Dominic Chambers

McGill University
M.Sc. candidate

Supervisor: Sylvie de Blois
Start: 2008-09-01
End: 2011-06-22

Project

Modeling tree abundance in eastern North America in response to climate change.
Trees are expected to modify their distribution and abundance in response to climate change with important consequences on forestry management practices and forest diversity. Whereas Species Distribution Models have been commonly used to relate known occurrence of species to the current climate as a first step to project future suitable environmental space, modeling abundance patterns using Species Abundance Models (SAMs) remains a challenge. This research aimed to: 1) evaluate the predictive performance of SAMs in predicting the current abundance of 105 tree species in eastern North America in response to climatic, topographic and edaphic predictors, and 2) explain the variation in SAMs’ performance among species. The relative importance values of 105 tree species were first related to environmental predictors using Random Forest. The predictive performance of SAMs for each tree species was then assessed using the coefficient of determination (R²). Finally, multiple linear regression was performed to explain the variation in SAMs’ performance among species (R²) using biogeographical and spatial attributes of species as explanatory variables. Predicting the current relative abundance of tree species using a combination of climatic, topographic, and edaphic variables was only partially successful. The coefficients of determination (R²) for all SAMs ranged from 0.000 to 0.857 with a mean of 0.258 and a standard deviation of 0.18. Black spruce (Picea mariana) had the best predictive model and Florida maple (Acer barbatum) the worst. Forty-one species out of 105 (39 %) had R² ≥ 0.3. These species had climate as the best and/or second best environmental predictor, except for Quercus macrocarpa, Pinus rigida, Pinus resinosa, and Ulmus alata, which were best predicted by non-climatic variables. The variation in the performance of SAMs among species was best explained by the range of relative abundance values and the spatial aggregation of species. This study highlighted the challenge in accurately predicting the relative abundance of trees in relation to current and therefore future climate, and identified species for which modeling approach worked best and for which abundance patterns would likely respond to climate change.

Keywords

climate change, forest management, tree abundance, predictive performance, Species Distribution Model, Random Forest, Quebec