reality: “A1B describes rapid economic development and growth,
with balanced technological development across all sources, i.e.
neither fossil intensive nor all non-fossil sources” (Cameron et al.,
2012).
We used the Maxent algorithm, which estimates a target probability
distribution by finding the probability associated with
the maximum entropy (or the closest to a uniform distribution)
(Phillips et al., 2006). The Maxent approach is useful primarily
because it can be applied to analyze small and presence-only
datasets (Wisz et al., 2008). The area under the curve (AUC) of the
receiver-operating graph (ROC) was used to estimate the success
and failure of the prediction during the modeling process with a
set of test data (30% of the data) (Fielding and Bell, 1997). The AUC
values vary from 0 to 1. A value of 1 indicates the highest accuracy.
The output of Maxent represents the occurrence probability for
each grid cell used in the model. We classified these probabilities
into three categories: high probability, corresponding to the cells
that show a predicted probability of presence of more than 75%;
medium probability, corresponding to a range of probabilities of
50–75%; and minimum probability, representing probabilities less
than 50%. We classified all of the models of all of the species using
these three classes.
We first compared the total frequency of pixels with the maximum
occurrence probability (>75%) obtained for the current and
2050 scenarios per species. Based on this information, we evaluated
the change in the amount of suitable area for each species based on
the future scenario of climate change. Because our objective was
to determine future suitable areas to protect species, we used the
highest probability of occurrence to avoid spending resources to
preserve species in unsuitable areas (Araújo and Peterson, 2012).
We then overlapped the areas of maximum probability for all of
the bee species to build the final models; we performed the same
overlapping with the models of all of the plant species. The purpose
of this procedure was to build a representation of the total
area that shows the highest suitability both for all passion fruit
pollinator bees and for their forage plants. We considered the same
procedure to forecast the future potential occurrence according to
the postulated scenario of climate change.
Finally, we overlapped the current and future models of bees
and plants to identify the areas in which both groups of species
are potentially present now and the areas in which they will
occur under the future scenario. This procedure aimed to identify
potential areas for preserving these species. All of the procedures
involving the maps obtained through modeling were performed
with ArcGIS 10 (Esri Inc.).