TROLL 4.0: representing water and carbon fluxes, leaf phenology, and intraspecific trait variation in a mixed-species individual-based forest dynamics model – Part 2: Model evaluation for two Amazonian sites

Abstract. TROLL 4.0 is an individual-based forest dynamics model that jointly simulates the structure, diversity and functioning of tropical forests, including their water balance, carbon fluxes and leaf phenology, while accounting for intraspecific trait variation for a large number of species. In a companion paper, we describe how the model represents the physiological and demographic processes that control the tree life cycle in a one-metre-resolution spatially-explicit scene and uses plant functional traits measurable in the field to parameterize such processes across species and individuals (Maréchaux et al., submitted companion paper). Here we evaluate the performance of TROLL 4.0 for two Amazonian sites with contrasting soil and climate properties. We assessed the model's ability to represent forest structure and composition using lidar-derived canopy height distributions and forest inventories combined with information on plant functional traits. We also evaluated the model's ability to represent carbon and water fluxes, as well as leaf area variation, at daily and fortnightly resolution over a decade, using detailed information from on-site eddy covariance towers, satellite data and ground-based or air-borne lidar data. We finally compared the responses of carbon and water fluxes to environmental drivers between simulated and observed data. Overall, TROLL 4.0 provided a realistic representation of forests at both sites. The simulated canopy height distribution showed a high correlation coefficient (CC) with observed aerial and satellite data (CC>0.92), while the species and functional composition were well represented (CC>0.75). TROLL 4.0 also realistically simulated the seasonal variability of carbon and water fluxes (CC>0.46) and their responses to environmental drivers, while capturing temporal variations in leaf area (CC>0.76) and its partitioning in leaf age cohorts. However, TROLL 4.0 overestimated annual gross primary productivity at both sites (mean RMSEP=0.94 kgC m^-2 yr^-1) and evapotranspiration at one site (mean RMSEP=0.75 mm day^-1), likely due to an underestimation of the soil water depletion and stomatal control during the dry season. This evaluation highlights the potential of TROLL 4.0 to represent ecosystem fluxes and the structure and diversity of plant communities at a fine resolution, paving the way for model predictions of the effects of climate change, fragmentation and forest management on forest structure and dynamics.