RT Journal Article
SR Electronic
T1 The evolution of open habitats in North America revealed by deep learning models
JF bioRxiv
FD Cold Spring Harbor Laboratory
SP 2021.09.03.458822
DO 10.1101/2021.09.03.458822
A1 Tobias Andermann
A1 Caroline Strömberg
A1 Alexandre Antonelli
A1 Daniele Silvestro
YR 2021
UL http://biorxiv.org/content/early/2021/09/05/2021.09.03.458822.abstract
AB Open vegetation today constitutes one of the most extensive biomes on earth, including temperate grasslands and tropical savannas. Yet these biomes originated relatively recently in earth history, likely replacing forested habitats as recently as the second half of the Cenozoic, although the timing of their origination and the dynamics of their expansion remain uncertain. Here, we present a new hypothesis of paleovegetation change in North America, showing that open habitats originated between 25 and 20 Ma in the center of the continent, and expanded rapidly starting 8 Ma to eventually become the most prominent vegetation type today. To obtain space-time predictions of paleovegetation, we developed a new Bayesian deep learning model that utilizes available information from fossil evidence, geologic models, and paleoclimate proxies. We compiled a large dataset of paleovegetation reconstructions from the peer-reviewed literature, which we used in combination with current vegetation data to train the model. The model learns to predict vegetation based on the learned associations between the vegetation at a given site and multiple biotic and abiotic predictors: fossil mammal occurrences, plant macrofossils, estimates of temperature and precipitation, latitude, and the effects of spatial and temporal autocorrelation. Our results provide a new, spatially detailed reconstruction of habitat evolution in North America and our deep learning model paves the way for a new quantitative approach to estimating paleovegetation changes.Competing Interest StatementThe authors have declared no competing interest.