Beta diversity and kīpuka size
NMDS plots of zOTUs and OTUs showed a clear separation of the arthropod community composition in our different sites (Fig. 4A & C). The first axis separated the continuous forests from kīpuka, while the two continuous forests in Stainback and Kona Hema were well-separated by the second axis. The NMDS plots also showed a separation of kīpuka center and edge habitats and barren lava. The differentiation of center and edge habitat was particularly pronounced in larger kīpuka. While the five smallest kīpuka (area 555 – 4,212 m2) grouped closer to the edge samples, the remaining eight (area 5,458 – 100,081 m2) were clearly distinct and formed a separate cluster. In comparison to sites in each continuous forest area, kīpuka were considerably more dispersed in the NMDS plots. Moreover, sites in kīpuka centers grouped closer to continuous forest sites than to the edge sites, suggesting that kīpuka center communities are more similar to continuous forest communities. While zOTU and 3 % radius OTU based NMDS supported similar patterns of differentiation, the differentiation between the edges and centers of large kīpuka was more pronounced for the 3 % radius OTU dataset.
The separation of different areas shown in the NMDS plot was also supported by patterns of beta diversity for zOTUs and 3 % radius OTUs (Fig. 4B & D). With a notable exception of sites in barren lava and the center of kīpuka, which showed the highest dissimilarity, the average within-area dissimilarity (average zOTU|OTU beta Stainback: 0.57|0.55; Kona Hema: 0.60|0.58; kīpuka center: 0.78|0.76; kīpuka edges: 0.69|0.64; Lava: 0.76|0.72) was significantly lower than between-area comparisons (average zOTU|OTU beta Stainback-Kona: 0.85|0.79; Stainback-center: 0.86|0.83; Stainback-edge: 0.85|0.83; Stainback-lava: 0.92|0.91; Kona-center: 0.92|0.88; Kona-edge: 0.91|0.88; Kona-lava: 0.95|0.94; center-edge: 0.79|0.76; center-lava: 0.86|0.84; edge-lava: 0.76|0.70). Within-area dissimilarity in continuous forest was significantly lower than in kīpuka habitats. This held true particularly for kīpuka centers, which showed a higher within-area differentiation than edges. Hence, the edges of different kīpuka were more connected than their corresponding centers. These patterns held true for zOTUs and 3 % radius OTUs, but the difference between kīpuka and continuous forest areas was slightly higher for zOTUs. The higher within-area differentiation of kīpuka centers compared to continuous forest and kīpuka edge held up even when only the eight largest kīpuka, which formed a separate cluster closer to continuous forest on the NMDS plot, were compared to each other (average zOTU|OTU beta Stainback: 0.59|0.55; Kona Hema: 0.61|0.58; kīpuka Center: 0.79|0.75; kīpuka Edges: 0.71|0.66; Lava: 0.79|0.72).
Centers and edges of the same kīpuka harbored distinct arthropod communities, particularly for larger kīpuka (> 5,000 m2; Suppl. Fig. 4). In fact, the dissimilarity between the center and edge of each kīpuka showed a pronounced association with the area of that kīpuka. Community dissimilarity between centers and edges steadily increased with kīpuka size for small kīpuka, but began saturating for kīpuka larger than 5,000 m2.
Our previous NMDS analysis revealed two separate communities in kīpuka centers: kīpuka so small they behaved more like kīpuka edges (< 5,000 m2) and larger kīpuka with core communities more similar to continuous forest communities (> 5,000 m2). To explore the association of beta diversity with distance between sites, we thus generated two kīpuka datasets. The first dataset included all kīpuka, the second consisted of only the eight largest kīpuka (5,458 – 100,081 m2), which formed a well-separated cluster from edge habitats in our NMDS plots, and the third one consisted of the five smallest kīpuka (555 – 4,212 m2). An analysis of within-area beta diversity for these datasets showed a significantly higher community dissimilarity between centers of large kīpuka than those of small kīpuka. At the same time, no significant difference of community dissimilarity for edges of large and small kīpuka was found.
We next analyzed patterns of community isolation by distance for continuous forests, kīpuka centers and edges. The continuous forests both showed a significant and quite similar linear association of logarithmic geographic distance and community dissimilarity (Fig. 5; r2Kona Hema= 0.16; r2Stainback= 0.74; LM p < 0.05). We also found a positive linear association of beta diversity and logarithmic geographic distance for the kīpuka sites (Fig. 5; r2center = 0.12; r2edge = 0.30; LM p < 0.05). Compared to continuous forest, the overall dissimilarity was much higher for kīpuka, even at short geographic distances. This pattern was particularly strong for kīpuka centers, while kīpuka edge habitats appeared to be more connected. Community dissimilarity between kīpuka centers saturated at very low geographic distances. The community differentiation between kīpuka centers isolated by only a few 100 m of barren lava was comparable to the differentiation between continuous forest sites isolated by over 10 km. These patterns also held up when different arthropod orders were analyzed separately (Suppl. Fig. 5 & 6). Nearly all orders showed a saturated or nearly saturated beta diversity with distance for kīpuka centers, while an association with geographic distance was found for edge habitat. Diptera was the only order for which both center and edge showed similar patterns of isolation by distance.