Effects of land- use change on avian taxonomic, functional and phylogenetic diversity in a tropical montane rainforest

Aim: Although land use change is a leading cause of biodiversity loss worldwide, there is scant information on the extent to which it has affected the structure and composition of bird communities in the Afrotropical region. This study aimed to quantify the effects of habitat transformation on taxonomic, functional and phylogenetic diversity in Afrotropical bird communities. Methods: Data on bird occurrence, abundance and functional traits were collected in 2017/18 using point counts. We also collected data on habitat and morphological traits for all bird species recorded. We quantified bird diversity using a range of metrics, including the inverse Simpson index, functional dispersion and the standardized effect size of mean nearest taxon distance. Results: In comparison with forest even low levels of land use change altered species composition and reduced diversity. Although overall functional diversity and were similar across types, we found a significant contraction of trophic and locomotory trait structures of in cultivated respectively. Soil moisture, elevation and lower vegetation height were major factors influencing taxonomic, functional and phylogenetic dimensions although their effects varied across these dimensions. use


| INTRODUC TI ON
Tropical ecosystems provide essential ecosystem services ranging from carbon sequestration to climate and water regulation, and they also harbour 91% of terrestrial avian species (Barlow et al., 2018;Diaz et al., 2019;Phillips et al., 2017). Tropical montane forest ecosystems in particular have distinct taxonomic and phylogenetic assemblages, reflecting their high environmental (especially climatic) diversity (Quintero & Jetz, 2018;Rahbek et al., 2019). Despite the multitude of services provided by tropical forests, they are facing increasing biodiversity loss due to land use change, mainly in the form of conversion of natural habitats to agricultural areas (Diaz et al., 2019;Turubanova et al., 2018). Global projections by Jetz et al. (2007), based on a low environmental degradation scenario, but incorporating the combined effects of climate change and land use change, showed that 900 land bird species (from a total of 8,750) will experience range contractions of at least 50% by the end of this century. The study indicated a particularly bleak future for tropical montane endemic species, due to accelerated anthropogenic habitat conversion (see also Ayebare et al., 2018;Plumptre et al., 2003Plumptre et al., , 2007. At the habitat level, land use change often entails the modification of the physical environment and vegetation structure, with impacts on many bird species. For instance, the removal of mature trees (large and tall) affects canopy foragers and cavity nesters (Bonaparte et al., 2020;Lindenmayer, 2017), while changes in microclimatic and edaphic conditions, such as soil water content, affect the abundance of soil invertebrates, which avian insectivores consume (Cifuentes-Croquevielle et al., 2020;Sutherland & Green, 2004).
The clearing of forest in the tropics is often concentrated at lower elevation habitats, leading to the loss of specific vegetation types and their associated bird assemblages (Franklin et al., 2019;Harris et al., 2014). Although birds have been the focus of many land use change studies worldwide, research coverage in the Afrotropics remains poor and is dominated by the evaluation of taxonomic diversity, which is not always sufficiently informative about ecosystem processes and responses to environmental disturbances (Cadotte et al., 2011;Mouillot et al., 2013;Petchey & Gaston, 2002;Walker et al., 1999).
To fill this gap, functional trait-based approaches are gaining prominence, wherein functional traits are defined as a quantifiable aspect of an organism that influences its interaction with the environment (Flynn et al., 2009). A growing number of studies have used these approaches to investigate the impacts of habitat loss and fragmentation on tropical forest bird communities (e.g. Bregman et al., 2016;Cannon et al., 2019;Chapman et al., 2018), including preliminary studies in the Afrotropics (e.g. Ulrich et al., 2016). The use of measurable morphological traits, such as beak dimensions, can help to characterize the bird trophic niche and thus enables inferences concerning ecosystem functions of bird communities such as seed dispersal and pest control (Bregman et al., 2016;Pigot et al., 2020;Tobias et al., 2020). In the tropics, functional diversity has been found to decline with increasing intensity of land use change (Flynn et al., 2009;Şekercioğlu, 2012). Yet, comparatively little is known about how different forms of anthropogenic disturbances impact different trophic processes (but see Bregman et al., 2016;Trisos et al., 2014), especially in the Afrotropics.
Phylogenetic differences reflecting evolutionary divergence among taxa can potentially provide further information about a range of hidden ecological or behavioural properties that may not be evident from the morphological traits analysed . Thus, phylogenetic diversity (PD) is often estimated in conjunction with functional diversity and is increasingly used to determine community responses to environmental disturbances (e.g. Aguirre-Gutiérrez et al., 2020;Helmus et al., 2010;Rolland et al., 2012). Comprehensive assessment of the impacts of land use change should therefore make use of different facets of biodiversity-including both morphological and phylogenetic data-to guide conservation strategies (Cannon et al., 2019;Chapman et al., 2018;Devictor et al., 2010;Le Bagousse-Pinguet et al., 2019).
Here, we assess how land use change (modification from primary forest to other land use types) has affected bird communities within Nyungwe forest (a key landscape in the Albertine Rift) and adjoining agricultural farms in south-west Rwanda, a region that is underrepresented in ecological and biogeographical research. In particular, we ask how bird species composition and taxonomic, functional and phylogenetic diversity vary with land use and related environmental properties. Following the intermediate disturbance hypothesis (Connell, 1978), we predict that (a) taxonomic, functional and phylogenetic diversity levels will be highest in moderately disturbed secondary forests and lowest in habitats heavily degraded by fire or agriculture. Since the structural complexity of vegetation is known to drive species diversity (MacArthur & MacArthur, 1961) and anthropogenic disturbances may catalyse habitat filtering processes resulting in reduced avian diversity Flynn et al., 2009), we also predict that (b) communities in highly degraded land uses will comprise sets of closely related species and (c) increased vegetation structural complexity will drive increases in each diversity component.

K E Y W O R D S
birds, functional diversity, land use change, Nyungwe forest, phylogenetic diversity, Rwanda, taxonomic diversity, tropical forest 2 | ME THODS

| Study area
The study area of Nyungwe landscape in south-west Rwanda comprises Nyungwe National Park (Nyungwe NP) and surrounding agricultural areas (Figure 1). Nyungwe NP is a tropical montane rainforest covering an elevational range of 1,600-2,950 m, with an area of 1,019 km 2 , mean annual rainfall of 1,500-2,500 mm and average minimum and maximum temperature of 10.9 and 19.6°C, respectively (Seimon, 2012;Sun et al., 1996). Nyungwe NP was gazetted as a forest reserve in 1933, gaining National Park status in 2004. The forest is considered one of the six key landscapes of the Albertine Rift in terms of the number of plant and animal species that are endemic to this region, and those that are considered globally threatened based on the IUCN Red List (Plumptre et al., 2007).
Changes to the Nyungwe landscape within the past 50 years have involved clearance for agriculture, settlements and roads, gold and F I G U R E 1 Location of the Nyungwe landscape within Central and Eastern Africa (a). Nyungwe forest (the dark shaded area towards the left of the image) lies in south-west Rwanda and is contiguous to Kibira National Park, Burundi (b). The distribution of points sampled in six different land use types within the Nyungwe landscape, Rwanda (c). PR: primary forest, SC: secondary forest, FD: fire-disturbed area, RS: restored areas, PT: plantation of non-native tree species and CT: cultivated areas. The points were separated by 200 m. Each point was sampled in both wet and dry seasons. Ten plots sampled within primary forest at Mt. Bigugu within Nyungwe NP (d). A high number of points in restored areas are close to the main road because one of the criteria considered by the restoration project was visibility from the road for tourism and aesthetic purposes (Masozera & Mulindahabi, 2007). Apalis personata, an Albertine Rift endemic bird that is commonly encountered in the Nyungwe forest (e). Base map source: WCS-Rwanda and Google Earth [Colour figure can be viewed at wileyonlinelibrary. com] sand mining, and the harvesting of trees and non-timber products, such as medicinal herbs and honey (Masozera & Alavalapati, 2004).
To delineate the forest from community land holdings, a buffer zone of non-indigenous trees was established in the 1980s, creating a pine and Eucalyptus-dominated forest clearly distinct from the native forest. Fires largely resulting from honey collection and hunting practices have contributed to the degradation of at least 12% of the forest, in some areas impeding tree growth for decades due to the colonization of opportunistic ferns (Masozera & Mulindahabi, 2007).
Assisted-natural regeneration activities were initiated in 2003 by the Nyungwe NP management and key conservation stakeholders to restore extensively fire-degraded areas.

| Study design
Based on the land use history, largely recorded through the park's ranger-based monitoring programme since the late 1990s, we identi- four-day pilot study to test bird and plant sampling methods and to train three field assistants. Sampling was conducted in two phases: 14/11/17-09/02/18 and 06/06/18-25/8/18, corresponding to a short-wet and a long-dry season, respectively. We sampled 10 days per land use type, bringing the total to 60 days in each season.

| Avifaunal sampling
A point-count transect method was used to record the occurrence and abundance of birds in the Nyungwe Landscape. We selected random starting points in a predefined site and set up 10 circular stations (plots) of 100 m radius. The distance between plots was at least 200 m to minimize the risk of double counts and to maintain statistical independence (Ralph et al., 1995). 100 points per land use type were established, collectively surpassing the 300 points recommended when sampling rainforests (Ralph et al., 1995). At sites where trails were present, they were followed for practical reasons due to the rough and steep relief of the sites. Establishing points along reconnaissance trails gives maximum attention to bird observation, while inducing little disturbance to habitats (Bibby et al., 2000).
Where possible plots were positioned at right angles to the path and 30 m into the habitat (Gregory et al., 2004). Sampling of the same set of points was replicated in the second season, amounting altogether to 1,200 point counts. At each point, 1 min was allowed to elapse, in a bid to let birds settle before taking records of birds seen or heard. Their distance from the centre of the station was measured using a laser range finder. The count period was of 10 min duration (Buskirk & McDonald, 1995). All bird recordings were conducted by one observer with 30 years' experience of bird monitoring in the Nyungwe landscape. Sampling usually started at 5:45 and finished at 10:30 a.m. and covered 10 point counts in one land use type.

| Functional traits
For all species recorded on surveys, we collected eight biometric measurements, including four bill dimensions (length from tip to skull along culmen, length from tip to nares, width at anterior nares and depth at anterior nares), tarsus length, tail length, wing length and Kipp's distance (the length from the tip of the longest primary and the first secondary, on a folded wing). The size and shape of the bill are associated with the trophic niche in birds, while wing, tail and tarsus are indicative of locomotory, flight and foraging strategies (Grant & Grant, 2006;Hutchinson, 1957;Pigot et al., 2020;Schoener, 1965). Kipp's distance provides additional information on the wing morphology relevant to flight and dispersal abilities (Baldwin et al., 2010;Sheard et al., 2020). See Appendix S1 for further details on the link between avian traits and vegetation attributes.
Morphological traits were measured from specimens using a food categories, which we combine to form the following five dietary categories: invertivores (consuming terrestrial and aquatic invertebrates), herbivores (consuming plant matter and seeds), frugivores/ nectarivores (consuming fruits and nectar), carnivores/scavengers (consuming vertebrates, fish and carrion) and omnivores (none of the four categories exceeds 50% of the bird's total diet).

| Habitat assessment
For each point count location, a smaller plot of 20 m radius was demarcated to record vegetation structure and composition, soil and weather in order to capture fine-scale habitat conditions. Elevation was recorded using a GPS. The ground cover (the extent of the ground covered with vegetation) was visually estimated and recorded in percentages. The depth of the litter was measured with a thin metallic ruler, and samples were taken and averaged from 4 different random points. The height of undergrowth <3 m (lower vegetation height) was measured with a 3-m folding rule. Tree species (for individuals with diameter at breast height (DBH, 1.3 m) >5 cm) were identified and counted, and their DBH was recorded in classes of 5-14, 15-50, 51-100, 101-200 and >200 cm. Tree height was measured with a laser range finder. Canopy cover was estimated from the average of four readings, all taken with a spherical canopy densiometer from the four cardinal directions (Cook et al., 1995;Jennings et al., 1999;Strickler, 1959). Temperature and soil moisture were recorded using a portable data logger and a soil moisture probe, respectively.

| Statistical analysis
All analyses were conducted at the site level, where each site comprised 10 point counts. To explore differences in species composition across land use types, a detrended correspondence analysis (DCA)-a multivariate technique that explores the model responses of species along predominant ecological gradients (Hill, 1979)was conducted ( Figure 2). The strength of DCA lies in its ability to correct distortions in ordination axes that are generated by other ordination methods, and it also provides an option to downweight the influence of rare species (Hill, 1979;Hill & Gauche, 1980). To test whether sites within each land use category exhibited greater species compositional similarity than expected by chance, we performed an analysis of similarity (ANOSIM) in the CAP program (Seaby et al., 2014).
To determine how land use change affected taxonomic, functional and phylogenetic diversity of bird communities, we computed the inverse Simpson diversity index, functional dispersion index (Laliberte & Legendre, 2010) and the standardized effect size of mean nearest taxon distance (ses.MNTD) using the "Vegan," "FD" and "Picante" R packages, respectively (Kembel et al., 2010;Laliberté et al., 2014;Oksanen et al., 2010). These three metrics were chosen on the basis of a low correlation between them (Pearson's r < .35) and because they allowed the incorporation of abundances. FDis and ses.MNTD were weakly correlated with species richness (Pearson's r = .25, and .27, respectively), while the inverse Simpson diversity index was strongly positively correlated with species richness (Pearson's r = .89). Additional metrics were used for comparison (below).
The inverse Simpson index is a derivative of the Simpson's index that measures the probability that two individuals drawn randomly from a large community belong to the same species. It is weighted by the abundances of the most common species, and it is thus less sensitive to rare species, which frequently occur in rainforest systems.
A high value of the inverse Simpson index denotes higher diversity.
The inverse Simpson index is more robust, intuitive and mathematically sound in comparison with many other species diversity indices (Chao et al., 2014;Magurran, 2004).
Functional dispersion measures the overall spread of species trait values in a multi-dimensional space (Laliberté & Legendre, 2010).
The functional dispersion index (FDis) is less sensitive to species F I G U R E 2 Detrended correspondence analysis (DCA) of bird community composition showing the distribution of sites (samples) (N = 120) belonging to the six land use types within the Nyungwe Landscape, Rwanda, along the first and second ordination axes. The categories were primary forest (PR), secondary forest (SC), fire-disturbed areas (FD), restored areas (RS), plantation of non-native tree species (PT) and cultivated areas (CT). The first sampling was conducted over the wet season, and a replication was done over the dry season. Circles and diamonds denote the first (wet) and second (dry) seasons, respectively. The plot was generated using the CAP 5 program richness than many other functional diversity metrics and allows the comparison of samples with less than three observations, which can occur within avian point counts (Laliberté & Legendre, 2010).
Higher levels of FDis correspond to assemblages with greater functional dissimilarity. To circumvent the problem of considerable intraspecific trait variations, and correlation among avian morphological traits influenced largely by body size, a two-step principal components analysis (PCA) was performed using the eight morphological traits (Trisos et al., 2014). Following Trisos et al. (2014) andBregman et al. (2016), the first PCA was conducted on bill variables, and its second component was retained as an index of trophic processes.
The second PCA was performed on tarsus, wing and tail variables, and its second component was retained as an index of locomotive abilities. To obtain an index of body size, a third PCA was performed on the first axes obtained from the previous PCAs. The hand-wing index, an additional index associated with flight and dispersal abilities, was computed from the wing measurements as described in Sheard et al. (2020). These indices, the dietary classification, and foraging strata were then used to compute the overall functional diversity metrics in the R package "FD" (Laliberté et al., 2014). Due to the presence of categorical data, we used Gower's distance, and we standardized avian traits by the range. We also assessed the functional diversity indices indicative of the trophic, dispersal (hand-wing index), locomotory and size niche axes separately.
The phylogenetic trees were pruned to match our study species.
We then built a consensus tree and converted it into a distance matrix from which phylogenetic diversity measures were calculated using the "ape" and "picante" R packages (Kembel et al., 2010;Paradis et al., 2004). We calculated the standardized effect size of mean nearest taxon distance (ses.MNTD), which quantifies the average phylogenetic distance separating individuals from their closest relatives, corrected for species richness using a null model that randomizes the tip labels of the phylogeny (999 iterations) (Webb et al., 2002). Low ses.MNTD values indicate high phylogenetic clustering of closely related individuals.
For comparison purposes, we computed additional metrics, including (a) the functional richness metric (FRic), which measures the volume of functional trait space occupied by a species assemblage (Villéger et al., 2008) Faith's (1992) phylogenetic diversity metric (PD), which is the sum of branch lengths of a phylogenetic tree encompassing a given sample; and (c) the standardized effect size of mean pairwise distance (ses.MPD), which is the average pairwise phylogenetic distance among individuals in an assemblage (Webb et al., 2002), corrected for species richness using the same null as for ses.MNTD.
Statistical significance of differences in diversity between primary forest and other land use types was tested with analysis of variance (ANOVA) where the metrics fulfilled the test assumptions, otherwise with the Kruskal-Wallis test. The tests were followed by a Tukey's post hoc test and Dunn's tests, respectively. Due to the repeated measure nature of the study (temporal replication) and the randomness of site placements, three linear mixed-effects models were first conducted to determine the influence of habitat attributes on inverse Simpson diversity index, FDis and ses.MNTD. Site identity was used as the random variable. A likelihood-ratio test showed no significant contribution of the random variable (Kuznetsova et al., 2017). Additionally, a test for spatial autocorrelation of the model residuals, using Moran's I coefficient, was non-significant for the metrics considered; hence, a standard multiple linear regression analysis was performed instead. Canopy height, litter and temperature were not included in the analysis due to their high correlation with other habitat attributes (Pearson's r > .7) (see Appendix S2 for further details on their exclusion). The regression analyses contained the following habitat variables: elevation, canopy cover, number of trees, DBH, lower vegetation height, ground cover and soil moisture. Model selection was carried out based on BIC, due to the high number of predictors, using the "MuMIn" R package (Barton, 2019).
Models within ΔBIC < 2 of the model with the lowest BIC value were then averaged. All analyses, except where otherwise mentioned, were performed in the R 3.6.1 environment (R Core Team, 2019).

| RE SULTS
We individuals (20 and 406, respectively), followed by the secondary forest (18 and 304, respectively). Fire-disturbed areas had more AR endemic species (18), but fewer individuals (296) than restored areas (14 and 405, respectively). The lowest counts were in non-native forest (11 and 104, respectively) and cultivated areas (4 and 13, respectively). One-third of AR endemics were at least twice as abundant in the primary forest as in other land use types. However, a few species were found only in other habitat types, notably Caprimulgus ruwenzorii, which was recorded in the secondary forest and in firedisturbed forest, and Geokichla tanganjicae, which was only recorded in the fire-disturbed forest. Two individuals of the rare species Apalis argentea were only recorded in the primary forest. Phylloscopus laetus was mostly found in the restored areas (N = 107), where its abundance was at least double compared to the other land uses.
We encountered most endemic species expected to occur in the Albertine Rift forest belt, with the exception of Glaucidium albertinum, Indicator pumilio, Cinnyris rockefelleri and Cryptospiza shelleyi.  (Table 1). The primary forest harboured species that were not observed in other land uses, particularly in cultivated areas and restored areas. These two land uses shared only 49% and 28% of their species with the primary forest. The closest land use to primary forest, in terms of similarity of species composition, was the secondary forest. Nonetheless, unshared species account for 34% (N = 122) of all the species recorded in the two habitats (Table 1).

| Species diversity
Species diversity, as measured by the inverse Simpson diversity index, decreased with increasing anthropogenic modification (Figure 3).

| Functional diversity
The pooled functional diversity, measured by mean functional dispersion (FDis), remained fairly stable across the land use types for the locomotion axis were only discerned in cultivated areas, TA B L E 1 Pairwise comparisons of similarity of bird communities among study land use types in the Nyungwe landscape, Rwanda. Sample size is 20 per land use. A sample comprised 10 point counts conducted within the same land use type. Raw shared species numbers are below the diagonal, and ANOSIM sample R above. Theoretically, R values range from +1 to −1. Values close to +1 indicate high within group similarity and therefore a highly dissimilar pair. All the R values for each pairwise comparison were significant (p = .001), reflecting compositional differences among land use types   where the changes were negative relative to levels in primary forest Dunn's test, p = .02 and p = .013, respectively).

| Phylogenetic diversity
Values of ses.MNTD (Figure 3) and ses.MPD were comparable across land uses. Faith's (1992) Phylogenetic diversity metric (PD), which does not incorporate abundances, showed declines in firedisturbed areas, restored areas and non-native forests, and steeper losses in cultivated areas compared to primary forest (Appendix S8).

| Habitat attributes
The regression model selection analyses revealed elevation as a primary negative influence on taxonomic diversity and FDis (Table 2a; Figure 4a). Soil moisture and the lower vegetation height had contrasting effects, in that they influenced taxonomic diversity positively and phylogenetic diversity (ses.MNTD) negatively (Table 2a-

| Altered patterns of species composition and diversity
Our analyses reveal that increasing levels of land use change alter species composition and reduce taxonomic diversity. In general, the patterns detected align with those reported by previous studies that have concluded that bird species diversity within tropical forested landscapes is reduced with intense human habitat transformations, particularly from primary forest to agricultural uses (Albanesi et al., 2014;Bregman et al., 2016). The bird diversity patterns noted here are also consistent with the predictions of the intermediate dis- turbance hypothesis (Connell, 1978), which is a relevant theoretical construct given that it describes diversity responses at the patch scale across landscapes.
Although primary and secondary forest had almost equal species richness, there were differences in species composition between the plots from the two land use types, highlighting the sensitivity of some forest species, even to low levels of disturbances, and the crucial roles that primary vegetation plays in maintaining distinct communities, particularly of forest specialist (often locally endemic) species (Barlow et al., 2007;Gibson et al., 2011; O'Dea & F I G U R E 4 Relationships between bird taxonomic diversity (a-d) and phylogenetic diversity (e-f) and habitat parameters of study samples (N = 10 per land use) in the Nyungwe landscape. Taxonomic diversity and phylogenetic diversity are quantified using inverse Simpson diversity index (inverse Simpson index) and standardized effect size of mean nearest taxon distance (ses.MNTD), respectively. Results were obtained from a linear model fitted on unscaled raw data. Model estimates obtained from scaled data are given in Table 2. The grey band represents ± 95% confidence interval. Habitat attributes with lower relative importance values (<0.5) and only found in models with low adj.R 2 (<0) are not displayed [Colour figure can be viewed at wileyonlinelibrary.com] (f) Whittaker, 2007). Although a higher species diversity is often reported in secondary forests, the gain usually originates from the influx of generalists and gap-tolerant species (Coetzee & Chown, 2016;Foord et al., 2018;Moura et al., 2013). For instance, in this study, Platysteira concreta and Ploceus insignis, known closed-forest species, were only recorded in primary areas, while open habitat and generalist species in the genera Estrilda, Crithagra and Ploceus were mostly recorded in secondary forests. The degree of disturbance and the time elapsed since release from disturbance may also determine the avian diversity supported by secondary forests. Areas restored following burning within the last 1-20 years supported only slightly more than half the taxonomic diversity found in the primary areas.
Similar findings were noted in restored Australian tropical rainforest, where the rapid recovery of certain vegetation attributes, such as the stem density and canopy closure, led to the attainment of half the number of forest-dependent species supported by old-growth rainforests within a decade, and the similarity in species composition increased with the age of the restored habitat (Catterall et al., 2012;Dunn, 2004).

| Differential response of functional diversity to land use change
Overall functional diversity (measured using the FDis index) failed to produce clear patterns of difference between land uses ( Figure 3); however, analysis at the level of individual trait axes revealed effects of varying magnitude (Appendix S7). Restored areas were the most affected land use type. The registered declines in the trophic and size axes in restored areas relative to primary forest can be explained by the absence of large-bodied birds such as those in the Accipitridae family, typical forest raptors such as Aquila africanus, as well as large invertivores (e.g. Apaloderma vittatum), and large frugivores (e.g.

Bycanistes subcylindricus).
Reforested areas tended to be dominated by small-sized insectivorous birds and exhibited poor representation of other guilds (Appendix S9); granivores were the least-represented guild, as exemplified by the absence of any of the eight forest and non-forest Ploceus species recorded within the study. The reduced niche breadth in restored areas can be attributed to the high degradation of the sites by anthropogenic fires before the restoration activities started (Masozera & Mulindahabi, 2007 Contrary to what is often reported (Bregman et al., 2016;Flynn et al., 2009;Frishkoff et al., 2014) The maintenance of functional diversity in secondary forests and fire-disturbed areas, despite the loss of taxonomic diversity, could be explained by the functional redundancy of tropical forest ecosystems (Cooke et al., 2019). In our study, half of the species (N = 85) were small-sized (<28.3 g), and about three-quarters of these were invertivores. Feng et al. (2020) also noted that functional redundancy characterized the avian assemblages of the Amazon and Andean forests. Our findings in this regard agree with patterns often observed in other studies in tropical regions, showing stable functional diversity of forest species in moderately disturbed habitats, but steep declines in highly disturbed habitats, such as agricultural land (Bregman et al., 2016;Flynn et al., 2009;Sayer et al., 2017;Şekercioğlu, 2012).

| Stable community phylogenetic structure across the land use types
Although the loss of species with land use change culminated in lower phylogenetic diversity (Faith's PD) compared to primary forest levels, the phylogenetic structure (ses.MNTD and ses.MPD) remained stable. The decreases in Faith's PD are thus attributed to the decline in species richness and abundance along the disturbance gradient. Bird communities across the land use types including primary forest were characterized by shorter average phylogenetic distances between species than expected, as shown by negative values of ses.MNTD and ses.MPD, which imply high co-occurrence of closely related lineages. An explanation for the general phylogenetic clustering in the study area may be environmental filtering due to historical disturbances. The Nyungwe landscape forms part of the Albertine Rift, which has served as a refugium for rainforest birds during the alternating extreme climatic cycles over the Pleistocene (Kahindo et al., 2017;Lyons et al., 2015;Prigogine, 1988). Historical anthropogenic disturbances of the Nyungwe landscape may also have left legacy effects on the vegetation and associated fauna, resulting in phylogenetic clustering.
We were expecting to find filtering of certain clades with increasing intensity of habitat change, and especially in agriculture areas, as reported by other studies focusing on an array of taxa (Egorov et al., 2014;Frishkoff et al., 2014;Prescott et al., 2016). For instance, Frishkoff et al. (2014)

| Habitat attribute effects on taxonomic, functional and phylogenetic diversity
Fine-scale habitat heterogeneity, particularly in terms of the vegetation structure, has been reported to drive species, functional and phylogenetic diversity patterns in tropical birds (Foord et al., 2018;Karp et al., 2012;Maseko et al., 2019;Prescott et al., 2016

| Conservation implications
This study has shown how land use change, particularly in the form

PEER R E V I E W
The peer review history for this article is available at https://publo ns.com/publo n/10.1111/ddi.13364.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support findings of the functional diversity component of this study are available in the Dryad Digital Repository: https:// doi.org/10.5061/dryad.41ns1 rndd Restrictions apply to the availability of the rest of the data due to co-ownership with a third party, so the data are not publicly available.