Williams-Climate-change-refugia-for-terrestrial-biodiversity_0

Figure 52: Distribution of rainforest types, and names of regions used in this
study.
We assessed stability for all rainforest in Figure 49 and – because the climatic limits for
rainforest would not be consistent across this range – separately for each of the 5
rainforest regions shown, based on the method of Graham et al. (2010), as follows. We
modelled the environmental niche of the rainforest pixels using maximum entropy
modelling (Maxent; Phillips et al. 2006) and the eight bioclim variables for present-day
climate. This model was then projected to each of the 62 time slices to estimate the
distribution of suitable habitat at each time, using clamping to limit prediction into
climate space beyond the range of conditions available to train the present-day model.
For each rainforest model, we calculated stability as the mean of the negative log of
suitability through time for each cell of the surface. We took the exponent of this value
to give to a habitat stability value (ranging from 0 to 1). This represents the degree to
which that cell has continuously provided suitable climate for rainforest (Graham et al.
2010). At the extremes 1 indicates continuous high suitability, and 0 indicates that the
area was highly unsuitable for rainforest during some periods. The value cannot be
interpreted as an average cell residence time. It is a multiplicative measure, so a period
of low suitability (i.e. a temporal gap in suitable habitat) downweights the index
strongly, to reflect the degree of continuity. The result of this analysis was a 120 000year
habitat stability surface for all rainforest and for each rainforest region.
An area of suitable habitat that functions as a refuge over geological time need not be
a static location. It may instead move to track suitable conditions, for example, by
shifting downslope in cooler times. We allowed continuity between suitable locations by
specifying a maximum velocity of movement, to represent the ability of rainforest to
expand into newly suitable areas nearby. This model, described in detail by Graham et
al. (2010), estimates suitability at each time period, but also allows dispersal between
Climate change refugia for terrestrial biodiversity 79