Study area
Mount Cameroon (South-Western Province, Cameroon) is the highest mountain in West/Central Africa. This active volcano rises from the Gulf of Guinea seashore up to 4095 m asl. Its southwestern slope represents the only complete altitudinal gradient of primary forests from lowland up to the timberline (~ 2200 m asl.) in the Afrotropics. Belonging to the biodiversity hotspot, Mount Cameroon harbour numerous endemics45,46,47. With > 12,000 mm of yearly precipitation, foothills of Mount Cameroon belong among the globally wettest places42. Most precipitation occur during the wet season (June–September; > 2000 mm monthly), whilst the dry season (late December–February) usually lacks any strong rains42. Since 2009, most of its forests have become protected by the Mount Cameroon National Park.
Volcanism is the strongest natural disturbance on Mount Cameroon with the frequency of eruptions every ten to thirty years. Remarkably, on the studied southwestern slope, two eruptions in 1982 and 1999 created a continuous strip of bare lava rocks (in this study referred as ‘the lava flow’) interrupting the forests on the southwestern slope from above the timberline down to the seashore (Fig. 1a).
A small population of forest elephants (Loxodonta cyclotis) strongly affects forests above ca. 800 m asl. on the southwestern slope28,45. It is highly isolated from the nearest populations of the Korup NP and the Banyang-Mbo Wildlife Sanctuary, as well as from much larger metapopulations in the Congo Basin48. It has been estimated to ~ 130 individuals with a patchy local distribution28. On the southwestern slope, they concentrate around three crater lakes representing the only available water sources during the high dry season, although their local elevational range covers the gradient from lowlands to montane grasslands just above the timberline28. They rarely (if ever) cross the old lava flows, representing natural obstacles dividing forests of the southwestern slope to two blocks with different dynamics. As a result, forests on the western side of the longest lava flow have an open structure, with numerous extensive clearings and ‘elephant pastures’, whereas eastern forests are characteristic by undisturbed dense canopy (Fig. 1). To our knowledge, the two forest blocks are not influenced by any extensive human activities, nor differ in any significant environmental conditions28,45. Hereafter, we refer the forests west and east from the lava flow as disturbed and undisturbed, respectively. Effects of forest elephant disturbances on communities of trees and insects were investigated at four localities, two in an upland forest (1100 m asl.), and two in a montane forest (1850 m asl.).
Tree diversity and forest structure
At each of four sampling sites, eight circular plots (20 m radius, ~ 150 m from each other) were established in high canopy forests (although sparse in the undisturbed sites), any larger clearings were avoided. In the disturbed forest sites, the plots were previously used for a study of elevational diversity patterns40,42. In the undisturbed forest sites, plots were established specifically for this study.
To assess the tree diversity in both disturbed and undisturbed forest plots, all living and dead trees with diameter at breast height (DBH, 1.3 m) ≥ 10 cm were identified to (morpho)species (see40 for details). To study impact of elephant disturbances on forest structure, each plot was characterized by twelve descriptors. Besides tree species richness, living and dead trees with DBH ≥ 10 cm were counted. Consequently, DBH and basal area of each tree were measured and averaged per plot (mean DBH and mean basal area). Height of each tree was estimated and averaged per plot (mean height), together with the tallest tree height (maximum height) per plot. From these measurements, two additional indices were computed for each tree: stem slenderness index (SSI) was calculated as a ratio between tree height and DBH, and tree volume was estimated from the tree height and basal area49. Both measurements were then averaged per plot (mean SSI and mean tree volume). Finally, following Grote50, proxies of shrub, lower canopy, and higher canopy coverages per plot were estimated by summing the DBH of three tree height categories: 0–8 m (shrubs), 8–16 m (lower canopy), > 16 m (higher canopy).
Insect sampling
Butterflies and moths (Lepidoptera) were selected as the focal insect groups because they belong into one of the species richest insect orders, with relatively well-known ecology and taxonomy, and with well-standardized quantitative sampling methods. Moreover, they strongly differ in their habitat use29. In conclusion, butterflies51 and moths52 are often used as efficient bioindicators of changes in tropical forest ecosystems, especially useful if both groups are combined in a single study. Within each sampling plot, fruit-feeding lepidopterans were sampled by five bait traps (four in understory and one in canopy per sampling, i.e. 40 traps per sampling site, and 160 traps in total) baited by fermented bananas (see Maicher et al.42 for details). All fruit-feeding butterflies and moths (hereinafter referred as butterflies and fruit-feeding moths) were killed (this is necessary to avoid repetitive counting of the same individuals53) daily for ten consecutive days and identified to (morpho)species.
Additionally, moths were attracted by light at three ‘mothing plots’ per sampling site, established out of the sampling plots described above. These plots were selected to characterize the local heterogeneity of forest habitats and separated by a few hundred meters from each other. To keep the necessary standardisation, all mothing plots at both types of forest were established in semi-open patches, avoiding both dense forest and larger openings. Moths were attracted by a single light (see Maicher et al.42 for details) during each of six complete nights per elevation (i.e., two nights per plot). Six target moth groups (Lymantriinae, Notodontidae, Lasiocampidae, Sphingidae, Saturniidae, and Eupterotidae; hereafter referred as light-attracted moths) were collected manually, killed, and later identified into (morpho)species. The three lepidopteran datasets (butterflies, and fruit-feeding and light-attracted moths) were extracted from Maicher et al.42 for the disturbed forest plots, whilst the described sampling was performed in the undisturbed forest plots specifically for this study. Voucher specimens were deposited in the Nature Education Centre, Jagiellonian University, Kraków, Poland.
To partially cover the seasonality54, the insect sampling was repeated during transition from wet to dry season (November/December), and transition from dry to wet season (April/May) in all disturbed and undisturbed forest plots.
Diversity analyses
To check sampling completeness of all focal groups, the sampling coverages were computed to evaluate our data quality using the iNEXT package55 in R 3.5.156. For all focal groups in all seasons and at all elevations, the sampling coverages were always ≥ 0.84 (mostly even ≥ 0.90), indicating a sufficient coverage of the sampled communities (Supplementary Table S1). Therefore, observed species richness was used in all analyses57.
Effects of disturbance on species richness were analysed separately for each focal group by Generalized Estimated Equations (GEE) using the geepack package58. For trees, species richness from individual plots were used as a ‘sample’ with an independent covariance structure, with disturbance, elevation, and their interaction treated as explanatory variables. For lepidopterans, because of the temporal pseudo-replicative sampling design, species richness from a sampling day (butterflies and fruit-feeding moths) or night (light-attracted moths) at individual plot was used as a ‘sample’ with the first-order autoregressive relationship AR(1) covariance structure (i.e. repeated measurements design). Disturbance, season, elevation, disturbance × season, and disturbance × elevation were treated as explanatory variables. All models were conducted with Poisson distribution and log-link function. Pairwise post-hoc comparisons of the estimated marginal means were compared by Wald χ2 tests. Additionally, species richness of individual families of trees, butterflies, and light-attracted moths were analysed by Redundancy Analyses (RDA), a multivariate analogue of regression, based on the length of gradients in the data59. All families with > 5 species were included in three RDA models, separately for the studied groups (the subfamily name Lymantriinae is used, because they are the only group of the hyperdiverse Erebidae family of the light-attracted moths). Fruit-feeding moth families were not analyzed because 83% of their specimens belonged to Erebidae and all other families were therefore minor in the sampled data. Species richness of individual families per plot were used as response variables, whilst interaction of disturbance and elevation were applied as factorial explanatory variable (for butterflies and light-attracted moths, the temporal variation was treated by adding season as a covariate).
Differences in composition of communities between the disturbed and undisturbed forests were analysed by multivariate ordination methods59, separately for each focal group. Firstly, the main patterns in species composition of individual plots were visualized by Non-Metric Multidimensional Scaling (NMDS) in Primer-E v660. NMDSs were generated using Bray–Curtis similarity, computed from square-root transformed species abundances per plot. Subsequently, influence of disturbance on community composition of each focal group was tested by constrained partial Canonical Correspondence Analyses (CCA) with log‐transformed species’ abundances as response variables and elevation as covariate59. Significance of all partial CCAs were tested by Monte Carlo permutation tests with 9999 permutations.
Finally, differences in the forest structure descriptors between the disturbed and undisturbed forests were analysed by partial Redundancy Analysis (RDA). Prior to the analysis, preliminary checking of the multicollinearity table among the structure descriptors was investigated. Only forest structure descriptors with pairwise collinearity < 0.80, i.e. tree species richness, number of dead trees, mean DBH, mean height, maximum height, mean SSI, and higher canopy coverage, were included in these analyses. Their log‐transformed values were used as response variables59. RDA was then run with disturbance as explanatory variable and elevation as covariate, and tested by Monte Carlo permutation test (9999 permutations). All CCAs and RDAs were performed in Canoco 561.
Species distribution range
To analyse if the elephant disturbance supports rather range-restricted species or widely distributed generalists, we used numbers of Afrotropical countries with known records of each tree and lepidopteran species as a proxy for their distribution range; we are not aware of any more precise existing dataset covering all studied groups for the generally understudied Afrotropics. Because of the limited knowledge on Afrotropical Lepidoptera, we ranked only butterflies and light-attracted Sphingidae and Saturniidae moths (the latter two analysed together and referred as light-attracted moths). This distribution data were excerpted from the RAINBIO database for trees62, Williams63 for butterflies, and Afromoths.net for moths64; all considered as the most comprehensive databases. Two non-native trees (Persea americana and Cecropia peltata) and three tree species not included in the RAINBIO database and all morphospecies were excluded from these analyses. In total, 73 species of trees and 71 species of insects (50 butterflies and 21 moths) were included in the distribution range analyses.
To consider the relative abundances of individual species in the communities, the distribution range of each species was multiplied by the number of collected individuals per sample and their sums were divided by the total number of individuals recorded at each sample. These mean distribution ranges per sample were then compared between disturbed and undisturbed forest sites by GEE analyses (with normal distribution; independent covariance structure) following the same model design as for the above-described comparisons of species richness.
Source: Ecology - nature.com
