Island selection: We included the largest inhabited islands, using the minimum size criteria of at least 25,000 km2 in area24 and a minimum population size criteria of at least 100,000 people (in 2022). Both were arbitrary thresholds but were designed to make this study more relevant to considering islands with some potential capacity in terms of size and population to allow for surviving “nodes of persisting complexity”15. We included Australia in our list of islands even though it is a “continental” island. Also included were islands that are jurisdictionally complex in the modern era eg, have parts governed by separate nation states (eg, the islands of: Borneo, Hispaniola, Ireland, Isla Grande de Tierra del Fuego, New Guinea, and Timor).
Literature search
Literature searches were conducted during April to August 2022 using Google Scholar and the search term of “Tambora and 1815” and specific searches for each island using the search terms “Tambora and [island name/s]”. Historical studies of food prices and famines in each island were also searched for. Such searches were also conducted using Scopus and using “Elicit.org” (a digital research assistant for literature searches using an artificial intelligence system and access to 175 million articles). Specific island name searches were also conducted in the digital versions of five key texts, ie, those by Harington25, Wood26, Brönnimann and Krämer19, Klingaman and Klingaman27, and Behringer21.
Impact definitions
In terms of likely impacts of the Tambora eruption on islands in the 1815 to 1817 period, we considered weather/climate impacts to be those involving anomalous temperature and/or precipitation changes (as measured with instrumentation, documented by observers at the time, or from palaeoclimate studies eg, of tree-rings or coral). For adverse food production impacts we defined these as where crop failures or reduced harvests were reported or where food prices rose. For adverse food insecurity impacts we defined these as reports of increased hunger, increased begging, and reported famines. We did not automatically assume that new epidemics (eg, of typhus) reflected increased malnutrition, but documented the occurrence of these, given that they might reflect underlying malnutrition.
Reconstructed climate
We included results from a climate reconstruction, EKF400v228, which uses atmospheric-only general circulation model simulations (with sea-surface temperatures, land cover, and external forcings prescribed from reconstructions)29. The reconstruction estimates monthly climatological data for the 1600 and 2005 time period and builds on an earlier version (EKF400) published in 201730. The EKF400v2 reconstruction has performed well in describing the Central European drought of 1726–1728 and provides insights into the climate dynamics leading up to this extreme dry period28. In another case study, it also performed well in reconstructing El Niño Southern Oscillation (ENSO) effects in the 19th century28. The earlier version of this reconstruction (EKF400), has also informed the impact of volcanic eruptions on the late phase of the Little Ice Age31.
Island-specific data inputs into the EKF400v2 reconstruction covering the time of the Tambora eruption were available for just over half of the islands (51.6%, 16/31). These were mainly from tree-ring studies (51.6%, 16/31), but also from instrumental records (9.7%, 3/31), and other sources (16.1%, 5/31; eg, documentary sources and coral data). The mean number of data inputs per island was 1.1, range: 0 to 6 (see Table S1 in the Supplementary Information File 1 for details). But where such island-specific observational evidence was lacking, the reconstruction output was driven by the underlying model simulations.
In this analysis we focused on just the temperature anomalies given that these were statistically far more likely to be related to the volcanic forcing from Tambora, than precipitation anomalies (see Introduction)22. For each island we used the reconstructed temperature data for a single geographic coordinate, the latitude and longitude for the most populous city on the island (listed in Table 2). The estimates presented in the results were for the temperature anomalies relative to the 1779 to 1808 period. The latter was selected as the closest 30-year period which had no major known global volcanic forcing (ie, there was the 1783 Laki eruption in Iceland, but this was largely tropospheric). Each annual result was the mean of the monthly anomalies for that year, with each monthly result being the ensemble mean of 30 model realisations.
In addition to island-specific analyses, the impacts were analysed by: hemisphere, ocean, tropics/temperate zones, and proximity to a continent. For the latter we measured the distance from the nearest two points of land (ie, from the island to the nearest continent-based country), using the “measure distance” tool on Google Maps. As part of this we defined the main part of Australia as an island continent (ie, for measuring distances to Tasmania and both main islands of New Zealand). In the statistical analyses we used ANOVA or the Kruskal-Wallis test if the data was not normally distributed. Excel files of the reconstruction data are available in the Supplementary Information File 2 and File 3.