Adventures in Global Vegetation Modelling

Trained as a physicist, I cut my post-PhD teeth in the 1990s trying to model the response of vegetation to global climate change at the then new, Hadley Centre for Climate Prediction and Research. The model I designed, called “TRIFFID”, was used in the first climate model projections to include vegetation and the carbon cycle as interactive elements. Those projections suggested that climate change could lead to dieback of the Amazon rainforest, and predicted an acceleration of climate change due to the release of carbon from soils and vegetation under global warming (Cox et al., 2000).

That result was alarming to us and many others, but we consoled ourselves with the fact that our model produced a much larger land carbon cycle feedback than the other models that were developed shortly afterwards (Friedlingstein et al., 2006). We could also see evidence that our model produced unrealistically large variations in the year-to-year increase in atmospheric CO2, which was consistent with our modelled tropical forests being too sensitive to climate (Cox et al. 2013). My colleagues and I gave a metaphorical sigh of relief.

But things just got scary again. Firstly, ground-based measurements of tree growth-rates suggested that the Amazon forest was absorbing carbon at a decreasing rate (Hubau et al., 2020). Then the latest generation of Earth System Models started to show more agreement that climate change could result in abrupt loss of forest or ‘localised dieback’ (Parry et al., 2022). Large parts of the real Amazon also suffered their worst recorded drought in 2023. Many would now agree the Amazon rainforest is very vulnerable to the combination of climate change and deforestation (Flores et al., 2024).

Nevertheless, accurately predicting how the Amazon rainforest will respond to climate change undoubtedly requires much more sophisticated vegetation models than we have used to date. First generation models like “TRIFFID” attempt to model the large-scale properties of the vegetation (e.g. vegetation carbon, leaf area index), without explicitly modelling the growth or death of individual plants. In fact, those models know nothing at all about the varying size of trees in forests, and therefore cannot make use of incredibly detailed forest inventory data.

My research team at Exeter set out to remedy that situation, by producing the ‘Robust Ecosystem Demography (RED)’ which models the number of tree as a function of tree mass (Argles et al., 2020). RED is therefore able to make sense of forest inventory, and can be used to interpret the size structure of trees in a forest in terms of the ratio of tree mortality to tree growth-rate (Moore et al., 2020). In fact this ratio is remarkably constant across different forest types and in different locations, a universality that we call ‘demographic optimality’ (Moore et al. 2023). We are still trying to understand how that optimality is arrived at, and what it might mean for the resilience of tropical forests.