Study sites
Our study sites are located on the Yatta Plateau in south-eastern Kenya. This region is characterized by dry savannahs. Annual rainfall (average: 810 mm) occurs during two periods, from March to May (average: 330 mm) and from October to January (average 480 mm) (c.f. Jaetzold et al.37). The commonest soil types are ferralsols and luvisols, which are of low fertility37. 97.1% of the human population in our study region depend on subsistence crop farming38, and the population has almost doubled in number from 1999 to 200938. Consequently, fallow periods for fields are omitted, which further decreases soil fertility, and increases pressure on pristine habitats.
The dry savannah landscape is traversed by temporary (seasonal) rivers. These rivers are bordered by riparian vegetation, consisting of a diverse and unique plant community. However, this vegetation is frequently exploited for timber, charcoal and brick production39,40. The region is further affected by climate change, with an increase in rainfall variability and mean temperature37. These factors lower the reliability of agricultural production and food security, hence leading to severe destruction of pristine habitats.
We selected two study sites, affected by different anthropogenic pressures, but which are subject to identical biotic and abiotic preconditions (including seasonality): Firstly, a highly degraded anthropogenic landscape along Nzeeu River, south of Kitui city. Secondly, a largely intact dryland environment along Kainaini River located near the university campus of the South Eastern Kenya University, north of Kitui city (Fig. 1). The landscape along Nzeeu River is densely populated by subsistence farmers. Thus, the original riparian and savannah vegetation has been mostly transformed into arable fields for the cultivation of maize, sorghum, peas, and mangos. Furthermore, the riparian vegetation, where it still exists, has largely been replaced by invasive exotic plant species (e.g. Lantana camara)12. The landscape of our second study site along Kainaini River represents a still largely intact riparian forest with adjoining dry savannahs. It remains mostly undisturbed, except for some moderate live-stock pasturing by nearby subsistence settlers.
Butterfly assessments
We counted butterflies in both habitat types along line-transects, each 150 m long. We set 24 transects along each of the two rivers, with eight transects along the river bank, eight 250 m distant to the river, and another eight 500 m distant to the river (in total: 2 × 24 transects = 48 transects). The minimum distance between transects was at least 200 m, to minimize spatial autocorrelation. Exact GPS coordinates of each transect are given in Appendix S2.
We recorded all butterflies encountered during transect counts (species, number of individuals of each species). Each transect was visited eight times during the dry season (August/September 2019) and eight times during the rainy season (January/February 2020). Data collection was performed between 9 a.m. and 4 p.m. Each butterfly individual within 5 m of the transect line (horizontally to vertically) was recorded by visual observation and, if needed, a butterfly net (see Pollard15, with modifications). While recording butterflies, the observers walked very slowly and spent about 15 min per transect. Species were identified either immediately while the butterfly was on the wing, or individuals were netted and then determined in the field. Individuals of species for which ad hoc identification was critical (e.g. many blues and skippers) were caught with the net, photographed (upper and under wing side) and released again. The photograph-based identification of these individuals was performed later using literature25. Apart from species and number of individuals per species, we recorded cloud cover during each transect walk (classified as: clear, slightly cloudy, mostly cloudy, overcast), exact time, and date. Field teams comprised two observers and one person making notes of all observations. Transects are displayed in Fig. 1. All butterfly data collected are compiled in Appendix S3.
Traits
The occurrence of a species in a specific environment strongly depends on its ecology, behaviour, and life-history41. Therefore, we considered these characteristics for each butterfly species recorded in the field. These trait data were compiled from Larsen25 and web-sites (e.g. www.gbif.org, www.lepiforum.de/non-eu.pl). We considered the following characteristics: wing span (mm), ratio length/width of the forewing (relative), ratio forewing length/thorax width (relative), geographic distribution (4 categories), savannah index (5 categories), forest index (5 categories), tree index (3 categories), wetness index (3 categories), habitat specialisation (3 categories), larval foodplant specialisation (3 categories), larval food plant type (dicotyledonous, monocotyledonous), and hemeroby index (4 categories). Detailed classifications are provided in Appendix S4.
Habitat parameters
Habitat structures impact species´ occurrence, abundances and community structures42. In our study, we considered habitat structures for each transect. Habitat parameters were recorded (counted and estimated) every 20 m along each transect. We estimated the following habitat parameters: Canopy cover (percentage of leaf cover vs. sky measured with the CanopeoApp); herb, shrub and tree cover (percentage coverage of each layer within a radius of 3 m); flowers on herbs, shrubs and trees (estimated within a radius of 3 m, and subsequently allocated to the classes 0, 1–10, 11–50, 51–100 and > 100 flowers); occurrence of Lantana camara shrub, and exotic trees (estimated coverage within a radius of 3 m, and subsequently allocated to the classes 0 (no), 1 (rare), 2 (present) and 3 (dominant), respectively); and water availability (presence/absence) within a radius of 3 m. All raw data of habitat parameters are provided in Appendix S5.
Statistics
We first arranged the raw data in three matrices: a 71 × 14 species × trait matrix T, a 71 × 96 species × transect matrix M, and a 6 × 96 habitat characteristics × transect matrix H. Matrix multiplication of E = T−1MA−1, where A is the vector of total abundances in the transects, returned a matrix E of average trait expression in each transect.
To answer the first research question, we compared species richness, abundances, and trait expression between the transects and used general linear modelling (glm) to detect differences in richness and trait expression with respect to the study sites (i.e. the two river systems with their different land-use patterns), season, distance from the rivers, as well as to environmental variables. Some of the habitat variables and trait expressions were highly positively correlated (Appendix S1). Consequently, the glm included only variables correlated by less than r = 0.7 (i.e. shrub cover, tree cover, habitat specialisation, savannah index, larval foodplant specialisation, and hemeroby).
To infer differences in community structure between transects (second research question), we first calculated the two most dominant eigenvectors, which explained 91.5% and 3.5% of variance, of a principal components analysis of the M matrix. These eigenvectors cover differences in species composition between and within transects. We used glm and two-way Permanova to relate these differences to season, distance to river, and study sites (i.e. different land-use types in the two river systems). Additionally, we assessed the degree of β-diversity among sets of transects with the proportional turnover metric of Tuomisto43: (beta =1-frac{alpha }{gamma }); where α denotes the average species richness per transect and γ the corresponding total richness.
To infer species spill-over effects from the riparian forests into the adjoining savannah (third research question), we calculated the Bray–Curtis similarities for three groups of transects within each season and study site. First, we compared average pairwise Bray–Curtis values between transects of intermediate and greater distance with the near-river transects within each study site. Second, we calculated the average Bray–Curtis similarities between all transects within each study site (2)—season (2)—distance class to river (3) combination. Third, we calculated the average within-transect Bray–Curtis similarity for the rainy season, to infer small scale compositional variability. The latter calculations were impossible for the dry season, due to the overall low number of recorded species. Calculations were done with Statistica 12.
Source: Ecology - nature.com
