Anthropogenic Climate Change will create a new rainfall regime, a shift in the
mean annual temperature gradient and an increase in extreme climatic events.
Modelling the effects of these changes on biodiversity requires that we integrate data
from many sources at many spatial and temporal scales. This is often difficult as error
becomes incorporated in the model rather than explicitly considered and quantified.
We examine the techniques of Structural Equation Modelling to address these
problems and explicitly quantify the error and influence attributable to different data
layers. The technique calculates the influence of several variables on the variance of
a resultant variable. By employing Structural Equation Modelling it is possible to
conduct natural experiments that exhibit the same causal explanatory power as
controlled experiments whilst incorporating an understanding that not all processes in
ecology can be directly measured. We give an example of a Structural Regression
Modelling approach to determine the possible changes in aquatic biodiversity in the
Sabie River, South Africa.
The Sabie River is one of the main rivers contributing to the biodiversity of
the Kruger National Park. It has a natural longitudinal temperature gradient that
enables space-for-time experiments that are useful in elucidating how species
assemblages might respond to changes in water temperature. A Structural Regression
Model was constructed that determines the influence of variables at one spatial scale
on those at another. We considered the influence of regional changes in precipitation
and temperature on the conditions at the catchment and channel scale. We also
determined how fish assemblages might change from their present state and found
that cold water species would retract their range up the catchment whilst warm water
species would expand their range into the void. The Structural Regression Model
indicates the magnitude of the error associated with any data source in its relationship
with any other latent or manifest variable thus allowing us to be explicit about the
confidence we have in our model.