Fig 1.
Construction of a trait similarity network from binary trait data.
(a) Trait incidence matrix for 12 example species across seven binary traits. A value of 1 indicates trait presence and 0 indicates trait absence. (b) Trait similarity network constructed from the incidence matrix in panel (a). Each circle (node) represents a species, and each line (edge) connecting two species indicates that those species share a similar combination of traits, for example, both being large-bodied, nocturnal, and specialist feeders. Only species pairs with sufficiently similar trait profiles are connected, meaning species that share more traits in common are more likely to be linked. The thickness of each line reflects how similar the two connected species are, with thicker lines indicating a greater number of shared traits. (c) The same trait similarity network with nodes colored by ecological strategy groups: large-bodied specialists, small-bodied generalists, and medium-bodied opportunists. Clusters of densely connected nodes correspond to species sharing recurring combinations of life-history traits, representing distinct ecological strategies occupying similar regions of multidimensional trait space.
Table 1.
Continuous variables excluded prior to construction of the species–trait incidence matrix. Continuous morphological measurements, monitoring-derived distribution and trend metrics, phenological shift statistics, and hostplant Ellenberg indicator values were excluded because species similarity in this study was defined using categorical ecological trait states representing life-history strategies. Variables reflecting monitoring coverage, temporal change, morphological magnitude, or environmental proxy gradients were retained separately for descriptive interpretation but were not incorporated into the trait-based similarity network.
Table 2.
Summary of binary conservation status targets derived from regional Red List assessments for Great Britain and Ireland.
Table 3.
Sensitivity analysis for selection of neighborhood size (k) in species similarity network construction. For each value of k, we report basic graph properties and bootstrap community stability measured by the ARI. The final value (k = 10) was selected as the smallest neighborhood size yielding a predominantly connected network and maximal stability.
Fig 2.
Species similarity network constructed from Lepidoptera trait data.
Nodes represent species and edges represent weighted trait similarity (Jaccard-based, union-kNN with k = 10). Node colors indicate Leiden communities, revealing clear modular structure with nine detected groups. Communities are described in Table 4.
Table 4.
Summary of Leiden network communities derived from the species similarity network. For each community, the number of species and the number (and percentage) classified as threatened are shown for Great Britain (GB) and Ireland (IE), with some counted in both regional summaries because they are classified as threatened in both GB and IE. Ecological profiles are summarized using the most common traits per community, based on prevalence relative to the full species pool.
Table 5.
Community-level associations between Leiden network communities and threatened status in Great Britain. Odds ratios compare each community against all others combined. Community 8 is omitted due to insufficient sample size for contingency-based statistical testing.
Fig 3.
Subnetwork corresponding to Community 7 (multivoltine, externally developing species).
Nodes represent individual species within Community 7 and edges represent weighted trait similarity among species in the community. Node size indicates degree. Green nodes are threatened.
Fig 4.
Enrichment of high-betweenness species by community.
For each Leiden community, the fraction of species falling within the top 5% and top 10% of betweenness centrality values is shown. Horizontal dashed lines indicate the expected fractions under a uniform distribution.