Deep learning of fossil pollen morphology reveals 25,000 y of ecological change in eastern African grasslands.

Deep learning of fossil pollen morphology reveals a 25,000-year record of ecological change in eastern African grasslands, including shifts in species diversity and photosynthetic types.

• Why it matters: Understanding grassland evolution is limited by traditional microscopy's inability to differentiate pollen species, hindering insights into past ecological dynamics and climate impacts.
• What they did: A semisupervised deep-learning approach using convolutional neural networks was applied to superresolution images of modern and fossil grass pollen, estimating taxonomic diversity and photosynthetic type over time.
• The result: The method uncovered a significant decline in grass species diversity during the last glacial period and a gradual decrease in C4 grasses after the glacial-Holocene transition, linking ecological shifts to climate and fire activity.