The predominance of stand composition and structure over direct climatic and site effects in explaining aspen (Populus tremuloides Michaux) site index within boreal and temperate forests of western Quebec, Canada

Existing models that use the site-index concept (dominant canopy height of a tree species at a reference stand age) are fundamentally stand-level models that do not account for stand dynamics, limiting their use to only a part of successional trajectories. Given that stand dynamics is influenced by both large and fine scale processes, we took a multi-level look at aspen (Populus tremuloides Michaux) productivity by determining landscape- and plot-level factors related to productivity as rated with site index. The study area extends from latitude 45° to 50°N in western Québec, from which were sampled 62 landscapes made up of 4948 plots, 25% of which had aspen as dominant and/co-dominants in the canopy. There, aspen is most often found in mixed stands. A stepwise procedure with forward selection was used in building landscape- and plot-level models and models were then arranged hierarchically such that (a) predicted estimates of the landscape model were inputs to the plot-level model (top-down) and (b) significant landscape variables were added to selected plot level variables (bottom-up). For the plot-level model, none of the climate variables considered were selected but at the landscape level, annual sum of degree–days was only the third to enter. In both cases, aspen site index was more related to the proportion of spruce (Picea mariana (Mill.) B.S.P. and Picea glauca (Moench.) Voss). At the level of landscapes, this observation might be due to the existence of particular vegetation mosaics, of which spruce proportion could be a surrogate. At the level of plots, influence of spruce on aspen site index is probably indicative of niche sharing with aspen. A high random variability was associated with the plot-level model but not with the landscape-level model. The similarity in drivers of aspen site index at both levels and the fact that both top-down and bottom-up approaches provided the same information, suggest that the use of landscape variables when modelling site index in mixed stands may help distinguish peculiarities shared by plots located in a landscape and improve the signal between site index and explanatory variables by reducing the random noise observed at the level of plots.