Medieval Irish chronicles reveal persistent volcanic forcing of severe winter cold events, 431–1649 CE

Our record of cold events comes from an exhaustive new survey of the rich corpus of medieval Irish chronicles, known generally as the Irish Annals and referred to hereafter as 'the Annals'. These comprise annually arranged written entries describing major events as reported by educated scribes in monastic centres up to the thirteenth century CE and by hereditary professional historians employed by the Irish aristocracy thereafter (McCarthy 2008). The Annals contain over 1 136 000 words of text and more than 40 047 distinct written entries, with contemporary reporting starting as early as the fifth century after the arrival of Christianity to Ireland (McCarthy 2008). The general reliability of even this early content can be tested by comparison to independently-known events. These include the spatially extensive sixth century plague of Justinian and the severe climatic downturn expressed in Northern Hemisphere tree-rings, c.536–550 CE, both credibly reported in the Annals (Baillie 1994, McCarthy 2008).

The large volume of accurately reported astronomical phenomena (e.g. solar and lunar eclipses) has led scholars to conclude that recording of natural phenomena in medieval Ireland was both systematic and sustained (McCarthy and Breen 1997a). One likely motivation arose from the ecclesiastical context in which the Annals were compiled and was an interest in foretelling the coming of the 'Last Days' as forewarned in the Bible by calamitous natural phenomena (McCarthy and Breen 1997b). This helped ensure the creation of a detailed record of the incidence of extreme weather for medieval Ireland.

Because the Annals begin during the 'Dark Ages' of the first millennium CE after the fall of Rome, when comparable European sources are scarce (Lamb 1995), their climatic content is of great significance, covering the entirety of the 'Medieval Climatic Anomaly' (c.900–1300) and the 'Little Ice Age' for over half of its commonly cited duration (c.1350–1850), as well as the transition between both. Ireland, moreover, occupies a climatically important location in the Northeast Atlantic, sensitive to ocean circulation changes and major modes of climatic variability such as the North Atlantic Oscillation (NAO) (Sweeney 1997). Yet the Annals have rarely been systematically exploited in palaeoclimatic research, despite extensive use of similar annals and chronicles elsewhere. A number of encouraging exceptions include McCormick et al (2007) and Kerr et al (2009) who examine weather reports in the Annals from 750–950 and 650–900, respectively, and Hickey (2011) who examines an extensive time-period but draws upon the early surveys of Wilde (1851) and Britton (1937), now updated by Ludlow (2010).

The abundant meteorological reports in the Annals adhere to established categories of useful documentary climatic data, including directly observed meteorological phenomena and conditions, such as snowfall, and indirect or proxy phenomena, such as ice on water bodies, from which underlying weather conditions may be inferred (Pfister et al 1999, Brázdil et al 2005). To derive our record of cold events, we compile all reports indicating severe cold occurrence for Ireland's mild maritime climate (examples in table 1). A representative report for the 855/856 winter season describes 'much ice and frost so that the principal lakes and rivers of Ireland could be crossed by people on foot and on horse-back from the ninth of the kalends of December (27 November, AD855, Gregorian Calendar) to the seventh of the ides of January (11 January, AD856, Gregorian Calendar)' (Annals of Ulster). These conditions are clearly severe for Ireland, being longer in duration and appearing comparable in severity to the anomalous cold of December 2000 in which temperatures fell to −14 °C in some locations in Ireland (Graham 2004).

Having preserved twelve centuries of record, the Annals cease in the mid-1600s when English custom and colonial rule disrupted their centres of recording.

Documentary climatic data require careful methodological preparation (Brázdil et al 2005), but this is rarely afforded space for effective demonstration in the literature. Historical sources may fabricate, exaggerate, misunderstand or misreport natural phenomena, and because extreme events are by definition rare, the inclusion of relatively few unreliable events can significantly distort analyses.

Understanding the historical context in which documentary sources were created allows potentially dubious reports to be identified when contexts that motivate unreliable reporting are discerned (MacNeil 2000). For 1022 the Annals of Clonmacnoise report that 'there fell a great wonderful snow...before the battle of Sleive Grott' in the Galtee Mountains, southeast Ireland. The conspicuous timing of this snow and its emphasis as 'great [and] wonderful' suggests its function as an omen of the impending conflict. The use of portentous natural phenomena to stress the significance of major human events is a stylistic technique employed elsewhere in the Annals and in such cases casts doubt on the reliability of the specific weather reported. Such reports are valuable examples of how past societies perceived and exploited natural phenomena for political, theological and literary purposes, but they must be excluded from palaeoclimatic analyses.

In total, 83 unique cold events are reported in the Annals, of which we consider 65 reliable. Were the 18 unreliable events included in the absence of assessment, a 27.7% inflation of event totals would result. Unreliable events are also unevenly distributed in time (figure 1) and the resulting distortion is thus concentrated in certain periods (e.g. before 750 when unreliable events comprise 53.8% of the total). This provides a brief but salutary example of the importance of source assessment before palaeoclimatic analyses.

Zoom In Zoom Out Reset image size

Pioneering research on the chronology of the Annals by McCarthy (1998, 2001, 2005) has corrected errors in their dating as verified by comparison to historical and astronomical events of independently-known date. We apply these corrections in the following analyses. Towards the end of our period up to 1649 we supplement the declining record of the Annals with a survey of the increasingly diverse sources becoming available for Ireland at this time, including personal letters and accounts of military campaigns (table 1, supplementary data (available at stacks.iop.org/ERL/8/024035/mmedia) gives a primary source bibliography). This survey identifies five additional well-dated cold events that we deem historically reliable, for 1588, 1601/1602, 1603, 1635 and 1641 CE. We include these in the following analyses for 70 cold events in total.

Our cold events involve the occurrence of three classes of reported phenomena and conditions, namely: (1) heavy snowfalls, (2) periods of prolonged frost and/or ice on water bodies, and (3) general observations of abnormally cold weather (figure 2(a) depicts the frequency of reporting of these classes in 20-year periods from 431 CE). Heavy snowfall is reported on 52 occasions and is the most numerous (constituting 55.3% of reports). Prolonged frost and ice are reported on 29 occasions (30.9%) and observations of abnormal cold on 13 occasions (13.8%). We note that these values sum to 94, suggesting a mismatch with the value of 70 cold events. This arises because distinct phenomena and conditions can be reported as occurring together (e.g. snowfall and frost) or immediately consecutively (e.g. snowfall then frost) and are counted as single cold events in such instances.

Zoom In Zoom Out Reset image size

The reports from which our record derives may indicate durations of days, weeks, months, seasons and occasionally entire years for the phenomena and conditions described, but may also fail to provide any indication of duration, even when giving otherwise detailed descriptions of severity. As a rule we identify cold events as periods of any duration that appear to represent discrete contiguous episodes of abnormal cold distinct in time from the next nearest events. Some phenomena and conditions are reported as spanning entire winter seasons and cold events can thus cross between calendar years. Supplementary table 2 (available at stacks.iop.org/ERL/8/024035/mmedia) gives the dates of all 70 cold events and their constituent phenomena and conditions.

Figure 2(b) depicts the frequency of the 70 cold events in 20-year periods from 431. Events range from 0 to 4 per period, with a median of 1 per period (mean 1.15). From 740 onwards, the series is characterized by long spans at or above the median (e.g. nine consecutive 20 year periods from 740 to 919), punctuated by smaller spans without events (e.g. two consecutive periods from 1340 to 1379). Before 740 the series exhibits frequencies at or below the median, which likely result from data paucity. Before c.600 the number of manuscripts providing coverage is reduced, as are the number of written entries in those that do provide coverage. Additional sources from Ireland are unavailable to compensate here, and the likelihood that cold events have gone unreported must be regarded as heightened for this earliest period.

Volcanic sulfate deposition in ice-cores allows a reconstruction of the incidence of explosive volcanism over past millennia (Hammer 1977, Robock and Free 1995, Robock 2000, Zielinski 2000, Gao et al 2008, Sigl et al 2013). To derive a history of explosive volcanism, we employ the non-sea-salt sulfate record from the GISP2 ice-core, which has been used extensively for this purpose (Zielinski et al 1994, 1996, Zielinski 1995, 2000, Gao et al 2008). Atmospheric sulfate is deposited in the GISP2 ice-core in all annual layers, but only a subset of layers register deposition from major explosive eruptions with the potential for large-scale climatic impact. To identify explosive volcanic events we posit the existence of two separable processes, each resulting in sulfate deposition in the GISP2 ice-core. Background deposition, we posit, arises from numerous relatively small sulfate sources, including quiescent volcanism. We approximate this background distribution as Gaussian. Major explosive eruptions, we posit, will result in sulfate deposition that falls outside this background distribution.

The GISP2 sulfate (SO₄²⁻) time series has a variable sampling resolution with a reported 2.5 year minimum (Mayewski et al 1997). 586 sulfate deposition values are available for our 1219 year period of interest with a mean SO₄²⁻ value of 36.9 ppb and assigned dates ranging from 431.80 to 1648.32 on the Meese–Sowers GISP2 chronology (Meese et al 1997). Starting with the largest of the 586 values, we iteratively apply the generalized extreme studentized deviate test to each SO₄²⁻ value between these dates. This is conceptually equivalent to testing whether the observed value is consistent with a background process Gaussian SO₄²⁻ distribution as represented by the other observed values. Each value identified as an outlier (at p < 0.05) is considered to register an explosive 'volcanic event' and removed from the distribution for the purpose of testing the remaining values. This method identifies 48 (8.2%) of the 586 SO₄²⁻ sample values as registering explosive volcanic events, with a minimum SO₄²⁻ value of 55.16 ppb distinguishing these values from those likely to register sulfate from background deposition only.

Using this record of volcanic events, we test whether explosive volcanism is associated with cold event occurrence. Because of potential recording biases that may impact upon the fidelity with which cold events are reported, we do not assume that the record of the Annals is continuous or complete, and hence seek a test for which intermittent reporting will not tend to bias the result. We consider that reported cold events correspond to volcanic events when occurring up to 5 years of each other. This window is comparable to that employed by Salzer and Hughes (2007) and accounts here for the reported 2.5 year minimum resolution of GISP2 SO₄²⁻ samples as well as the possibility of climatic impact before the date of notable sulfate deposition, and the possibility of lagged or multi-year climatic impact after eruptions. We adopt a null hypothesis stating that dates of cold event occurrence are unassociated with dates of volcanic events. We build a distribution for the number of correspondences under this null hypothesis by randomly reordering the 586 SO₄²⁻ observations 1 000 000 times. We omit from this exercise the time between GISP2 sample dates at 541.2 and 613.8 when GISP2 SO₄²⁻ values are unavailable due to core damage. After each reordering we sum the number of correspondences between cold and volcanic event dates falling within the above window and observe a mean value of 24.9 random correspondences.

We find in actuality that 37 (53.6%) of our 69 cold events correspond to volcanic events within this window (figure 3) (conversely 38 (79.2%) of 48 volcanic events correspond to cold events). We use only 69 of 70 cold events because no GISP2 data are available at the time of the 586 CE cold event, but we note in passing a major volcanic event at 585 in the Dye 3 Greenland ice-core (Clausen et al 1997). Employing our distribution of random correspondences under the null hypothesis, we find that the probability of observing 37 correspondences by chance alone is <0.003 (i.e. the observed correspondence is statistically significant at the 99.7% confidence level). Note that the mean number of years covered by the randomized ± 5 year windows from which we build our distribution of random correspondences (and from which we observe the mean of 24.9 random correspondences) is much less than the 528 total years in which correspondences could be counted if all 48 windows were maximally distributed (i.e. (2 × 5 + 1) × 48 = 528). This is because it is exceedingly unlikely that the 48 volcanic event dates will have no overlapping windows when independently distributed.

Zoom In Zoom Out Reset image size

To further evaluate the robustness of the observed correspondence between volcanic events and cold events, we conduct several further tests. We first repeat our analysis by iteratively varying the width of the correspondence window to include all integer values of 1 ≤ N ≤ 5. While the number of observed correspondences varies in each window, the null hypothesis that the correspondence arises purely by chance is rejected, in all cases, at p < 0.05 (i.e. >95% confidence).

In table 3, supplementary data (available at stacks.iop.org/ERL/8/024035/mmedia), we show that multiple cold events at times correspond to single volcanic events, and vice versa. This is physically plausible given the potential multi-year climatic impact of major eruptions, and the compounding impact of multiple closely-spaced eruptions. However, as an additional robustness test we examine whether the inferred non-randomness of the on-aggregate correspondence between cold events and volcanic events is dependent upon the correspondence of multiple cold events with single volcanic events. We repeat our analysis, enforcing a restriction that each volcanic event can only 'explain' one cold event within its window. For parity, we impose this restriction on both the re-count of actually-observed correspondences and the model of the null hypothesis, which we invoke 1 000 000 times to build a distribution of the number of such correspondences expected randomly. Under this restricted definition of correspondence, the significance of the result is found to be >99.9% (p < 0.001).

For the above analyses we employed an objective test to identify significantly outlying values in the GISP2 SO₄²⁻ series reflecting the incidence of major explosive eruptions, but previous estimates using independent methods also exist. Repeating our analysis using the somewhat smaller list of volcanic events identified from the GISP2 by Zielinski (1995), we again find a correspondence between volcanic events and cold events (± 5 year window) that is statistically significantly non-random at p < 0.01 (i.e. >99% confidence). As a final check, we repeat our analysis using the 21 October 2012 (Version 2) update to the Gao et al (2008) multi-ice-core reconstruction of global explosive volcanism, using a correspondence window of ± 2 years to account for the greater number and presumed accuracy of the cores contributing to this reconstruction. We again find a statistically significant correspondence between volcanic events and cold events at p < 0.01 (i.e. >99% confidence).

The calendar months in which cold events occurred are cited or identifiable in the Annals for 38 (54.3%) of our 70 events (figure 4). 102 months are cited in total because single events can span multiple months. January is most often cited, with 23 (or 22.5%) of the 102 total citations, followed by February with 19 (18.6%). More evenly cited are December with 11 (10.8%), November with 10 (9.8%), March with 11 (10.8%) and April with 10 (9.8%). Other months are rarely cited, with only 2 citations each for June and July. There is thus a clear winter-season bias in cold event occurrence, a bias that persists for cold events that both do and do not correspond to GISP2 volcanic events under the ± 5 year window. 47 (46.1%) of 102 months are cited for cold events without, and 55 (53.9%) are cited for cold events with a volcanic correspondence. January and February remain the most frequently cited, with 11 (20.0%) and 8 (14.5%) of the 55 months cited for cold events having a volcanic correspondence.

Zoom In Zoom Out Reset image size

While the dominant citation of winter months for cold events that correspond to volcanic events is revealing of a role for volcanism in winter cooling of the Northeast Atlantic region represented by Ireland, the infrequent citation of summer cold should not be taken as evidence of its absence here. Oppenheimer (2011) describes, for example, the human impacts in Ireland of the cold wet summer following the 1815 Tambora eruption. Instead, our record acts primarily as a proxy for anomalous winter-season cold, because the phenomena and conditions such as heavy snow and frozen water bodies that were most impressive to the human observers compiling the Annals (and which more objectively indicate the occurrence of anomalous cold) are very rare in summer in Ireland's mild maritime climate (Sweeney 1997, Dukes and Eden 1997). That the summer cooling expected after major eruptions has in fact occurred for at least some of our 48 volcanic events is suggested given that the only citations for June and July occur for cold events with a volcanic correspondence, as well as the marginally greater number of such citations for other non-winter months.