Abstract
Process-based models and empirical modelling techniques are frequently used to (i) explore the sensitivity of tree growth to environmental variables, and (ii) predict the future growth of trees and forest stands under climate change scenarios. However, modelling approaches substantially influence predictions of the sensitivity of trees to environmental factors. Here, we used tree-ring width (TRW) data from 1630 beech trees from a network of 70 plots established across European mountains to build empirical predictive growth models using various modelling approaches. In addition, we used 3-PG and Biome-BGCMuSo process-based models to compare growth predictions with derived empirical models. Results revealed similar prediction errors (RMSE) across models ranging between 3.71 and 7.54 cm2 of basal area increment (BAI). The models explained most of the variability in BAI ranging from 54 % to 87 %. Selected explanatory variables (despite being statistically highly significant) and the pattern of the growth sensitivity differed between models substantially. We identified only five factors with the same effect and the same sensitivity pattern in all empirical models: tree DBH, competition index, elevation, Gini index of DBH, and soil silt content. However, the sensitivity to most of the climate variables was low and inconsistent among the empirical models. Both empirical and process-based models suggest that beech in European mountains will, on average, likely experience better growth conditions under both 4.5 and 8.5 RCP scenarios. The process-based models indicated that beech may grow better across European mountains by 1.05 to 1.4 times in warmer conditions. The empirical models identified several drivers of tree growth that are not included in the current process-based models (e.g., different nutrients) but may have a substantial effect on final results, particularly if they are limiting factors. Hence, future development of process-based models may build upon our findings to increase their ability to correctly capture ecosystem dynamics.