Bioclimatic variables
Yearly and seasonal patterns in temperature and precipitation are known
to influence behaviour and life history in bats around the world
(Conenna et al., 2019; Gorman et al., 2021; Kohyt et al., 2021; Weinberg
et al., 2022). To capture these conditions and test the hypothesis that
macroclimatic effects can determine occurrence of bats, the 19 standard
current bioclimatic variables were sourced at 2.5 arc-minute resolution
from the WorldClim v2.1 database for the period of 1970 to 2000 (Fick &
Hijmans, 2017). We focused on this resolution (~4x4 km)
considering, in the absence of detailed species-level information, that
such distances are within the average foraging range and mobility of
these species. Our study covers large taxonomic, spatial, and temporal
scales that aim to capture broad environmental effects that are best
detected using moderately coarse resolutions (Wiens et al., 2009). Finer
resolutions are more suitable to detect smaller-scale behavioural
effects (Pulliam, 2000) including movement, territoriality, and
inter-species interactions in mixed colonies.
Four future climate predictions were obtained combining two different
models and two different scenarios to capture uncertainty. We considered
the Canadian Earth System Model 5 (CanESM5; Swart et al., 2019) and
Hadley Centre Global Environment Model 3 (HadGEM3; Good, 2019; Good,
2020) available in the Coupled Model Intercomparison Project 6 (CMIP6)
from which we obtained 2.5 arc-minute climate predictions for a
near-future time (2050, averaged from 2041-2060). From each model, data
were obtained for the shared socioeconomic pathways 2 and 5 (SSP2,
equivalent to Representative Concentration Pathway RCP4.5; and SSP5,
equivalent to Representative Concentration Pathway RCP8.5). These
pathways, used by the International Panel on Climate Change,
characterise an optimistic ‘middle of the road’ socioeconomic scenario
representing an ideology towards sustainable development, and a
pessimistic ‘fossil-fuelled development’ scenario of climate change
based on development almost entirely based on fossil fuels in the
future, and little development towards sustainability and an emphasis on
resource- and energy-intensive lifestyles (Kriegler et al., 2017). The
two CanESM5 models are hereafter referred to as Can2-45 for SSP2-RCP4.5
and Can5-85 for SSP5-RCP8.5, and the two HadGEM3 models as Had2-45 for
SSP2-RCP4.5 and Had5-85 for SSP5-RCP8.5.
To allow for accuracy in model transfer and clarity in interpretation,
bioclimatic variables in ENM models are usually selected a priorito avoid multicollinearity and variance inflation. The bioclimatic
variables were first filtered using an assessment of ecological
relevance based on field knowledge and literature (Stones & Wiebers,
1965; Grindal et al., 1992; Bates & Harrison, 1997; Appel et al., 2019;
Corro et al., 2021). Then the variables were further filtered based on
collinearity - noting that available ENM algorithms are not greatly
impacted or can account for correlation and interaction between
variables (Muñoz & Felicísimo, 2004; Dormann et al., 2012; Junior &
Nobrega, 2018; Feng et al., 2019), and in some cases, correlated
variables may be used if they are considered ecologically relevant. An
analysis of multicollinearity was conducted using a combined variance
inflation factor (VIF) and pairwise correlation test in the usdmpackage (Naimi et al., 2014) in R. Variables with an absolute pairwise
Pearson’s r < 0.85 were selected for the analysis. When
a variable pair had high correlation (r 0.85|), we removed the
variable deemed to have least ecological relevance, but if both
variables were considered ecologically relevant, the variable with the
highest VIF was excluded instead. Additionally, collinearity shifts
between current and future predictions were measured using paired
t-tests of VIF scores for each variable, as these tend to impact model
transferability (Feng et al., 2019). All climate data were cropped and
masked to the extent of the study area defined above. A final set of 10
climate variables was selected for the analysis, which were considered
ecologically relevant for South Asian bats, and showed no significant
collinearity shifts between current and future scenarios (Supplementary
Material 1). The same variables were used for all species for consistent
interpretations.