Numerous current efforts seek to improve the representation of ecosystem ecology and vegetation demographic processes within Earth System Models (ESMs). These developments are widely viewed as an important step in developing greater realism in predictions of future ecosystem states and fluxes. Increased realism, however, leads to increased model complexity, with new features raising a suite of ecological questions that require empirical constraints. Here, we review the developments that

Summary Tree mortality rates appear to be increasing in moist tropical forests (MTFs) with significant carbon cycle consequences. Here, we review the state of knowledge regarding MTF tree mortality, create a conceptual framework with testable hypotheses regarding the drivers, mechanisms and interactions that may underlie increasing MTF mortality rates, and identify the next steps for improved understanding and reduced prediction. Increasing mortality rates are associated with rising temperature and vapor pressure deficit, liana abundance, drought, wind events, fire and, possibly, CO2 fertilization‐induced increases in stand thinning or acceleration of trees reaching larger, more vulnerable heights. The majority of these mortality drivers may kill trees in part through carbon starvation and hydraulic failure. The relative importance of each driver is unknown. High species diversity may buffer MTFs against large‐scale mortality events, but recent and expected trends in mortality drivers give reason for concern regarding increasing mortality within MTFs. Models of tropical tree mortality are advancing the representation of hydraulics, carbon and demography, but require more empirical knowledge regarding the most common drivers and their subsequent mechanisms. We outline critical datasets and model developments required to test hypotheses regarding the underlying causes of increasing MTF mortality rates, and improve prediction of future mortality under climate change.

We describe an implementation of the Ecosystem Demography (ED) concept in the Community Land Model. The structure of CLM(ED) and the physiological and structural modifications applied to the CLM are presented. A major motivation of this development is to allow the prediction of biome boundaries directly from plant physiological traits via their competitive interactions. Here we investigate the performance of the model for an example biome boundary in eastern North America. We explore the sensitivity of the predicted biome boundaries and ecosystem properties to the variation of leaf properties using the parameter space defined by the GLOPNET global leaf trait database. Furthermore, we investigate the impact of four sequential alterations to the structural assumptions in the model governing the relative carbon economy of deciduous and evergreen plants. The default assumption is that the costs and benefits of deciduous vs. evergreen leaf strategies, in terms of carbon assimilation and expenditure, can reproduce the geographical structure of biome boundaries and ecosystem functioning. We find some support for this assumption, but only under particular combinations of model traits and structural assumptions. Many questions remain regarding the preferred methods for deployment of plant trait information in land surface models. In some cases, plant traits might best be closely linked to each other, but we also find support for direct linkages to environmental conditions. We advocate intensified study of the costs and benefits of plant life history strategies in different environments and the increased use of parametric and structural ensembles in the development and analysis of complex vegetation models

The NERC and CEH trademarks and logos ('the Trademarks') are registered trademarks of NERC in the UK and other countries, and may not be used without the prior written consent of the Trademark owner.

Abstract. Plant functional traits determine vegetation responses to environmental variation, but variation in trait values is large, even within a single site. Likewise, uncertainty in how these traits map to Earth system feedbacks is large. We use a vegetation demographic model (VDM), the Functionally Assembled Terrestrial Ecosystem Simulator (FATES), to explore parameter sensitivity of model predictions, and comparison to observations, at a tropical forest site: Barro Colorado Island in Panama. We define a single 12-dimensional distribution of plant trait variation, derived primarily from observations in Panama, and define plant functional types (PFTs) as random draws from this distribution. We compare several model ensembles, where individual ensemble members vary only in the plant traits that define PFTs, and separate ensembles differ from each other based on either model structural assumptions or non-trait, ecosystem-level parameters, which include (a) the number of competing PFTs present in any simulation and (b) parameters that govern disturbance and height-based light competition. While single-PFT simulations are roughly consistent with observations of productivity at Barro Colorado Island, increasing the number of competing PFTs strongly shifts model predictions towards higher productivity and biomass forests. Different ecosystem variables show greater sensitivity than others to the number of competing PFTs, with the predictions that are most dominated by large trees, such as biomass, being the most sensitive. Changing disturbance and height-sorting parameters, i.e., the rules of competitive trait filtering, shifts regimes of dominance or coexistence between early- and late-successional PFTs in the model. Increases to the extent or severity of disturbance, or to the degree of determinism in height-based light competition, all act to shift the community towards early-successional PFTs. In turn, these shifts in competitive outcomes alter predictions of ecosystem states and fluxes, with more early-successional-dominated forests having lower biomass. It is thus crucial to differentiate between plant traits, which are under competitive pressure in VDMs, from those model parameters that are not and to better understand the relationships between these two types of model parameters to quantify sources of uncertainty in VDMs.