Exploring avian biodiversity dynamics: insights from a sub-Saharan African reservoir

Biodiversity in tropical landscapes is influenced by vegetation, disturbance, and seasonal change, but their combined effects are not well known. We studied bird communities at seven sites around the Shiroro Reservoir, Nigeria; a hydropower landscape with different levels of disturbance. Species richness, abundance, and functional richness were highest at reservoir-edge habitats (DYKE, CTMC) and lowest at heavily disturbed sites (COOJ, POWH). Species richness and Shannon diversity did not vary with season, but functional evenness was higher in the dry season, showing that trait distribution responds more to seasonal change than species numbers. Ordination analysis showed that location and season strongly shaped bird communities, while vegetation structure had weaker effects. Sites close to infrastructure consistently supported fewer species. Our results show that semi-natural habitats act as refuges for birds in disturbed landscapes. Protecting vegetation complexity, keeping dead trees, and limiting settlement expansion are important for conserving birds and the services they provide in African savanna reservoirs. These findings underscore the complex interplay between human activities and ecological dynamics, highlighting the urgent need for sustainable management strategies to mitigate further biodiversity loss and habitat degradation in this critical ecosystem.

Freshwater reservoirs are among the most widespread forms of human-modified landscapes, providing essential ecosystem services such as hydropower, irrigation, water supply and flood control (ltinbilek, 2002). However, they also profoundly alter local habitats, reshaping vegetation structure, hydrology, and disturbance regimes, with cascading effects on biodiversity (Nilsson et al., 2005; Sekercioglu, 2012).

Birds are not only integral parts of ecosystems but also act as indicators of environmental health and biodiversity; they are important surrogate measuring instrument of the health of planet earth (Fraixedas et al., 2020). Birds are sensitive indicators of ecological changes, responding not only to habitat loss and fragmentation but also to the structural complexity of vegetation and temporal fluctuations in resource availability (Bregman et al., 2014; Iniunam et al., 2025; Keith, 2020). The alterations of natural landscapes and hydrological patterns caused by dams can strongly influence bird populations, habitats, and migration routes (Wu et al., 2019). Understanding these impacts is crucial for developing strategies that balance the advantages of reservoir infrastructure with the conservation needs of avian species.

In tropical savanna systems, habitat heterogeneity arises from the interplay of topography, vegetation structure, and seasonal dynamics, producing spatially and temporally variable resources for avifauna. While species richness often reflects broad patterns of habitat occupancy, functional diversity, measuring the distribution of traits related to feeding, nesting, and movement, captures how ecological processes are maintained across landscapes (Mouillot et al., 2013). Assessing both taxonomic and functional dimensions of biodiversity is therefore critical for understanding community assembly and ecosystem resilience in human-modified systems.

Hydropower reservoirs, such as the Shiroro Reservoir in Niger State, Nigeria, represent complex socio-ecological mosaics where human infrastructure, residential settlements, and semi-natural habitats coexist. These landscapes provide a unique opportunity to explore how spatial variation, habitat features, and seasonal dynamics shape bird communities. Yet, few studies in African reservoir systems have integrated both taxonomic and functional perspectives across multiple sites and seasons, limiting our understanding of how infrastructure development interacts with natural processes to influence biodiversity.

Reservoirs can have both positive and negative effects on bird conservation. On the one hand, reservoirs created behind dams can create new aquatic habitats, attracting waterfowl, wading birds, and other water-dependent species such as Actophilornis africana, Actitis hypoleucos, Ardea alba, Bubulcus ibis, Dendrocygna viduata, Egretta garzetta, and Quelea (Liang et al., 2022; Erin et al., 2020; Edegbene, 2019; Ringim et al., 2015; Agbelusi and Afolayan, 2010). Moreover, reservoir shorelines and riparian zones can become crucial breeding, foraging, and nesting sites for various bird species. For instance, the Shiroro reservoir has generated new aquatic habitats, attracting waterfowl, wading birds, and other water-dependent species (Erin et al., 2020; Olayemi et al., 2014; Agbelusi and Afolayan, 2010). These reservoir shorelines and riparian zones serve as vital breeding, foraging, and nesting sites, significantly contributing to local avian biodiversity (Bennett et al., 2014).

Our study explored the intricate relationship between environmental dynamics and avian ecology around the Shiroro reservoir. Acknowledging the inherent value of bird life, we assessed alpha diversity across various locations and advocated for innovative strategies that prioritize ecological stability in the context of reservoir development. We tested whether bird diversity, functional structure, and species composition vary in response to seasonal dynamics, vegetation structure, and anthropogenic disturbance across contrasting habitats. Specifically, we predicted that 1) community metrics (Shannon–Wiener diversity, species richness, and abundance) would differ among sites; 2) functional diversity (Rao’s quadratic entropy, functional richness, and functional evenness) would be shaped by both seasonality and site characteristics; 3) species composition would be structured by season, location, and habitat features. These predictions are grounded in the expectation that seasonal variation alters resource availability, vegetation structure mediates niche diversity, and human activities modify habitat quality and connectivity. Understanding these responses is essential for conservation, as it reveals the environmental drivers that sustain avian biodiversity and ecosystem functioning, thereby guiding management strategies in landscapes facing increasing anthropogenic pressures.

Study area

The Shiroro reservoir is situated in Niger State, Nigeria, approximately 50 km northeast of Minna, the state capital (Figure 1). Geographically, it is located at latitude 9°53’N and longitude 6°21’E. The reservoir is nestled within the scenic landscapes of the Kaduna River, a tributary of the Niger River. The region experiences a tropical savanna climate characterized by distinct wet and dry seasons. The wet season typically spans from April to October and has high rainfall, while the dry season occurs from November to March and has relatively lower precipitation (Olayemi et al., 2014). The average annual temperature ranges from 25 °C to 30 °C. The annual rainfall in the area varies between 1100 mm and 1600 mm. The dam and its reservoir, covering an area of approximately 312–320 km 2 (Adegun et al., 2018; FAO, 2012), create a mosaic of aquatic habitats, including open water, submerged vegetation, and marshy areas. The major river in the study area is the Kaduna River, on which the Shiroro reservoir was built. The reservoir lies completely within the Guinea Savanna vegetation zone consisting of coarse grasses, shrubs and woodland made up of high closed canopy in its undisturbed state dominated with Khaya senegalensis, Daniella sp (Clayton, 1975). Other common tree species found in this region include Adansoni digitata, Afzelia Africana, Adenium obesum, Annonaz senegalensis, Bridelia ferruginea, Calotropis procera, Cochlospermum tinctorium, Combretum molle, Combretum nigericans, Detarium microcarpum, Diospyros mespiliformis, Gardenia aqualla, Isoberlina doka, Nauclea latifolia, Parkia biglobosa, Prosopis africana, Piliostigma recticulatum, Piliostigma thonningii, Sterculia setigera, Strychnos spinosa, Terminalia macroptera, Vitellaria paradoxa, Vitex doniana, and Ximenia Americana (Arbonnier, 2004; POWO, 2018; WFO, 2018). The area is characterized by its high lying topography with dotted hills at various locations and with Zumba Hills where the dam is located rising to 600 m above sea level (Adeogun et al., 2023; Ladokun et al., 2019; Kolo, 1992). The hills are mostly made up of the basement complex of the North-Western Nigeria, which are resistant to erosion.

In this study, the reservoir surveyed was stratified into three sampling areas; stratifications were based on density of human settlements around the reservoirs: Charles T. Main Camp (CTMC), Employee Camp (EMPC), Secondly, proximity to the water body: Water-inlet-Italian Camp (ICWT), Riverside-First-Gate (FGSR), Dyke (DYKE) and thirdly, proximity of the dam/barrier: Coordinators junction (COOJ), Power house (POWH. These seven transects were each, 2km long. COOJ is a busy road flanked by the switch yard, coordinator’s block and transmission station offices. CTMC is one of the reservoir operators’ residential areas; 6km to the west of the powerhouse. DYKE is a road with less activity, used only by securities and reservoir records custodians. It is located on the edge of the reservoir, not far away from the 388m elevated dam crest which is 7.5m in width. EMPC is 2km west of the powerhouse; this is also a staff residential area. FGSR is a 7.5km busy road linking the residential camps to the power station. ICWT is another residential settlement 2km west of the powerhouse; POWH is an active road, from the entry gate to the massive reinforced concrete structure of which most are within the rock (GCNL, 2006). The transects were coded with locations acronyms with start and end points marked with a Global Positioning System (GPS) recording coordinates on Garmin 64 ® GPS unit amidst the characterized high topography dotted hills at various locations and of which Zumba Hills is raised up to 600m above sea level (Adeogun et al., 2023; Ladokun et al., 2019; GCNL, 2006).

Bird sampling

The data for this study was collected between January 2018 and December 2019. The line transect bird census technique (Bibby et al., 2000) was used in this study. The exact coordinates and elevation for each of the seven transects can found in Appendix (S7). Each survey day started at 6:30 a.m. to 9:30 a.m. for the morning bird sampling sessions and concluded with the evening sessions at 3:30 p.m. to 6:30 p.m. Each day four transect surveys were conducted with two bird survey sessions holding per day. All transects were sampled once a month. Line transects were walked at a steady pace, stopping every few steps to hear and observe birds more. Bird viewing was aided by the use of a pair of Bushnell Power View 10 × 42 binoculars and species identification followed the field guide to the birds of western Africa (Borrow and Demey, 2014). Along each transect, birds that were seen and heard within a 100 m radius of the line of observation were recorded. Whenever a bird was spotted outside the radius, it was noted but was not used for analysis. A typical 2km transect was steadily walked in 90 minutes while 90 minutes was used to record predictor variables in three categories: (i) vegetation/environmental (ii) weather and (iii) anthropogenic data. These three levels recorded percentage plant cover, numbers of trees, shrub cover, presence of logs and snags, percentage bare ground, numbers of rocks, altitude, distance to water sources, distance to nearest human settlement of each transect. At anthropogenic level the research took records of agriculture, bush burning, grazing, logging and mining activity within and around each transect as bird habitat records.

Bird habitat assessment

Vegetation variables were estimated at every 200m intervals along each transect in 10 × 10m quadrats (Manu and Cresswell, 2007). The number of trees and the number of dead trees standing (snags) and those lying on the floor (logs) were also counted. The values from counts were used to calculate the mean tree and dead tree densities at each census station. Percentages of plant cover (e.g. grasses and litter covers were estimated by the use of 2×2m quadrats which were randomly placed within the quadrats in order to estimate the numbers of all tall and short herbaceous vegetation (shrubs), number of rocks, percentage bare ground and litter cover in the larger plot were recorded.

Cloud cover was estimated to the nearest 10% (Manu and Cresswell, 2007). Rainfall (mm), temperature (°C), humidity (%), and wind speed (Km/h), which constituted the weather data in this study, were sourced from the meteorological stations of each reservoir. Weather data of only survey days were included for analysis. Altitude (m), distances to water sources and human settlements were recorded from the GPS unit. These measurements were recorded with reference to waypoints marked on transects and the nearest bank of the reservoirs or human habitation or settlement.

Observations of logging, mining, agriculture, grazing and bush burning activities were made by visual assessment. Following the study of Ringim and Harry (2017), the intensities of the anthropogenic activities were categorized as low, medium or high based on the severity with regard to its percentage coverage within the 400 m section of the 2km transect on which observation is made.

Functional trait analyses

To assess functional diversity, we compiled data on 12 species-specific traits for all birds recorded across the Shiroro Reservoir. Traits were selected following previous studies to capture variation in body size, diet, foraging strategy, and ecological roles relevant to ecosystem functioning (Iniunam et al., 2025; Lee et al., 2021; Child et al., 2009). Body mass (g) was included as a continuous trait. Categorical traits encompassed dietary preferences (vertebrate prey, scavenger, herbivore, insectivore, frugivore, granivore, piscivore, aquatic diet), ecological functions (ecosystem engineer, pollinator), and habitat specialization (habitat specialist). Each trait was coded as 1 (presence) or 0 (absence) for each species (Danmallam et al., 2025; Iniunam et al., 2025). These traits were not mutually exclusive; consequently, a species could simultaneously be classified, for example, as an insectivore, frugivore and granivore. This representation therefore captured the multidimensional nature of species’ functional roles rather than assigning each species to a single feeding category.

Because the trait dataset contained both continuous and binary variables, functional dissimilarity among species was calculated using Gower distance, which is appropriate for datasets containing mixtures of quantitative and qualitative trait types (Gower, 1971; Laliberté and Legendre, 2010). Before calculating the distance matrix, body mass was log10-transformed to reduce the influence of its strongly right-skewed distribution and to prevent species with exceptionally large body masses from disproportionately influencing the functional distances. The binary traits were retained as symmetric binary variables, such that shared presence and shared absence contributed to similarity. No differential weighting was assigned to individual traits, so each trait contributed equally to the overall Gower dissimilarity. The resulting pairwise functional dissimilarities ranged from 0 to 0.615, with 0 indicating identical trait profiles and larger values indicating increasingly dissimilar functional profiles.

The Gower distance matrix was subsequently subjected to principal coordinates analysis (PCoA) to represent the multidimensional functional relationships among species in Euclidean space. PCoA provides a distance-based representation of functional trait relationships and permits functional diversity indices to be calculated from a reduced multidimensional functional space (Villéger et al., 2008; Laliberté and Legendre, 2010). The first four positive PCoA axes were retained as the functional trait space used for subsequent calculations. These axes represented the principal dimensions of variation in the combined trait dataset while reducing the dimensionality of the original trait space. Importantly, the four axes were selected a priori for subsequent functional-diversity calculations to provide a multidimensional representation of species’ functional relationships rather than using the original 12-dimensional trait matrix directly.

Functional diversity was then calculated for each sampling occasion from the species-by-sampling-unit community matrix and the four-dimensional functional trait space. Species abundances were retained in the community matrix, allowing the resulting functional indices to incorporate differences in the relative representation of species among sampling units. Three complementary aspects of functional diversity were quantified: functional richness (FRic), functional evenness (FEve), and Rao’s quadratic entropy (Rao’s Q). FRic represents the volume of multidimensional functional space occupied by the species occurring in a sampling unit, and therefore reflects the extent of functional trait space represented by the community (Villéger et al., 2008). FEve measures the regularity with which species abundances are distributed within the occupied functional space, with higher values indicating a more even distribution of abundance across the functional space (Villéger et al., 2008). Rao’s Q measures the expected functional dissimilarity between two individuals randomly selected from the community and therefore incorporates both differences in species’ functional traits and their relative abundances (Rao, 1982; Laliberté and Legendre, 2010).

FRic requires sufficient numbers of species relative to the dimensionality of the functional space because it is based on the volume of the convex hull enclosing the species in multidimensional trait space (Villéger et al., 2008). The final functional space therefore comprised four PCoA axes, and sampling occasions with insufficient species to define a four-dimensional convex hull produced undefined FRic values and were retained as missing observations rather than being assigned arbitrary values. This resulted in nine missing FRic observations. FEve also produced two undefined values, while Rao’s Q was calculated for all sampling occasions. These missing values arose from the mathematical requirements of the functional indices rather than from missing trait information. Functional indices were calculated at the same sampling-unit level as the underlying community matrix, after which they were linked to the corresponding location, transect, section, year, month and season information for subsequent modelling (Appendix S8). Functional diversity metrics were computed using the fundiversity package in R (Grenié and Gruson, 2023).

Statistical analyses

We used R 4.3.1 (R Core Team, 2023) to conduct all analyses. Individual and coverage-based rarefaction and extrapolation curves were computed based on bird abundance using the iNext 2.2.2.0 R package (Hsieh et al., 2016; Chao and Jost, 2012). Rarefaction analysis standardizes comparisons of species richness by accounting for differences in sampling effort and abundance amongst sites. Individual-based rarefaction curves were constructed from bird abundance data to estimate expected species richness as a function of number of individuals sampled, whereas coverage-based rarefaction curves standardized comparisons based on sample completeness rather than sample size. Sampling coverage was calculated to assess the proportion of the total bird community represented in the samples. Coverage values approaching 1 indicate high sampling completeness and suggest that most species present within a location were detected during surveys (Chao and Jost, 2012). Extrapolation was extended beyond the observed size to evaluate whether additional sampling effort would likely yield substantial increases in species richness. 95% confidence intervals were generated using bootstrap resampling to access uncertainty around richness around richness estimate.

We assessed spatial and seasonal variation in avian community metrics (Shannon diversity, species richness, and total abundance), by fitting a linear mixed-effect models using the nlme package. Location (COOJ, CTMC, DYKE, EMPC, FGSR, ICWT, POWH) and season (dry and wet) were included as fixed effects, while transect and section were incorporated as hierarchical random effects to account for the repeated sampling structure. Species richness and total abundance were log-transformed to improve normality and homoscedasticity of residuals.

The functional diversity indices were analyzed using mixed-effect models with the same hierarchical sampling structure used for taxonomic diversity. Location and season were specified as fixed effects, while transect and section were incorporated as random effects to account for the non-independence arising from repeated sampling of the same transects and sections through time. Thus, the functional-diversity analyses tested whether the functional structure of bird communities differed among locations and seasons while explicitly accounting for repeated observations within the seven transects and their constituent sections. Model assumptions were evaluated visually using residuals-versus-fitted and Q-Q plots. Type II ANOVA test was conducted using ‘car’ package (Fox and Weisberg, 2018). Where significant effects of location were detected, we computed estimated marginal means (EMMs) using the emmeans package and performed Tukey-adjusted pairwise comparisons to identify specific differences among sites (Lenth et al., 2019). All estimates are reported as means ± standard error (SE) on the original or transformed scale, with 95% confidence intervals (CI) where relevant.

Community composition was analyzed using Non-metric Multidimensional Scaling (NMDS) to visualize patterns in species assemblages (Laliberté et al., 2014). The analysis was performed in R using the vegan package (Oksanen et al., 2016). Before the NMDS, the species abundance matrix was cleaned to remove rare or missing entries. NMDS was conducted with Bray–Curtis dissimilarity, two dimensions (k = 2), and a maximum of 100 random starts (trymax = 100) to ensure convergence. The stress value was used to assess the goodness of fit of the ordination.

To assess the influence of environmental and anthropogenic variables on bird community composition, a permutational multivariate analysis of variance (PERMANOVA) was conducted using the adonis2 function in the vegan package (González-Megías et al., 2011). PERMANOVA partitions variation in a multivariate dissimilarity matrix among explanatory variables and obtains significance through permutation, making it suitable for ecological community data without require multivariate normality (Anderson, 2001). Bray-Curtis dissimilarities were calculated from species abundance matrix and significance was assessed using 999 permutations. Predictor variables included season, location, and environmental and anthropogenic characteristics comprising shrub cover, logs, snags, rocks, litter cover, plant cover, altitude, distance to water sources, distance to nearest human settlement, logging, grazing, and bush burning. To avoid dependence of individual predictor effects on the order in which variables were entered into the model, predictors were evaluated using marginal tests (by = “margin”). Thus, the contribution of each predictor was assessed after accounting for the other predictors included in the model. This approach provides estimates of the marginal contribution of individual variables within the multivariate model rather than sequential (order-dependent) sums of squares.

Prior to PERMANOVA, correlations among continuous environmental predictors were examined to identify redundant predictors. Strongly correlated variables were not retained simultaneously in the final model to reduce redundancy and improve the interpretability of individual environmental effects. Categorical variables (season, location, logging, grazing and bush burning) were treated as factors. Because significant PERMANOVA results can potentially arise from differences in within-group multivariate dispersion rather than differences in the location of group centroids alone, homogeneity of multivariate dispersion was additionally assessed using the betadisper procedure (Anderson, 2014). Multivariate dispersion was calculated from the same Bray-Curtis dissimilarity matrix, and differences in dispersion among locations and between seasons were assessed using permutation tests with 999 permutations. Results of the dispersion analyses were considered when interpreting the corresponding PERMANOVA effects, because heterogeneity in within-group dispersion can contribute to significant PERMANOVA results.

Sampling completeness

A total of 21,418 individual birds representing 179 species (Eighteen Orders and Fifty-nine Families) were recorded across the seven locations surveyed around the Shiroro Reservoir. Individual-based rarefaction and extrapolation curves revealed clear differences in species richness among locations (Figure 2A). DYKE (140 species) exhibited the highest estimated species richness, while ICWT (87 species) supported the lowest across observed and extrapolated samples. Most curves approached asymptotes, suggesting that sampling effort was sufficient to capture the majority of species occurring within each location. Similarly, coverage-based rarefaction curves showed high sample coverage across all locations, exceeding 80% in all cases (Figure 2B). The relatively limited increase in estimated richness under extrapolation suggests that additional sampling would likely contribute only marginally to total species detection. Collectively, these results indicate robust sampling coverage and provide confidence in subsequent comparisons of avian diversity among locations.

Overall, spatial variation in both taxonomic and functional diversity was pronounced, with DYKE and CTMC consistently supporting higher species richness, abundance, and functional richness. Functional evenness, however, displayed a contrasting seasonal pattern, suggesting that community assembly processes and trait distribution may be modulated by seasonal environmental dynamics rather than species counts alone. These results highlight the importance of considering both taxonomic and functional dimensions to understand patterns of biodiversity across heterogeneous landscapes.

Taxonomic diversity

Shannon diversity differed significantly among the seven sampling locations after accounting for repeated observations within sections and temporal variation between years and seasons (linear mixed-effects model: F6,28.18 = 6.91, p < 0.001). In contrast, neither season (F1,865.19 = 0.001, p = 0.977) nor year (F1,873.83 = 0.059, p = 0.809) had a significant effect on Shannon diversity. Model-estimated Shannon diversity was highest at DYKE (estimated marginal mean = 2.07, 95% CI: 1.90–2.23), followed by CTMC (1.81, 95% CI: 1.65–1.98). Estimated diversity was lower at EMPC (1.57, 95% CI: 1.40–1.73), POWH (1.56, 95% CI: 1.37–1.74), ICWT (1.54, 95% CI: 1.37–1.71), FGSR (1.52, 95% CI: 1.36–1.68), and COOJ (1.50, 95% CI: 1.34–1.66) (Figure 3A). Tukey-adjusted pairwise comparisons indicated that DYKE had significantly greater Shannon diversity than COOJ (p < 0.001), EMPC (p < 0.01), FGSR (p < 0.001), ICWT (p < 0.01), and POWH (p < 0.01). The difference between DYKE and CTMC was not significant (p = 0.31). Likewise, COOJ did not differ significantly from CTMC (p = 0.11), and no significant differences were detected among EMPC, FGSR, ICWT and POWH (all p ≥ 0.997) (see S1 for more details).

Species richness varied significantly among locations after accounting season and year (F 6,28.71 = 7.95, P < 0.001). Neither season (F 1,860.90 = 0.0001, P = 0.991) nor year (F 1,875 = 0.36, P = 0.551) had a significant effect on richness. Back-transformed estimated marginal means indicated that predicted species richness was highest at DYKE (9.87, 95% CI: 8.17 – 11.85), followed by CTMC (7.92, 95% CI: 6.69 – 9.40). Predicted richness was lower at ICWT (5.81, 95% CI: 4.90 – 6.96), POWH (5.70, 95% CI: 4.76 – 6.89), EMPC (5.53, 95% CI: 4.66 – 6.62), COOJ (5.42, 95% CI: 4.62 – 6.42) and FGSR (5.36, 95% CI: 4.57 – 6.36) (Figure 3B). Turkey-adjusted pairwise comparisons showed that DYKE had significantly greater richness than COOJ (P < 0.001), EMPC (P < 0.001), FGSR (P < 0.001), ICWT (P < 0.01), and POWH (P < 0.001). Richness was also significantly higher at CTMC than at COOJ (P = 0.040) and FGSR (P = 0.036) (see S2 for more details).

Bird abundance varied significantly among locations (F 6,32.09 = 10.31, P < 0.001) and between years (F 1,878.90 = 12.05, P < 0.001), whereas the effect of season was not significant (F 1,871.07 = 1.27, P = 0.260). Back-transformed estimated marginal means indicated that abundance was highest at DYKE (25.2, 95% CI: 21.2 – 30.0), followed by CTMC (22.8, 95% CI: 19.2 – 27.1). Abundance was lower at ICWT (16.2, 95% CI: 13.5 – 19.4), and POWH (16.2, 95% CI: 13.2 – 19.7), and lowest at COOJ (13.4, 95% CI: 11.4 – 15.8), FGSR (13.3, 95% CI: 11.2 – 15.7) and EMPC (13.2, 95% CI: 11.0 – 15.7) (Figure 3C). Turkey-adjusted pairwise comparisons showed that abundance at COOJ was significantly lower than at CTMC (ratio = 0.588, P = 0.002) and DYKE (ratio = 0.532, P < 0.001). CTMC also had significantly greater abundance than EMPC (ratio = 1.732, P = 0.002) and FGSR (ratio = 1.718, P = 0.002). DYKE had significantly greater abundance than EMPC (ratio = 1.915, P < 0.001), FGSR (ratio = 1.900, P < 0.001), ICWT (ratio = 1.559, P = 0.016), and POWH (ratio = 1.562, P = 0.024) (see S3 for more details).

Functional diversity

Rao’s quadratic entropy (RaoQ) also varied significantly among locations, but between seasons. Locations had a significant effect on RaoQ (F 6,25.26 = 2.58, P = 0.043), whereas the effect of season was not significant (F 1.223.00 = 0.44, P = 0.508). Estimated marginal means indicated that RaoQ was highest at DYKE (0.205, 95% CI: 0.194 – 0.217), followed by FGSR (0.190, 95% CI: 0.178 – 0.2020), whereas ICWT had the lowest estimated RaoQ (0.176, 95% CI: 0.165 – 0.188) (Figure 4A). However, Tukey-adjusted pairwise comparisons showed that the only significant difference was between DYKE and ICWT, with RaoQ being higher at DYKE (P = 0.019) (See S4 for more details).

Functional richness (FRic) varied significantly among locations and between seasons. Location had a significant effect on FRic (F 6, 22. 44 = 3.67, P = 0.011), while FRic was also significantly higher during the wet season than the dry season (F 1, 213. 03 = 19.43, P < 0.001). Estimated marginal means back-transformed to the original FRic scale indicated that FRic was highest at DYKE (0.00079, 95% CI: 0.00054–0.00115), followed by FGSR (0.00049, 95% CI: 0.00033–0.00071) and CTMC (0.00041, 95% CI: 0.00027–0.00061), whereas ICWT had the lowest estimated FRic (0.00026, 95% CI: 0.00018–0.00039) (Figure 4B). Tukey-adjusted pairwise comparisons indicated that FRic at DYKE was significantly higher than at EMPC (P = 0.046), ICWT (P = 0.01), and POWH (P = 0.04) (See S5 for more details). FRic also differed markedly between seasons, with estimated FRic of 0.00029 (95% CI: 0.00023–0.00035) during the dry season and 0.00054 (95% CI: 0.00044–0.00067) during the wet season (Figure 5A). The wet-season estimate was significantly higher than the dry-season estimate (log-scale difference = 0.638 ± 0.145 SE, t 218 = 4.41, P < 0.001).

Functional evenness (FEve) varied significantly among locations and between seasons. Location had a significant effect on FEve (F 6, 251 = 3.90, P < 0.001), while FEve was also significantly higher during the dry season than the wet season (F 1, 251 = 15.75, P < 0.001). Estimated marginal means indicated that FEve was highest at ICWT (0.567, 95% CI: 0.538–0.596), followed by EMPC (0.561, 95% CI: 0.531–0.592) and POWH (0.555, 95% CI: 0.522–0.587), whereas DYKE had the lowest estimated FEve (0.494, 95% CI: 0.466–0.522) (Figure 4C). Tukey-adjusted pairwise comparisons showed that FEve was significantly higher at EMPC and ICWT than at DYKE (P = 0.036 and P = 0.017, respectively (See S6 for more details). Seasonal differences were also evident, with FEve estimated at 0.555 (95% CI: 0.540–0.571) during the dry season compared with 0.512 (95% CI: 0.497–0.528) during the wet season (Figure 5B). The dry-season estimate was significantly higher than the wet-season estimate (difference = 0.043, t 225 = 3.97, P < 0.001).

Community composition

The NMDS ordination converged with a stress value of 0.137, indicating an acceptable representation of community structure in two-dimensional space (Figure 6). Bird community composition differed significantly among locations and between seasons, although the proportion of variation explained by these factors was modest. Marginal PERMANOVA based on Bray–Curtis dissimilarities showed a significant effect of location (F 6, 877 = 2.69, P = 0.001), which accounted for 1.72% of the variation in community composition. Season also showed a significant association with community composition (F 1, 877 = 5.55, P = 0.001), but explained only 0.59% of the variation (Table 1). Among the habitat variables, shrub cover (F 1, 877 = 1.42, P = 0.022), snag density (F 1, 877 = 2.95, P = 0.001), plant cover (F 1, 877 = 1.63, P = 0.011), and distance to the nearest human settlement (F 1, 877 = 1.65, P = 0.010) were significantly associated with community composition. Litter cover (P = 0.062) and distance to water (P = 0.078) showed marginal associations, whereas logs, rocks, altitude, logging, grazing, and bush burning were not significant (P > 0.05; Table 1). Collectively, the predictors accounted for a modest proportion of the total variation in bird community composition, with the majority remaining unexplained by the variables included in the model. However, tests of multivariate dispersion indicated significant heterogeneity among locations (F 6, 890 = 8.20, permutation P = 0.001) and between seasons (F 1, 895 = 7.79, permutation P = 0.013). Thus, the significant PERMANOVA effects for location and season should be interpreted cautiously, as differences in within-group dispersion may contribute to the observed differences in community structure.

Our study examined patterns of avian taxonomic and functional diversity across multiple habitats surrounding the Shiroro Reservoir and evaluated the ecological and anthropogenic factors shaping bird assemblages within a tropical hydropower landscape. Overall, bird diversity and community composition varied significantly among locations, whereas seasonal variation had comparatively weak effects on taxonomic diversity. Semi-natural habitats, particularly DYKE and CTMC, consistently supported higher species richness, Shannon diversity, abundance, and functional richness relative to disturbed sites such as COOJ and POWH. Environmental variables associated with vegetation structure and habitat complexity significantly influenced avian community composition, highlighting the importance of habitat heterogeneity in maintaining biodiversity within reservoir ecosystems.

The higher diversity observed in DYKE and CTMC likely reflects the ecological importance of structurally complex habitats within the reservoir landscape. Habitat heterogeneity is widely recognized as a key driver of avian diversity because heterogeneous environments provide a wider range of foraging substrates, nesting opportunities, microclimatic conditions, and refuge resources capable of supporting species with varying ecological requirements. Semi-natural reservoir-edge habitats may additionally function as ecological transition zones where aquatic and terrestrial resources interact, thereby increasing habitat suitability for both generalist and specialist bird species. Similar relationships between vegetation complexity and avian diversity have been reported in tropical and riparian ecosystems where structurally diverse habitats support greater niche availability and ecological stability (Bennett et al., 2014; Erin et al., 2020; Xu et al., 2022). In contrast, disturbed sites associated with infrastructure, settlement activities, and intense human-use tended to support lower diversity and reduced functional richness, suggesting that anthropogenic modification may simplify habitat structure and reduce ecological opportunities for sensitive species.

The observed spatial differences in species richness and abundance further suggest that hydropower-associated landscapes can simultaneously create both ecological opportunities and conservation challenges. Although dams and reservoirs are frequently associated with habitat fragmentation, hydrological alteration, and biodiversity loss (Nilsson et al., 2005; Wu et al., 2019), our findings indicate that certain habitats within modified reservoir systems may retain considerable conservation value when vegetation structure and habitat integrity are maintained. The persistence of relatively high diversity in semi-natural sites may therefore reflect the capacity of some reservoir-edge habitats to provide substitute ecological functions for bird communities within otherwise human-modified landscapes. Similar findings have been reported from other reservoir ecosystems where remnant vegetation and riparian habitats act as biodiversity refugia despite surrounding disturbance (Liang et al., 2022; Menger et al., 2024).

Functional diversity metrics provided additional insight into how bird communities responded to environmental gradients across the reservoir landscape. Rao’s quadratic entropy and functional richness were highest in the least disturbed habitats, indicating that these areas supported a broader range of ecological strategies and functional traits. Functionally diverse communities are often associated with greater ecosystem resilience because they enhance complementarity in resource use and increase the capacity of ecosystems to maintain ecological processes under environmental change (Mouillot et al., 2013; Danmallam et al., 2024). In contrast, lower functional richness in heavily modified sites may indicate environmental filtering, whereby only species possessing traits tolerant of disturbance are able to persist (Iniunam et al., 2025). Such filtering processes may contribute to functional homogenization and reduced ecosystem functioning in anthropogenically altered habitats.

Interestingly, functional evenness displayed an opposite spatial pattern to Rao’s quadratic entropy and functional richness across locations. Sites such as DYKE and CTMC, which supported the highest functional richness and trait dissimilarity, exhibited comparatively lower FEve values, whereas more disturbed sites tended to show higher functional evenness despite reduced functional richness. This pattern suggests that species-rich habitats supported a broader range of ecological strategies, but that trait distributions within these assemblages were uneven, potentially reflecting dominance by particular functional groups or ecological guilds. In contrast, disturbed habitats may have undergone stronger environmental filtering, resulting in communities composed of fewer but more evenly distributed functional traits. Similar contrasting responses among functional diversity indices have been reported in disturbed ecosystems where habitat simplification reduces niche diversity while simultaneously homogenizing species abundances and trait distributions (Mason et al., 2005; Villeger et al., 2008; Mouillot et al., 2013; Danmallam et al., 2025). These findings emphasize that different dimensions of functional diversity capture distinct aspects of community organization and should therefore be interpreted complementarily rather than interchangeably.

Functional diversity and seasonal variation

Functional diversity exhibited contrasting seasonal responses, with functional richness (FRic) higher during the wet season, whereas functional evenness (FEve) was higher during the dry season. The higher wet-season FRic suggests that bird assemblages occupied a broader range of functional trait space during the wetter period, potentially reflecting greater availability or diversity of resources and habitat conditions. In contrast, the higher dry-season FEve indicates a more even distribution of species within the available functional space. These contrasting responses suggest that seasonal variation may influence both the breadth of ecological strategies represented in the community and their distribution within functional space, even where changes in taxonomic diversity are less pronounced.

The higher dry-season FEve may also partly reflect seasonal turnover associated with Palearctic migrants, whose occurrence in the region overlaps substantially with the dry season. The arrival of migratory species can introduce additional functional traits into local assemblages and alter the relative distribution of species within functional space. Consequently, the observed increase in dry-season FEve should not necessarily be attributed solely to changes in local resource availability or habitat conditions. Rather, it may reflect a combination of seasonal environmental variation and changes in community membership associated with migration. Because migratory status was not explicitly incorporated into the present analysis, however, this remains a plausible explanation rather than a demonstrated mechanism.

Overall, the contrasting patterns of FRic and FEve demonstrate that seasonal changes in bird communities may involve changes in functional breadth and functional structure, rather than simply changes in the number of species present. This supports the value of combining taxonomic and functional measures when assessing biodiversity responses to environmental variation in reservoir landscapes (Mouillot et al., 2013; Keith, 2020).

Community composition and habitat structure

Variation in bird community composition across the reservoir landscape was associated with aspects of vegetation structure and human influence, particularly shrub cover, snags, plant cover, and proximity to human settlements. The associations with vegetation structure are consistent with the role of habitat complexity in providing nesting, foraging, and shelter resources for birds (Fraixedas et al., 2020). Similarly, the relationship with distance to human settlements is consistent with evidence that bird species differ in their tolerance of anthropogenic pressures, which can alter local assemblage structure (Marjakangas et al., 2024). Infrastructure and associated human activities may further modify habitat through fragmentation, vegetation alteration, and edge effects (Laurance et al., 2015).

Nevertheless, the relatively small explanatory contribution of these variables suggests that community composition around Shiroro Reservoir is shaped by a broader set of factors than those measured here. Moreover, significant differences in multivariate dispersion among locations and seasons indicate that the observed PERMANOVA patterns may partly reflect differences in within-group variability. Thus, habitat structure and human influence appear to contribute to spatial variation in bird communities, but should not be regarded as the sole determinants of assemblage composition.

Limitations of the study, recommendations and conclusion

From a conservation perspective, our findings highlight the importance of maintaining structurally diverse habitats within hydropower landscapes. Semi-natural habitats surrounding the Shiroro Reservoir supported the highest levels of both taxonomic and functional diversity, indicating that these areas provide important refugia for avian communities within modified ecosystems. Retaining tree cover, preserving riparian vegetation, maintaining standing deadwood, and minimizing further habitat degradation may therefore be essential for sustaining ecosystem functioning and biodiversity in the region. Reducing anthropogenic pressures such as logging, agricultural expansion, uncontrolled burning, and settlement encroachment could further enhance the ecological integrity of reservoir-edge habitats. Because birds contribute important ecosystem services including seed dispersal, pollination, and insect regulation, conserving functionally diverse avian communities may also support broader ecosystem resilience within tropical savanna landscapes (Gaston, 2022; Sekercioglu, 2012).

Although several environmental variables significantly influenced bird assemblages, the overall proportion of explained variation was relatively low. This suggests that avian communities within reservoir ecosystems are likely shaped by multiple interacting processes operating across different spatial and temporal scales. Bird communities are highly dynamic and may additionally respond to factors not explicitly measured in this study, including food resource availability, microclimatic variation, interspecific interactions, predation risk, short-term climatic fluctuations, acoustic disturbance, and fine-scale vegetation structure. Future studies integrating hierarchical modelling approaches, passive acoustic monitoring, and additional ecological variables would further improve understanding of avian responses to tropical reservoir systems.

In conclusion, the Shiroro Reservoir landscape illustrates both the conservation challenges and ecological opportunities associated with hydropower development in tropical Africa. Bird diversity and community composition were strongly associated with habitat structure and anthropogenic disturbance, while functional diversity analyses revealed important seasonal and ecological dynamics not captured by taxonomic metrics alone. These findings demonstrate that structurally complex semi-natural habitats can retain substantial conservation value within modified reservoir ecosystems. Integrating habitat protection, vegetation management, and sustainable land-use planning into hydropower development strategies will therefore be essential for conserving biodiversity and maintaining ecosystem functioning within Nigeria’s rapidly changing landscapes.