Earth System Modelling and the evolutionary history of South America

Hiromitsu Sato1,2, Douglas I Kelley3

1 University of Toronto, 2 Ontario Forest Research Institute, 3 UK Centre for Ecology and Hydrology

While ESMs often look to the future, Earth system modelling techniques can also provide insights into past vegetation distributions and tackle some of the biggest paleoecology and human development questions. The South American refugia hypothesis, for example, describes how the growing and shrinking of forest and savannas over glacial cycles (moving in and out of ice ages) explain modern day biodiversity. This hypothesis has been hugely controversial since it was first put forward over 50 years ago, and a lack of proxy observations means it will likely remain unanswered – unless modelling techniques can help inform what past vegetation looked like.

Hiro Sato, from the Ontario Forest Research Institute, explains the refugia hypothesis’ background and describes how we combine dynamic vegetation modelling with the fossilised pollen record to reconstruct forest and savanna locations during the Last Glacial Maximum. He found that the combined fire and CO2 deprivation stress drove savannafication during the last ice. A corridor linked central and Southern American savannas through a destabilised Amazon forest. Hiro finishes by describing future work – including brand new machine learning algorithms for bias correcting the simulated land surface against observations that can directly tackle the existence of Refugia, as well as the mysteries around human and cultural emergence in Africa.

Refugia hypothesis

Jurgen Haffer, a geologist by trade and ornithologist by passion, developed the Refugia Hypothesis (1969) to explain the rich diversity of birds in the Amazonian rainforest. Based on species distributions and precipitation patterns, Haffer hypothesised that the vast and continuous rainforest fractured and splintered into ‘islands’ of forest known as refugia, separated by savanna during glacial periods. Populations of rainforest species would then be isolated, gene flow would be blocked, and diversification would occur through allopatric speciation (speciation that happens when two populations of the same species become isolated from each other due to geographic changes). During interglacials, forests would expand, and refugia would re-connect, leading to a range expansion of their newly diversified species, establishing a kind of Pleistocene[1] species pump. In the nineties and early noughties, key palynological studies (pollens and spores) suggested an Amazonia relatively stable against glacial environmental changes. This contradicted central ideas of the Refugia hypothesis and stimulated ongoing debate on the subject.

A lack of paleo data

Biologists continue to seek understanding of the dynamics of the past environments of their species of interest to understand their evolutionary history better. Phylogenetic methods[2] can give insight into roughly when and where diversification occurred. However, they must then be reconciled with knowledge of their past environments to elucidate the specific processes that led to differentiation. Reconstructing environments that long predate human measurement is a complex task and typically approached within two distinct realms of research: palaeoecology and Earth System modelling. Proxy methods, while empirically based, are often sparse and irregularly distributed due to specific preservation requirements. Fossilised pollen, for example, which tells us what used to grow in an ecosystem, only preserved in lake sediment and peat bogs with the correct chemical composition. Dynamic Vegetation Models (DVMs) can generate continuous reconstructions of past terrestrial environments but are rarely adapted for biogeographical analyses or made accessible to non-specialists.

[1] 2,580,000 to 11,700 years ago, spanning the Earth’s most recent period of repeated glaciations.

[2] Used to establish evolutionary relationships among biological entities

Modelled vegetation reconstructions for the LGM

Figure 1: Modelled vegetation reconstructions for the LGM. Colours represent different biomes. Simulations include fire impacts and low CO2 – the configuration that best matches pollen-based reconstructions of past vegetation is located in circles on the map. Arrows show corridors of open vegetation formations that may have facilitated dispersal during glacial periods, while refugia are stable forested regions that remained robust against past environmental change. Thf = tropical humid forest, Tdf = tropical dry forest, Ts = tropical savanna, sw = sclerophyll woodland, tp = temperate parkland, bp = boreal parkland, dg = dry grass/shrubland, hd = hot desert, st = shrub tundra, t =tundra

Fusing model and proxy observations

The initial aim of our research was to produce reconstructions of vegetation cover of the Neotropics during the Last Glacial Maximum (LGM) that would utilise both advanced modelling techniques and the available body of appropriate pollen records. We generated twenty model reconstructions of vegetation through five climate reconstructions of the LGM for four scenarios designed as a factorial experiment to test the effects of low CO2 and wildfire on vegetation cover. Comparing these model reconstructions against forty-two pollen cores gave local estimates of vegetation during the LGM. This task involved developing a method to quantitatively compare model output with pollen records through the ecological closeness of biome types.

Fire and CO2 deprivation drove savannafication

The reconstruction that best fits with empirical data showed major savannafication and forest dieback, indicating possible instability of Amazonian forest to the more arid glacial climate, lower CO2, and wildfire (Fig. 1). Though our reconstructions did not show clear signs of Haffer’s refugia, they showed complex mosaics of open and closed biomes that also ran contrary to the notion of a stable Amazonian forest. This work also provided evidence for the past existence of two savanna formations known as ‘dry corridors’ that are hypothesised to have existed based on distant and disjoint savanna species.

The results of this work have been published in Nature Geoscience (Sato et. al, 2021) and have stimulated direct collaboration in phylogenetics research in Amazonia (Buainain et al, 2020). We are now working to quantitatively integrate species distribution data to formalise notions of niche and connectivity, which has thus far been absent in Earth System modelling. While our best reconstruction showed a good fit against empirical data, there is still a mismatch that we can correct for. Transforming the model’s bioclimatic output into species style niches will also allow an assessment of connectivity of possible forest islands – a direct test of Haffer’s refugia. We are also excited to apply our newly developed methods toward another context where savanna-forest dynamics were crucial over long timescales: early human movement and evolution. We have already published the first study (Ecker et al., 2020) with a major project planned for the coming year, which will utilise data from early human settlements.

What I enjoy most about this work is taking part in new interactions between researchers from very different fields of study. The big questions that may be a driving force in one area of research, such as the origins of biodiversity, may benefit significantly from applying tools from another field, such as climate science. I believe that it is beneficial and necessary to make these connections, given the vastness of these questions. This is why I was most interested to learn of Haffer’s eclectic research background, which combined understanding of the large scales of geology with tropical ecology and his methodological specialisation with his subject of interest.


Buainain, N., Canton, R., Zuquim, G., Tuomisto, H., Hrbek, T., Sato, H., & Ribas, C. C. (2020). Paleoclimatic evolution as the main driver of current genomic diversity in the widespread and polymorphic Neotropical songbird Arremon taciturnus. Molecular Ecology, 29(15), 2922-2939.

Ecker, M., Kelley, D., & Sato, H. (2020). Modelling the effects of CO 2 on C 3 and C 4 grass competition during the mid-Pleistocene transition in South Africa. Scientific reports, 10(1), 1-8.

Sato, H., Kelley, D. I., Mayor, S. J., Martin Calvo, M., Cowling, S. A., & Prentice, I. C. (2021). Dry corridors opened by fire and low CO2 in Amazonian rainforest during the Last Glacial Maximum. Nature Geoscience, 14(8), 578-585.

Download this Newsletter as a PDF (2015KB)