Context-dependent hierarchy of plant community predictors from the Oregon Coast Range to the eastern slopes of the Cascade Mountains

Background/Question/Methods

Hierarchy theory from the field of landscape ecology implies that different variables may emerge as strong correlates with vegetation pattern at differing spatial scales.

We investigated the hierarchical nature of vegetation-environment relationships in forests across the western mountain ranges of Oregon, comparing the relative strength of climate, topography, local disturbance history, and imagery variables in predicting vegetation-types (Nature Serve’s Ecological Systems classification). Our dataset included 76,388 plots from regional vegetation inventories. Each plot was also associated with 140 explanatory variables that include modeled climate predictors (Daymet), elevation, and associated topographical descriptors, LANDSAT imagery and associated derivatives and summaries, disturbance history (insect/disease, harvest and fire), and soil parent material. We built Random Forest models (an extension of classification trees, Breiman 2001) with subsets of the plots, selected to fall within nested variable radius zones (60,000ha, 120,000ha, and by ecoregion). The nested zones were stratified among three ecoregions that span a broad range of climatic variability for Oregon’s forests (the Coast Range, the Western Cascades, and the Eastern Cascades). Within each model, we ranked explanatory variable strength using the GINI index of variable importance.

Results/Conclusions

At the broadest observational extent (ecoregion), climate variables and elevation were often stronger predictors than imagery, local topography (e.g., slope, topographic position index), disturbance, and soil parent material. However, the most important climate variables differed among the ecoregions. In the East Cascades, the strongest climate variable was temperature variability (August maximum, minus December minimum). In the West Cascades, elevation was the strongest explanatory variable, and the strongest climate variable was annual solar radiation. In the Coast Range, stratus clouds, and fog, which reduce moisture stress during the growing season, were the strongest predictive variables.

In the West Cascades, the relative predictive strength of the variables varied with the scale of observation. At the ecoregion-scale of observation, elevation was the strongest predictor variable, while at the 60,000 ha scale, it was ranked much lower and local topographic position indices were strong. At all spatial extents, climate variables were strong, but within the smaller analysis extent, annual precipitation was the most important of the climate variables, rather than annual solar radiation.

Our work highlights potential effects of observational scale on modeling vegetation-environment relationships, and also describes the climate gradient that shapes Oregon’s forest vegetation.