Global models and predictions of plant diversity based on advanced machine learning techniques

Despite the paramount role of plant diversity for ecosystem functioning, biogeochemical cycles, and human welfare, knowledge of its global distribution is still incomplete, hampering basic research and biodiversity conservation. Here, we used machine learning (random forests, extreme gradient boosting, and neural networks) and conventional statistical methods (generalized linear models and generalized additive models) to test environment-related hypotheses of broad-scale vascular plant diversity gradients and to model and predict species richness and phylogenetic richness worldwide. To this end, we used 830 regional plant inventories including c. 300 000 species and predictors of past and present environmental conditions. Machine learning showed a superior performance, explaining up to 80.9% of species richness and 83.3% of phylogenetic richness, illustrating the great potential of such techniques for disentangling complex and interacting associations between the environment and plant diversity. Current climate and environmental heterogeneity emerged as the primary drivers, while past environmental conditions left only small but detectable imprints on plant diversity. Finally, we combined predictions from multiple modeling techniques (ensemble predictions) to reveal global patterns and centers of plant diversity at multiple resolutions down to 7774 km² . Our predictive maps provide accurate estimates of global plant diversity available at grain sizes relevant for conservation and macroecology.