In the field of biodiversity modelling, mechanistic or process-based models have been recently advocated as a response to the conceptual and practical limitations of correlative approaches (such as species distribution models). In particular, the inclusion of biotic interactions into those approaches among species is still a challenge that has yet to be solved to provide reliable predictions at small spatial scale. Ecological theory has produced multiple mechanistic models that can act as a basis for that purpose; they have been extensively used to demonstrate theoretical properties of communities and provide insights into the causes of biodiversity structure. However because they often include a the large number of parameters to estimate, their direct fit to empirical data has often been limited to simplified experimental settings, thus strongly limiting their use for more ambitious predictive biodiversity modelling projects. Here we propose a way forward by presenting a modelling framework that use functional trait data to estimate the parameters of a community model. Our approach use a transfer function between the empirical functional traits and the phenomenological species traits from the model. The transfer function is then calibrated to maximize the fit to species abundance data using distribution sampling algorithms (MCMC).
Results/Conclusions
We demonstrate the use of such model on a dataset of 22 plant communities spread along an elevation gradient in the French Alps. Traits for 148 species (Height, SLA, LDMC, LNC, LCC, δ13C and δ15N) were sampled on site to construct the species trait space. Species abundances were estimated using pin contacts. We used a simplified Lotka-Volterra model of competition to predict species abundance. The model includes temperature-dependent growth and sensitivity to neighboring competition. We show that the calibrated transfer function suggests links between phenomenological model parameters and functional traits that conform expectations drawn from the functional trait literature: for instance, the model calibrated a positive relationship between species growth rate in the model with Height and reproduced the turnover of species along the elevation gradient. We then discuss the limits of the approach and illustrate the applicability perspectives of this method. Our approach outlines a flexible methodology to more easily fit and compare mechanistic models on species-rich systems by drastically decreasing the number of parameters to estimate while making good use of increasingly available trait data.