Figure 1. Range of ecosystem properties available for modelling
forest ecosystem patterns in our study. Data encompass constituent and
aggregate ecosystem properties occurring across biotic (labels 1 and 2
– green font) and abiotic (labels 3, 4, and 5 – brown font) domains.
Properties marked with an asterisk were calculated from field data
records. Tree and shrub graphics (Natural Resources Canada 2015).
physical and chemical abiotic variables (e.g., exposed ground, rain
throughfall, and soil temperature, bulk density, and nitrogen). This
inclusive approach helps underscore reciprocal relationships between
biotic and abiotic properties. It also strengthens recognition of the
joint contributions biotic and abiotic properties make to ecosystem
patterns. We contend these two premises are key to improving predictions
of ecosystem spatial patterns and we have incorporated them into our
integrative modelling approach. What is more, our approach is applied at
a regional extent which is the extent where many conservation decisions
are made (Nicholson et al. 2019).
Our two primary study objectives are to (1) model independent and
shared, biotic and abiotic, field-collected response variables to
geographic and remotely-sensed environmental predictors; and (2) employ
continuous spatial turnover in modelled responses to resolve ecosystem
patterns at landscape extents. Model outcomes reveal emergent
relationships among ecosystem properties, across biotic, abiotic, and
ecosystem levels of ecological organization. The approach also provides
a basis for partitioning the relative contributions of biotic, abiotic,
and geographic predictors to ecosystem spatial patterns.
To quantify variation in ecosystem patterns across space, we predict
biotic-abiotic dissimilarities between pairs of ecosystem survey sites
using generalized dissimilarity modelling. Generalized dissimilarity
modelling (GDM) is an extension of matrix regression developed by
Ferrier et al. (2002) for spatial biodiversity modelling. Ferrier and
Guisan (2006) highlight GDM as an analytical tool to ‘assemble and
predict together’ , one of three general strategies they proposed for
spatial prediction of community patterns. With this strategy, community
constituents (i.e., species, traits) are modelling simultaneously to
predict biotic turnover in space or time (Ferrier et al. 2007). Here, we
extend this strategy to model ecosystem patterns, arising from
simultaneous prediction and mapping of both biotic and abiotic
properties. We show how predicted shifts in dissimilarity among
combinations of these properties can be used to map spatial ecosystem
patterns across landscapes.