Combining statistical and mechanistic models to identify the drivers of 1 mortality within a rear-edge beech population 2 3

32 Since several studies have been reporting an increase in the decline of forests, a major issue 33 in ecology is to better understand and predict tree mortality. The interactions between the different 34 factors and the physiological processes giving rise tree mortality, as well as the individual variability 35 in mortality risk, still need to be better assessed. 36 This study investigates mortality in a rear-edge population of European beech (Fagus sylvatica 37 L.) using a combination of statistical and process-based modelling approaches. Based on a survey of 38 4323 adult beeches since 2002 within a natural reserve, we first used statistical models to quantify 39 the effects of competition, tree growth, size, defoliation and fungi presence on mortality. Secondly, 40 we used an ecophysiological process-based model (PBM) to separate out the different mechanisms 41 giving rise to temporal and inter-individual variations in mortality by simulating depletion of carbon 42 stocks, loss of hydraulic conductance and damage due to late frosts in response to climate. 43 The combination of all these simulated processes was associated with the temporal variations 44 in the population mortality rate. The individual probability of mortality decreased with increasing 45 mean growth, and increased with increasing crown defoliation, earliness of budburst, fungi 46 presence and increasing competition, in the statistical model. Moreover, the interaction between 47 tree size and defoliation was significant, indicating a stronger increase in mortality associated to 48 defoliation in smaller than larger trees. Finally, the PBM predicted a higher conductance loss 49 together with a higher level of carbon reserves for trees with earlier budburst, while the ability to 50 defoliate the crown was found to limit the impact of hydraulic stress at the expense of the 51 accumulation of carbon reserves. 52 We discuss the convergences and divergences obtained between statistical and process-based 53 approaches and we highlight the importance of combining them to characterize the different 54 processes underlying mortality, and the factors modulating individual vulnerability to mortality. 55