Including biotic information in SDMs generally improves model
performance
In this study, we showed that the addition of the biotic interaction
increases model performance under all metrics and this increase is the
highest for the cleptoparasitic bees, followed by the oligolectic and
polylectic bees. An increase in model performance by the addition of the
host of the cleptoparasitic bee has been observed (Giannini et al.
2013), however the increase in model performance for oligolectic bees
and their host plants at a resolution of 10 km was often not significant
(Giannini et al. 2013). This highlights once again the importance of
resolution. Furthermore, the importance of including biotic interactions
is not limited to plant-pollinator interactions (Heikkinen et al 2007;
Kissling et al. 2007; Bateman et al. 2012; Leach et al 2016; Mpakairi et
al. 2017; Roslin et al. 2017; Atauchi et al. 2018; Herrera et al. 2018;
Mathieu‐Bégné et al. 2021) and biotic interaction can play a role in the
distribution range edges of species even at a larger scale (Paquette &
Hargreaves 2021; Freeman et al. 2022).
The null models with interactions of random pollinated plants or host
bees, revealed how the specificity of the interaction (e.g. specialist
versus generalist) influences which interacting species could be used. A
higher specificity was observed for the oligolectic and cleptoparasitic
bees than the polylectic bees, whose models benefitted from a range of
different flowering plants. The high performance of the specific
interacting species in the models of the cleptoparasitic bees confirmed
how important their host species are for modelling their distribution.
Another contributing factor can be the biases in the data source: the
distribution of the cleptoparasitic interacting species is sourced from
the same wild bee occurrences database and therefore, shares similar
collection biases to the modelled species. In contrast to the
distribution of the plant species which likely have their own separate
collection biases. In the case that data on the interacting species is
lacking, an option would be to use information from a co-occurring
species (Briscoe Runquist et al. 2021). Our study showed that the
inclusion of other visited plants can also improve model performance.
We found that a higher degree of flower specialization for the flower
visiting bees and a narrower distribution of the interacting species for
both the cleptoparasitic and flower visiting bees were related to a
higher importance of the biotic factor in the SDMs. The dependence
between the distribution of two organisms, each at one side of the
biotic interaction, has been shown in many different studies (Fauchald
et al. 2000, Byholm et al. 2012, Atauchi et al. 2018) and even at a
macroecological scale (Araújo & Luoto 2007). If the distribution of the
interacting species is narrow, it is more likely to be a limiting
factor, delimiting the boundaries of the potential distribution of the
modelled species. Other studies have found that narrow distributions,
although not too narrow, mean more accurate models and high importance
for certain key habitat factors (Tsoar et al. 2007; Syphard & Franklin
2010) and that specialist species yield better models than generalist
species (Marshall et al., 2015; Grenouillet et al. 2011). We found that
if the biotic relationship is strong, e.g. parasite-host relationships,
then that becomes by far the most important factor.