2.5 Similarity distance-decay and landscape connectivity
We are interested in the effect of landscape features on the variation
of arthropods community composition, but considering the
multi-hierarchical approach of Baselga et al. (2013). For this, we
tested for distance decay of similarity across the different clustering
levels, considering as independent variable geographic distance (i.e.,
IBD) and effective distance, that is, resistance to dispersion
considering the landscape features: slope, altitude and vegetation types
(i.e., IBR). We focused on Diptera and Collembola assemblages, because
they were the two sampled orders showing better completeness and it is
expected they show strong differences in dispersal potential (Figure
S3). As response variable we used pairwise similarity of assemblages at
the haplotype, all CLs (0.5%, 1.5%, 3%, 5% and 7.5% lineages) and
putative molecular species by GMYC, among all sites. The analysis was
done both with the entire dataset and at finer geographical distances
within the West (AAB and ASB), and East (SJH, TLC) sampling points of
our study area (Figure 1a). At each CL, we calculated patterns of
similarity between pair of sites using the means of the Simpson’s
similarity index (1- pairwise βsim = a/[a +
min(b,c)] diversity) with R using the package betapart(Baselga & Orme,
2012), where a is the number of species present in both
territories, and b and c the number of species unique to
one or another, respectively.
To estimate effective distances we used the programcircuitscape-4.0(McRae et al., 2013)
to calculate “resistance distances” between pairs of nodes on a raster
grid. The input files for circuitscape are a raster of the study
area in which each cell is assigned a conductance value corresponding to
the relative probability of the arthropods moving through it; and a list
of focal nodes, that is the geographic coordinates of our sampling
blocks. In our study, we assessed whether community similarity in
different arthropod groups can be explained by landscape features, so we
used conductance grids representing different levels of resistance to
dispersion depending on vegetation heterogeneity, altitude and slope. To
compare the effect of these landscape variables against the effect of
distance alone, we also generated a ‘flat’ landscape; that is, a
landscape in which all cells have equal conductance. This is equivalent
to Euclidean distance but accounts for the finite size of the input
landscape being analyzed, so therefore is more appropriate for
comparison with the models using other grids
(Lee-Yaw et al.,
2009).
We acquired a 10 m-resolution digital land cover map of the Nevado of
Toluca defining the Abies religiosa forest (sampled vegetation),Pinus hartwegii forest, alpine grassland, agriculture,
urbanization, water in the crater and Nevado de Toluca’s crater
(González-Fernández et
al., 2018; Sunny et al., 2017). After cross-validation with our
sampling records, minor modifications were performed to adjustAbies forest distribution in areas of high topographic complexity
(see Supporting Material S2 for details). Then, we built the conductance
grids for the vegetation heterogeneity analyses assigning different
conductance values to each vegetation type (maximum value of conductance
is 1, meaning no resistance to dispersion) as detailed in Table S2.
To build the conductance grids for altitude and slope, we followed a
similar approach, assigning different conductance values to altitudinal
and slope ranges. In total 30 conductance grids were tested (Table S2;
Figure S5). Additionally, we tested for distance decay of similarity at
smaller geographic distances, for which we separated the West (AAB, ASB)
and East (SJH, TLC) sites and performed the same analyses of above, but
only with the “flat” raster and the one with highest explanatory level
on the previous test.
We used a negative exponential function ‘decay.model’ included inR package betapart (Baselga & Orme, 2012), to adjust a
negative exponential function to a generalized linear model (GML). We
used Simpson similarity (1 – βsim ; Baselga,
2010) as a response variable, the pairwise effective distances of each
resistance surface as predictor, log link and Gaussian error
(Arribas et al., 2020;
Gómez-Rodríguez & Baselga, 2018). Finally, we evaluated the fractal
pattern (i.e., self‐similar systems; Baselga et al., 2015) by a log–log
Pearson correlation of the haplotype level and (1) number of lineages,
(2) initial similarity (i.e., intercept), and (3) mean similarity
independently in each group (Collembola and Diptera) for both our entire
sampling and at the East and West sections of our sampling. High
correlation values are indicative of self-similarity in lineage
branching (i.e., number of lineages) and/or spatial geometry of lineage
distributional ranges (i.e., initial and mean similarity; Baselga et
al., 2015). Analyses and graphical representations of data were
performed with R using the packages vegan(Oksanen et al.,
2019) betapart(Baselga et al.,
2018) and ecodist(Goslee & Urban,
2020).