Effects of extreme global warming on the distribution of terrestrial plants

In the last century, humans have driven carbon dioxide concentrations beyond levels experienced in millions of years. While climate models are essential tools for predicting future climate changes, they remain uncertain in their projections of temperature and rainfall on continents – the aspects that most directly impact human society.

For decades, paleoclimate scientists have tested Earth System Models (ESMs) by running them under past greenhouse conditions. However, comparisons between simulated paleoclimate and fossil data consistently show a mismatch: models simulate drier, more seasonal continental interiors than plant fossils suggest. Sparse fossil data have limited the strength of these tests.

The project aims to significantly increase available continental fossil data using fossil pollen, the most widely preserved and climatically sensitive proxy for continental paleoclimate. The research team will image extensive fossil pollen collections housed at the Smithsonian National Museum of Natural History and the University of Melbourne's Paleobiology Laboratory.

By combining advanced slide-scanning microscopes with cutting-edge AI technology, researchers will, for the first time, be able map plant species that existed during greenhouse climate intervals tens of millions of years ago – periods that reflect similar conditions to today's rising carbon dioxide concentrations. These detailed maps of past plant distributions will allow scientists to reconstruct climate conditions across different regions and periods.

The team will then test whether current state-of-the-art climate models can accurately simulate the historical temperature and rainfall patterns inferred from the fossil data. The pollen-based paleoclimate reconstructions will be tested against simulations made by partners at the National Center for Atmospheric Research using the Community ESM. This will provide the most thorough test yet of a climate model's ability to simulate regional climate patterns during a greenhouse climate period.

MDAP's expertise is essential to this collaboration. They will develop artificial intelligence routines to locate and identify pollen grains on scanned slides, potentially implementing supervised or unsupervised learning algorithms for classification. The research team will validate the AI assignments and align provisional categories with existing taxonomic nomenclature. The AI model will undergo further training to accurately identify prevalent pollen types, allowing human experts to focus on uncommon grains.

This iterative process of model testing and refinement will significantly accelerate generation of comprehensive pollen data – enabling more accurate climate model testing and improved future climate projections. The more precisely we can predict upcoming climate changes, the better societies can plan and adapt to them.