The Arabuko Sokoke Forest (ASF) is the largest forest fragment remaining within the Kenyan coastal forests covering an area of 42,000 ha, the second being Shimba Hills Forest (25,300 ha; Fig. 1A; Burgess and Clarke (2000); Fungomeli et al. (2020a)). It is globally valued as a world biodiversity hotspot of the Eastern Arc and Coastal Forests of Kenya and Tanzania (Myers et al. 2000). ASF is a centre of endemism, hosting a conspicuous number of threatened and endangered species and recently declared a UNESCO Biosphere Reserve (UNESCO 2019). Anthropogenic pressure and biodiversity loss, together with climate change are heavily impacting tropical forests, such as the Arabuko Sokoke Forest in Kenya (Burgess and Clarke 2000; Newton and Echeverría 2014; FAO 2018; Fungomeli et al. 2025).

As a dry lowland coastal forest, ASF spreads between the cities of Kilifi in the south and Malindi in the north, positioned between 39°48'E and 40°00'E longitude and between 3°11'S and 3°29'S latitude (Fanshawe 1995; Muchiri et al. 2001). It lies on a flat coastal plain at sea level and the area is divided by a low escarpment which crosses the forest from south-west to north-east (Moomaw 1960; Fanshawe 1995).

The climate consists of rainy and dry seasons, with two rainfall seasons of long and short rains. The long rainy season occurs from April to July; short rains from October to December, while the dry season lasts from December to March and in August/September (Burgess and Clarke 2000; Omenge 2002). The annual rainfall ranges from 600 to 1,000 mm, with rainfall decreasing from east to west within the Forest (Omenge 2002; Habel et al. 2017). Temperature ranges from monthly averages of 24 °C and 30 °C and humidity is about 60% annually (Burgess and Clarke 2000). Several water pools exist within the Forest during the rainy season with most drying out in the dry season and there are no rivers within the forest (Fungomeli et al. 2001; Kanga 2002; Muriithi and Kenyon 2002).

A defining characteristic of the Arabuko-Sokoke Forest is the presence of distinct vegetation types, usually referred to as forest types (Fig. 1B–E). These forest types are closely associated with the underlying soil characteristics. The area features two predominant soil types: light, white sandy soils and heavier, red clay soils (Fanshawe 1995; Muchiri et al. 2001). These contrasting soil conditions have significantly influenced the distribution and composition of the forest’s vegetation communities. The four distinct forest types are the Cynometra forest, Brachystegia forest, Mixed forest and the forest edge (Fig. 1B–E).

Figure 1. 

A. Map of Arabuko Sokoke Forest, Kenya, showing the distribution of the 108 studied butterfly transects within the four forest types: Brachystegia, Cynometra, Mixed forest and Forest edge. B–E. The four forest types study sites within Arabuko Sokoke Forest, Kenya showing. B. Cynometra forest; C. Brachystegia forest; D. Mixed forest; E. Forest edge. Photo credits: Maria Fungomeli.

Cynometra forest

The Cynometra forest, which occupies the western sector of the ASF, is the most extensive of the forest types, accounting for over 50% of the forest area (Fanshawe 1995). This zone is found predominantly on red clay soils and is characteried by a dense, low-statured canopy, composed mainly of Cynometra webberi, Cynometra suhalensis and Manilkara sulcata, with occasional emergents, such as Euphorbia candelabrum (Fig. 1). The undergrowth is sparse due to the closed canopy and low light penetration, creating a relatively stable microclimate with reduced temperature fluctuations and high humidity.

Brachystegia forest

Brachystegia forest is located centrally within the ASF, on the nutrient-poor, white sandy soils and covers approximately 18% of the forest (Fig. 1; Muchiri et al. 2001). It is dominated by Brachystegia spiciformis, characterised by an open canopy structure interspersed with grasses and shrubs, resulting in a more sunlit and drier environment than other forest types. The floristic composition is adapted to dry conditions and the habitat is important for butterfly species, reptiles and birds that depend on high light availability and open understoreys. Additionally, the openness of this zone creates microhabitats favourable for thermoregulation and basking, crucial for many invertebrate and herpetofauna species.

Mixed forest

The mixed forest type is located in the eastern part of the Forest, where it grows on grey sandy soils that retain more moisture than the white sands, but are lighter than the red clay soils. This forest type represents about 17% of the Reserve and is notable for its high plant species richness and vertical stratification (Fanshawe 1995; Muchiri et al. 2001; Arabuko Sokoke Management Team 2002). It is dominated by mixed plant species of Afzelia quanzensis, Hymenaea verrucosa, Newtonia hildebrandtii and Manilkara sansibarensis.

Forest edge

The forest edge was selected along the transition from the mixed forest of the Arabuko-Sokoke Forest (ASF) to adjacent agricultural lands. This ecotone represents a gradual shift from intact forest ecosystems to human-modified landscapes and coastal habitats (Fig. 1; Muchiri et al. 2001). It is characterised by a heterogeneous mosaic of land uses, including subsistence farms, scattered shrubs and open grassy clearings. Structurally, it exhibits increased light penetration, higher temperature fluctuations and reduced canopy cover compared to the forest interior. These conditions create a distinctive microclimate that supports a unique assemblage of flora and fauna adapted to edge environments.

Field sampling and data collection were conducted during the dry season months (January-April) of 2019 across the four forest types of ASF: Cynometra forest, Brachystegia woodland, mixed forest and forest edge (Fig. 1). Butterflies were sampled by using a standard number of 27 transects within each vegetation type leading to a total of 108 transects. Each transect measured 10 m × 100 m. Butterflies were recorded in each transect by using a standard count technique performed by walking at slow constant pace for approximately 15 min. All butterfly species seen on both sides of the path were recorded. Each transect was surveyed once per day, every day, throughout the four-month dry season (January-April 2019), ensuring comprehensive and exhaustive species detection in accordance with established butterfly monitoring protocols (Pollard 1977; Pollard and Yates 1993).

Butterflies were identified and recorded at species level. Specimens that could not immediately be identified in the field were caught with a butterfly net and placed in numbered envelopes or photographed for further identification in the lab. Identification was carried out using the butterfly references for the area (Larsen 1991) and supported by taxonomic counter checks from published sources. All transects were geo-referenced, with details of date, hour of start and end.

Vegetation field sampling was performed by using 27 plots each measuring 10 m × 100 m (same used for butterfly transects hereafter referred to as plots) and internally subdivided into 20 subplots of 10 m × 5 m. Each vegetation plot corresponded to a butterfly transect. Within the plots and subplots, we identified and measured the height and diameter at breast height (DBH) for each individual woody plant species (trees, lianas and shrub) with DBH ≥ 5 cm. Plants with DBH < 5 cm, such as small shrubs, were identified in two subplots of each plot (see Fungomeli et al. 2020a, 2020b).

Butterfly traits: wingspan sizes

We compiled and obtained wingspan sizes for our sampled butterfly species from published data sources of Woodhall (2005), Woodhall (2020), Schmitt (pers. comm), Barcode of Life Data System database (https://v3.boldsystems.org) and from the collection of the Senckenberg German Entomological Institute, Müncheberg.

All butterflies encountered along the transects were classified according to their ecological traits and distribution. The larval diet of each species was determined, based on host plant use and categorised into one of three trophic breadth classes: (1) monophagous, restricted to a single host plant genus; (2) oligophagous, restricted to host plants within a single plant family; or (3) polyphagous, utilising host plants from multiple plant families. A further classification into endemic status was assigned to species according to Larsen (1996).

A community matrix was prepared for the butterfly species abundances and another matrix was prepared for the woody plant species across the forest types.

Butterfly species diversity

Butterfly species diversity was analysed in terms of species richness, Shannon index and Simpson index across forest types:

Shannon Index: $$H=\sum _{i=1}^k{p}_1\ln \left(p_1\right)$$

Simpson Index: $$D=1-\sum _{i=1}^k{p}_1^2$$

For both indices, k represents the total number of species, while p_i indicates the relative abundance of each species that is calculated as n_i/N (in which n_i indicates the number of individuals of the i-species and N indicates the number of individuals of all species within the transect.

Butterfly species richness and mean abundance distributions across forest types were visualised using boxplots. Mann-Whitney U test for pairwise comparisons was used to compare species richness and abundance amongst forest types. Additionally, rank-abundance curves were constructed for each forest type to illustrate patterns of species dominance and evenness (Whittaker 1965).

Rarefaction curves and species diversity estimation

To assess and compare species diversity across forest types, we employed sample coverage-based rarefaction and extrapolation within the Hill numbers framework (Chao et al. 2014). We analysed abundance data for each forest type using the iNEXT() function from the iNEXT package in R (Hsieh et al. 2016), specifying datatype = “abundance”. This function performs both interpolation (rarefaction) and extrapolation of diversity estimates, based on individual-based abundance data. Confidence intervals (95%) were derived using 1,000 bootstrap replicates. Following Chao et al. (2014), extrapolation was constrained to a maximum of twice the reference sample size to maintain estimate stability.

Correlation between butterfly and plant species diversity

We applied a symmetric Co-correspondence analysis (CoCA) to quantify relationships between the plant species community with the butterfly species community across the forest types. Co-correspondence analysis is useful for comparing biological communities where observations have been made at the same locations (Braak and Schaffers 2004). We did this by a weighted average of species abundance values for plant species and separately for butterfly species within each of the forest types. We used ‘coca’ function of the ‘cocorresp’ R package (Simpson 2009) to correlate the butterfly and plant communities using the ‘symmetric’ method. All graph plotting was performed using R package ggplot2 (Wickham 2016) and ggrepel (Slowikowski 2020).

Butterfly species composition

We square-root transformed butterfly community abundances prior to the analysis to reduce effects of dominant species. Transformed community abundances were then used to generate a Bray-Curtis dissimilarity matrix (Bray and Curtis 1957). We tested for species composition differences in the butterfly community structure amongst forest types by an analysis of similarities (ANOSIM) using the ‘anosim’ function of the ‘vegan’ R package (Oksanen et al. 2020). We also tested for significant differences between forest types using the permutational analysis of variance (PERMANOVA), using the ‘adonis’ function of the ‘vegan’ R package. All tests were conducted using 999 permutations. Butterfly species contributing to similarities across forest types were determined using similarity percentages analysis (SIMPER). P-values were adjusted using the Benjamini-Hochberg test to control the false discovery rate. This approach was selected for its ability to limit type I errors, while retaining greater statistical power than more conservative methods, such as the Bonferroni correction.

Beta diversity partitioning

To assess whether variability in butterfly community composition differed amongst the four forest types (Brachystegia, Cynometra, mixed forest and forest edge), we performed a permutation test for homogeneity of multivariate dispersions (PERMDISP), based on Bray-Curtis dissimilarities using the ‘betadisper’ function in the vegan R package. We then evaluated beta diversity amongst the four forest types. We quantified multi-site beta diversity using the abundance-based extension of the ‘betapart’ framework (Baselga et al. 2017), implemented via the ‘beta.multi.abund’ function in the R package betapart (v. 1.6). This method partitions Bray-Curtis dissimilarity into three components: (i) total beta diversity (β_total); (ii) balanced variation in abundance (β_balanced) and (iii) abundance gradients (β_gradient).

We also assessed spatial heterogeneity of community composition within forest types by testing homogeneity of multivariate dispersion. Bray-Curtis dissimilarities, calculated from species abundance data, were used to compute distances of individual plots to their respective group centroids using the ‘betadisper’ function in the R package vegan (Oksanen et al. 2020). These distances reflect within-group variation in community composition. Differences in dispersion amongst forest types were tested using ANOVA, followed by Tukey’s Honest Significant Difference (HSD) post-hoc tests.

Butterfly composition ‒ NMDS

To visualise differences in butterfly species composition amongst forest types, we conducted a non-metric multidimensional scaling (NMDS) analysis, based on Bray-Curtis dissimilarities (Kruskal 1964). Prior to analysis, butterfly species abundance data were square-root transformed to reduce the influence of highly abundant species. NMDS was performed using the ‘metaMDS’ function from the vegan package (Oksanen et al. 2020) in R (R Core Team 2020). This function conducts automatic data standardisation, multiple random starts and iteration to ensure a stable and optimal ordination solution. Groupings by forest type were visualised using convex hulls.

Butterfly traits: wingspan sizes

Using Pearson correlation, we correlated butterfly wingspan sizes across forest types, by first correlating for total abundances in all forest types and then second within each vegetation type. Following confirmation of normality, pairwise t-tests were conducted to compare average wingspan sizes across different forest types. To account for multiple comparisons and control the false discovery rate, p-values were adjusted using the Benjamin-Hochberg test.

Butterfly species diversity was primarily dominated by the Nymphalidae family, which had the highest number of species, followed by Pieridae, Papilionidae, Lycaenidae and Hesperiidae (Appendix 1). Analysis on the most abundant butterfly species revealed Phalanta phalantha, Appias epaphia, Catopsilia florella, Hypolimnas misippus and Coeliades forestan, as the most frequent across all forest types (Suppl. material 1: fig. S1). A strong positive correlation was observed between species richness and abundance across the forest types (R² = 0.89). The distribution of butterfly larval feeding habits showed oligophagous and polyphagous species being dominant in all forest types (Suppl. material 1: fig. S2).

Rarefaction curves and species diversity estimation

Rarefaction curves revealed the mixed forest exhibited the highest species richness, followed by Brachystegia forest, forest edge and Cynometra forests (Fig. 2). Rarefaction curves for all forest types approached asymptotes, indicating sufficient sampling and species richness capture, except in the Brachystegia forest, where the non-asymptotic curve suggests that further sampling may uncover additional species.

Butterfly species richness and abundances across forest types showed that the mixed forest had the highest cumulative species richness, followed by Brachystegia and forest edge, while Cynometra had the lowest value (Table 1). Average species richness and abundance per plot varied across forest types, with significantly higher abundance values observed at the forest edge and in mixed forest habitats (ANOVA, P < 0.05; Fig. 3). According to ANOVA, differences amongst species richness per transect were at the significance threshold amongst forest types (P = 0.05), while species abundances per transect were significantly different (P = 0.001; Fig. 3). Pairwise comparisons of butterfly abundance revealed significant differences between Cynometra forest and forest edge (P = 0.001), Brachystegia forest and forest edge (P = 0.013), as well as between Brachystegia forest and forest edge (P = 0.001; Fig. 3).

Figure 2. 

Rarefaction curves showing species richness as a function of number of individuals across the sampled forest types of Brachystegia, Cynometra, Mixed forest and Forest edge in the Arabuko Sokoke forest, Kenya. Each solid line represents actual sampled species (interpolated), and the dashed-line represents extrapolated individuals (extrapolated). Shaded areas represent 95% confidence interval.

Figure 3. 

Boxplots showing butterfly species richness and abundances comparison across the four forest types of Brachystegia, Cynometra, Forest edge and Mixed forest in the Arabuko Sokoke Forest, Kenya. Forest type with different letter denotes statistical significant differences (P < 0.01).

The diversity indices indicated relatively similar levels of butterfly diversity across vegetation types. Shannon index values ranged from 2.91 ± 0.40 at the forest edge to 2.82 ± 0.42 in the mixed forest, while the Simpson index values ranged from 0.93 ± 0.04 to 0.92 ± 0.03.

Beta diversity partitioning

Multivariate dispersions revealed significant difference in multivariate dispersion across the four forest types (F = 3.893, P = 0.007). Therefore, variation in butterfly community composition may in part be influenced by differences in within-forest type heterogeneity. Multi-forest type beta diversity analysis (β) revealed consistently high total dissimilarity across forest types, with β_total values ranging from 0.885 to 0.910 (Table 2). The largest contribution to beta diversity came from balanced variation in species abundance (β_balanced: 0.803‒0.837), indicating that community compositional differences were primarily driven by species turnover in abundance rather than nestedness (β_gradient: 0.067‒0.096) (Table 2). Brachystegia forest exhibited the highest total beta diversity (β_total = 0.910), suggesting substantial heterogeneity in community composition across plots. Forest edge had the lowest β_total (0.885), while Cynometra and Mixed forests showed intermediate values (Table 2).

Spatial dispersion of community composition

Spatial heterogeneity within forest types, assessed via multivariate dispersion, varied significantly (Table 2). Brachystegia forest exhibited the highest dispersion (mean distance = 0.45), indicating greater spatial variability in species composition. These results reflect substantial spatial heterogeneity in species abundances in Brachystegia. In contrast, forest edge had the lowest dispersion (median ~ 0.33), indicating more homogeneous community structure. Tukey HSD tests confirmed dispersion in Brachystegia was significantly greater than at forest edges (P = 0.006), while other pairwise comparisons were not statistically significant (Table 2). These results reflect that forest edge communities are more similar to each other, while Brachystegia forest is more ecologically diverse. Overall, beta diversity results indicate that abundance-based species turnover drives community differentiation across forest types, with Brachystegia forests supporting more spatially heterogeneous communities, while forest edges harbour more homogenised assemblages.

Co-correspondence analysis (CoCA) revealed a strong correlation between the species composition of plants and butterflies across the forest types. The correlation coefficients for Axis 1 and Axis 2 between the butterfly and plant communities were 0.991 and 0.994, respectively. The eigenvalues for the first and second axes indicated the contribution of each axis to the total inertia, with values of 0.022 and 0.012, representing a variance of 57.3% and 32.6%, respectively. This resulted in a total explained variance of 89.9% (Fig. 4), highlighting a robust and highly significant correlation between the community matrices of plants and butterflies.

Figure 4. 

Symmetric co-correspondence analysis (CoCA) ordination bi-plots showing correlations between (a) butterfly and plant species and (b) plant and butterfly species within the four forest types of Brachystegia, Cynometra, Mixed forest and Forest edge in Arabuko Sokoke Forest, Kenya. The Axis-1 eigen value of 0.022 explains a variance of 57.3% and Axis-2 eigen value of 0.012 explains a variance of 32.6%. Total explained variance by Axis 1 and 2 is 89.9%.

The NMDS analysis of butterfly species composition across forest types revealed considerable overlap, with no clear separation observed amongst the different forest types, with Cynometra forest covering a wider NMDS space that overlaps Brachystegia, forest edge and mixed forest (Fig. 5, Table 2). Butterfly assemblages across different forest types showed substantial overlap, with many species occurring in more than one forest type within the ASF. A pairwise permutational multivariate analysis of variance (PERMANOVA) revealed statistically significant compositional differences in butterfly assemblages amongst the forest types (R² = 0.07; P = 0.006). Additionally, SIMPER results show species composition differences amongst forest types that contribute up to 70% of the observed dissimilarities (Appendix 2).

Figure 5. 

Non-metric multidimensional scaling (NMDS) for butterfly species composition within the four forest types of Arabuko Sokoke Forest, Kenya. Different colours represent different forest types as follows: Brachystegia (red), Cynometra (blue), Mixed forest (green), Forest edge (yellow).

Average wingspan sizes were significantly larger at the forest edge compared to the Cynometra forest (P < 0.01; Fig. 6). However, no significant correlation was found between wingspan size and species abundance across forest types (Fig. 7).

Figure 6. 

Butterfly average wingspan sizes across the four forest types of Brachystegia, Cynometra, Mixed forest and Forest edge in Arabuko Sokoke Forest. Forest type with different denotes significant difference (P < 0.01).

Figure 7. 

Butterfly wingspan sizes correlation across the four forest types of Arabuko Sokoke Forest, Kenya. Showing wingspan correlation for (a) The total correlation in the four forest types (b) the correlation for each forest type of Brachystegia, Cynometra, Mixed forest and Forest edge.

The majority of larvae from the species encountered were classified as oligophagous, followed by polyphagous species and a smaller number classified as monophagous (Suppl. material 1: fig. S2). Overall, Oligophagous species constituted the largest share of individuals across all forest types, accounting for 63.4% of butterflies at the forest edge and reaching up to 67.5% in the mixed forest (Suppl. material 1: fig. S2).

Butterfly assemblages across different forest types showed substantial overlap, with many species occurring in more than one forest type within the Arabuko Sokoke Forest. This pattern of overlap suggests a degree of functional connectivity between habitats, where butterflies may move freely across forest types for resource availability, to access nectar sources, host plants or suitable microclimates for thermoregulation and breeding (Aguirre-Gutiérrez et al. 2017). Such overlap is ecologically significant, as it implies that, while habitat types may differ in structure and floristic composition, they are not functioning as isolated ecological units for butterflies. Rather, these habitats may form a complementary mosaic that supports butterfly movement and persistence. This could be particularly important in fragmented landscapes, where connectivity between habitat patches is critical for maintaining viable populations.

As our study was limited to a single dry season, we acknowledge that seasonal variation could play a significant role in shaping butterfly community structure in ways that our data do not currently capture. A study by Habel et al. (2018) in ASF found that butterfly communities showed reduced community overlap amongst habitats during the rainy season. This discrepancy may reflect seasonal shifts in resource availability and changed microclimatic conditions. For instance, increased vegetation growth during the wet season may alter habitat structure, increase flowering plants and fruits for both nectar and fruit feeders and accessibility of host plants, leading to more specialised or stratified butterfly distributions (Castor and Espinosa 2015). In addition, species with narrower ecological requirements might emerge or become more dominant during the rainy season, thereby reducing the community-level overlap. Long-term or seasonally repeated surveys would be essential to determine how stable these patterns are over time and to what extent butterfly habitat associations fluctuate in response to environmental seasonality. Incorporating a temporal dimension into future studies could also help disentangle the relative importance of spatial versus seasonal drivers of diversity and overlap in tropical butterfly communities.

Our findings revealed a strong positive correlation between plant community composition and both butterfly diversity and assemblage structure across distinct forest types. This underscores the critical role of floristic diversity and vegetation architecture in shaping insect communities and confirms a fundamental ecological relationship between butterflies and the plants that comprise their habitats (Larsen 1991). This finding points to the foundational role that vegetation plays in structuring butterfly communities. Given that butterflies are closely tied to their host plants for larval development and to nectar-bearing and fruiting plants for adult foraging, this floristic diversity strongly predicts butterfly richness and composition. Importantly, what is the strength and clarity of this correlation within a relatively small spatial scale and across naturally occurring forest types within ASF is noteworthy. Such robust cross-taxa associations are rarely reported in ecological studies conducted in the Tropics, where insect diversity often responds to a complex interplay of biotic and abiotic factors (Checa et al. 2014). In many systems, insect-plant relationships are masked by noise from microclimatic variability, dispersal dynamics or seasonal shifts (Checa et al. 2014). Our results, however, suggest that in Arabuko Sokoke, vegetation composition acts as a strong and consistent environmental filter, shaping butterfly assemblages in a predictable way. This reinforces the importance of plant diversity not only as a foundation for ecosystem structure, but also as a practical proxy for broader biodiversity patterns with significant implications for conservation planning and biodiversity monitoring.

Butterfly species richness and abundance varied notably across forest types. The mixed forest exhibited the highest cumulative species richness, followed by the forest edge and Brachystegia forest, while Cynometra forest supported the lowest species richness and abundance. Butterflies, as herbivorous insects, often exhibit narrow larval host plant specificity and selective adult nectar preferences, making them highly responsive to changes in plant community composition. This strong trophic linkage means that variation in plant diversity and abundance directly affects butterfly community structure. As previously observed by Larsen (1991), different butterfly species display distinct preferences for larval host and less for nectar plants, particularly during breeding and foraging periods. In ASF, the high plant species richness likely supports a broad range of butterfly guilds, contributing to the overall high butterfly diversity observed in the region. Moreover, our findings are consistent with broader ecological theory, which hypothesise that plant-insect interactions are mediated not only through direct resource relationships, but also via indirect mechanisms, such as vegetation structure and associated microclimatic conditions (Steffan-Dewenter and Tscharntke 1997; Aguirre-Gutiérrez et al. 2017; Oliveira et al. 2021). Thus, structurally diverse vegetation can buffer environmental extremes and provide varied microhabitats, enhancing survivability and niche partitioning amongst butterfly species.

The significance of these findings is particularly notable in the East African context, where community-level studies examining cross-trophic linkages remain limited. The novelty of this study lies not only in the strength of the observed plant-butterfly correlation, but also in the implication that butterfly assemblages could serve as sensitive bioindicators of plant community composition and forest integrity. Given their rapid response to habitat changes, butterflies offer a valuable, cost-effective means for monitoring ecological health, particularly in data-deficient tropical regions, such as ASF. In light of increasing anthropogenic pressures on East African forest ecosystems, including habitat fragmentation, selective logging and land-use conversion, our results emphasise the need to conserve floristically rich and structurally complex vegetation to maintain both plant and insect biodiversity. Integrating butterfly monitoring into conservation planning could enhance early detection of ecosystem degradation and provide information for adaptive management strategies.

This study highlights differences in butterfly community structure across forest types, emphasising the distinct ecological contributions of each habitat to local and landscape-scale diversity. Butterfly abundance varied significantly amongst the different forest types, while species richness did not show significant difference. Beta diversity analyses revealed consistently high community dissimilarity across all forest types, driven predominantly by balanced variation in species abundances rather than nestedness. Consequently, species turnover, rather than richness differences, is the primary mechanism structuring butterfly communities at the landscape scale in ASF. Notably, the highest species abundance was recorded at the forest edge; however, they also exhibited reduced species turnover and lower community dispersion. The mixed forest supported the highest cumulative species richness, while Brachystegia forests, despite lower local species richness and abundance, contributed most to beta diversity and spatial heterogeneity. The high species turnover observed suggests that Brachystegia forest maintains a diverse and spatially variable butterfly community, potentially driven by microhabitat differences or floristic specialisation (Legendre et al. 2005). Overall, abundance-based species turnover is a key driver of butterfly community differentiation across forest types, highlighting the conservation value of structurally complex and spatially heterogeneous habitats like Brachystegia forest.

High species abundance at the forest edge can be explained by the possible presence of microhabitat niches of biotic and abiotic resources that support the availability of food and flowering plants for nectar feeding, especially during the dry season when most forest vegetation is not flowering. This pattern is largely attributed to the ecological blending of habitat types at forest edges, where forest edges often support both forest-dependent species and species adapted to open or semi-open environments. In addition, forest edges are typically associated with increased levels of disturbance, including greater exposure to light, wind and human activity, which can significantly alter habitat structure and microclimatic conditions (Habel et al. 2025). In ASF, this trend is evident across the forest types surveyed. Forest-interior specialists, such as Euxanthe wakefieldi, Amauris niavius and Amauris ochlea, were predominantly observed within the mixed forest interior, where micro-climatic stability and specialised host plants are more likely to be present. In contrast, habitat generalists like Hypolimnas misippus, Phalanta phalantha and Catopsilia florella, which tolerate a wider range of ecological niches and host plants, were more frequently recorded at forest edges. As a result, overall species abundance may be elevated in edge habitats, not due to the presence of rare or conservation-priority taxa, but because of the additional contribution of widespread, disturbance-tolerant generalists (Blair and Launer 1997; Bobo et al. 2006).

Moreover, forest edges host a synergy of cultivation around the forest edge, where agricultural activities promote herbaceous species that cannot be found in the forest during the dry season (Habel et al. 2025). These herbaceous plants provide a variety of flowers for butterflies to feed on during the dry season (Bonebrake et al. 2010). This characteristic makes the forest edge a place mainly devoted to feeding, while other vegetation habitats may be preferred for reproduction and oviposition. Consequently, relying solely on species richness or overall diversity as a metric for conservation planning can be misleading. High diversity in edge or disturbed habitats may mask the loss or decline of forest specialists, whose presence is a more accurate indicator of ecological integrity and long-term ecosystem health. Thus, assessing conservation value requires a species-specific approach that prioritises the occurrence, abundance and habitat fidelity of specialist taxa, particularly those confined to forest interiors.

In addition, the mixed forest within ASF appears to maintain relatively moist and stable microclimatic conditions, owing to its tall canopy cover, sub-canopy layers, variable understorey and favourable soil characteristics (Fanshawe 1995; Muchiri et al. 2001). These environmental conditions create a wide range of ecological niches which supports several shade-dependent species, including forest-dependent butterflies, likely contributing to its ability to support a higher diversity of butterfly species. Notably, Papilio dardanus, a species typically more abundant during the rainy season, was recorded exclusively in the mixed and Brachystegia forest types during this survey, i.e. during the dry season. This unusual seasonal presence suggests that the mixed forest offers sufficiently favourable microhabitats to buffer against seasonal fluctuations in temperature and humidity. A seasonality study by Habel et al. (2018) reported higher butterfly species richness in ASF during the rainy season, highlighting the influence of seasonal conditions, particularly rainfall on butterfly richness. Supporting the occurrence of P. dardanus, the plant survey revealed a high abundance of its known host plants, particularly Clausena anisata, Vepris amaniensis and Vepris trichocarpa, within the mixed forest (Fungomeli et al. 2020a, 2020b). The availability of these larval host plants, combined with shaded, humid microclimates, likely creates optimal conditions for the survival and reproduction of P. dardanus.

Furthermore, the mixed forest is a preferred habitat for elephants, likely due to its food resources and canopy shade (Habel et al. 2025). Elephant activity contributes indirectly to butterfly persistence: their dung is a critical source of salts and minerals, especially during the dry season and is frequently used by butterflies for nutrient supplementation (Larsen 1996; see also Suppl. material 1: fig. S3). These combined factors, host plant abundance, favourable microclimate and access to nutrient-rich resources, help explain the relatively stable and diverse butterfly assemblages observed in the mixed forest. Given its structural complexity, ecological stability and richness of butterfly species, the mixed forest can be regarded as a key biodiversity ‘hotspot’ within ASF. Its conservation is therefore of particular importance, especially for sustaining specialist species and buffering biodiversity against climatic and anthropogenic pressures.

Butterfly wingspan sizes across forest types were significantly different at the forest edge compared to Cynometra forest. This can be attributed to the open-canopy state of the forest edge compared to the closed and dense forest type of the Cynometra forest (Hill et al. 2001). Concordantly, butterflies with larger wing sizes can be more abundant in forest gaps which exhibit a wide flight area favouring large-size butterflies. This argument supports our results and explains the high abundance of larger wing-size butterfly at the forest edge (Hill et al. 2001).

Oligophagous species constituted the largest share of individuals across all forest types, accounting for 67.5% of butterflies at the forest edge and reaching up to 63.4% in the mixed forest. The higher relative abundance of oligophagous butterfly species at the forest edge likely reflects the favourable ecological conditions typical of ecotonal environments. Oligophagous species, whose larvae feed on a restricted range of host plants, usually within a single plant family or a few related genera, combine a degree of specialisation with moderate flexibility in host use. This feeding strategy allows them to exploit the structurally and floristically diverse vegetation that characterises forest edges. Such areas generally support a greater diversity and density of sun-loving host plants and nectar resources compared to the shaded interior of Brachystegia and Cynometra forests. Similar patterns have been reported in other tropical forest systems, where edge habitats harbour a higher abundance of moderately specialised butterflies due to increased resource heterogeneity and microclimatic variation (Bonebrake et al. 2010).

This study underscores the importance of preserving diverse forest types to conserve both local and regional butterfly diversity. While mixed forest enhances richness and abundance, Brachystegia forest contributes disproportionately to beta diversity, emphasising their value in maintaining broader ecological variability. Protecting habitat heterogeneity across the landscape is therefore essential to sustaining butterfly diversity and ecosystem function. This study also highlights the strong positive relationship between vegetation composition and butterfly diversity across different forest types within ASF. Our findings demonstrate that vegetation structure and microclimatic conditions play a pivotal role in shaping butterfly community composition.

Notably, the higher butterfly abundance observed at the forest edge reflects the predominance of ecologically flexible, generalist species rather than greater habitat quality, while the mixed forest functions as a reservoir of host plants for breeding. Nonetheless, while forest edges showed elevated species abundance, this was largely due to the presence of generalist species and did not necessarily reflect high conservation value of specialist butterflies. These results underscore the limitations of using species richness alone as a conservation metric and emphasise the importance of beta diversity and species-specific assessments in evaluating habitat quality. These findings also reinforce the potential use of butterflies as ecological indicators for forest integrity and plant community health.

Given increasing anthropogenic pressures and habitat fragmentation in coastal East Africa, conserving structurally complex and floristically rich forest habitats, like those in ASF, is essential. Such efforts are vital not only for maintaining butterfly diversity, but also for preserving broader ecosystem functionality and resilience. In addition, ASF is known for its butterfly farming activities, conducted by communities adjacent to the forest for conservation and educational purposes. Our results provide valuable insights into the relationship between butterfly diversity and forest types. This will help guide the responsible utilisation and conservation of ASF’s resources, while preserving its ecological integrity.