Identifying the determinants of insular biodiversity at large scales remains a question in biogeography. We conducted a global test of island biogeography theory by evaluating the importance of island physical, environmental, and historical characteristics on mammal species richness and endemism. We quantified the effects of island characteristics while accommodating variation among biogeographic realms by fitting generalized linear and mixed models. Analyzes were also performed separately for bats and non-volant mammals. Diversity patterns were most consistently influenced by the physical characteristics of the islands. Area positively affected mammal diversity, in particular the number of non-volant endemics. Island isolation, both current and past, was associated with lower richness but greater endemism. Flight capacity modified the relative importance of past versus current isolation, with bats responding more strongly to current and non-volant mammals to past isolation. Environmental effects on biodiversity were more variable, with a tendency for greater effects on endemism than on richness. Unexpectedly, climate change velocity was positively associated with endemism. In line with island biogeography theory, we found that area and isolation were among the strongest drivers of overall mammalian biodiversity. Moreover, our results support the growing evidence on the importance of past conditions on current patterns, particularly on non-volant species.

We derived a global database of mammalian insular biodiversity by using the Global Administrative areas version 3.6 (GADM, 2018) to subset spatial polygons of all land masses smaller than Greenland (2,166,000 km²) that are surrounded by salty water and overlapping it with the mammalian range maps from IUCN (IUCN, 2017). We carefully inspected and manually corrected any other alignment inconsistencies using QGIS 3.6 (Open Source Geospatial Foundation Project, 2019). We opted for a highly conservative approach of excluding any island with the slightest doubt about species attribution and ignoring islands where no mammal species occurs according to the IUCN data (i.e., our dataset only includes islands with at least one species). For example, regions with clusters of nearby islands – e.g., Patagonia and Scandinavia. We removed introduced species from the database by excluding (1) species polygons recorded as introduced by IUCN and (2) species listed as invasive for each particular island in the Database of Island Invasive Species Eradication (DIISE, 2015). We also removed fully aquatic and marine semi-aquatic species because they are not expected to respond to the characteristics of the islands in the same way as terrestrial species. We ensured that native species that were extinct due to human activity were included in the database by adding occurrence records from (Faurby & Svenning, 2016; Upham, 2017; Faurby et al., 2018). The presence and absence matrix of species per island is available in Appendix 1. From this matrix, we calculated richness of native species, number of single island endemics (SIE), and proportion of SIE (Appendix 2).

For each island in the database, we gathered environmental and physical characteristics expected to influence biodiversity: mean annual temperature (in degrees Celsius), annual precipitation (in millimeters), standard deviation of mean annual temperature and precipitation within the island, standard deviation in elevation within the island, area (in km²), surrounding landmass proportion (SLMP), island connectivity to the mainland during the last glacial maximum (GMMC), and climate change velocity in temperature since the last glacial maximum (CCVT, in meters/year). We derived temperature and precipitation data from CHELSA using monthly estimates across the years 1979 to 2013 (Karger et al., 2017) and elevation from the Global Digital Elevation Model GTOPO30 (USGS, 1996) and calculated mean and standard deviation per island using QGIS 3.6 (Open Source Geospatial Foundation Project, 2019). We obtained island area, SLMP, GMMC and CCVT from a public island characterization database (Weigelt et al., 2013) by matching the centroid coordinates to the island polygons. SLMP is a proxy of island isolation with great predictive power and was calculated as the log₁₀-transformed sum of the proportion of surrounding landmass within buffer distances of 100, 1,000 and 10,000 km around each island perimeter (Weigelt & Kreft, 2013). GMMC is a binary descriptor of historical isolation that uses past and present global bathymetry data to infer if islands were connected to the continent during the last glacial maximum by assuming the estimated sea level decrease of 122m at 18,000 years ago (more details in Weigelt et al., 2013). We multiplied SLMP by –1 and coded GMMC as 0 being connected and 1 being disconnected to the mainland during the LGM, so both metrics represent isolation (i.e., higher SLMP and GMMC represent greater isolation). Hereafter we will refer to those variables as “current isolation” and “past isolation”. CCVT over the past 21,000 years was calculated by dividing the difference in mean annual temperature between past and present by the spatial change in present mean temperature (Loarie et al., 2009; Weigelt et al., 2013). CCVT is interpreted as the speed at which the organism would have to move to keep pace with historical temperature change, assuming no change in topography (Loarie et al., 2009; Weigelt et al., 2013). Islands were classified into the 12 global mammalian zoogeographical regions (Holt et al., 2013), hereafter “realm”. We removed 505 islands from the dataset because it was not possible to derive all environmental variables or to assign a realm with confidence, usually because they were small (< 1km²) or located on a biogeographical boundary. The final dataset comprised 5,592 islands (out of the ~17,000 islands larger than 1km² worldwide; Weigelt et al., 2013) of which 123 contained single-island endemics.

• DIISE (2015) The Database of Island Invasive Species Eradications, developed by Island Conservation, Coastal Conservation Action Laboratory UCSC, IUCN SSC Invasive Species Specialist Group, University of Auckland and Landcare Research New Zealand.
• Faurby, S., Davis, M., Pedersen, R., Schowanek, S.D., Antonelli, A. & Svenning, J.C. (2018) PHYLACINE 1.2: The Phylogenetic Atlas of Mammal Macroecology. Ecology, 99, 2626.
• Faurby, S. & Svenning, J.-C. (2016) Resurrection of the Island Rule: Human-Driven Extinctions Have Obscured a Basic Evolutionary Pattern. The American Naturalist, 187, 812–820.
• GADM (2018) GADM database of Global Administrative Areas, version 3.6.
• Hanna, E. & Cardillo, M. (2014) Island mammal extinctions are determined by interactive effects of life history, island biogeography and mesopredator suppression. Global Ecology and Biogeography, 23, 395–404.
• Hébert, K., Millien, V., Lessard, J. & Masters, J. (2021) Source pool diversity and proximity shape the compositional uniqueness of insular mammal assemblages worldwide. Journal of Biogeography, jbi.14156.
• Holt, B.G., Lessard, J.-P., Borregaard, M.K., Fritz, S.A., Araújo, M.B., Dimitrov, D., Fabre, P.-H., Graham, C.H., Graves, G.R., Jønsson, K.A., Nogués-Bravo, D., Wang, Z., Whittaker, R.J., Fjeldså, J. & Rahbek, C. (2013) An update of Wallace’s zoogeographic regions of the World. Science, 339, 74–78.
• IUCN (2017) IUCN Red List of threatened species – mammal range polygons.
• Karger, D.N., Conrad, O., Böhner, J., Kawohl, T., Kreft, H., Soria-Auza, R.W., Zimmermann, N.E., Linder, H.P. & Kessler, M. (2017) Climatologies at high resolution for the earth’s land surface areas. Scientific Data, 4, 1–20.
• Lavery, T.H., Olds, A.D., Seddon, J.M. & Leung, L.K.-P. (2016) The mammals of northern Melanesia: Speciation, ecology, and biogeography. Mammal Review, 46, 60–76.
• Loarie, S.R., Duffy, P.B., Hamilton, H., Asner, G.P., Field, C.B. & Ackerly, D.D. (2009) The velocity of climate change. Nature, 462, 1052–1055.
• Meiri, S., Cooper, N. & Purvis, A. (2008) The island rule: made to be broken? Proceedings of the Royal Society B: Biological Sciences, 275, 141–148.
• Millien-Parra, V. & Jaeger, J.-J. (1999) Island biogeography of the Japanese terrestrial mammal assemblages: an example of a relict fauna. Journal of Biogeography, 26, 959–972.
• Open Source Geospatial Foundation Project (2019) QGIS 3.6.
• Upham, N.S. (2017) Past and present of insular Caribbean mammals: Understanding Holocene extinctions to inform modern biodiversity conservation. Journal of Mammalogy, 98, 913–917.
• USGS (1996) GTOPO 30 - Global Digital Elevation Model. Sioux Falls, SD: US Geological Survey.
• Weigelt, P., Jetz, W. & Kreft, H. (2013) Bioclimatic and physical characterization of the world’s islands. Proceedings of the National Academy of Sciences, 110, 15307–15312.
• Weigelt, P. & Kreft, H. (2013) Quantifying island isolation - insights from global patterns of insular plant species richness. Ecography, 36, 417–429.

We excluded from the dataset those islands where no mammal species occur so that the dataset only includes islands with at least one species. The name of the rows in the presence and absence matrix of species per island (Appendix 1) is paired to the ID column in the database containing island physical, environmental and biotic variables (Appendix 2).