Phylogenetic community patterns suggest Central Indian tropical dry forests are structured by montane climate refuges

We used an eco‐phylogenetic approach to investigate the diversity and assembly patterns of tropical dry forests (TDFs) in Central India. We aimed at informing conservation and restoration practices in these anthropogenically disturbed forests by identifying potential habitats of conservation significance and elements of regional biodiversity most vulnerable to human impact and climate change.


| INTRODUC TI ON
Understanding the mechanisms that shape the distribution of species diversity across space and time is essential for developing effective strategies to manage and conserve regional biodiversity. By examining how species respond to large-scale environmental gradients, we can gain insights into the processes that structure regional biodiversity and identify elements that may be most vulnerable to global change. Elevation gradients often encompass a wide range of abiotic changes across relatively narrow geographical distances and may therefore be particularly suitable for exploring the influence of environmental variables on regional biodiversity (Körner, 2007;Lomolino, 2001).
The complex topography that shapes these abiotic gradients can lead to the persistence of climatically stable high-elevation habitats, which more closely reflect past conditions in a landscape (Birks & Willis, 2008;Hilbert et al., 2007;Stewart & Lister, 2001;Tzedakis et al., 2002). These habitats can act as refugia, sheltering elements of biodiversity during periods of climate instability across evolutionary or geological timescales (Habel & Assmann, 2009;Tzedakis et al., 2002;Weber et al., 2013), contrasting with species refuges that operate over shorter time frames, ranging from minutes to decades (Keppel et al., 2011). Mosaic disturbances may occur during periods of climate instability, potentially contributing to the formation of refuges or refugia in areas that remain isolated from disturbance regimes (Jasinski & Angelstam, 2002;Keppel et al., 2011;Krawchuk et al., 2020;Niklasson et al., 2002). Species' ranges may contract towards refugia during periods of climate instability and expand back into the landscape during periods of stability Fehlberg & Ranker, 2009;Keppel et al., 2011).
Several contemporary community patterns may reflect refugial habitats, including greater species richness and environmental stability compared with the surrounding landscape, the presence of relict taxa and complex topography (see Keppel et al., 2011).
Furthermore, forest refuges may exhibit a greater basal area of trees and seedling recruitment due to a combination of more stable conditions for primary productivity and reduced disturbance. Given the historical role of these habitats in sheltering biodiversity during periods of climate instability, they may be important conservation targets for mitigating future species losses.
Climatically stable habitats may harbour species from both new and old lineages that survived past climate fluctuations (García et al., 2022). In this context, phylogenetic analysis can provide a useful framework for detecting refugial habitats within landscapes.
Older species might belong to clades with greater extinction and lower speciation rates, making them phylogenetically distinct from the regional community Wu et al., 2021).
These species disproportionately contribute to community phylogenetic diversity and structure (Shooner et al., 2018) and can be detected through measures of evolutionary distinctiveness (e.g. García et al., 2022). Thus, phylogenetic analysis can reveal the impact of biogeographical history on contemporary biodiversity gradients and identify unique elements of biodiversity within communities, such as relict taxa that persisted within refugia during historical climate instability (Mastrogianni et al., 2019;Shooner et al., 2018;Wu et al., 2021).
Investigating the phylogenetic divergence of communities can also provide valuable insights into the ecological processes shaping community assembly and biodiversity gradients (Cavender-Bares et al., 2009;Mayfield & Levine, 2010;Webb et al., 2002). These processes include a combination of niche-based, neutral and historical factors (see Cavender-Bares et al., 2009). For instance, species with shared evolutionary histories are likely to have similar traits and ecological requirements, resulting in stronger competition among closely related species and weaker competition between distant relatives (Hutchinson, 1959;MacArthur & Levins, 1967;Webb, 2000;Webb et al., 2002). Assemblages structured by competition are therefore expected to exhibit decreased species relatedness (i.e. increased phylogenetic dispersion), while increased relatedness (i.e. phylogenetic clustering) is expected under environments that select for specific physiological tolerances or traits (Elton, 1946;Simberloff, 1970;Webb et al., 2002;Williams, 1947). However, patterns of phylogenetic clustering may also result from competition if traits that provide greater competitive ability (fitness) within an environment are phylogenetically conserved (Cahill et al., 2008;Davies, 2021;Freckleton & Jetz, 2008;Mayfield & Levine, 2010;Venail et al., 2014). Furthermore, historical biogeography can shape species distributions in ways that lead to nonrandom divergence, even in the absence of strong community structuring processes (Warren et al., 2014). Consequently, identifying processes that structure community assembly from phylogenetic patterns alone may be challenging (Davies, 2021).
Considering axes of beta diversity alongside measures of phylogenetic community structure may improve inferences about the processes structuring community assembly (Fine & Kembel, 2011;Myers et al., 2012;Zhang et al., 2013). Taxonomic beta diversity (i.e. species turnover) measures differences in the species composition of assemblages across spatial, environmental or temporal gradients connectivity between these habitats may provide a range of climatic conditions for species to retreat to or persist within as climates change.

K E Y W O R D S
beta diversity, biodiversity, community ecology, disturbance, elevational gradient, phylogenetic diversity, phylogenetic structure, refugia, species richness, tropical dry forests (Whittaker, 1960), while phylogenetic beta diversity reflects changes in species' phylogenetic relationships across sites (i.e. phylogenetic turnover). These compositional changes can be linked to ecological or evolutionary processes structuring diversity in the regional species pool (Graham & Fine, 2008;Swenson et al., 2011). By combining beta diversity analysis with phylogenetic structure and biodiversity patterns, we may be able to better detect community assembly processes that shape regional biodiversity patterns. For instance, a combination of taxonomic and phylogenetic turnover with a trend towards reduced species richness and increased phylogenetic clustering of assemblages along environmental gradients would suggest species filtering along a given environmental axis. Identifying such processes can be useful for detecting abiotically sensitive species or potential refuges within a landscape.
In this study, we investigate the species richness, stem density, basal area, phylogenetic structure and turnover of tree assemblages across 117 plots located throughout the Central Indian state of Madhya Pradesh (Figure 1). The region is characterized by tropical dry deciduous forests (TDFs), which are common to the Indian subcontinent and account for approximately 35% of total forest cover (Reddy et al., 2015). However, these forests currently face threats from frequent anthropogenic disturbances, including deforestation, wildfires, livestock grazing and the collection of forest products by local people. Despite their importance for regional biodiversity and local livelihoods, the biogeographical, ecological and evolutionary processes structuring these forest systems remain underexplored.
By examining TDF community patterns along environmental axes, we aimed at (1) infering potential processes structuring their assembly, (2) identifying biodiversity gradients in the landscape and (3) detecting habitats that may be most vulnerable to human impact and climate change. Our analysis may inform future research and conservation efforts within the region and is the first to examine Central Indian TDFs through an eco-phylogenetic framework.

| Study location and data collection
We surveyed 117 50 × 50 m (0.25 ha) plots randomly distributed throughout the natural forests of Madhya Pradesh (Figure 1).
Our study location encompasses parts of the lower Vindhyan and Satpura ranges, which are characterized by topographically complex and discontinuous landscape features, running parallel from east to west and divided by the Narmada River Rift Valley. The climate in this region features an extended dry season from October to mid-June, with wet seasons during southwest monsoon rains between June and September. The mean precipitation of the plots in our study ranged from 5 to 388 mm during the driest and wettest months, with a mean annual precipitation of 1235 mm. Mean temperatures varied between 10 and 40°C between the coldest and warmest months, with a mean annual temperature of 25°C F I G U R E 1 Map of 117 (0.25 ha) forest plot locations throughout Madhya Pradesh, India. Unnumbered points indicate single plots, while numbered points indicate multiple plots within the given area. Elevation ranges from ~230 to ~940 m above sea level, with darker shades representing lower elevations.
(https://www.world clim.org). Leaf litter accumulation from deciduous trees is a significant fuel source in Central Indian TDFs, which leads to frequent forest fires during the dry season. Approximately 90% of these fires are of anthropogenic origin, mainly resulting from agricultural practices aimed at improving fodder production for livestock and the collection of nontimber forest products (Saha, 2002).
To facilitate sampling, each of the 117 50 × 50 m (0.25 ha) plots was subdivided into 25 10 × 10 m grids. All woody plants within plots with a height ≥1.3 m were counted and identified to the species level.
Additionally, the stem diameter of each individual was recorded at the height of 1.3 m. Specimens were prepared and identified at the Department of Botany, Dr. Harisingh Gour Central University, Sagar.
GPS coordinates and elevation were recorded through a handheld GPS (Garmin-750), and mean annual temperature and precipitation were assigned to each plot from the WorldClim data set (https:// www.world clim.org) using the sp and raster packages in R (Bivand et al., 2013;Hijmans, 2021;Pebesma & Bivand, 2005). To quantify the environmental stress within each plot, we used Chave's stress index (Chave's E;Chave et al., 2014). Chave's E is a covariable commonly used to determine the diameter-height relationships of tropical trees and is defined by: where TS is temperature seasonality, CWD is climate water deficit (mm/year) and PS is precipitation seasonality as defined in the WorldClim data set. Plots with high Chave's E are considered dry sites with large annual variations in temperature and precipitation.

| Phylogeny reconstruction
To generate a species-level phylogeny of all tree species identified in plots (110 angiosperm species), we used the phylo.maker function of the R package V.PhyloMaker with default arguments (Jin & Qian, 2019; see Appendix S1). This package utilizes the timecalibrated mega tree GBOTB.extended.tre as a backbone for phylogenetic reconstruction, which combines Smith and Brown's (2018) phylogeny of seed plants (GBOTB) and Zanne et al.'s (2014) phylogeny for pteridophytes, with corrections, updates and expansions (Jin & Qian, 2019). All 34 families and 83 genera from our study were present on the backbone tree; however, 20 of the 110 regional species were absent (see Data S1). The phylo.maker function's default scenario (Scenario 3) for generating phylogenetic hypotheses adds phylogenetically uncertain species to the tree as polytomies by binding them to the crown node of their corresponding genus (the node that connects all members of the clade; Jin & Qian, 2019). Such polytomies may bias phylogenetic structure metrics, generally towards type II errors (Molina-Venegas & Roquet, 2014); however, since our analysis relies on relative comparisons of phylogenetic structure among regional assemblages, unresolved phylogenetic relationships may be less impactful.

| Community metrics
To quantify the phylogenetic structure of assemblages, we calculated several indices using the R package picante (Kembel et al., 2010).
We assessed the total phylogenetic diversity within assemblages using Faith's phylogenetic diversity (PD), which sums the total phylogenetic branch lengths connecting all taxa within an assemblage (Faith, 1992). Because PD is highly correlated with species richness (Tucker & Cadotte, 2013), we used the ses.pd function to calculate the standardized effect size of PD (ses.PD), setting the null.model argument to taxa.labels and number of runs to 999 randomisations.
This function compares PD with a null distribution generated by randomly shuffling the tips of the phylogeny with taxa from the regional species pool (defined by all species identified in plots), generating a z-score (Kembel et al., 2010). Assemblages with positive ses.PD values (z-scores) have higher PD values than expected, given their richness.
In addition to ses.PD, we used the standardized effect size of mean pairwise distance (ses.MPD) and mean nearest taxon distance (ses.MNTD) as complementary metrics of plot-level phylogenetic divergence (Kembel et al., 2010;Webb et al., 2002). ses.MPD measures the mean phylogenetic distance between all species in an assemblage, capturing deeper evolutionary divergences in the phylogeny, while ses.MNTD measures the extent of terminal clustering by taking the average of phylogenetic distances between each taxon and its nearest neighbour in the phylogeny (Webb et al., 2002). ses.
MPD and ses.MNTD can therefore provide unique yet complementary information about the phylogenetic structure of assemblages, which can aid in detecting ecological processes operating at different evolutionary depths (Mazel et al., 2016). Abundance-weighted metrics are less sensitive to bias from rare taxa, so we also estimated abundance-weighted ses.MPD and ses.MNTD (ses.MPD ab and ses. MNTD ab ). Each phylogenetic divergence metric (ses.MPD, ses. MPD ab , ses.MNTD and ses.MNTD ab ) is expressed as a standardized effect size relative to a null distribution generated using the same arguments as specified for ses.PD. Positive z-scores indicate community assemblages that are phylogenetically overdispersed relative to null expectations, and negative z-scores indicate phylogenetically clustered assemblages.
To measure evolutionary distinctiveness at the community level, we calculated the standardized mean evolutionary distinctiveness (ses.cED) of each plot, following García et al. (2022). Unlike divergence metrics, which quantify the phylogenetic relatedness of species in an assemblage (i.e. the phylogenetic distance between species), ses.cED measures how phylogenetically isolated species are within a regional phylogeny. Species belonging to more singular lineages with few or no descendants contribute more to ses.cED.
Briefly, we computed the evolutionary distinctiveness of each species using the equal.splits method from the evol.distinct function in the R package picante (Kembel et al., 2010; see Data S2 for ED values). We then summed the ED values of species within each plot (cED) and generated null distributions by randomly shuffling the community matrix while keeping the number of species constant.
We calculated the standardized effect size of community-level ED (ses.cED) as follows: where cED obs is the observed cED of an assemblage, cED rand is the mean of the distribution of the null model and cED sd is the standard deviation of the null model distribution. Assemblages with positive ses.cED values contain a higher proportion of species that are more evolutionarily distinct within the regional phylogeny (i.e. species that belong to more singular lineages) than expected by chance, given their species richness.
Lastly, to better understand the relationship between environmental factors and the distribution of tree species, we measured the stem density and basal area of trees within plots. Analysing how these density measures correlate with environmental axes may be informative for detecting habitats with relatively low disturbance frequencies and/or more favourable abiotic conditions for seedling establishment and productivity.
To examine patterns of taxonomic and phylogenetic turnover, we generated distance matrices from presence-absence transformed community data utilizing the taxonomic pairwise dissimilarity measure Sørensen dissimilarity (Dice, 1945;Sørensen, 1948) and the phylogenetic measure UniFrac (Lozupone et al., 2006;Lozupone & Knight, 2005). We then decomposed both beta diversity matrices into their 'true turnover' (β sim ) and 'nestedness' (β sne ) components using the R package betapart (Baselga & Orme, 2012). β sim represents the component of dissimilarity resulting from species or phylogenetic branch length replacement, while β sne results from the addition or subtraction of species or phylogenetic branch lengths (Baselga & Orme, 2012).

| Statistical analysis
We used linear regression to explore the relationships between community metrics and environmental variables, including elevation, mean annual temperature, mean annual precipitation and Chave's E. Regressions were performed independently between each community-level and environmental variable. However, a central assumption of regression is that observations are independent, and clustered sampling designs in ecological studies could violate these assumptions due to spatial autocorrelation, inflating type I errors (Peres-Neto & Legendre, 2010). To assess whether spatial autocorrelation influenced our regression parameter estimates, we generated Moran's eigenvector maps (MEMs; Dray et al., 2006) using the function quickMEM (Borcard et al., 2018; see Appendix S2). Moran's eigenvector maps translate spatial relationships between data points into explanatory variables, which describe positive spatial correlation at different spatial scales (Peres-Neto & Legendre, 2010). We found two significant MEMs that were added to partial regressions along with explanatory (community data) and predictor variables (environmental data).
We calculated the proportion of total phylogenetic (UniFrac) and taxonomic (Sørensen) beta diversity between assemblages that was explained by true turnover (UniFrac β sim and Sørensen β sim , respectively). Partial Mantel tests-using the mantel function in vegan (Oksanen et al., 2020)-were then used to assess the relationship between taxonomic and phylogenetic turnover with environment, correcting for the geographical distance between plots. We also performed Mantel tests on Sørensen dissimilarity and UniFrac distances to explore the relationship between both metrics. All statistical analyses were conducted using the R statistical programming software, version 4.0.5 (R Core Team, 2021).
Species richness increased with elevation and precipitation, and decreased with temperature and Chave's E (Figure 2 and Table 1).
Although weaker, the relationships between the basal area of assemblages and environmental variables followed similar trends to species richness. Stem density correlated positively with elevation and precipitation, but relationships with other environmental variables were not significant. We identified two significant MEMs (MEM2 and MEM4), indicating positive spatial autocorrelation between plots. However, partial regression, including MEMs, did not qualitatively shift the relationships between species richness, stem density or basal area relationships with environmental variables (see Appendix S2).
Phylogenetic structure metrics revealed that, on average, assemblages were more phylogenetically diverse, evolutionarily distinct and overdispersed than null expectations (see Appendix S2). Partial regressions with MEMs affected the relationships between phylogenetic structure and environment (see Appendix S2). Therefore, we report results controlling for spatial autocorrelation by including MEM2 and MEM4 in the regressions (Figure 3 and Table 2). Although weaker, ses.PD exhibited similar trends to species richness, increasing with elevation and under more abiotically favourable conditions. Metrics of terminal phylogenetic divergence (ses.MNTD and ses.MNTD ab ) trended towards increased overdispersion with higher precipitation and reduced dispersion with Chave's E. Overdispersion increased with elevation; however, the relationship was nonsignificant when MEM2 was included in the regression. Regressions between basal divergence metrics (ses.MPD and ses.MPD ab ) and environment were generally nonsignificant after correcting for spatial autocorrelation (see Appendix S2). The evolutionary distinctiveness of assemblages (ses.cED) trended towards decreasing distinctiveness with increasing elevation and precipitation, and increasing distinctiveness with higher temperature and Chave's E (Figure 3 and Table 2).

| Taxonomic and phylogenetic turnover
The true turnover components of beta diversity (Sørensen β sim and UniFrac β sim ) explained ~98% of the total taxonomic (Sørensen) and phylogenetic (UniFrac) beta diversity between assemblages (see Appendix S4). Taxonomic and phylogenetic beta diversity were strongly correlated (Mantel r = .92, p < .001; β sim Mantel r = .91, p < .001; see Appendix S4). Both taxonomic and phylogenetic beta diversity were associated with changes in elevation, and these relationships remained significant after adjusting for the geographical distance between plots in the partial Mantel tests (Table 3). We detected true turnover (β sim ) along elevation gradients, indicating that dissimilarity along this environmental axis is not solely the result of changes in species richness or PD gradients. However, we were largely unable to detect true turnover components of beta diversity along other environmental axes, likely due to the low statistical power of the Mantel model and more subtle gradients in community composition (see Appendix S4).

| DISCUSS ION
We examined the species richness, stem density, basal area, phylogenetic structure and community turnover of forest assemblages distributed throughout the Central Indian state of Madhya Pradesh.
High-elevation sites-characterized by milder, more abiotically favourable conditions-supported greater species richness, PD, stem density and basal area of trees compared with lower elevations. F I G U R E 2 Species richness, stem density and basal area relationships with gradients in elevation, temperature, precipitation and Chave's E of 117 forest plots within Madhya Pradesh, India. Regression lines and 95% confidence intervals are displayed.

TA B L E 1
Pearson's correlation coefficients between species richness and density metrics with environmental variables of 117 forest plots within Madhya Pradesh, India.
Significant community turnover and a trend towards lower dispersion and species richness at lower elevations suggest environmental constraints filter species and lineages from harsher abiotic environments along the elevational gradient. Low-elevation assemblages were more evolutionarily distinct, indicating that they contain a greater proportion of species belonging to more isolated lineages within the regional phylogeny. These species may belong to clades that experienced higher extinction rates. We hypothesize that these community patterns result from high-elevation habitats acting as refuges for species maladapted to the harsher climatic conditions of low elevations, which were drier, hotter and had increased temperature and precipitation seasonality (Chave's E). To preserve unique elements and areas of high biodiversity in the Central Indian landscape, we recommend forest management strategies that prioritize both high-and low-elevation habitats. Establishing connectivity between both habitats may provide high-elevation refuges for drought-sensitive species to retreat to if warming trends continue.
Seasonal drought can impose significant stress on plants (Choat et al., 2012), which may influence regional diversity patterns by imposing physiological constraints on species (Engelbrecht et al., 2007;Krishnadas et al., 2021;Pockman & Sperry, 2000). Our findings indicate that a climatic stress gradient exists across the elevational range of Madhya Pradesh, characterized by hotter, drier and more seasonal precipitation and temperature conditions at low elevations.

F I G U R E 3
Relationships between phylogenetic structure metrics and gradients in elevation, temperature, precipitation and Chave's E of 117 forest plots within Madhya Pradesh, India. Regression lines and 95% confidence intervals are displayed.
Greater species richness, PD, stem density and basal area in higher elevation sites may indicate that high-elevation habitats provide more favourable growing conditions for tree species, acting as a ref- uge from the harsher climates of low elevations. The increasing basal area of trees with elevation might also reflect increased canopy cover, which could further moderate high-elevation microclimates by limiting light penetration to the forest floor. This reduction in light penetration could lead to reductions in temperature, solar radiation and increased moisture (Elliott et al., 1998;von Arx et al., 2013), leading to more favourable microhabitats for seedling establishment (von Arx et al., 2013). Increased topographical complexity may also enhance the persistence of unique microclimates, for example, by increasing soil moisture during the dry season (Becker et al., 1988;Daws et al., 2002).
Community phylogenetic divergence and turnover support the presence of high-elevation refuges within Madhya Pradesh.
On average, assemblages were phylogenetically overdispersed; however, we found that the magnitude of dispersion decreased as plots became drier and temperature and precipitation seasonality increased. Species lacking trait adaptations necessary to persist under harsher climate conditions may therefore be filtered by the environment across the elevational range, leading to relatively more phylogenetically clustered assemblages at low elevations. Our analyses and previous findings indicate a general trend towards increased phylogenetic clustering as precipitation becomes more limited and seasonality increases in Indian forests (Bose et al., 2019;Divya et al., 2021;Shivaprakash et al., 2018). Significant community turnover combined with reductions in species richness provides further support for species and lineage filtering across the elevational range.
Greater phylogenetic dispersion and species richness at high elevations along the elevational gradient might imply that increased niche partitioning contributes to species coexistence under more climatically stable high-elevation habitats. Community divergence patterns were more distinct when measured through tip-level metrics (e.g. ses.MNTD and ses.MNTD ab ), which could suggest that most of the differences in species' habitat preferences or abiotic tolerances that structure assembly are relatively shallow on the phylogenyreflecting more recent trait adaptations among lineages. However, we note that our interpretations rest on untested assumptions that species' ecologies are phylogenetically conserved, and nonrandom phylogenetic structure may occur in the absence of niche-based processes (Warren et al., 2014). Moreover, the relationship between divergence and elevation was weak and nonsignificant when adjusting for spatial copredictors (i.e. MEM2). Therefore, the contribution of such processes to community assembly along the elevational biodiversity gradient we observed may be only weak, although, a stronger relationship was detected between phylogenetic dispersion and both precipitation and climatic stability (lower Chave's E).
We detected only moderate differences between weighted and unweighted metrics of phylogenetic divergence, indicating that our results are not biased by the presence of rare species in assemblages. Note: β sim explained ~98% of the total Sørensen and UniFrac beta diversity between assemblages (see Appendix S4). Elevation and distance represent the pairwise differences between 117 forest plots within Madhya Pradesh, India.
The persistence of stable high-elevation climates across evolutionary timescales can provide critical refugia for species following regional climate change (Hewitt, 2004;Stewart et al., 2010). The greater evolutionary distinctiveness we detected in lower elevation assemblages and those under more stressful abiotic conditions may reflect the imprint of extinction events that occurred in regions of historical climate instability. Bose et al. (2016Bose et al. ( , 2019 suggested that increased climate instability in more seasonal areas of the Western Ghats, India, occurred following fluctuations in monsoon rains during the Quaternary. Similar climatic changes may also have occurred within the more seasonal environments of Central India. Following such changes, species preadapted to the novel conditions may have been able to persist within more climatically variable environments, while others went extinct or shifted their range distributions towards more suitable habitats.
Such extinction events would lead to a greater signal of increased evolutionary distinctiveness in assemblages within harsher abiotic environments, consistent with our findings. The lower ses.cED we detected in high-elevation assemblages should not be misconstrued as a complete absence of evolutionarily distinct species; rather, it suggests they occur less frequently. We caution, however, that our community-level measure of evolutionary distinctiveness is relative to our regional phylogeny, and results may vary with the scale of analysis.
Additional disturbances include anthropogenic forest fires, which are regarded as one of the principal factors leading to the degradation of TDFs in Madhya Pradesh (Chandra & Bhardwaj, 2015), with previous studies demonstrating a reduction in the taxonomic diversity of forests exposed to recurrent burn events (Kodandapani et al., 2008;Sathya & Jayakumar, 2017;Verma & Jayakumar, 2015). Such fires may select for species with fire-tolerant traits such as thickened bark and resprouting, shifting community composition towards fire-tolerant species (Saha & Howe, 2003;Sathya & Jayakumar, 2017;Verma & Jayakumar, 2015). If traits for fire tolerance are phylogenetically conserved, fire may also alter the phylogenetic structure of communities (Nóbrega et al., 2019), leading to phylogenetically clustered assemblages. Additionally, fire may influence forest canopy structure (Karna et al., 2020), favouring the recruitment of pyrophytic grass species, which increase the susceptibility of forests to future burn events (D'Antonio & Vitousek, 1992). It is likely that the increased aridity, temperature and seasonality we detected in low-elevation habitats could increase the frequency and intensity of forest fires, for example, by influencing fuel load and flammability (Mondal & Sukumar, 2016), while high-elevation habitats remain relatively isolated from such disturbances. Thus, historical increases in fire regimes may be an additional factor, which has left an imprint on the evolutionary distinctiveness and dispersion (i.e. clustering) of low-elevation assemblages.
As regional climates change, species may respond by adapting to novel conditions or shifting their range distributions to regions with climates that more closely reflect the conditions under which they evolved (Hewitt, 2004;Stewart et al., 2010). Our analysis suggests that increased abiotic stress at low elevations throughout Madhya Pradesh has favoured the recruitment of maladapted tree species to wetter, cooler and more stable high-elevation habitats. We propose that many species found in these habitats may belong to lineages that persisted under stable climates during the general establishment of more seasonal climate regimes across India. Our results indicate that high-elevation habitats may provide important refuges for species unable to persist within harsher abiotic environments; however, we emphasize that our inferences into the niche-based processes structuring biodiversity gradients remain a hypothesis that requires further validation.
Underlying the conservation of evolutionary diversity is the notion that phylogenetic distance between species reflects an accumulation of unique traits (morphological and physiological) and ecologies, which contribute to biological variability (Cadotte et al., 2008;Faith, 1992). This variability may be crucial for sustaining ecosystem processes and the adaptive capacity of communities; however, evolutionary diversity has often been overlooked in conservation planning (Mace & Purvis, 2008;Winter et al., 2012). While high-elevation habitats in Madhya Pradesh are species-rich and phylogenetically diverse, low-elevation assemblages support a larger proportion of evolutionarily distinct species. These species may possess unique traits and ecologies that disproportionately contribute to the biodiversity of Central Indian forests; however, if anthropogenic disturbances are indeed more prevalent in low-elevation habitats, these species would also be most vulnerable to their impact. Highand low-elevation assemblages thus contain different components of biodiversity, and practices that prioritize both habitats should be encouraged. Establishing connectedness between seasonal low-and stable high-elevation reserves would additionally offer a range of environmental conditions for species with varying physiological tolerances to persist within or retreat to as regional climates change. Such connectedness may become increasingly important for droughtsensitive species if current warming trends continue, as species may move upwards in elevation to escape increasing drought conditions (Chen et al., 2011;Lenoir & Svenning, 2015).
Through an eco-phylogenetic framework, we describe potential refugial dynamics in a region that has not been influenced by glacial ice sheets. Such work will help extend our understanding of the role that species refuges may have in biodiversity distributions and species persistence. Future work in the region should incorporate ecological niche and trait-based approaches to strengthen inferences about the assembly processes structuring Central Indian TDFs and evaluate the impacts of anthropogenic disturbances, such as grazing and fire, across the abiotic stress gradients we have identified.