Fig. 3. Machine learningpredictshowincreased samplingcouldchangeour understandingof antrichness andrarity centers. Random Forestmodelswere trained topredict ant speciesrichness andrarity values asafunction of climate (7 vars.),topography,biogeographic realm, vertebratebiodiversity, andsampling density. Wethen used the models to predict (A) richness and rarity values under a "universal high sampling" scenario, revealing which areas may drop out of the top 10% with increased global sampling (red), which are robust to sampling (purple), and which centers are predicted to enter the top 10% with increased sampling (blue). The latter represents a treasure map indicating areas that should be prioritized for future sampling. The top 10% areas for vertebrates are indicated by hatched regions. (B) Overlap fractions for empirical and projected center designations for richness and rarity, and Spearman's correlations continuous richness and rarity values.

Published as part of Kass, Jamie M., Guénard, Benoit, Dudley, Kenneth L., Jenkins, Clinton N., Azuma, Fumika, Fisher, Brian L., Parr, Catherine L., Gibb, Heloise, Longino, John T., Ward, Philip S., Chao, Anne, Lubertazzi, David, Weiser, Michael, Jetz, Walter, Guralnick, Robert, Blatrix, Rumsaïs, Lauriers, James Des, Donoso, David A., Georgiadis, Christos, Gomez, Kiko, Hawkes, Peter G., Johnson, Robert A., Lattke, John E., MacGown, Joe A., Mackay, William, Robson, Simon, Sanders, Nathan J., Dunn, Robert R. & Economo, Evan P., 2022, The global distribution of known and undiscovered ant biodiversity, pp. 1-17 in Science Advances 8 (31) on page 4, DOI: 10.1126/sciadv.abp9908, http://zenodo.org/record/7007296