1. Introduction

Biodiversity loss is a major crisis, worsened by climate change (Zeng et al., 2023). Biodiversity hotspots, especially those rich in endemic biota (e.g., Western Ghats, India), face unprecedented habitat loss and anthropogenic pressures, driving species decline and extinction risks (Mittermeier et al., 2011; Doré et al., 2021). A primary goal of community ecology is to assess biodiversity. Accordingly, community ecologists have used data on floristic structure and composition to define and map forest types, while improving our knowledge of the links between species and their habitat (Condit, 1996; Couteron et al., 2003). Understanding patterns of forest biodiversity and the factors that drive these patterns has remained a major challenge in ecology (Palmer, 1994), however, clarifying these patterns would help develop targeted climate-smart conservation strategies–especially in protected and human-modified landscapes (Babu et al., 2023a).

Several measures of biological diversity have been used to characterize forest ecosystems. α-diversity measures species richness in a community, while β-diversity compares diversity between communities (Whittaker, 1960). β-diversity has more recently been decomposed into contributions to community β-diversity (Sor et al., 2018) and local contributions to beta diversity (LCBD; Legendre and De Cáceres, 2013). These metrics facilitates precise assessments of the ecological distinctiveness of a site and community across space and time (Legendre and De Cáceres, 2013; Ruhí et al., 2017). Despite these advances, ongoing research predominantly emphasizes α-diversity patterns (Pyšek et al., 2005; Bhat et al., 2020), with β-diversity being less frequently addressed especially in studies pertaining to the Indian tropics.

Previous studies have indicated that multiple processes are crucial for determining forest species composition and diversity (Barczyk et al., 2023). Topographic factors (e.g., elevation) have been shown to affect species richness (Zhang and Zhang, 2011; Zhang et al., 2012; Irl et al., 2015; Zellweger et al., 2015; Dar and Parthasarathy, 2022) and plant composition in forests, as well as factors that may have indirect effects on forest diversity (e.g., resource availability) (Macek et al., 2019; Oldfather and Ackerly, 2019; da Silveira et al., 2022). Environmental variation (e.g., soil physicochemical properties and nutrients) and climate variability have also been shown to play important roles in determining forest composition and diversity (Jiménez-Alfaro et al., 2018; Bona et al., 2020; Dantas de Paula et al., 2021; Boyle et al., 2021). Moreover, the role of climate and environmental variation on forest diversity has been conjectured to become more important in response to climate change (Ntirugulirwa et al., 2023). Interactions among species and competition for resources (e.g., light, water) have also been shown to modify local taxonomic composition, abundance patterns, and richness (Vilmi et al., 2017; da Silva et al., 2018; Chen et al., 2024). For instance, stand density/basal area may affect species richness via competition and β-diversity by adjusting the proportion of common and/or shared species within and/or among communities (Yao et al., 2020). However, few studies agree on which of these factors is most important to forest biodiversity.

Human disturbance also alters the structural and functional processes of forest ecosystems, affecting plant species dynamics and conservation (Shrestha et al., 2013). Specifically, human activities such as fuel-wood collection, grazing, and road construction degrade forests by altering stem density, basal area, and biomass (Foley et al., 2005). Such modifications significantly shape and determine species diversity directly, through species loss, and indirectly by altering local environmental conditions (Fischer and Lindenmayer, 2007; Buba, 2015). Thus, studies of biodiversity must distinguish between environmental factors and human-mediated disturbances. Moreover, while existing studies have examined the effect of individual predictors on forest composition and diversity (Zhang et al., 2016; Báez et al., 2022), these variables are mutually entwined, rendering it difficult to ascertain their influence on community structure (He et al., 2022). Elucidating these interactions is essential for understanding species abundance and distribution predictions in changing environmental conditions (Angert et al., 2013; Chu et al., 2018). Importantly, these mechanisms are often site-specific, shaped by unique climates and management histories (Li et al., 2020).

The Shettihalli landscape represents a characteristic tropical forest ecosystem in the Western Ghats, India, one of the most important biodiversity hotspots in Asia. In this study, we used 170 forest inventory plots in the Shettihalli landscape to identify the direct and indirect factors that shape tree α- and β-diversity in the central Western Ghats. Specifically, we (1) characterized tree species composition and diversity across forest types and (2) identified factors that shaped these diversity patterns, including environmental conditions (e.g., topography and soil properties), biotic factors, climate, spatial factors, and human disturbances. Our characterization of diversity patterns in the Western Ghats should advance an in-depth, comprehensive understanding of the underlying mechanistic factors that shape tree diversity, providing insights crucial for ecological management and conservation.

2. Materials and methods 2.1. Study area

The present study was carried out in Shettihalli Wildlife Sanctuary and the adjacent territorial divisions together called "Shettihalli landscape" in the central Western Ghats, India. The Shettihalli landscape, covering an estimated area of 750 km², lies between 13°40′1.2″N to 14°4′58.8″N and 75°10′1.2″E to 75°34′58.8″E. The Shettihalli landscape is spread over three taluks in the Shivamogga district, Karnataka (Fig. 1; Babu et al., 2023a). The topography of the landscape is relatively plain in the eastern regions (600 m) compared to the undulating terrain of the western regions (850–1050 m) with steep hills such as Shankargudda (1031 m) (Kanda et al., 2021). As a result, temperature (Min: 17.50–19.38 ℃ and Max: 27.76–29.92 ℃) and precipitation (1044–3076 mm) vary in the landscape (https://worldclim.org/). The landscape is a mosaic of natural forest, plantation forest, and human-dominated land uses (Babu et al., 2023a).

The Shettihalli landscape is also of great floral and faunal importance, harboring substantial spatial coverage of endemic tree species such as Terminalia paniculata, Lagerstroemia microcarpa (Babu et al., 2024). The forests are dominated by Southern tropical moist mixed deciduous, dry mixed deciduous, and tropical semi-evergreen forests (Kanda et al., 2021). Interspersed within the protected area is a network of five dominant tree plantations, viz. Tectona grandis, Eucalyptus grandis, Acacia auriculiformis, A. mangium, and Pinus roxburghii. Although the biodiversity of these tree plantations is poor compared to natural forests, they have great carbon storage ability and harbor hardwood trees of economic importance (Babu et al., 2023a).

The landscape has a documented history of seasonal forest fires, which can become pronounced in the summer, from February to May (Babu et al., 2023b). Additionally, the rapid proliferation of invasive species in the vegetated areas has become a fire hazard and a constant menace to both locals and the forest department. The presence of high human density within the landscape, with over 124 hamlets and 32 enclosures, and an extensive interior road network poses a threat to the biodiversity of the region due to easy accessibility.

2.2. Sampling design and data collection

From 2019 to 2022, we used stratified random sampling to establish 170 georeferenced (Garmin GPS 66SR < 5 m accuracy) 0.1 ha (31.63 m × 31.63 m) vegetation plots across the Shettihalli landscape (Fig. 1). These included 22 plots in semi-evergreen forests, 71 in moist deciduous forests, 27 in dry deciduous forests, and 50 in plantation forests. All plots were laid out with a consistent directional alignment, starting from the southwest corner and oriented towards north, to standardize layout and measurements across sites. Random points were generated in ArcGIS, with a 500 m minimum distance between adjacent plots to ensure maximum landscape coverage. The topography, climate, canopy density, and forest types of the plots were recorded.

In each plot (Fig. 1), trees ≥ 30 cm girth at breast height (GBH; defined at 1.3 m aboveground) were measured for both girth and height using measuring tape and a laser range finder (Nikon Forestry Pro) for selected individuals, followed by visual estimations for the rest of the individuals. Trees with multi-stems were measured separately, and for trees with buttresses, GBH was measured above the point of measurement (i.e., 1.3 m; Pearson et al., 2005). Plant identification was supervised in the field and later confirmed against the reference database of the Herbarium of the French Institute of Pondicherry (HIFP) and Kanda et al. (2021). Voucher specimens were collected, methodologically processed, numbered, and deposited in HIFP. Taxonomic names were standardized using the World Flora Online database (https://www.worldfloraonline.org/) through the U.Taxonstand Online web application (https://ecoinfor.shinyapps.io/UTaxonstandOnline/; Zhang et al., 2025), which implements automated matching and validation procedures. Species classification into families follows APG Ⅳ (Chase et al., 2016). Comprehensive details about the flora are available in our previous study (Bai et al., 2021).

2.3. α- and β-diversity metrics

Species abundance data were used to derive α-diversity parameters for each plot (Fig. 2). α-diversity metrics, including species richness, Shannon index, Simpson index, Fisher's α, and Pielou evenness, were computed following Magurran (2004) using Paleontological Statistics (PAST) software (v.3.26) (Table 1).

Local contributions to β-diversity (LCBD) shows the ecological uniqueness of species assemblages, with higher values indicating larger contributions to species composition variation. We evaluated LCBD in the Shettihalli landscape by quantifying the total variation in species composition, where LCBD reflects the unique contribution of each plot to overall β-diversity (Legendre and De Cáceres, 2013). We then estimated LCBD for each plot across the landscape using both abundance (method = "Hellinger") and incidence data (method = "jaccard") with beta.div function in 'adespatial' R package (Legendre and De Cáceres, 2013; Legendre, 2014). Given the lack of significant difference (Mann–Whitney W = 15,040; p = 0.51; Fig. S1) between abundance- and incidence-based LCBD scores, further analyses were conducted using abundance-based scores to maintain consistency with the abundance-based α-diversity metrics.

2.4. Quantifications of environmental predictors

To identify the main drivers of tree diversity patterns in the Shettihalli landscape, we examined whether tree diversity is correlated with 27 explanatory variables, including environmental factors, human effects, and stand structure (Table 2). Environmental factors included climatic, edaphic, spatial, and topographic factors.

2.4.1. Topography

Landscape complexity influences species diversity. We examined the following topographic variables: slope, aspect, and elevation. These data were derived from the Shuttle Radar Topography Mission Digital Elevation Model (SRTM DEM) with 30 m spatial resolution via Google Earth Engine (GEE) platform clipped to the study area. Aspect data were transformed into sine and cosine, representing the eastness and northness of each plot (Hirzel et al., 2002) using ArcGIS. Topographic Position Index (TPI) was calculated using DEM data in SAGA GIS.

2.4.2. Stand structure

The following forest structural attributes were calculated: stand density (ha⁻¹), canopy cover (%), coefficient of variation of diameter at breast height (%), and height (%), Mean Normalized Difference Vegetation Index (NDVI), and Mean Enhanced Vegetation Index (EVI). Stand density was assessed by enumerating the number of stems per plot. Coefficient of variation of diameter at breast height (hereafter referred as DBH) and coefficient of variation of height (hereafter referred as height) quantify the variability of DBH and height measurements within plots as percentages, determined by dividing their standard deviation by the mean. We used on-site visual estimations to incorporate canopy cover information in the plots following Ayushi et al. (2023). We procured cloud-free (< 10%) Sentinel-2 data for the peak growth period (November and December) from 2016 to 2021 and computed mean NDVI and mean EVI using GEE. The average values for the six years were taken for the analysis.

2.4.3. Edaphic parameters – soil sampling and laboratory analysis

Edaphic properties were examined for each plot as follows. A representative soil sample was collected by homogenizing four subsamples taken randomly with a metallic soil corer 15 cm below plant debris. To analyse the physical properties, soil cores are were duplicated at the same depth for consistency. pH measurements were based on a 1:2.5 dry soil suspension (soil/deionized water) following the apparatus (PETS 372, Systronics India Ltd.) calibration with buffer. Soil organic carbon (SOC; %), total nitrogen (%), and C: N ratio were measured from a 5 ± 0.05 mg soil sample using the dry combustion method (Semi Macro Elemental Analyzer). Available phosphorus (kg ha⁻¹) was determined by extraction with a 0.5 M NaHCO₃ solution at pH 8.5, followed by color intensity measurements with a photoelectric colorimeter (Olsen et al., 1954). Likewise, available potassium (kg ha⁻¹) was assessed by exchanging K⁺ ions (in the sample) with NH₄⁺ (in the solution), thereby measuring K⁺ ion concentration with Flame Photometer (Metson, 1956). Gravimetric soil moisture was determined by calculating the weight loss of soil samples following drying at 105 ℃ for 24 h (FAO, 2020). A detailed description of soil sampling and analyses is provided in Babu et al. (2023a).

2.4.4. Climate

To quantify climate filtering, we retrieved plot-level mean annual temperature (MAT) and precipitation (MAP) from the WorldClim database (http://www.worldclim.org/) with 30 arc seconds (~1 km²) spatial resolution, using "raster" package (Hijmans et al., 2015) in R software. Potential evapotranspiration (PET) was computed from MAP and MAT following Holdridge et al. (1971). All variables were standardized to a 30 m resolution using bilinear interpolation for uniformity.

2.4.5. Disturbance

Disturbances were characterized by distance to roads, stump-to-tree ratio (Cutstem_ratio), grazing intensity, and invasive species coverage. Road networks were tracked through field surveys and spatial techniques (Babu et al., 2023b). Grazing intensity was determined by observing recent signs and assigning values between 0 and 4 depending on the extent of grazing within the plot. A similar protocol was employed for the estimation of invasive cover. The cut stem ratio was calculated by dividing the number of cut stems by the total number of stems in the plot and then multiplying the result by 100 to express it as a percentage.

2.4.6. Spatial attributes

Geographical distances between plots were transformed into orthogonal spatial variables through PCNM in the R package "vegan" (Oksanen et al., 2013). The generated eigenvectors were considered spatial factors (i.e., PCNMs), capturing unmeasured broad-scale patterns in the present-day environment or historical processes (Svenning et al., 2009). Large eigenvalue PCNMs depict broad-scale patterns, while those with small eigenvalues characterize fine-scale patterns. Finally, 85 generated PCNMs were subjected to dimension reduction (principal component analysis; PCA) to identify the principal component (i.e., spatial variable) contributing to the variation, enabling further analyses with predictor variables.

2.5. Vegetation composition and drivers of α- and β-diversity

Statistical differences in diversity attributes among forest types were analysed using the non-parametric Kruskal–Wallis test at the 5% significance level. We used boxplots to visualize the plot-level distribution of diversity attributes among the forest types. Significant mean differences were followed by post-hoc Dunn's test for multiple comparisons to reveal differences between forests. We used balanced resampling to avoid bias in LCBD scores due to unequal plot representation, ensuring fair comparisons across forest types. This was necessary because LCBD measures each plot's deviation from the overall centroid, and the disproportionate representation of forest types can bias the centroid and, consequently, the LCBD scores. To visualize and interpret floristic patterns (dis/similarity) among forests, NMDS with Bray–Curtis distance in two-dimensional space was performed using the "vegan" package (Oksanen et al., 2013). PERMANOVA using Bray–Curtis dissimilarity with 9999 permutations was performed with vegan 'adonis2' function to quantify the strength of species association among forest types. The 'pairwise.perm.manova' function was used to compare the pairwise compositional dissimilarities between forest types.

2.5.1. Variable selection

We used 27 predictor variables (i.e., 5 topographic, 7 structural, 7 edaphic, 3 climate, 4 disturbance, and spatial variables) to model α- (Shannon diversity) and β-diversity (LCBD) in the Shettihalli landscape. To avoid model overfitting and multicollinearity among independent predictors, which commonly undermine statistical accuracy (Hall, 1999; Odebiri et al., 2020), we removed non-relevant and redundant variables (Guyon and Elisseeff, 2003). Pearson's correlation (r ≥ ±0.6) helped identify seven highly correlated (multicollinearity) variables for removal (Fig. S2) (Cerrejón et al., 2020; Singh et al., 2022; Gotta et al., 2023). Variable selection was further optimized by the established machine learning technique Least Absolute Shrinkage and Selection Operator (LASSO). LASSO (Tibshirani, 2011), a penalized machine learning technique, characterizes diversity relative to plot-level covariates. By adjusting the tuning parameter, LASSO achieves a parsimonious and interpretable model by penalizing predictor coefficients, thereby driving noisy coefficients to zero to prevent overfitting and using the non-zero coefficients for prognostic modeling. The optimal regularization value, minimizing misclassification error, is chosen through grid search coupled with 10-fold cross-validation. LASSO identified 15 and 13 variables as most important for Shannon and LCBD, respectively (Fig. S3). LASSO and Pearson correlation were implemented in "caret" v.6.0–93 (Kuhn, 2008) and "metan" v.1.17.0 (Olivoto and Lúcio, 2020) packages in R software.

2.5.2. Structural equation modeling

We fit structural equation modeling (SEM) in AMOS (SPSS, IBM, Inc.) to model (α/β) diversity with predictor variables across large-scale ecological gradients in tropical forests. SEM facilitates concurrent testing of both direct and indirect relationships among observed variables while determining the relative importance of predictor variables influencing the response variable (Grace and Bollen, 2005; Mensah et al., 2018). To determine the effect of forest types, we created k−1 dummy variables representing each non-reference category (e.g., Semi-evergreen forest = 1 if Forest type = Semi-evergreen forest, else 0). The results were interpreted relative to the reference category, i.e., Dry deciduous forest. Following recommendations (Grace et al., 2012), we initially specified a meta-model based on known theoretical constructs, integrating hypothesized multiple paths derived from literature and background knowledge within the SEM framework (Fig. S4). The standardized path coefficient, allowing comparison among pathways, measures the sensitivity of the response variable to the predictors (Grace and Bollen, 2005). Indirect effects were defined as the interaction between the standardized direct effects of a predictor on a mediator and the mediator's direct effect on a response variable; total effects were summed from standardized direct and indirect effects (Ali et al., 2016; Grace et al., 2016). All variables were log-transformed (ln) and standardized (mean = 0 and SD = 1) before SEM for model convergence and interpretability to ensure normality and homoscedasticity. The final SEM model was assessed using standard goodness-of-fit measures: Chi-square (χ²) test, Tucker–Lewis index (TLI), comparative fit index (CFI), root mean square error of approximation (RMSEA), and standardized root mean square residual (SRMR; Grace and Bollen, 2005). In SEM, lower and nonsignificant χ² (p > 0.05) indicates a better fit, assuming consistency between observed and expected covariance matrices. Similarly, higher values of TLI and CFI (close to 1) and lower values of RMSEA and RMR (close to 0) suggest a good model fit (Rosseel, 2012).

3. Results 3.1. Species composition and diversity patterns

Altogether 8120 individual trees representing 175 species (including two morphospecies) and 46 families were inventoried across four forest types in Shettihalli landscape. Of these the highest number of species were recorded in SEF (117) (Table 3). The most represented families were Fabaceae (16 species, 9.14%), Moraceae (15 species, 8.57%), and Lauraceae and Rubiaceae (10 species each, 5.71%). A comprehensive list of all species is included in Table S1.

Forest types differed in α-diversity and LCBD (Table 3 and Fig. 3). While α-diversity varied significantly across forest types, it was highest in semi-evergreen forests, then in dry deciduous forests, moist deciduous forests, and lastly plantation forests. There were no significant differences in α-diversity between dry deciduous forests and moist deciduous forests or between dry deciduous forests and semi-evergreen forests. LCBD, in contrast, was higher in plantation forests and semi-evergreen forests, than in dry deciduous forests and moist deciduous forests. There were no significant differences in LCBD between plantation forests and semi-evergreen forests.

Forest types also differed in tree density and abundance (Table 3). The mean tree density and abundance were highest in plantation forests (69.84 individuals ha^−0.1; 43% of total abundance), followed by semi-evergreen forests (43.95; 11.91%), moist deciduous forests (37.87; 33.12%), and dry deciduous forests (36.04; 11.97%).

NMDS clearly distinguished four forest types (Fig. 4), corresponding to semi-evergreen forests, dry deciduous forests, moist deciduous forests, and plantation forests. Forest composition varied among forests (F = 18.04, p < 0.001) and between forest types (p < 0.001). The species distribution in NMDS ordination space also demonstrated a consistent variation in species composition across forest types. Plots in close proximity suggest that moist deciduous forests and dry deciduous forests share species composition, as do the moist deciduous forests and semi-evergreen forests.

3.2. Effect of predictor variables on α- and β-diversity

SEM accounted for 79% and 45% of the total variance in α- and β-diversity patterns (Figs. 5 and 6; Tables S2 and S3). Specifically, SEM indicated that α- and β-diversity patterns in the Shettihalli landscape are driven by several direct and indirect factors, including stand structure, edaphic features, topography, disturbance, and climate (Figs. 5 and 6; Tables S2 and S3). However, α-diversity is driven by different factors than is β-diversity.

α-Diversity can be largely explained by forest type, followed by disturbance and structure. Of these, both height and tree density directly affected α-diversity, although the effect of height was positive and that of tree density negative (Fig. 5 and Table S2). In contrast, other factors, including distance to road, soil moisture, and C: N ratio, exhibited no significant direct effects. Instead, these factors had an indirect effect on α-diversity. For example, distance to the roads had a positive effect on α-diversity, mediated through its significant positive/negative effect on the forest type (plantation forests, β = −0.35; semi-evergreen forests, β = 0.33), soil moisture (β = 0.25), height (β = 0.22), and C: N ratio (β = −0.16). Similarly, tree density and soil moisture indirectly and negatively affected α-diversity by positively affecting mean NDVI (β = 0.54) and height (β = 0.12), and negatively affecting C: N ratio (β = −0.10 and −0.23). Moreover, mean NDVI positively influenced soil moisture (β = 0.35), ultimately affecting α-diversity (Fig. 5 and Table S2).

Several factors partially mediate the effect of forest type on diversity. For example, distance to road directly influenced Shannon diversity scores in semi-evergreen forests (β = 0.20), which are located in the interior, away from roads, and in plantation forests (and −0.59), which are located at roadsides. Similarly, our estimation indicate that Shannon diversity was indirectly increased by soil moisture (β = 0.01), but indirectly decreased by mean NDVI, height, and tree density decreased (β = −0.19). The standardized total effects (direct + indirect) of each factor were as follows: distance to the roads (0.43) > height (0.15) > tree density (−0.10) > C: N ratio (0.09) > soil moisture (0.02) > mean NDVI (0.01) (Table S2).

β-Diversity can be explained by MAP, tree density, SOC, grazing, available phosphorous, and elevation (Table S3). β-diversity was directly increased by tree density (β = 0.42), forest type (semi-evergreen forest, β = 0.34), mean annual precipitation (β = 0.24), and available phosphorus (β = 0.13). β-diversity was directly decreased by SOC (Fig. 6a and Table S3). Several factors indirectly affected β-diversity by influencing tree density, available phosphorus, and SOC. These indirect factors include MAP (β = 0.22), grazing (β = 0.15) and tree density (β = 0.09). Elevation also indirectly increased β-diversity through its effects on mean annual precipitation, SOC, forest type (semi-evergreen forest) and tree density; however, the effect of elevation on β-diversity was weak and statistically insignificant. Forest type also directly and indirectly increased LCBD (e.g., semi-evergreen forest vs. dry deciduous forest), with the indirect effect mediated by grazing and tree density.

Based on the exploratory analyses (Fig. 3), a separate model incorporating plantation forest only was also tested. The model interpreted 39% (p > 0.10) of the variation with model-fit metrics well with the recommended values (Fig. 6b). MAP had the strongest positive direct effect (β = 0.38), followed by tree density (β = 0.24), plantation forest (β = 0.18), and available phosphorus (β = 0.17), whereas variables implying direct negative effect include SOC and grazing (Fig. 6b and Table S4). Tree density (β = 0.06) and grazing (β = 0.06) indirectly affected β-diversity by directly influencing SOC and tree density and SOC, respectively (Fig. 6b). In contrast to its effect on Shannon diversity, the plantation forest relative to dry deciduous forest exerted a strong positive influence on LCBD (β = 0.18). The total effect on β-diversity in the decreasing order follows MAP, plantation forest, tree density, SOC, available phosphorus, and grazing (Table S4).

4. Discussion 4.1. Patterns of tree diversity, density, and composition

In the present inventory, 175 species were recorded from 46 families across four forest types of Shettihalli landscape, part of the Western Ghats biodiversity hotspot, India. Despite efforts to understand differences in diversity and composition patterns among tropical forests, variations in plot dimensions, shapes, girth threshold, and forest type often complicate direct comparisons of results with others (Dar and Parthasarathy, 2022). However, regardless of the area sampled, the observed species richness (175 species) is comparable with those reported across the tropics. For instance, our results closely coincide with the floristic enumeration from the Agumbe region adjacent to Shettihalli, wherein authors reported 185 tree species (Rao and Krishnamurthy, 2021), tropical riparian forest along Cauvery River Basin (177 species; Sunil et al., 2010), tropical forest from Eastern Ghats, Odisha (185 species; Reddy et al., 2007), and tropical forest in Little Andaman (186 species; Rasingam and Parthasarathy, 2009). Similarly, the reported species richness is well within the observed richness from tropical rainforests in Upper Amazonia, Brazil (89–283 species; Gentry, 1988), and tropical humid forests in South America (141–163 species; Sullivan et al., 2017).

We recorded 47.76 stems ha^−0.1 across the Shettihalli landscape, considerably lower than the seasonally dry tropical forest in northwestern Costa Rica (Kalacska et al., 2004) and Tanzanian tropical forest (Huang et al., 2003) at the comparable scale (0.1 ha). However, when scaled to the hectare level, the recorded tree density aligns well with the typical range for tropical forests (245–859 stems ha⁻¹; Richards, 1952). On a continental scale, the recorded mean tree density is 10.90% less than in African (Lewis et al., 2013) and 24.11% greater than in South American tropical forests (Lewis et al., 2004). Variations in plot size, methodology, and geographical extent likely explain this outcome. The mean tree densities observed in the present study for the deciduous dry forest (360 stems ha⁻¹), moist deciduous forest (378 stems ha⁻¹), semi-evergreen forest (439 stems ha⁻¹) and plantation (698 stems ha⁻¹) were very much lower than those observed for dry deciduous forest and moist deciduous forest (739 stems ha⁻¹; Kishore et al., 2024) in Mudumalai Tiger Reserve, Western Ghats, India, and tropical dry forest, Ethiopia (Adem et al., 2014).

We found that plant species composition and diversity metrics vary significantly among forest types (Figs. 3 and 4). The forest type with the highest tree species richness and diversity metrics was semi-evergreen forest, while the forest with the lowest scores was plantation forest. These findings may reflect the greater range of suitable microhabitats and resource availability within semi-evergreen forests. Mean species richness was more or less similar between moist and dry deciduous forests, although the total species richness was higher in moist deciduous forests, likely due to its larger geographical area. Lower evenness in plantation reflects an uneven abundance distribution among species, resulting in the dominance of only a few species.

Our study revealed that LCBD was lower in moist and dry deciduous forests than in semi-evergreen and plantation forests, suggesting a greater ecological uniqueness and distinctive species composition in semi-evergreen and plantation forests. Although substantial within-plot variations in diversity metrics were observed in the plantation forests, this variation is likely driven by differences in species composition coupled with post-establishment management across the Shettihalli landscape. As detailed in Babu et al. (2023a), these plantations vary in species composition (monodominant vs. mixed), stand age (10 to > 40 years), management history (e.g., Wildlife vs. Territorial Divisions), disturbance regimes, and ecological conditions (e.g., elevation, soil, and rainfall). Particularly, plantation forests comprising teak mixed category with native species (e.g., Terminalia paniculata, Lagerstroemia microcarpa, Dalbergia latifolia) exhibit greater species diversity, whereas monodominant stands of Pinus, Acacia, and Eucalyptus are more uniform, enhancing the ecological variability among plantation types. Moreover, legacy effects from differing silvicultural objectives further shape current diversity patterns (Babu et al., 2023a).

4.2. Drivers of α- and β-diversity

Changes in external conditions greatly influence the spatial distribution of ecological communities, leading to dynamic shifts over time (Zhou et al., 2019). The mechanisms underlying shifts in plant diversification (i.e., niche or neutral processes) have long been debated (Jones et al., 2007; Liu et al., 2016; Murphy et al., 2016), although recent findings suggest that both contemporary environment and spatial factors greatly influence plant diversity, with varying effects across study scales and habitat types (Blundo et al., 2016; Liu et al., 2016). Our findings are consistent with these studies. SEM analysis corroborated the well-established notion that topography, soil, climate, disturbance, and stand variables govern species diversity (Baldeck et al., 2013; Wu et al., 2017; Yan et al., 2018). This demonstrates that niche and neutral processes are not mutually incompatible but rather integrate to foster species' coexistence and diversity (Gravel et al., 2006; Chase, 2007; Legendre et al., 2009). Yet, depending on the diversity measure, the corresponding effects of the processes differ in magnitude and direction. The relatively weak effect of predictor variables on β-diversity (R² = 0.45) than α-diversity (R² = 0.79) may be attributed to the broader range of niche breadths exhibited by tree species not fully captured by the variables analysed. Alternatively, this study did not account for all environmental elements that may have potentially influenced diversity patterns. Overall, our findings emphasize that distinct ecological mechanisms influence α- and β-diversity patterns.

4.2.1. Topography

Elevation substantially affects local environmental conditions, including soil processes, hydrology, and light availability (Tateno and Takeda, 2003; Mascaro et al., 2011), altering plant growth and dynamics (mortality and recruitment), thus contributing to variations in species diversity (Moeslund et al., 2013; Rumpf et al., 2018). Our analysis suggests a multifaceted effect of elevation on LCBD by directly shaping the environmental processes such as soil properties (SOC), microclimatic variations (MAP), stand structure (tree density), and community distribution. Alternatively, higher elevation might have influenced β-diversity by fostering conditions conducive to a wide range of species, resulting in increased β-diversity. In line with our findings, Wu et al. (2017) and Bai et al. (2021) also demonstrated the importance of elevation as the significant determinant of species composition in subtropical forests, with aspect and slope showing no influence. This is further corroborated by studies indicating that environments characterized by constrained conditions, such as high elevations, typically display higher β-diversity (Tan et al., 2019). The redistribution of resources (light, temperature, water, or nutrients) caused by topographic heterogeneity (Tateno and Takeda, 2003; Mascaro et al., 2011) might have satisfied species' demands for diverse resource conditions, resulting in distinct species assemblages (Andersen et al., 2014; Jucker et al., 2018). Indirectly, elevation can also affect β-diversity by increasing SOC (β = 0.38) and decreasing tree density (β = −0.12). At higher elevations, temperature and precipitation tend to be cooler and higher, which helps maintain soil moisture and promote organic matter accumulation, thereby increasing SOC (Hewins et al., 2018). Moreover, environmental limitations at higher elevations can inhibit plant growth, potentially resulting in reduced tree density. In contrast to the present study, prior research suggested that α-diversity increases with elevation (Fried et al., 2008; Pál et al., 2013). The lack of relationship between α-diversity and elevation observed in our study suggests the overriding influence of other drivers, particularly resource availability.

4.2.2. Stand structure

Stand structure is the outcome of autogenic succession, including regeneration, competition, and consequently, self-thinning, as well as past management actions like disturbances (Lei et al., 2009; Keith et al., 2009). Similar to previous research (Heino et al., 2017; da Silva et al., 2018; Valizadeh et al., 2023), forest structural characteristics emerged as strong predictors of diversity over large ecological gradients in forest ecosystems, corroborating the significance of stem density and height variations in species diversity patterns. Stand density and height determine resource acquisition and biotic competition (Ali et al., 2019; Forrester, 2019), altering tree species composition. Additionally, stand density influences species coexistence by altering light and community structure (Shu et al., 2021). α-diversity was negatively correlated with stand density, suggesting that forest stand density reduced species diversity. Dominant plant species in habitats with moderate resources often prioritize resource competition rather than facilitating species richness (Bertness and Callaway, 1994; Angelini et al., 2011). Besides, dominant species exhibiting high productivity limit space and nutrients accessible to other plants (Grime, 1998). However, by fostering unique species through micro-environmental conditions or competitive interactions, tree density might have increased LCBD, promoting regional diversity. Conversely, tree height variations (directly and indirectly) contributed to α-diversity through niche complementary by creating a complex forest structure with multiple strata (e.g., emergent, canopy, sub-canopy), supporting diverse species and habitats.

4.2.3. Climate

Climate interactively shapes the realized niches of trees and ecological communities across multiple spatial extents (Clarke and Gaston, 2006). Although temperature affects tropical forests, rainfall quantity and distribution frequently exhibit a stronger direct effect on tree species composition, growth, and transpiration owing to water availability (Hu et al., 2017). In turn, precipitation plays a significant role in shaping the distribution and composition of plant species (Engelbrecht et al., 2007; Mykrä et al., 2010; Dakhil et al., 2019). This is particularly relevant in tropical forests where rich biodiversity and nutrient-deficient soils intensify the competition for water, serving as a critical selective pressure. Similar to previous findings (Liu et al., 2016; Konig et al., 2017), we found that MAP better explained β-diversity patterns in tropical forests, emphasizing the need to understand precipitation variations to predict changes in forest composition and biodiversity under shifting climates. Precipitation in tropical forests is essential for determining tree ecological uniqueness, as it drives various phenological and eco-physiological processes. Studies have shown that dynamic precipitation patterns can shift the onset and coordination of flowering and fruiting, key elements of the plant life cycle and ecological role (Singh et al., 2024). Generally, changes in precipitation alter soil water availability, favouring species with specific traits, such as deep-rooted species, thereby restructuring species composition (Yu et al., 2016). Similarly, Khaine et al. (2017) observed significant shifts in dominant species across varying precipitation gradients. However, at the mesoscale in central Western Ghats, Ramesh (2001) found that temperature was more influential than the dry season and rainfall duration in shaping floristic differences in wet evergreen forests. Similarly, across the Western Ghats biodiversity hotspot, Krishnadasan et al. (2021) suggested that seasonal drought regulates tree species distribution and community assembly. The heterogeneity in environmental, soil, and microclimatic variables, accentuated by Western Ghats relief and variations in spatial extent across studies, might have contributed to discrepancies observed.

4.2.4. Edaphic parameters

Edaphic features are crucial locally, filtering plant community assembly based on resource availability gradient (Balvanera et al., 2011; Prada et al., 2017). Soil chemical properties, including SOC and AP, determine nutrient availability, while soil texture governs water availability, both affecting tree diversity and functioning (Paoli et al., 2005; Bu et al., 2019). The influence of soil was apparent in both α- and β-diversity, emphasizing the significance of soil attributes in driving diversity patterns. The decline in β-diversity (β = −0.26) with increasing SOC supports Huston's hypothesis (Huston, 1980) that higher nutrient availability promotes competitive species, leading to the exclusion of others, as evidenced by the positive association (p = 0.022) of dominant species such as Terminalia paniculata and Lagerstroemia microcarpa with SOC. Further, the abundance of dominant species correlates significantly with β-diversity patterns, shaping their distribution (Pearson correlations, p < 0.001). Likewise, Peña-Claros et al. (2012) reported comparable findings in Bolivian moist and dry tropical forests. A high C: N ratio was associated with diverse microbial communities proficiently decomposing complex organic materials (Wan et al., 2015). Enhanced microbial diversity improves soil nutrient cycling and fosters diverse plant growth (Chen et al., 2024), increasing habitat heterogeneity. The resultant ecosystem sustains diverse plant species, each with distinct nutritional demands and ecological niches. These findings emphasize the critical significance of biodiversity conservation in maintaining ecosystem functionality (Duffy et al., 2017; Augusto and Boča, 2022).

Soil moisture is critical in determining plant growth and survival while maintaining α-diversity in tropical forests (Tavili et al., 2009). Our findings are consistent with studies that showed soil moisture influences plant species distribution in Southern Sinai, Egypt (El-Ghani and Amer 2003). Similarly, soil moisture contributed to understanding the distribution of evergreen species in the low-elevation forests of the southern Western Ghats (Swamy et al., 2000). We found, however, that high soil moisture levels indirectly lower α-diversity by increasing C: N ratio. The negative correlation between soil moisture and C: N ratio (β = −0.23) likely reflects enhanced microbial decomposition under moist conditions, accelerating carbon loss through respiration, thereby reducing C: N ratio. Similarly, we found that optimal soil moisture improves NDVI, a key indicator of plant health, by accelerating plant growth, thereby gradually increasing NDVI.

In tropical forests, phosphorus is frequently cited as the primary limiting nutrient (Vitousek et al., 2010; Laliberté et al., 2013). Increasing soil phosphorus typically stimulates plant growth by altering soil characteristics, thus enhancing species diversity (Xu et al., 2016). The results of the SEM analysis indicated that soil phosphorus significantly influenced β-diversity, highlighting the prevalence of deterministic processes over neutral ones across the landscape (Yang et al., 2015). Similarly, Xu et al. (2016) identified phosphorus availability as a strong determinant of tree species diversity in the Xishuangbanna tropical seasonal rainforest. This may be because the woody plants in low soil fertility communities are often co-limited by nitrogen and phosphorus, reflecting competition for limited nutrient resources driving species richness (Gravel et al., 2011; Sanaei et al., 2021). However, contrasting results were reported from various studies (Olde Venterink, 2011; Wu et al., 2019; Wang et al., 2024).

4.2.5. Disturbance

The primary anthropogenic activities that cause disturbance in tropical forests include highway or road development, illegal tree felling, and grazing, alongside practices like harvesting tree branches for fodder and firewood collection (Babu et al., 2023a). Consistent with previous findings (Santos et al., 2008; Osuri et al., 2020; Dar et al., 2024), disturbances in the Western Ghats decreased species diversity and adversely affected stand features, likely setting off cascading impacts on ecosystem functionality besides services beyond carbon sequestration. Our SEMs highlighted a multifaceted relationship between disturbances and tree species diversity loss, corroborating Liu et al. (2022) that both direct and indirect effects of disturbance drive diversity loss in eastern China.

While the consequences of disturbance are likely to vary depending on the nature of the disturbance (grazing or distance from the roads), diversity metrics typically decline with increasing disturbances (Fig. 5, Fig. 6). Increased accessibility via the road network could explain the substantial impact of disturbance on vegetation composition. Further, road accessibility might degrade ecosystems by directly causing habitat loss and resource depletion, besides intensifying various forms of human disturbances (Foggin, 2016). Greater distance from roads is associated with greater variation in tree height (β = 0.22) and higher soil moisture (β = 0.25), likely due to reduced human disturbance and more intact forest structure. This environment promotes organic matter accumulation and microbial activity, which may accelerate decomposition and nitrogen mineralization, thereby lowering the C: N ratio (β = −0.16). Together, these factors create a more heterogeneous and resource-rich environment that supports an overall increase in tree diversity (β = 0.43). The increase in tree density with grazing intensity in LCBD models could be due to reduced competition from understorey palatable species since high-intensity grazing substantially lowers their biomass, thus allowing trees greater access to space and resources for growth (Schulz et al., 2018). In our study, the direct effects of disturbances and indirect effects as a consequence of the strong negative impact on biotic and edaphic features are stronger on α-diversity.

4.2.6. Variation among forest types and implications for management

Our study demonstrates that multiple factors exert distinct and often synergistic effects on different facets of forest communities. These findings highlight the necessity of integrated, multidimensional strategies for effective management and conservation (Ali et al., 2019). One approach that proved effective for identifying conservation priority sites was our partitioning of diversity metrics. This allowed us to identify plots with both high α- and β-diversity. For example, semi-evergreen forest, which inhabits inaccessible high-elevation areas, are characterized by high Shannon diversity, high soil moisture, and high LCBD (Fig. 6), indicating relatively intact conditions that foster complex forest structure and unique species assemblages. These characteristics, together with low grazing intensity and significant environmental heterogeneity, highlight the need to prioritize this forest type for stringent conservation, as it likely serves as a biodiversity refuge and maintains essential ecological processes.

Our analysis of plantation forests identifies conservation actions specific to these forest types. Although plantation forests had high stand density, these forests had low Shannon diversity, high proximity to roads, high grazing intensity, low NDVI, and low C: N ratio. Loss of nutrients from forest ecosystems mainly results from converting primary to secondary forests, harvesting practices, soil disruptions, thinning operations, grazing, litter and biomass collection for fodder, fuel, or bedding, and fire (Mayer et al., 2020). The low C: N ratio of plantation forests may have been caused indirectly by human mulching with conifer litter in ginger cultivation. Moreover, residents in certain parts of the study area have poor socioeconomic status and rely heavily on harvesting dry biomass to satisfy their energy requirements (Babu et al., 2023a). Together, these activities have homogenized and perhaps degraded forest structure, as characterized by disturbance-tolerant species such as Acacia and Eucalyptus. These findings corroborate earlier studies and underscore the significant influence of local management on species diversity in the Western Ghats (Dar et al., 2024; Babu et al., 2023a). Our findings, thus, indicate that management interventions in plantation forests should prioritize regulating grazing, enhancing canopy cover, and increasing species composition via aided regeneration. Consequently, efficient conservation planning in protected areas necessitates a thorough assessment of grazing effects, with stocking rates per hectare serving as a critical management parameter (da Silveira et al., 2022).

Dry deciduous and moist deciduous forests have moderate levels of grazing and structural complexity. These factors make these forests ideal for low-intensity management approaches such as aided natural regeneration and community-based conservation. The distinct interplay of various factors in determining diversity in these forests requires customized management plans that address specific challenges in each forest type. Overall, sustaining biodiversity and fostering long-term ecological resilience in these forest ecosystems requires enhancing structural complexity, maintaining forest integrity, and avoiding anthropogenic disturbances.

5. Conclusion

In conclusion, we provide an in-depth quantitative landscape-level understanding of species richness, spatial diversity patterns, and ecological drivers of diversity metrics, focusing on tree species in the Shettihalli landscape in the central Western Ghats. The observed differences in the diversity and richness indices among forests emphasize the significance of maintaining environmental heterogeneity to sustain species diversity. The high richness and diversity in semi-evergreen forests indicate the need to conserve these forests without undermining the necessity of protecting the dry deciduous and moist deciduous forests. Using a SEM framework capable of analysing both direct and indirect relationships, we determined the principal drivers of tree diversity in tropical forests. Our results demonstrate that complex interactions among topography, stand structure, edaphic properties, climate, disturbance, and spatial constraints directly and indirectly shaped the tree diversity in Shettihalli forests, integrating environmental filtering and neutral processes. However, the variability in the direction and magnitude of these processes on diversity metrics emphasizes the need for multifaceted conservation approaches. Disturbance indicators predominantly affected α-diversity, with climate exerting a strong influence on uniqueness when considering both direct and indirect pathways. Our study highlights the dominance of direct effects over indirect ones, illustrating the complex ecosystem responses to seemingly minor changes in individual factors. Understanding these key drivers enables better predictions of ecosystem reactions, especially considering the variability observed across ecosystems. The insights into diversity dynamics along the environmental gradient in the Shettihalli landscape provide a foundation for targeted strategies aimed at forest management and biodiversity conservation in this globally important biodiversity hotspot.

Acknowledgements

This research was carried out as part of a project, "Biodiversity characterization at community level in India using Earth observation data". We thank the Karnataka Forest Department for granting the necessary permissions for project execution. We thank the anonymous reviewers for their insightful comments and valuable suggestions, which greatly enhanced the clarity and quality of this manuscript. This work was supported by the Department of Biotechnology, Ministry of Science and Technology, Govt. India, under grant No. BT/Coord.II/10/02/2016/22.03.2018. The first author thanks the Indian Council of Social Science Research, New Delhi, India, for providing a short-term doctoral fellowship (RFD/Short-Term/2022-23/ENV/ST/66).

CRediT authorship contribution statement

Kanda Naveen Babu: Conceptualization; Methodology; Data curation; Formal analysis; Investigation; Validation; Writing-original draft; Writing–review & editing. Ashaq Ahmad Dar: Methodology; Formal analysis; Validation; Writing-original draft; Writing–review & editing. Kurian Ayushi: Validation; Visualization; Writing–review & editing. Narayanan Ayyappan: Project administration; Funding acquisition; Resources; Supervision; Validation; Writing–review & editing. Narayanaswamy Parthasarathy: Investigation; Supervision; Writing–review & editing.

Data availability statement

Data will be made available from the corresponding authors upon request.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary data

Supplementary data to this article can be found online at https://doi.org/10.1016/j.pld.2025.09.009.