Bulletin of Volcanology

, Volume 67, Issue 1, pp 15–26

Mortality in England during the 1783–4 Laki Craters eruption

Authors

    • Department of GeographyUniversity of Cambridge
  • C. Oppenheimer
    • Department of GeographyUniversity of Cambridge
Research Article

DOI: 10.1007/s00445-004-0357-7

Cite this article as:
Witham, C.S. & Oppenheimer, C. Bull Volcanol (2004) 67: 15. doi:10.1007/s00445-004-0357-7

Abstract

1783/4 has been recognised as a mortality “crisis year” in the population history of England. This demographic incident coincides with the Laki Craters eruption, Iceland, which began in June 1783 and fumigated many parts of Europe with volcanic gases and particles. Many reports and proxy climate records implicate the volcanic cloud in meteorological anomalies, including notably hot 1783 summer conditions in England and a severe subsequent winter. We present here a detailed analysis of the geographical and temporal trends in English mortality data, and interpret them in the light of the climatological records and observations of the pollutant cloud. We show that there were two distinct crisis periods: in August-September 1783, and January-February 1784, which together accounted for ~20,000 extra deaths. In both cases, the East of England was the worst affected region. Possible causes for the two crisis periods are considered and we conclude that the timing and magnitude of the winter mortality peak can be explained by the severe cold of January 1784. The late summer mortality followed 1–2 months after the very hot July of 1783 and may also have been related to the weather, with the time lag reflecting the relatively slow spread of enteric disease or the contraction of malaria. However, it is hard to explain the entire late summer anomaly by these high temperature causes. We therefore consider that fine acid aerosol and/or gases in the volcanic haze may also have contributed to the unusual August-September mortality. Given that complex radiative and dynamical effects of the volcanic cloud are implicated in the climatic anomalies in 1783–4, it is likely that the Laki Craters eruption did play a role in the English mortality crises of the same period.

Keywords

LakiEnglandMortalityHealth effectsAerosolsTemperature changeVolcanic hazard

Introduction

The eruption of the Laki Craters, Iceland began on 8 June 1783 and continued for 8 months. During this period an estimated 122 Mt of SO2, 7 Mt of HCl and 15 Mt of HF were released to the atmosphere from a series of explosive fissures and vents, and from voluminous lava flows (Thordarson and Self 2003; Thordarson et al. 1996). These gases were emitted into the troposphere and, during the most intense phases of magma discharge, probably the lower stratosphere. Thordarson and Self (2003) have proposed an eruption chronology in which over 90% of these emissions were released by nine separate eruption episodes during the first five months. The impacts of the eruption on Iceland were severe, with over 24% of the human population on the island – some 9,000 people – dying during the ensuing “haze famine”. Contemporary evidence suggests that the human impact may also have been more wide ranging. Reports from across Europe attest to the presence of an atmospheric haze or fog at various times in the summer and autumn of 1783. This phenomenon is believed to be pollution from the volcano (Grattan 1998; Stothers 1996; Thorarinsson 1981). In some instances, sources recall damage to vegetation coincident with the haze (Grattan and Brayshay 1995) and in some European cases an effect on human health in the form of headaches and respiratory discomfort is reported (Durand and Grattan 1999; Grattan and Pyatt 1999).

The earliest references to the appearance of haze over England date to 12 June 1783 (King and Ryskamp 1981) and its presence is recorded over periods of days to weeks hence. Details of mortality in England at this time are contained within the Population History of England data set (Schofield 1998), which has previously been analysed in terms of “harvest years”, where the twelve-month period begins in July. Mortality crisis months and years are recognised within the data (Wrigley and Schofield 1989), of which, very few major crises occurred after the 1720s. The period July 1783 to June 1784 was one of those recognised as a crisis year, ranking 26th in severity out of all the harvest years in the period 1540–1871. In terms of national death rate, mortality in 1783/4 was 16.7% above trend (Wrigley and Schofield 1989), equivalent to approximately 30,000 extra deaths. The cause of this mortality increase was unclear to the original workers, but was subsequently linked to the spread of volcanogenic pollution from the Laki Craters (Durand and Grattan 2001). Further analysis of the population data set in terms of summer (July-September) mortality by Grattan et al. (2003) revealed a notable mortality crisis in the summer of 1783, which they attributed to the Laki haze.

If Laki emissions were responsible for all the estimated English national mortality increase then three times as many people died in England as a result of the eruption than in Iceland itself. The aims of this paper are to advance our understanding of the consequences of the Laki Craters eruption and examine the mechanisms whereby morbidity and mortality could have been affected during this period. We provide an alternative and extended analysis of the English mortality data from the period during and after the appearance of the Laki haze over Europe. We use statistical Z-score techniques to analyse the mortality data and extend the time-range compared to previous work to look at winter mortality. We hypothesise that temperature extremes were responsible for some of the mortality increases and use regression analysis to test this. The influence of the eruption emissions on the regional climate and their possible direct impacts on human health are then considered and other potential causes of mortality at the time are discussed.

The Data

The primary source of information we use is the Population History of England data set (Schofield 1998; Wrigley and Schofield 1989). This data set comprises burial, marriage and baptism records for 404 English church parishes in 39 counties (Fig. 1) and spans the period 1538–1871. Parishes are areas that were served by a single church or clergyman, whereas counties are much larger territorial divisions of the country. The data represent only about 7% of the English population at the time and are not evenly distributed amongst the counties. This is due to the use of volunteers in the collection of records and the need for completeness in those included. Wrigley and Schofield (1989) explain in detail the uncertainties and limitations associated with the data and their methods of collection. The main points of note here are that the parishes were not randomly sampled and are not necessarily representative of the surrounding areas. Additionally, not all parishes contribute to the entire database, but the most comprehensive coverage is from 1662 to 1811, spanning the time period of most interest in this study. The time between death and burial was typically very short (within three days) so that burials are a very good proxy for mortality (Wrigley and Schofield 1989) and this term will be used henceforth. Unfortunately, details such as cause of death, and age and sex of the deceased, are not available in the data set, as such information was rarely recorded in parish registers. The lack of such demographical detail is a major limitation of epidemiological interpretation of the mortality patterns.
Fig. 1

The 39 English counties for which some parish data exist in the Population History of England data set (shaded). Data from the counties named are discussed in the text

Temperature data for the period of interest are taken from the database of mean monthly temperatures for Central England, originally compiled for the period 1659–1973 (Manley 1974), and updated since (Parker et al. 1992). This record is composited from observations made at different locations and at different times of the day using equipment appropriate to the period. As a monthly and regional average of probable 24-hour means, it does not capture daily or localised extremes of temperature, but it is considered a good proxy for general temperature trends.

Methods

Monthly totals of mortality for the whole data set were calculated first to indicate which months contributed most to the attribution of 1783/4 as a national crisis year. Next, the individual parish data were considered. Data on the population size of the individual parishes are unavailable for the eighteenth century, as the first national census was taken only in 1801. Patterns of in- and out-migration, religious non-conformity, and population growth in the period 1784–1901, mean that despite being only 18 years later, the 1801 census does not represent a good proxy for 1783 population distribution. Consequently, rates of mortality by parish cannot be calculated. This causes a problem in that the parishes were not of equal size (burial numbers imply at least two orders of magnitude difference in population size), so direct comparison of numbers of burials per month between parishes is not possible. In addition, the eighteenth century was a period of overall population growth, so any temporal analysis of mortality must account for the general trend.

To account for this underlying trend and overcome the problem of population size, we calculated mortality Z-scores for each parish:
$$ Z - score = \frac{{M - \overline{M} }} {\sigma } $$
(1)

Where M is the observed monthly mortality, and the monthly mortality means (M̄) and standard deviations (σ) apply to the fifty-year period of 1759 to 1808 for each individual parish. This fifty year period was considered adequate to remove the long-term trend in mortality whilst being unaffected by short-term fluctuations. It was centred on 1783 and 1784 to allow the same mean to be used for both years and make the results directly comparable. Such a technique has been used in other demographic studies (Charbonneau and Larose 1979; Dobson 1997). Mortality at the time varied naturally and significantly with season. Calculation of Z-scores on a monthly basis in 1783 and 1784 allowed these variations to be accounted for. Positive Z-scores demonstrate mortalities higher than the mean, and negative scores indicate mortality below the mean. We consider a Z-score >2 to represent a significant mortality event or “crisis” (Charbonneau and Larose (1979) also used this threshold in their investigation of historical mortality crises). The Z-scores were used to determine which parishes were the worst affected by crises and how the crises were distributed in space and time. Sensitivity tests using different Z-score thresholds and mean periods produced similar results to those presented here and are not discussed. The use of percentages as a means of inter-comparison was avoided due to the small number of burials recorded in some parishes, and the signal bias that this would cause. Additionally, percentage results give mortality in decimal numbers, which leads to problems in rounding to meaningful numbers of deceased during calculation stages.

We compiled data for each of the counties from the individual parish level records, and then calculated Z-scores in a similar manner, accounting for variations in numbers of contributing parishes where necessary. The use of percentage change would have been more valid on a county scale as the values involved are much larger, however the choice of percentage above which a crisis is defined is not clear. Wrigley and Schofield (1989) chose 10% when defining crises on an annual time period, but 25% on a monthly scale, whereas Dobson (1997) used 40% as the cut-off value. For consistency, the Z-score threshold of two was retained for the counties and similar analysis conducted.

To determine whether there was any significant geographical pattern in the distribution of parishes experiencing crisis at any time, we performed cluster analysis for mortality hot-spots using the SaTScan space-time scan statistics program (Kulldorff et al. 1998). The scan statistics analysis uses a Monte Carlo procedure to pinpoint the location of, and test the significance of, spatial clusters. A mortality hot-spot is a geographical cluster of parishes experiencing a significantly higher than expected number of crises in time and/or space. The presence of a hot-spot might indicate a regional control over mortality. The Bernoulli model version of the program was used with mortality cases defined as any parish and month where Z-score >2.

To examine the influence of temperature on mortality on a monthly scale in the eighteenth century, we correlated the Central England temperature data with detrended monthly totals of mortality for the fifty-year period 1759–1808 (exclusive of 1783 and 1784)(e.g. Fig. 2). We then used linear least-squares regression analysis to determine the magnitudes of the relationship between monthly temperature and mortality. Using these relationships, we were able to calculate expected changes in monthly mortality per degree of temperature change for each month. Mortality changes were calculated for temperature changes in the same month and for those in each of the preceding three months to examine whether there were any temperature lag effects. The linear least-squares regression relationships also allowed us to determine what proportion of the observed 1783/4 mortality increases could be attributed to the extreme July and January temperatures.
Fig. 2

Correlations between (i) January total mortality and January temperature and (ii) September total mortality and July temperature, for the latter half of the Eighteenth Century. Vertical dashed lines show temperatures in January 1784 and September 1783 respectively

Results

Mortality Data

Two mortality peaks are evident in the results (Fig. 3), the first in August-September 1783, followed by another in January-February 1784. The combined mortality for August-September is 40% higher than the mean and that for January-February 23% higher. Nationally this would have been equivalent to approximately 11,500 and 8,200 additional deaths in these periods, respectively. The fifty-year mean data (Fig. 3) show the natural cyclical seasonality of mortality that existed in England, with increased mortality in winter and decreased mortality in the summer. High mortality in January and February was therefore not unusual. Conversely, August and September were usually the months of minimum mortality so the peak at this time is particularly striking. Mortality levels return to the trend in June 1784, providing a cut-off date for the period of interest.
Fig. 3

Comparison of monthly mortality Z-scores (series have been normalised by month) for the period 1782–1785 and the 50-year mean, both calculated with respect to the calendar year mean. Arrows indicate the timing of significant phenomena related to the Laki eruption and the shading shows the duration of the eruption

Comparison of the monthly totals for 1783 and 1784 to those of the years in the fifty-year series reveal that the calendar year 1783 was particularly anomalous. It was the only year in the series in which September was the month of maximum mortality and where mortality in September was greater than that in January. It was also the only year in which the month of minimum mortality occurred in the first five months of the year. This unusual monthly pattern indicates the action of a forcing on mortality unrelated to the normal seasonal trend, potentially of a type unique to this year of the series.

In the period June 1783–March 1784, half of the parishes experienced at least one crisis month. Most of these crises were of short duration, on the order of one month, although in some parishes, crisis months occurred in succession. The worst affected parishes experienced six months of crisis in this period. The locations of the parishes experiencing more than three months of crisis are shown in (Fig. 4). September 1783 was the most severe crisis month with ~17% of all parishes recording a crisis (Table 1). June 1783 was about normal, however. A similar temporal pattern of crises occurred at the county level (Table 1). Both data series indicate that mortality remained above normal for the three months following the January-February mortality peak.
Fig. 4

Distribution of parishes experiencing (i) 3–6 and (ii) 1–2 crisis months in the period June 1783–March 1784. The four named parishes in (i) experienced five or six crisis months. Parishes not experiencing any crisis months in the period are depicted as open squares in (ii). Parish positions are plotted as the location of the parish church to the nearest 100 m

Table 1

Percentage of parishes and counties experiencing a crisis (Z-score >2) in each month in 1783 and 1784

Month

Parishes

Counties

1783

1784

1783

1784

Jan

3.2

7.9

2.6

10.3

Feb

1.7

8.2

0.0

12.8

Mar

3.5

7.2

0.0

10.3

Apr

2.0

5.2

0.0

5.1

May

3.0

7.7

0.0

10.3

Jun

4.0

3.7

2.6

7.7

Jul

6.5

3.5

0.0

0.0

Aug

9.5

5.0

20.5

2.6

Sep

17.2

4.2

23.1

0.0

Oct

5.0

5.7

10.3

2.6

Nov

7.7

4.7

10.3

5.1

Dec

7.7

4.0

5.1

0.0

August and September are the months of maximum above-trend mortality in half of the counties in 1783. In Bedfordshire, mortality in August and September 1783 was more than five times the mean, and September was so severe as to be the month with the highest mortality in the whole of the eighteenth century. Bedfordshire parishes predominate in both crisis periods (Fig. 4) and this county seems to have been more severely affected than others were.

The occurrence of crisis mortality varies strongly with location and its geography is dependent on the spatial scale used (parish versus county). The distribution of the crisis parishes in Fig. 4(ii) is representative of both mortality peak periods. This demonstrates that crisis parishes were not limited to one part of England during either period. The distributions of counties at crisis during the two mortality peaks (Fig. 5) reveal a southward shift in the main region of crisis between the two periods, with the eastern counties affected at both times. Comparison of Fig. 5 with Fig. 4 shows that the county level data hide this incidence of crises throughout the country. The counties registering the most prolonged crisis periods and highest Z-scores are Bedfordshire, Cambridgeshire and Leicestershire in 1783, and Cambridgeshire alone in 1784.
Fig. 5

Distribution of crisis occurrence at the county level. “Summer” counties experienced crisis during the summer 1783 mortality peak only, “Winter” counties experienced crisis during the winter 1784 peak only, and “Both” counties experienced crisis during both peaks. Counties for which data were unavailable are also shown. The labelled counties are referred to in more detail in the text

The statistical cluster analysis of all parishes in 1783/4 reveals the presence of a significant (p=0.001) mortality hotspot in central and east England in the period July–November 1783, inclusive (Fig. 6). Although the spatial scale is different to the county data (Fig. 5), it confirms the regional pattern recognised for the summer peak. The result suggests that a regional control may have influenced mortality during these months. No clusters were detected for the winter 1783/84 period, implying that crisis mortality at this time was more widespread.
Fig. 6

Result of hot spot cluster analysis. Filled squares are those parishes determined as being within the crisis mortality hot spot cluster. Open squares are the remaining parishes. The shaded circular area encompasses all the identified parishes and indicates that east central England was the most significant crisis region in the period 1783–1784

Temperature Data

Central England temperature data show that the summer of 1783 was particularly hot, and that the first months of 1784 were among the coldest on record (Fig. 7(i)). The mean temperature in July 1783 was 18.8 °C (the hottest July in the whole data set except for 1983), whereas the temperature for January 1784 was –0.6 °C, over 3 °C below the month’s 30-year average centralised on 1783/4. Meteorological records from the period show that a high-pressure system existed over Northern Europe for many weeks in the summer of 1783 (Kington 1988). This system may explain the high summer temperature. Fig. 7(ii) shows that the winter mortality peak coincided with the extreme January temperature and subsequent colder than average months, but that the summer mortality peak lagged behind the extreme high temperature by 1–2 months.
Fig. 7 (i)

Monthly temperatures (°C) in 1783 and 1784 compared to the mean of the 30-year period 1769–1798. (ii) Monthly temperature and burial anomalies in 1783 and 1784 with respect to the 50-year means

Results from the temperature and mortality correlations demonstrate that the most significant (p=0.001) correlations exist between January temperature and mortality in January (r=-0.68) (Fig. 2(i)) and February (r=-0.52), and July temperature and mortality in August (r=0.51) and September (r=0.54) (Fig. 2(ii)). The statistical significance of these correlations and the anomalous temperatures in July 1783 and January 1784 suggest that part of each of the two mortality peaks can be explained by a temperature-related effect. Fig. 8 shows the monthly mortality changes attributable to January and July temperatures. The plots show the effect of these temperatures on mortality in the same month (lag=0 months) and in the following three months (lag=1–3 months). Mortality response to a cold January temperature would have been immediate, whereas the response to a hot July temperature would have lagged by one to two months, having its greatest effect on mortality in September. Results from the regression show that the extreme July temperature can explain only ~30% of the observed variation above trend in August and September 1783, whereas the January temperature explains ~50% of the January and February 1784 above-trend mortality (Table 2). The extreme values of the two temperatures mean that they represent end points of the mortality-temperature series so caution should be exercised with these estimated mortality values, as to some extent they are extrapolations of the series. The confidence intervals on the predicted mortalities also give large error ranges to the explained above-trend mortality.
Fig. 8

Percentage changes in mortality (black lines) with 95% confidence intervals (dashed lines) for the month of the temperature (lag=0) and the following three months (i) per degree C increase in July temperature, (ii) per degree C decrease in January temperature (the February mortality variation series was not normally distributed so is included for reference only)

Table 2

Comparison of the mean monthly mortality, the total mortality predicted using the July 1783 and January 1784 temperatures in the regression equations, and the actual mortality recorded in 1783/4. The percentage of the above-mean mortality that can be accounted for by the extreme temperature (final column) is calculated from these values. Values shown are for all parishes in the database

Month

Temperature used

Detrended mean mortality

Predicted mortality (±95% CI)

Actual mortality

Percentage explained by temperature

Aug 1783

July 1783

918

1,025 (146)

1,282

29

Sep 1783

July 1783

938

1,073 (169)

1,376

31

Jan 1784

Jan 1784

1,162

1,331 (232)

1,525

47

Feb 1784

Jan 1784

1,209

1,314 (208)

1,415

51

Discussion

September 1783 was the most fatal month in the two-year period 1783–1784 in terms of both above-average mortality and the number of crisis parishes. It was followed, in decreasing order of severity, by August 1783, February 1784 and January 1784. The distribution of the crisis parishes and counties, and the cluster analysis shows that the two periods of mortality crisis were not felt equally over the whole country. The east of England bore the brunt of the increase in mortality and was at crisis during both mortality peaks. The different distributions of crisis by county in the two peak periods suggest that the causes of mortality at these times were different and unrelated.

Crisis mortality in 1783/4 could have been the result of a number of contributing factors including fever and “ague” epidemics, atmospheric pollution from the Laki plume, temperature extremes, weather-related increases in disease, and food shortages. Below average mortality in the 1782/83 harvest year (Fig. 3) may also have contributed to higher mortality in 1783/84, as more at-risk people may have survived than was usual. We consider next the evidence for the various potential contributing factors using accounts from historical sources, such as the Gentleman’s Magazine (a London-based monthly digest of letters and reports from around Britain, the British Empire and Europe) and diaries, to illustrate some of the causes of mortality reported at the time.

Weather

The months of July and August 1783 brought violent thunderstorms to England (Gentleman’s Magazine 1783; Grattan and Brayshay 1995) and there are many reports of damage caused to ships, properties and crops by these storms. Writing about the storms experienced in July, one correspondent in the Gentleman’s Magazine (Aug 1783) proposed that there was “...no year upon record when the lightning was so fatal in this island as the present...”. In addition to direct lightning strikes on people, causes of deaths attributed to the storms include collapse of buildings following storm damage or lightning strike, and deaths of ladies occasioned by terror at the lightning or convulsions in the bowels caused by thunder (Gentleman’s Magazine, Sep 1783). These accounts, whilst giving an intriguing glimpse into contemporary weather and culture, and suggesting unusual causes of mortality, do not adequately explain the magnitude of the mortality peak in August and September. A more plausible climatological candidate for the cause of the mortality peaks is the influence of temperature.

Modern epidemiological studies (Braga et al.2001) show that a relationship exists between daily temperature and mortality. In general, the effects of cold temperatures on mortality today persist for days, whereas those of high temperatures are restricted to the same day or the day after. The eighteenth century temperature-mortality relationships show that mortality would have increased immediately in response to the extreme January 1784 temperatures and that at least half of the excess mortality can be explained by the cold weather. This is consistent with the modern epidemiological pattern, although does not prove that causes of death then were similar to those now. The relationships also show that, unlike today, the hot July temperature would have had its greatest effect on mortality in September, coincident with the observed peak.

These findings are based on average temperature data so do not represent the true temperature variation that would have existed over all the parishes at any one time. They are sufficient though as a first order approximation of the effect of temperature on English mortality. Maximum temperatures on a local scale would have been much higher than the July mean of 18.8 °C. Cowper, for instance, records a midday temperature of 83 F (28 °C) on one day for central England (King and Ryskamp 1981). But it seems unlikely that these local conditions would have been of sufficient duration or magnitude to account for all of the unexplained variance in mortality, and that much of the summer mortality peak cannot be explained by temperature.

Anecdotal reports of the severity of the winter in England can be found in contemporary diary entries (Denning 1995; Turner 1936; Woodforde 1985). Some records indicate direct impacts of the cold weather on mortality: “In England men were found frozen to death on roads and in the open country and great apprehensions were entertained for the poor, who it was feared would freeze to death”(Gentleman’s Magazine, Jan 1784). Many deaths in England at this time were attributed to the “...inclemency of the weather” (Gentleman’s Magazine, Jan and Feb 1784), which included cold temperatures, deep snows, ice and hard frosts over much of the country. The above average mortality seen in the months following the 1784 peak (Fig. 3) could also be related to this cold period, which appears from documentary sources to have lasted until the end of April 1784 (Turner 1936). In addition to the direct affects of temperature and weather on mortality, diseases such as typhus were prevalent in the eighteenth century. Typhus was spread by parasitic lice and was passed through contact with other humans or cloth carrying the lice. Cold weather would have prompted closer contact between people trying to stay warm and increased the transmission of the disease and therefore mortality (Dobson 1997). Pneumonia, bronchitis, influenza and other respiratory tract diseases may also have brought forward deaths amongst the elderly and infirm (Wrigley and Schofield 1989).

The lag in mortality increase following July 1783 indicates that the main impact of high summer temperature on mortality was indirect during this period. The most likely explanation for the lag is that high summer temperatures heightened the transmission of certain disease vectors. For example, very hot summers increased the soil temperature, thereby raising the rate at which flies’ eggs hatched and so increasing the risk of infection from contaminated food. A contemporary quote from White in Selborne, England demonstrates the reality of the hot temperatures and suggests that insects may have played a role: “All the time the heat was so intense that butchers’ meat could hardly be eaten on the day after it was killed; and the flies swarmed so in the lanes...” (in Wood 1984). Hot temperatures also increased the breeding of mosquitoes and were necessary for the malaria they carried to become infectious (Dobson 1980). Consequently, burials from malaria were high in the autumn months following warm summers. Transmission of malaria was normally confined to areas of marsh and fenland in England, these being the breeding grounds of mosquitoes (Dobson 1980). Some of the eastern areas worst affected in the summer crisis were in these areas. However, Kent, normally one of the counties worst affected by malaria, did not experience a crisis during autumn 1783 suggesting that occurrence of malaria in the other counties may not have been substantially increased.

Typhoid and dysentery and other digestive tract diseases are strong contenders for the cause of mortality in infants and young children at this time, as their transmission was heightened by hot summer temperatures. Such diseases were debilitating and took longer to kill. They also spread more slowly than other illnesses, as food and water became progressively contaminated by either human or insect vectors (Wrigley and Schofield 1989). These “summer” diseases (Dobson 1997) are probably the cause of the lagged relationship between July temperature and August and September mortality seen in Fig. 8.

Laki Craters emissions

The mortality data presented were first examined due to the coincidence of the crisis year (on a harvest year time-scale) with the eruption of the Laki Craters. (It is of note that if the Population History of England had been first presented in calendar year totals as opposed to harvest years it is unlikely that 1783 or 1784 would have appeared as a crisis year and may not have been subjected to such investigation). In evaluation of historical evidence in such circumstances, it is important not to confuse coincidence with cause. In this instance, the timings of the VEI 4 eruption of Asama volcano in Japan (3 August 1783) and the VEI 3+ explosive island-forming eruption of Reykjaneshryggur (1 May to 15 August 1783) to the south-west of Iceland tie in much better with the first observed English mortality crisis than the intense phase of the Laki eruption in June 1783. The evidence for the Laki craters eruption being the cause of the historical hazes that appeared throughout Northern Europe at the end of June and in July 1783 is strong however (Grattan and Pyatt 1999; Stothers 1996; Thordarson and Self 2003). We now consider whether the contemporaneity of the eruption and the mortality peaks could have been more than coincidence.

Volcanic aerosol was the most likely cause of the haze over Europe, as none of the major volcanic gases absorb or scatter visible light significantly (Ammann and Burtscher 1993). A growing body of evidence (Baxter 2000; Delmelle et al.2002; Dockery et al.1992; Faive-Pierret and Le Guern 1983; Zanobetti et al.2002) on the effects of volcanic-gases and aerosols on human health suggests that if the Laki emissions were present at high enough concentration they could have been responsible for ill health and maybe mortality amongst populations at distance from the volcano where the haze was present. Some contemporary sources (summarised in Durand and Grattan (1999) report human health affects associated with the haze lending credence to this hypothesis. For the Laki emissions to have been a feasible cause of mortality in England, the following conditions would have needed to have been met at the time: (i) transport of the volcanic plume over England; (ii) presence of the plume in the surface boundary layer; and (iii) harmful concentrations of pollutants in the plume. Historical reports such as the following from the city of Lincoln: “A thick hot vapour had for several days before [July 10] filled up the valley...so that both the sun and moon appeared like heated brick-bars” (Gentleman’s Magazine, July 1783); together with previous work (Stothers 1996), indicate that conditions (i) and (ii) were satisfied during June and July 1783. It has been proposed (Grattan 1998; Thordarson and Self 2003) that the emissions were brought down into the lower troposphere over Europe by the high-pressure system mentioned above. The presence of such high-pressure conditions prevents vertical dispersion, so any emissions reaching the lower troposphere over Western Europe would have been trapped. If this system is considered to be the main way Laki emissions were brought to the lower atmosphere in the region, then they would only have been present during the periods of high-pressure in the summer months.

The first known observations of the haze in Western Europe were from England on 12 June 1783 (King and Ryskamp 1981) and France on 14 June 1783 (Thordarson and Self 2003). This provides a maximum travel time from Iceland to Europe for the pollutants of 5–7 days. Using this as a rough estimate of travel time (exact times would depend on precise meteorological conditions and the input height-range of pollutants to the atmosphere, neither of which are known to a suitable accuracy) and the eruption emissions time-line of Thordarson and Self (2003), we estimate that maximum concentrations would have occurred over Europe in the first few weeks in July. Observations from Mannheim, Germany (converted into a numerical haze index by Thordarson and Self (2003)) suggest that a second period of high haze concentration occurred in this area in the last week of August. There is no evidence as yet to support a similar occurrence over England, although this would coincide with the summer mortality peak.

Atmospheric modelling of a Laki-like plume by Stevenson et al. (2003) suggests that a large proportion of the sulphur dioxide emitted at the vents could have remained in its gaseous form in the troposphere before being deposited. High concentrations of sulphur dioxide (SO2) are hazardous to human health, but modelled surface boundary layer concentrations for sulphur dioxide were less than 20 ppb, too low to cause any adverse human response. Modelled sulphate aerosol concentrations were similarly below those considered hazardous to health. This model did not however use meteorological conditions representing those known to be present at the time of eruption or fully represent the eruption chronology.

Brugmans (1787) (in Grattan and Pyatt 1999) writing in mainland Europe reported a very strong smell of sulphur coincident with the haze, as did the Bristol Journal in England on 19 July 1783. The detectable odour range of SO2 is approximately 0.5 ppm to 1.5 ppm (Baxter 2000; Wellburn 1994). Although such concentrations may have caused health effects such as eye sensitivity and respiratory resistance, they would not have been responsible for death on their own. The detection of odours in England does not prove that gases were present in harmful concentrations or necessarily reflect their levels, but it does suggest that SO2 concentrations were more than order of magnitude higher than those modelled by Stevenson et al. (2003).

Adverse health effects from exposure to volcanic gases are generally acute in nature, so any impacts would be expected to occur and be noticed during the period of contact with the gas. It is therefore of some note that June and July 1783, when the haze seems to have been at its thickest in England, were about normal in terms of monthly mortality. Concentrations of sulphate might have been a more likely cause of mortality. There is considerable evidence linking atmospheric aerosols, measured as particulate matter with aerodynamic diameter <10 μm (PM10) and <2.5 μm (PM2.5), to daily mortality and hospital admissions (Schwartz 2001; WHO 1999). The estimated travel time of the Laki plume to Europe is sufficient for the formation of a considerable mass of sulphate aerosol through oxidation of SO2. Tropospheric volcanic sulphate aerosols are typically <2.5 μm in diameter (e.g Allen et al. 2002; Mather et al.2003) and it is recognised that the effects of an increase in fine particulate matter (PM2.5 or smaller) on health are worse than those of PM10. The scale of mortality increase predicted in modern studies is only a few percent however (e.g. WHO 1999), not the 40% seen here. The use of current epidemiological results in the context of eighteenth century mortality as given above and also used by Grattan et al. (2003) is fraught with caveats. Modern studies generally relate to pollution from anthropogenic emissions, which exert different toxic effects to gases and particles from volcanoes. In addition, they usually associate increased particulate matter with increases in mortality on the same or following day, and do not consider longer time periods. Applying the results from modern time series analyses to spatially-distributed historical monthly data therefore requires a high degree of caution.

The decrease in Laki emissions from the end of August (Thordarson and Self 2003) implies that, even if transport mechanisms had remained the same, atmospheric concentrations in early 1784 would have been much lower. This suggests that the January-February mortality peak is unlikely to have been caused directly by high pollutant concentrations. Provocation of respiratory and cardiovascular illnesses by increases in atmospheric particulate matter may have left people more susceptible to the affects of temperature at this time though.

Volcanoes and climate change

We have shown that some of the rise in mortality in 1783/4 can be explained by extremes in English weather. Volcanic eruptions are known to cause climatic change and it has been proposed (Grattan and Sadler 1999; Robock 2000; Thordarson and Self 2003) that the Laki Craters eruption products could have been responsible for these temperature extremes. Observations of English, European and North American temperature changes (Manley 1974; Thordarson and Self 2003) following the Laki eruption suggest that not only was the winter of 1783/4 particularly cold, but that mean surface temperatures remained below average for the following three years. However, very cold January temperatures were not unusual in England during the second half of the eighteenth century. 1763, 1780 and 1795 all experienced colder January mean temperatures than 1783, and had coincident mortality peaks of a similar magnitude. These other cold events are presumed not to have been caused by volcano induced cooling, suggesting that the 1784 mortality peak could have been unrelated to the Laki eruption.

Grattan and Sadler (1999) suggested that the extreme summer temperatures might also have been caused by the volcanic emissions, with increased coarse particulate matter in the atmosphere causing atmospheric heating. Alternatively, the presence of the high-pressure system over Western Europe during June and July 1783 (Kington 1988) would have caused high temperatures and may be the sole cause of the apparent localised extreme. Little modelling has been carried out on the climatic impacts of high-latitude, long-duration eruptions and further work is needed to resolve what proportion, if any, of these localised climatic changes could be attributed to the Laki eruption and hence its impact on mortality.

Epidemic

Epidemics of disease were common during the eighteenth century. In 1783, the threat from smallpox had been reduced due to the start of inoculation against the disease but other illnesses such as typhoid, dysentery and typhus were common, and fevers and agues were frequently reported causes of mortality. The spread of some diseases would have been increased by extremes of temperature as described above. A protracted period of ague in1779–1782 suggests that by 1783 the population of England may have already been in a weakened state and susceptible to any further forms of illness. The spread of disease in 1783/4 may have been hastened by the return and dispersal of soldiers returning from the American War of Independence, which ended with the Peace of Versailles on 3 September 1783. These men may also have introduced new diseases to the country.

Evidence for the prevalence of a fever epidemic in the autumn of 1783 comes from the memoirs of Rev. Charles Simeon of Cambridge (Carus 1847) who travelled through central England in August and September 1783. On his return to his Cambridge parish on 19 September 1783, Simeon wrote “...many whom I left in my parish well are dead, and many dying; this fever rages wherever I have been.” (Carus 1847). The memoir implies that the illness affected previously healthy persons. The diary of James Woodforde (in Norfolk) and the letters of William Cowper (in Bedfordshire) also allude to the widespread nature of a fever-type illness. On 2 September 1783 Woodforde records that all three of his male servants were “bad” and that: “Almost all the house ill in the present disorder... It is almost in every house in the village...” (Woodforde 1985). Cowper writing on the 8 September 1783 states that “The epidemic begins to be more mortal as the Autumn comes on” (King and Ryskamp 1981).

Food Shortages

No documentary evidence has been found to support a large-scale harvest failure in England in the summer of 1783, but sources do indicate that some damage to crops and vegetation coincided with the observation of the haze over England (Grattan 1998; Grattan and Charman 1994). Damage to crops may have caused localised food shortages and increased stress on the health of those living in these regions, but does not appear to be the overriding cause of the mortality increases. Harvest prices at the time were higher than usual (Dobson 1997; Post 1977), but statistical analysis has shown that the impact of high prices on mortality at this time was weak (Dobson 1997; Wrigley and Schofield 1989).

Conclusions

Through analysis of monthly burial data we have revealed that two periods of mortality crisis occurred in England during the Laki Craters eruption. The first mortality crisis peak occurred in August and September 1783, nearly two months after the start of the eruption and the first reported appearance of haze in England, and the second peak occurred in January and February 1784, with mortality remaining above normal in the following two months. If the parish data are assumed to be representative of England as a whole, then the peaks represent ~19,700 extra deaths in the country during this period.

Limitations in information from the period demand caution in ascribing causes to the two periods of crisis. A possible cause of the August-September 1783 peak was the anomalously high July temperature, which would have increased summer enteric diseases and the spread of malaria in subsequent months. We cannot be certain what the concentrations of Laki gases and/or aerosols were or what their role in the hot temperatures might have been, but they may have caused direct health effects. Documentary evidence suggests that the worst effected eastern counties may have suffered from a fever epidemic in early September and this might explain an apparent regional crisis mortality. Whether this epidemic was related to atmospheric conditions at the time is not clear. No evidence has been found for food shortages in England at this time. The exact variation in mortality explained by the different causes is hard to determine. The magnitude of the summer peak is highly anomalous in the long-term seasonality series suggesting that temperature alone is not responsible for the increase, but the large error range in the temperature-mortality relationship begs cautious interpretation. The evidence for the direct impact of Laki gases and aerosols on humans in England is similarly inconclusive. In particular, the lag between the haze observation in England and the mortality peak does not fit with the rapid health impacts observed today following air pollution incidents.

We attribute the timing and magnitude of the January-February 1784 peak to the severe winter temperatures. Documentary evidence proves that the winter weather directly caused an increased number of deaths. The cold temperatures would have also increased the spread of diseases such as typhus. There is no evidence for the presence of haze, epidemics or food shortages in this period, but people weakened by the summer phenomena would have been more susceptible to cold and disease at this time. The influence of the Laki emissions on the climate system may have aided the extreme winter temperatures and hence mortality, but the coincidence of other cold Januaries with mortality peaks suggests a complex scenario.

The roles played by different causes of mortality in the peaks of summer 1783 and winter 1784 may never be conclusively known. However, further work on modelling of the emissions from the Laki Craters eruption will help resolve the uncertainties in their role on regional temperatures and the concentrations and height profile of potentially harmful components of the haze. If the eruption were solely responsible for the two English mortality peaks, this would treble the number of known deaths resulting from it and make Laki the third most fatal eruption in history following Tambora 1815 and Krakatau 1883. Furthermore, England appears to have been under a haze for a shorter duration than other parts of Northern Europe, where reports of associated phenomena seem more severe. If the climatic situation that existed in England had prevailed in other Northern European countries (documentary evidence suggests that this was the case) and the causes of mortality proposed here are correct, corresponding and potentially greater increases in monthly mortalities could be expected in these countries in the period June 1783 to March 1784.

Acknowledgements

This manuscript has benefited from discussion with Richard Smith and Robert Haining. We thank Peter Baxter and John Grattan for incisive reviews of the manuscript, and Anna Hansell for additional comments and suggestions that were of great help in its revision. CW gratefully acknowledges support from the UK NERC (award number NER/S/A/2001/06108).

Copyright information

© Springer-Verlag 2004