In arthropod community ecology, species richness studies tend to be prioritized over those investigating patterns of abundance. Consequently, the biotic and abiotic drivers of arboreal arthropod abundance are still relatively poorly known. In this cross-continental study, we employ a theoretical framework in order to examine patterns of covariance among herbivorous and predatory arthropod guilds. Leaf-chewing and leaf-mining herbivores, and predatory ants and spiders, were censused on > 1,000 trees in nine 0.1 ha forest plots. After controlling for tree size and season, we found no negative pairwise correlations between guild abundances per plot, suggestive of weak signals of both inter-guild competition and top-down regulation of herbivores by predators. Inter-guild interaction strengths did not vary with mean annual temperature, thus opposing the hypothesis that biotic interactions intensify towards the equator. We find evidence for the bottom-up limitation of arthropod abundances via resources and abiotic factors, rather than for competition and predation.

Field sites and experimental design

We studied lowland temperate forests in Mikulcice (Czech Republic (CZE); 1 plot) and Toms Brook (Virginia, USA; 2 plots), and tropical forests in Papua New Guinea (PNG): lowland forest in Wanang (2 plots), mid-elevation forest in Numba (2 plots), and montane forest in Yawan (2 plots). All plots were located in old-growth forests (secondary forests, forest edges, plantations, stands with non-native vegetation, and large gaps were all avoided). For detailed information about location, climate, elevation, etc. at each site and individual plots see Fig. 1, Table S1, and Supporting Information: Supplementary Materials and methods.

Sampling methods

Each plot was 0.1 ha in size, rectangular, and chosen to represent the typical vegetation structure and species composition of local broadleaf forests. The vegetation of each plot was surveyed, and all stems with a diameter at breast height (DBH) ≥ 5 cm were tagged, mapped and identified to species. The plots were then gradually felled and sampled for arthropods (see detailed protocols in Volf et al. 2019). We took advantage of ongoing logging operations (CZE, USA) or subsistence shifting agriculture (PNG) at our sites to   avoid contributing to net deforestation. The felled trees were stripped of leaves and the total leaf biomass of the foliage was weighed. We estimated the total leaf area of the foliage using a ratio of leaf area to weight measured from randomly selected leaf samples for each tree (see details in Volf et al. 2019).

The felled trees were exhaustively surveyed for focal taxa (non-flying arthropods) by manually searching the foliage. Our focal taxa included all leaf-chewing lepidopteran larvae (free feeding and shelter-builders), leaf miners, spiders, and ants. We only used data on the number of live leaf miners since abandoned mines do not reflect the population size at the time of sampling. All live mines, caterpillars, and spiders were collected. For ants, we hand-collected foraging individuals during a standardised search immediately upon felling, beginning with the base of the trunk and working  up to the top of the canopy (Klimes et al. 2015). In addition, we sampled all ant nests found during the destructive sampling ( including cryptic nests inside tree tissues and the attached epiphytes and lianas, see details in Klimes et al. 2015; Mottl et al. 2019a; Plowman et al. 2019). The size of nests was visually estimated on a three-level categorical scale: 1) <100, 2) 100-1,000, and 3) >1,000 individuals. We used the values 50, 500 and 1,500 individuals, respectively, to represent these categories in our analyses. We calculated the total number of ant individuals sampled outside the nest for each tree (freely foraging on a tree; Foraging ants hereafter) and estimated the total ant abundance for each tree (foraging and nest ants summed; All ants hereafter). Unlike abundances of other arthropods, the overall abundances for ants are more likely to be underestimated, as our nest size estimates are probably conservative. However, this sampling allows the examination of each nest, including cryptic nests (in tree cavities, under epiphytes, etc.), which could easily be overlooked (Yanoviak et al. 2003). Therefore, it is more precise than other methods used for ant sampling (e.g. beating, fogging). Note that spiders were not sampled in Wanang (both plots) and the plot A in Yawan.