## Introduction

Explosive volcanic eruptions constitute a natural, external climatic forcing factor when their sulfate aerosol emissions reach the stratosphere and reflect solar irradiance. Short-term impacts of this radiative forcing are well documented over the modern instrumental period and lead to post-eruption cooling of global summer temperatures and reduce the amount of sunlight reaching the biosphere1. Tree-ring proxies constitute primary evidence for detecting the magnitudes of volcanic impacts prior to instrumental observations2. These annual data provide a precise record of climate anomalies and offer a way to examine the interaction of preindustrial society, climate change and other natural phenomena such as volcanic eruptions3,4. In addition to palaeoclimate data derived from tree-ring width/density chronologies, the long history of volcanism can be traced from deep ice cores5. Much interest on past volcanic impacts has centered on the mid-sixth century AD climate anomalies6,7,8,9,10,11,12, with recent proliferation of new data from the bipolar ice-core timescales and sulfur records.

The revised ice-core data reconstruct the radiative forcing from eruptions in AD 536 and 540 and imply that a combined volcanic signal from multiple North American eruptions preceded the AD 536 cooling event, whereas a second cooling phase from AD 541 until 550 was likely due to tropical eruptions2. Climate modelling indicates that the decadal radiative forcing of these events totaled larger than that of any other volcanic event in extra-tropical Northern Hemisphere in the last 1200 years, with longer duration of sulfate deposition from the second event13. Similar evidence available from high-resolution palaeoclimate data demonstrate climatic cooling on decadal and similar spatial scales12,14. Climate anomalies during these years have been linked to crop failure, high number of famines and societal crises in Eurasia15,16 and debated as the cause of several large human migrations in the Palearctic4.

Topically, the mid-sixth century cooling is suggested to have decreased the agricultural production in Europe, especially at high altitudes and latitudes13 where historical crop yields were temperature driven17,18. However, this effect appears more subtle over central and southern Europe13 where the primary production is in fact more constrained by solar radiation19 rather than temperature. As a consequence, the volcanic cooling may not constitute the only climatic factor to explain the human consequences and their nutritional background. Given that terrestrial photosynthesis is strongly limited by irradiance20, the human utilization of plant products ought to be similarly affected by post-eruption radiative forcing. It is essential that the AD 536 eruption is believed to have created a ‘dust veil’ that, according to historical accounts from the Mediterranean to eastern Asia, dimmed the sun for more than a year6,7,8,9,10,11. In addition to cooling global temperatures, the dust veil must have dramatically reduced irradiance and thereby photosynthetic products and their contemporary human utilization as a component of the post-eruption forcing. Clearly, quantifying this relationship would help to assess the impact of the explosive volcanic eruptions of AD 536 and 540 on human populations of Europe and elsewhere3,4,15,16.

In addition to temperature inferences traditionally obtained from tree-ring width/density data, irradiance proxies are obtained from the stable carbon isotope 13C/12C ratio in tree rings. This ratio (δ13C) is explained by the concentration of CO2 in the intercellular spaces in the leaf/needle, in relation to its concentration in the air, the ratio between the two varying through photosynthetic gas exchange as controlled by stomatal openings/closure and/or rate of CO2 assimilation by the plant21. Delineated by the Farquhar model of photosynthesis22,23, the δ13C is described to increase by high light intensity and/or moisture deficit, whereas the reductions in available light and water surplus lead to decreased δ13C. This being the case, the tree-ring samples taken from sites unlikely to suffer from drought ought to contain strong signals of irradiance, such conditions commonly prevailing near the northern timberline24,25,26,27. Further, the signal strength can be warranted by site selection for sampling the trees from riparian habitats where plants should not suffer from water deficits. To test the hypothesis of reduced irradiance during and after AD 536, we present tree-ring δ13C proxy data from subfossil and living pines (Pinus sylvestris L.) collected from timberline sites in northern Europe (Fig. 1). These samples represent a composite of sites (Table S1) from the region where the δ13C signal remains consistently sensitive to sunlight intensity during the summer (June–August) growing season27.

Calibrated against the instrumental records of irradiance, the modern δ13C data provide a model to reconstruct the signal over the mid-sixth century ad. Analysing the subfossil δ13C assemblage from riparian/upland sites we disentangle the irradiance-dependent δ13C variability from any remnant of drought signal. Our high-resolution assessment of the volcanic dust veils shows distinct, well-dated anomaly consistent with evidence in historical–documentary sources, to highlight irradiance as a new palaeoclimate parameter. We show that carbon isotopes of ancient wood offer a reliable proxy of the forcing event, as dated reliably on a calendric timeline analogous to historical records.

## Results

### Subfossil δ13C assemblage

Averaged chronology of subfossil δ13C series pinpoint the AD 536 event as a sharply negative excursion of 0.5–1.6 per mille (‰), followed by a 1–2 year recovery to pre-event δ13C values (Fig. 2). The second anomaly starts in AD 541 and is described by a more markedly negative departure with a sustained δ13C downturn of 1.6‰ until AD 544 ad. In AD 545 the mean δ13C exceeds the value of AD 541. Overall, the δ13C series show a strong correlation in their inter-annual variability during the suggested volcanic forcing (Fig. S1). A closer look at the δ13C series illustrates markedly negative δ13C for riparian samples (Fig. 2A) and most strongly for a tree from forested site during the event. These data exhibit the most distinctive volcanic signals according to their anomalous values. The strong signal in the riparian δ13C data contrasts with the upland δ13C series having the least depleted δ13C values, suggesting a relatively stronger dependence on soil moisture in upland data.

Based on these results and the photosynthesis model22,23, we first establish a palaeoecological interpretation for the δ13C event. That trees from forested site tend to exhibit most anomalous δ13C excursion likely indicates that such shaded individuals were those most stressed during the event. A large magnitude of the signal implies that shading from adjacent trees amplified the effects of reduced irradiance, leading to strongly negative δ13C values, in comparison to forest line sites where trees grow more widely spaced. The upland data appears least affected implying that the δ13C response was not reinforced by moisture anomalies as the combination of low irradiance and excess of available water had likely intensified the depletion of δ13C values in that habitat. Collectively, these findings point towards a dry fog event, such as a volcanic dust veil, as origin of the δ13C anomaly in that the conditions responsible for the event represent a combination of reduced irradiance without any noticeable change in hydroclimate.

Further validating the signals, the data of riparian chronology is regressed against the data of upland series, and the residuals from that model are retained as a new δ13C series (Fig. 2C), thus being statistically independent of the moisture effects (assumed to be present in upland data). We note that also the new, residual δ13C series illustrates an isotopic excursion in AD 536 and between AD 541 and 544 (Fig. 2D). The persistence of such anomalies adds credibility to the dry fog/dust veil assumption and thus to identify the anomaly as a volcanic signal.

The post- AD 536 irradiance anomalies can be reconstructed using a model calibrated with the modern δ13C record. These data provide an estimate of isotopic variance associated with irradiance measured over recent decades by twentieth-century technology (Fig. 3). The final reconstruction model (i.e., the transfer function built over the 1971–2011 period) explain more than half of the variance in the instrumental data (R2 = 0.522, p < 0.0001). Thus, our δ13C dataset exceeds the thresholds previously set for an acceptable δ13C proxy28,29. Furthermore, jack-knifed simulations of the original data retain statistically significant (p < 0.01) results and confirm the proxy-to-target agreement for a variable dataset over the instrumental period (Table S2).

The δ13C-based reconstruction quantifies the strongly reduced irradiance in 536 and especially AD 541–544 (Fig. 4). These findings contrast with the suggestion of enhanced photosynthesis, as suggested following the worldwide increase of diffuse radiation due to volcanic aerosols released during the AD 1991 Pinatubo eruption30. We also find no correlation between the δ13C and diffuse radiation data measured with instruments during recent decades (Fig. 5). Collectively, these data agree with the tree-ring observations from multiple sites in the Northern Hemisphere (AD 1300–1950) (ref.31) and suggest that the effects of overall light loss for photosynthesis have not been compensated by aerosol-driven changes in light composition during volcanic events. Thus, our results concur with biogeochemical models that have explained the enhanced post-Pinatubo CO2 sink by several other (non-photosynthetic) land and ocean sink mechanisms32,33.

Compared to the AD 536 event, the AD 541–544 drop in irradiation was substantially stronger, at least in absolute terms (Fig. 4A). These magnitudes were consistent with ice-core sulfur records with respect to the strength of global volcanic forcing and with the longer duration of sulfate deposition from the AD 540 eruption2,13. With regard to the pre-event (AD 519–535) values, the magnitudes of reconstructed irradiance change for AD 536 and AD 541–544 amount to 41 and 54 to 62 Wm−2 reductions, respectively. These losses represent values 2.5–3 standard deviations below the reconstructed overall mean values (Fig. S2). The strength of these anomalies is further demonstrated in that both the amplitude and duration of the AD 536 and AD 541–544 events respectively exceed the effects of eleven large eruptions34 experienced over the past two centuries by the living tree δ13C chronology (Figs 6; S3).

## Discussion

Surviving writings describing the veiling of the solar radiation during and after AD 536 largely originates from the Mediterranean sources once created by court historians and chroniclers. In these writings the sun was observed blue-colored, without brightness, spring without mildness, summer without heat10. The overriding reason for these anomalies was the mystery cloud, a persistent dry fog that darkened the sky, the cloud that was observed by contemporaries over wide areas across the Palearctic all the way from the British Isles to China11,38,39. Palaeoclimate literature has for long attributed the cloud to volcanic aerosol emissions6,7,8. Moreover, the anomalous climates during the event have been largely described in terms of cold summers8 such conditions having probably lasted as a protracted, at least decadal event much over the Northern Hemisphere12. An essential point is that the existing palaeoclimatic inferences have thus far been extracted from tree-ring width/density chronologies that are proxies for past summer temperatures. However, the temperature effects remain subordinate to the primary diagnosis, the opaque skies and the vastly reduced sunlight under them11. As a consequence, the survey of the climate processes during the event has remained, at best, one-sided, and somewhat biased towards its temperature characteristics which, albeit playing an important role, may actually represent secondary effects. By contrast, the δ13C data we have presented facilitate the first quantification of sunlight conditions from year to year during the dust veil episode and make it possible to reconstruct the markedly varying solar radiation from subfossil tree rings. Importantly, our results provide several aspects of volcanic summers independent of temperature effects and contribute to our understanding of the event as a multifaceted climate crisis during which the adverse effects of cold temperatures may have been reinforced by strong reductions in irradiance with the hardship of rapid climate change on human societies.

Yet another factor possibly causing the crop failures and famines was the drought described in the written sources, according to which the winter (conceivably that of AD 536/537), or possibly later part of the year, was rendered dry and without storms10. Our δ13C data remain sensitive to summer climate27 and cannot comment on any wintertime anomalies. Moreover, the coming of drought would not be consistent with hypothetical post-eruption hydroclimate summertime responses, expected to mimic the configuration of North Atlantic pressure fields during the negative phase of the East Atlantic (EA) teleconnection pattern44, and to result in wet conditions around the Mediterranean realm44,45,46. The EA pattern is the leading mode of climate variability over the North Atlantic and surrounding continents representing a north-south dipole of pressure centers47. The EA is negatively correlated with instrumental precipitation/soil moisture across the Mediterranean, the respective correlations in the region of our δ13C evidence remaining slightly positive but non-significant during the summer months44,48 (see Fig. S5 for correlations with could cover). Assuming that the mid-sixth century eruptions were followed by similar, negative EA phase (there are currently no palaeoclimate EA reconstructions available for the first millennium), the presented correlations44,48 would agree with our palaeoecological model, suggesting a negligible hydroclimatic response over the tree-ring sites during the event years.

Collectively, our data confirm abrupt changes to the growth seasons (i.e., summers) following the large volcanic eruptions in AD 536 and 541–544 in the form of cooling and, more importantly, strong reduction in incoming solar radiation. During and after these events, the cooling was likely driven by the dust veil and photosynthetic products were limited to such an extent that they likely affected food security and human immune system. These findings add to our knowledge of volcanic aerosol forcing and emphasize their importance with respect to the temperature-related scenarios frequently described in the literature. Understanding the multifaceted environmental impacts of ancient explosive eruptions requires the use of proxy data that are sensitive to variable irradiance levels and which faithfully track this vital component of a productive ecosystem. Our results underscore the pressing need for a database of tree-ring isotope chronologies and archaeological/historical records in order to investigate the relationship between irradiation and human society. Only these data can describe the spatial and temporal variation in volcanic aerosol emission forcing from one event to another and allow them to be compared to accounts from the regional to continental scale. Nevertheless, we endorse the need of combining the data of climate forcings with thorough analyses of political and economic structures15,16,49,50 of which determinants have almost certainly contributed to the event. Such a comparison would reveal the relative impact(s) of climatic forcing on agriculture, human health, urbanisation and movement during the first millennium – a period considered to contain the main ‘hinges’ of human history55.

## Methods

Our sampling area is situated in the Finnish Lapland in northern Europe. Tree-ring dated samples (see Fig. S6) were dissected from subfossil and living Scots pine (Pinus sylvestris L.) collected from the timberline and forest protection areas of Lapland56. Cross-dating of our tree-ring series against the existing mean chronology57 enables the dating of each ring to the exact calendar years. Tree-ring samples were processed to α-cellulose using two alkaline extractions (5–7% NaOH) and a chlorination step (NaClO2) in between58, homogenized using an ultrasonic probe59, and freeze-dried. These dry cellulose fibres (ca. 70 μg) were combusted for isotopic analysis on a DeltaPlusAdvantage isotope ratio mass spectrometer coupled to a CN2500 elemental analyzer at the Laboratory of Chronology, University of Helsinki. All samples were analyzed in duplicate, and randomly selected samples were subsampled for 10 replicate analyses to monitor efficiency of homogenization and result reproducibility. The δ-notation as per mille (‰) expresses the deviations from the VPDB standard. The mean reproducibility of analyses was ±0.1‰, estimated from sample replicates and 75 repeated analyses of an internal laboratory reference cellulose (Fluka-22181 cellulose powder, Sigma-Aldrich, Lot. 442654/1) analyzed alongside sample material. The raw δ13C data is made available at the National Centers for Environmental Information – National Oceanic and Atmospheric Administration (https://www.ncdc.noaa.gov/data-access/paleoclimatology-data). The δ13C data were corrected for changes in δ13C value of atmospheric CO2 due to the industrial revolution60 and discrimination rate changes by 0.0073‰ per ppmv CO2 (ref.61); the reliability and validity of which have been established27 and found to be consistent with these two independent methods of correction62,63. Trends related to tree age rather than climate variability were previously documented for δ13C data33 and removed here using the regional curve standardization64 known to preserve the full spectrum of short-to-long timescale climate information. Each δ13C value was standardized by subtracting age-dependent δ13C value from the δ13C value and by adding a constant (−24.9‰) representing the overall mean value of the δ13C data. Arithmetic mean was used to build the mean δ13C chronology.

The results were interpreted in keeping with the Farquhar model of photosynthesis22,23 describing the δ13C in wood material as

$${\delta }^{13}C={\delta }^{13}{C}_{ATM}-a-(b-a){c}_{i}/{c}_{a}$$
(1)

where δ13C ATM represents the δ13C value of CO2 in the ambient air, a the diffusional fractionation, b carboxylation fractionation, and c i /c a the ratio between the concentration of CO2 in intercellular spaces (c i ) and in the air (c a ). According to the model, more negative δ13C values are expected when factors increasing c i /c a such as low light intensity lowering the rate of photosynthesis lead to a rise in c i . Such conditions may be expected after explosive volcanic eruptions as their aerosol emissions reduce solar radiation, as observed in our subfossil δ13C data in AD 536 and 541–544. Less negative δ13C values are expected when drier conditions lead to a higher degree of stomatal closure and hence, according to the model, decreased c i . Consistent with this premise, the subfossil δ13C values of the upland site are less negative than those of the riparian sites and thus reflect the drier soil conditions at upland sites. We note that the riparian and upland δ13C data correlate positively (Fig. S1) and see no verifiable reason the riparian and upland pines had reacted disproportionately to factors other than moisture. Thus, the moisture signal present in the upland δ13C series may be statistically extracted from the riparian δ13C data using a regression model. Entering upland δ13C as independent and riparian δ13C as dependent data, we obtain a new δ13C series (δ13CNEW) in the form of residuals from the model as

$${\delta }^{13}{C}_{NEW}={\delta }^{13}{C}_{RIPARIAN}-(0.379\times {\delta }^{13}{C}_{UPLAND}-15.810)-24.9\textperthousand$$
(2)

where the parameters of 0.379 and −15.810 are the slope and intercept obtained respectively from fitting a linear regression model to the data and −24.9‰ represents the overall mean value of the data. We observed the anomalies in AD 536 and 541–544 also in the new δ13C data as evidence to suggest that this signal is unrelated to moisture fluctuations and represents a dry fog event akin to a volcanic dust veil.

$${I}_{t}=35.7\times {\delta }^{13}{C}_{t}+1059.5$$