This thesis aims to contribute to increased understanding and predictability of vegetation responses to changes in the environment. This is achieved via the development and interrogation of wood formation process formulations suitable for resolving hitherto missing growth processes in Dynamic Global Vegetation Models (DGVMs). Current understanding of environmental controls on plant carbon dynamics is encapsulated in DGVMs. These models are today largely ‘source-driven’, meaning that they assume growth to be the direct outcome of photosynthesis. Contrary to plant physiological evidence, they neglect that growth itself may be under stronger environmental controls than photosynthesis. If not considered in models, this may have large implications for projections of vegetation carbon dynamics and, ultimately, climate change. Testing the impact of including a growth sink in DGVMs has to date been hampered by the fact that we have limited understanding of sink processes and therefore lack an explicit sink representation for DGVMs. This thesis is a solution to this issue by providing a mechanistic approach, RINGSlite.1.0, to represent vegetation sink processes for trees in DGVMs, and so enables the source-driven paradigm to be challenged.

To provide this mechanistic approach I first investigate the validity of the sink hypothesis using a data-driven method wherein I derive a soil-moisture growth (sink) response curve on the basis of tree ring observations and soil moisture simulations. Comparison of this new sink response curve against source response curves from observations and those currently in use in DGVMs confirms the sink limitation hypothesis under water stress for trees. This highlights the need for separate representation of the source and sink processes in DGVMs. I then review past and contemporary wood formation models and identify five critical features for a wood formation model required for global application. I use these features to modify an existing model and create RINGSlite.1.0, which I compare with other wood formation models in a Bayesian model-intercomparison framework. More testing is still required against additional observations at environmentally diverse sites to verify the usefulness of RINGSlite.1.0 globally. Nevertheless, the comparison reveals that RINGSlite.1.0 is the most validated by data at the non-water limited test site.

The implications for resolving sink processes in DGVMs through wood formation are numerous. Firstly, it provides a more realistic sink response to the environment. Secondly, it allows for additional data sources to be used in model benchmarking, calibration and initialisation. Finally, through simulating wood formation responses to local environments, additional features such as ring width and density can serve as emergent functional traits and provide a mechanistic link to modelling ecological and hydrological processes in DGVMs.