How Does Plant Diversity Affect the Resilience of Forest Function to Disturbance in the Amazon Basin?

Abstract

Dynamic global vegetation models (DGVMs) struggle to reproduce the spatial gradients of above-ground biomass (AGB) in Amazonia and the observed decline in the Amazonian carbon sink in recent years, making predictions for its future uncertain. This thesis focuses on two interconnected challenges in modelling Amazonian carbon dynamics: the lack of plant diversity in models using plant functional types (PFTs), and the unrealistic representation of mortality for tropical PFTs.

First, coexistence between PFTs is enabled in RED (Robust Ecosystem Demography), a cohort DGVM with Lotka competition dynamics. Inter-PFT competition, represented by a universal, variable competition coefficient (c), is assumed to be weaker than intra-PFT competition, contrary to assumptions in current models. This allows a tunable number of PFTs to coexist in a parameter sparse, and analytically solvable approach. Also presented is the idea of an optimum c, which maximises the Shannon diversity index per unit resource, replacing the need to quantify the strength of PFT interactions at the global scale. The optimum c value depends on the number of PFTs input and saturates at 0.8. The responses of vegetation carbon and growth rate to increased PFT coexistence are modelled at a point level to demonstrate the potential of the new competition scheme. It is shown that, in RED, PFT diversity increases forest function and impacts resilience by influencing the properties of the most growth-efficient PFT sampled. These results are important for modelling regions like Amazonia, where biodiversity is chronically underestimated in top-down DGVMs.

Spatially heterogeneous plant mortality rates are an important predictor of AGB patterns in Amazonia. Therefore, in the second chapter, spatially variable mortality rates are enabled for the tropical tree PFT in TRIFFID, the DGVM currently coupled to the JULES land surface scheme, replacing a homogeneous mortality rate. Maps of Amazonian tree mortality are created using a well-known link with wood density and three independent wood density maps. Model outputs using the diversified mortality scheme, evaluated against observations of AGB, improve its representation in TRIFFID, mainly through a reduction in the model bias and an increase in the correlation with observed biomass.