bedrock as well as juvenile volcanic material and hence the dose received by individual quartz grains after the phreatomagmatic eruption is likely to vary from grain to grain. Provided that the dose response of individual grains is in the linear range, the average dose rate determined for the matrix combined with the arithmetic mean D_e should result in the best estimate for the true burial dose [38]. Potential spatial differences in β-radiation (mm-scale) within the maar tuff and therefore variance in doses received by individual quartz grains are anyway assumed to be averaged out by using comparatively large aliquots (~5 mm in diameter), hosting ~750 ± 200 grains (calculated with the function ‘calc_AliquotSize’ implemented in the R package ‘Luminescence’ [29–31]). Given the resultant D_e distributions, application of age models such as the Central Age Model (e.g., [39]) does not significantly change the final dose estimate within errors; even the Minimum Dose Model (e.g., [39]) would produce a dose estimate not substantially different from the arithmetic mean (maximal 9% deviation in the worst case scenario). Apart from the insignificant effect, the application of any age model does not appear to be justified considering the complexity of dose rate assessment.

The cosmic dose rate was computed by use of the function ‘calc_CosmicDoseRate’ (implemented in the R package ‘Luminescence’ [29–31]) in combination with topographic information of the samples (longitude, latitude, altitude, overburden). However, the temporal variation of overburden thickness relevant for the luminescence samples is unknown and has to be estimated from geometrical information available for the pyroclastic cone of Nyos maar. Field observations suggest that the initial surface of this cone at the Lake Nyos dam was ~1.5 m higher than today [4]. We therefore approximated the cosmic dose rate by adopting this number for the luminescence sampling sites and assuming constant erosion rate from an initial flat overburden of ~1.7 m to a final cover of ~ 0.2 m using a density of 2.2 g cm^-3. An assumed error of 10% is thought to represent most of the inaccuracies related to this model.

The saturation water content of representative sample material was determined in the laboratory (~13 wt.%; water weight over dry weight) and 75% of this value was considered to reflect the annual average moisture content in view of the tropical climatic conditions at Nyos maar. Increased error margins of 40% (relative to absolute moisture) should include all conceivable deviations (e.g. topographic effects or long-term change in precipitation), so that we adopted a value of 10 ± 4 wt.% for age calculations. The correction factors given in Aitken [26] served to transform dry dose rates to wet dose rates.

Quality of luminescence results

The successful UVTL heating plateau test validates that the luminescence signal was reset during the phreatomagmatic maar explosion, which represents a prerequisite for successful dating. This result supports earlier indications that phreatomagmatic eruptions are potentially capable of exerting sufficiently high temperatures and/or pressure to evict all trapped charge from dosimetrically relevant traps in the crystal lattice of fragmented country rock [11,42]. Further evidence that the dated event corresponds to the volcanic eruption is the agreement of TL ages of both samples within 1σ error margins. Had other agents of luminescence signal resetting been active in the past (e.g., post-depositional re-location of the volcanic deposits), it would be unlikely for them to cause congruent ages.

The accuracy of applied TL protocols was assured by checking three different approaches of D_e measurement against each other for one sample, resulting all in ages in good agreement based on 1σ uncertainties. We are thus confident that we have determined a robust D_e, provided that anomalous fading is absent in the studied samples what is supported by the conducted fading test. These laboratory tests, however, cannot detect long-term fading which would underestimate the dose and the age.

By contrast, dose rate determination for the two investigated samples is more prone to systematic errors. The most significant factors for dose rate uncertainties comprise the estimated moisture content of the sedimentary matrix over the burial period, the overburden of pyroclastic surge deposits since the time of eruption, potential non-uniformity of the β- and γ-radiation fields as well as the original size distribution of quartz fragments in the sampled matrix.

The moisture content of bulk samples was estimated based on laboratory measurement of the saturation water content (~13 wt.%), and the assumed representative value taken for age calculations spans a range considered as covering a large part of physically meaningful values (10 ± 4 wt.%). The uncertainty of this factor is hence reflected in the overall error margin of TL ages.

While in many settings the cosmic dose rate contributes only a minor proportion of the total dose rate (based on dose rate generated by sediment of average crustal composition, e.g. loess [26]), it makes up ~10% of the total dose rate for the Nyos maar samples. An erroneous calculation of cosmic dose rate thus causes a relatively larger impact on the final age. Although we considered the best available information, a systematic error inherent in cosmic dose rate assessment cannot be ruled out.

Comparatively large granite rocks (cm-scale) can occasionally be found within the Nyos maar tuff. Although we carefully tried to consider only material for luminescence measurements that was located as far as possible away from such granite rocks, their potential influence on the actual dose rate and hence age must be discussed. Unfortunately, it was not possible to conduct in-situ measurements of the γ-radiation field directly at the sampling spot. Previous analyses of similar granitoids in the region showed that with approximate radioelement contents of K ~ 3 wt.%, Th ~ 20 μg g^-1 and U ~ 2 μg g^-1 these rocks generate a stronger γ-radiation field than the surrounding maar tuff (see Table 1). In case such a piece of granite was located in the vicinity (within ~ 30 cm) of the sampling spot in the tuff during burial, the effective dose rate would have been higher, with the consequence that ages become younger. In that sense, the presented ages (Table 2) should be regarded as maximum estimates, at least in view of the potential impact of spatial γ-radiation heterogeneities in the tuff.

Due to attenuation of external β-radiation within the quartz grains, their size distribution plays an important role for accurate dose rate calculation. However, no sorting of grains after the phreatomagmatic eruption took place prior to deposition (as is the case for wind-blown and water-lain sediments), and bedrock structure as well as the spatio-temporal dynamism of the eruption might have led to strongly varying conditions of rock fragmentation for different locations within the maar diatreme. Hence, the range of quartz grain sizes found in the Nyos maar tuff is comparatively large. Given that the attenuation of β-radiation emitted by K, Th and U isotopes follows an approximately inverse linear relationship with grain size up to ~ 2 mm [40,43], it is expected that averaging effects circumvent major systematic errors in dose rate assessment, once a representative average grain size of quartz clasts has been determined.

Finally, it has to be noted that the success of luminescence methods in accurately dating phreatomagmatic processes critically relies on the type of bedrock being ejected during the volcanic eruption and incorporated into the volcanic deposits. While our study produced promising results for shattered granitic country rock, gneisses might be associated with quartz luminescence properties less advantageous for dating purposes. In such cases, focusing on feldspar as the target mineral could represent an alternative approach, although allowing for anomalous fading will then require additional measurement and correction procedures [44]. Also, the effects of high-pressure shock waves on the luminescence properties of quartz as experienced during a phreatomagmatic explosion are still poorly studied. Despite the encouraging findings obtained for the Nyos maar samples, further methodological studies are mandatory to fully explore the potentials and pitfalls of this still experimental technique.

Implications for hazard assessment

Despite all methodological challenges faced in dose rate assessment, we conclude that luminescence dating of phreatomagmatic eruptions in the CVL is a promising alternative to other dating techniques (e.g., ¹⁴C, K/Ar, U-series), which have produced ages for the Nyos maar differing by three orders of magnitudes [23]. It thus appears prudent to expand the application of luminescence methods to other sites of the CVL. Especially for Holocene and Late Pleistocene eruption sites too young for precise ⁴⁰Ar/³⁹Ar dates and lacking datable material for ¹⁴C analyses, luminescence dating is a valuable tool to complement and refine existing chronologies of active volcanism along the CVL. Such chronologies play a key role in risk evaluation of future eruptions [2,45].

Our new averaged maximum TL age estimate of 12.3 ± 1.5 ka is in rough agreement with the proposed eruption age of the Nyos maar between 5 and 10 ka ago (U-series dating [23]). The Nyos maar is the only one among the existing ~20 lake-bearing maars of the CVL, for which chronometric constraints are available, because of the aforementioned difficulties in determining ages for young volcanic activity. TL dating methods may thus also help to confine the age of the Monoun maar about 100 km south of the Nyos maar. A systematic dating of these maars can generate a better understanding of maar volcanic eruption cycles and possibly reveal indicative characteristics that may help classifying maar lakes of the CVL in terms of relative risk.