2.1 Study Area

The research was carried out under the conceptual framework of the ‘Provisioning of ecosystem services And cultuRAl values in the MOntane tropics’ project, which focused on how best to incorporate and optimise the combination of biodiversity, ecosystem services, and cultural values within natural resource management in Colombia. The study was conducted in Cundinamarca, Boyacá, Meta, and Santander, encompassing the eastern and western slopes of the eastern Andes cordillera, in Colombia. Natural habitats included remnants of Andean Forest, paramo forests (high-elevation forest > 3200 masl), and paramo shrublands and grasslands. Human-transformed habitats were areas that have undergone high local deforestation, where the natural habitat has completely been transformed mainly for cattle production, comprising Andean and paramo pastures utilized mainly for grazing. Habitats have experienced historical anthropogenic disturbances due to their accessibility via roads or footpaths, as confirmed by information provided by local field assistants and landowners. All forests were part of Colombia's protected areas network known as the Sistema Nacional de areas Protegidas (Ministerio de Ambiente y Desarrollo Sostenible 2023).

2.2 Sampling Design

We designed our sampling to represent two types of habitats depending on land transformation to capture the effects of land-use in community turnover. Across the study area, we sampled 18 landscapes composed of natural and transformed habitats (Figure 1). Each landscape was defined as an area of 9 km² where natural habitats were paired with transformed habitats. This approach allows us to have both habitats at approximately the same elevation and reduce the effects of topography and other landscape-scale and climate variables unaccounted for in the models. All natural habitat points were placed at least 50 m from the forest edge or roads keeping a minimum of 172 m distance apart (range = 172.3–2759 m). We set up transformed habitat plots at least 60 m away from the forest edge and at a minimum of 193 m apart between them (range = 193–4225 m; n = 114 plots in 22 pasturelands). Transformed habitats correspond to Andean pastures characterised by having sparse trees and grasses mainly of the Poaceae family, and Paramo pastures were dominated by grasses and lacked frailejones species (Asteraceae: Espeletiinae subtribe), which are a cluster of highly endemic, slow-growing species with high sensitivity to disturbance and narrowly restricted geographical distribution (Cortés et al. 2018).

We sampled 341 plots across a wide elevational range of 1163–3763 m (4°C–28°C and 879–3817 mm per year), spanning ~270 km in distance. Specifically, 206 plots were in natural habitats (148 forest, 48 paramo, and 10 paramo forest), and 135 plots in transformed habitats (90 pastures next to Andean forests and 45 plots in transformed Paramos, n = 341). We sampled birds, dung beetles, and orchids from January 2019 to March 2020, with sampling conducted simultaneously across taxa in time and space. Due to the nature of the cross-species sampling we used points for animal taxa and plots for orchids, and these terms will be used throughout the text. We sampled between 2 and 18 sampling points randomly placed within forests and kept a minimum of 200 m distance apart from each other (with only two cases where points were placed around 170 due to topography). Larger habitats had more plots to ensure representative cover. All sampling plots were placed at least 30 m from the forest edge to reduce the impact from cattle (Parra Sánchez et al. 2016). The sampling area of each sampling point differed between groups (orchids = 10 × 30 m plots; dung beetles = ~ 100 m radius; birds = 100 m radius; see survey explanation below).

2.3 Birds Survey

Bird communities were sampled using repeat-visit point counts at each sampling point (Mills et al. 2022). This methodology allows for detections within an estimated 100-m radius (Gilroy et al. 2014b). Points were visited on four consecutive days for counts of 10-min duration (sampling times between 05:00 and 12:00 pm). Visiting time to each point was randomised to ensure that each point was visited both early and late in the morning during sampling days. Counts were carried out by one person at a time, avoiding heavy rain, dense mist, or high winds, because these conditions reduce visibility, introduce noise to recordings, and alter species behaviour. Each survey was recorded using a recording device to record vocalizations. Species identification was done in the field and subsequent confirmation using recordings against online reference material (www.xeno-canto.org, and recordings deposited in the ‘Coleccion de Sonidos Ambientales’ from Instituto Alexander von Humboldt, Colombia). We omitted species that exhibit high mobility or transient behaviour making them unreliable indicators of local environmental conditions (e.g., non-breeding trans-continental migrants, large raptors and swifts). Non-breeding transcontinental migrants may only be present temporarily, large raptors often have large home ranges that extend beyond the surveyed areas, and swifts spend most of their time in flight, making detection inconsistent.

2.4 Dung Beetle Survey

Baited pitfall traps were used to sample dung beetles (Coleoptera: Scarabaeidae: Scarabaeinae). We placed one trap at each sampling point and insects were collected every 24-h across four sampling days. Traps were set with fresh human faeces as bait and rebaited after 2 days. This type of bait attracts a wide range of dung-feeding species to give the best representative sample of the community (Larsen and Forsyth 2005; Mora-Aguilar et al. 2023). Species determinations were carried out by Diego Martínez and F. Edwards at the Instituto Alexander von Humboldt (IvAH), Colombia. All specimens were deposited in the biological collections at IvAH.

2.5 Orchid Survey

Orchids were sampled in a 10 × 30 m plot at each sampling point, recording both terrestrial and epiphytic orchids within the plots, from the ground up to a height of 2 m (Parra-Sanchez et al. 2023b). All standing tree trunks, fallen tree trunks and branches, vines, lianas, leaves on standing trees or herbaceous plants, palm trees, tree ferns and cycads were sampled. Canopy orchid species could have been sampled from fallen branches, but because we used a standardized sampling approach across all sites, every plot had an equal chance of capturing canopy orchids. Only mature individuals with developed floral or reproductive structures, large clumps of stems, and prominent root systems were recorded. Plant individuals without flowers were translocated to nearby nurseries and subsequently visited for identification. Identification to the species or morphospecies level was carried out using specialized literature and consultation with local experts at the Herbarium VALLE in Palmira, Colombia.

2.6 Habitat and Landscape Predictors

We selected predictors that represent different determinants of community composition (Myers et al. 2000; Rahbek et al. 2019; Parra-Sanchez et al. 2023a). First, we used tree density and canopy cover to capture forest structure. Second, we used the proportion of forest cover in 1 km radius (hereafter forest cover), number of forest fragment per area (hereafter fragmentation), and disturbance to assess the effect of landscape composition and configuration on community turnover. Third, we used elevation and precipitation as natural biogeographic drivers of species richness. Finally, we used geographical distance between sampling points as a proxy linked to dispersal (Loke and Chisholm 2023). Geographical distance was calculated as spherical distance in the modelling process (Latombe et al. 2017).

At a local scale, we measured tree density (trees per plot) and percentage of canopy cover as predictors of forest structure on trees > 10 cm DBH (diameter at breast height). Predictors were quantified at each 10 × 30 m plot. Canopy cover was measured as the percentage of canopy cover in the sampling plot from the 30-m resolution global change forest map (hereafter referred to as the GCF map; Hansen et al. 2013).

At the landscape-scale, we quantified three predictors linked to different ecological processes: (1) percent of forest cover as a metric of habitat available in the landscape (Fahrig 2013); (2) the number of forest patches present in the local landscape as a proxy for fragmentation (Fahrig 2013) and (3) disturbance, calculated as the total of forest pixels that have undergone any change over a span of 18 years (2005–2022) using data from Global Forest Watch (2024). Percent of forest cover and number of forest patches were measured from GCF by 2010 (Hansen et al. 2013), which was converted into a binary forest/non-forest cover map (using the 50% threshold; non-forest <= 50%, and > 50% labelled as forest). Due to the spatial resolution of the GCF, analyses were constrained to forest fragments > 9 ha. Disturbance was defined as a single metric of forest disturbance that included both deforestation and degradation (forest degradation areas defined by Vancutsem et al. 2021). Disturbance included deforested land (permanent conversion from moist forest cover to another land cover), degraded forest (where disturbances were observed over a short period), forest regrowth (post-deforestation recovery), conversion to plantations, conversion to water, and afforestation (regrowth of areas initially classified as non-forest cover). These predictors were obtained at a 1000 m resolution, which also matches the climatic predictors (as the finest scale available), as we did not quantify the scale-of-effects (Jackson and Fahrig 2015).

Finally, elevation (m.a.s.l.) and precipitation (mm per year) were included as environmental predictors involved in driving biogeographical patterns. Mean annual precipitation was extracted from CHELSA (1 km resolution MAP; climatologies at high resolution for the Earth's land surface areas; Karger et al. 2017). Elevation was measured at each sampling point with a GPS garminC60 and validated with the 12.5-m ALosPalsar´s Radiometric terrain model using data from ASF DAAC (2015). Multicollinearity among climatic and landscape predictors was weak using the variance inflation factor (VIF range = 1.15–4.09, Table S1).

2.7 Data Analysis

2.7.2 Land-Use Change has Diverse, Pervasive Effects on Montane Community Structures

We further explore the zeta-diversity decline as the trend across sampling points of rare-to-widespread species in each taxon and habitat type. As the zeta order increases, the zeta values (ζi) decrease, indicating the rate of change in species turnover from rare to widespread species across space (McGeoch et al. 2019). As before, we rescaled zeta values using the equivalent of Simpson's dissimilarity index in the analyses to assess the effects of predictors on species turnover caused by species replacement and not richness difference (Jost 2007).

Zeta decline was tested using two spatial sampling schemes to capture the spatial structure of species distribution at the community level (McGeoch et al. 2019). First, the ‘all combinations’ (ALL) spatial arrangement considers combinations of points independently from their geographical position. Points that are far from each other are therefore less likely to share species than closer sites if isolation by distance structures the ecological communities. By contrast, the ‘nearest neighbours non-directional’ (NON) sampling scheme combines points based on a nearest-neighbour approach based on their spatial coordinates in the computation of the zeta values. The comparison between zeta-diversity declines computed with these two schemes therefore allows to unpack how communities are structured in space, and if the spatial structure differs for rare to widespread species.

We first hypothesized that major biophysical drivers of community turnover in natural habitats (e.g., elevation, precipitation, high forest cover and low fragmentation; Fahrig 2017; Watling et al. 2020; Arroyo-Rodríguez et al. 2020), would be replaced by local disturbance and high fragmentation in transformed habitats. For instance, communities in natural habitats would have a steady turnover along the elevational gradient, but this pattern would be shifted in transformed habitats where it is expected to have the highest turnover at mid-elevations (Figure 2). Our second hypothesis is that across taxa we expect more widespread species in transformed than in natural habitats, which reflect biotic homogenization (Kramer et al. 2023; Montràs-Janer et al. 2024).

2.7.2.1 Software

All analyses were performed in R studio (R Core Team 2021). Data manipulation was done with tidyverse (Wickham et al. 2019), landscape metric calculations with landscapemetrics (Hesselbarth et al. 2019), zeta analyses with zetadiv 1.1.1. (Latombe et al. 2017), and plotting using ggplot2 (Wickham 2016).

3.1 Land-Use Change Alters Drivers and Patterns of Community Turnover

The conversion of natural habitats to pastures disrupted the spatial structure of communities. Communities in natural habitats had 71.7%, 86.5%, and 166.6% higher species richness than communities in transformed habitats for birds, dung beetles, and orchids, respectively (Figure 3). These effects were statistically significant across taxa (all p-values < 0.001; Birds, beta = −0.60, CI [−0.77, −0.43]; Dung beetles, beta = −2.04, CI [−2.74, −1.41]; and orchids, beta = −0.99, CI [−1.45, −0.52]; Table S2). The explanatory power across models was robust (conditional R² = 0.62–0.79; marginal R² = 0.22–0.43).

Land-use change modified community composition in two ways. First, in natural habitats, elevation was the prevalent force shaping community turnover, but habitat transformation rewired this pattern (ζ_2-7; average splines over 50 replicates; Figure 4; Figures S1–S12; Appendix S1 MS-GDM model output). Second, transforming natural habitats into pastures changed the scale at which biophysical factors influence community structure.

For bird communities, we found that ζ₂ in natural habitats (beta-diversity) shifted from a consistent turnover of species along the whole elevation gradient to a rapid loss of species at low elevations in pasturelands (1100–1600 m; Pearson's R² = 0.45–0.57 in ζ₂; Figure 4a,b). For widespread bird species (ζ_{> 4}), turnover was high at lower (< 1500 m) and high elevation (> 2900 m), but low at intermediate elevations, as shown by the steep slope of the splines at low and high values and the plateau at intermediate values in natural habitats (R² = 0.19–0.52 in ζ₇; Figure 4b). This pattern changed in transformed habitats, where the highest turnover happens at low elevations (< 1600 m) and mostly disappeared at high elevations (< 3000 m), meaning low turnover of widespread species (steady curves; Figure 4d). In addition, the variance explained by MS-GDM decreased as the zeta order increased in transformed habitats, in contrast to natural habitats, consistent with the failure to identify environmental drivers of species turnover for widespread species (Supporting Information S1; Figure S13).

In natural habitats, ζ₂ of dung beetles (beta diversity, Figure 3) was steady at mid-elevations (1500–2500 m), then peaked at high-elevations (> 2500 m), but in transformed habitats turnover was low at low elevations (1100–1600 m), then switch to a rapid turnover at mid-elevation (~1600 m; Pearson's R² = 26–0.34 in ζ₂). For high orders (ζ₄), widespread dung beetle species across natural habitats increased at the same elevation (> 2500 m), but turnover remains steady in transformed habitats (Pearson's R² = 0.20–0.22 in ζ₅ showing low steady slopes ζ_{> 3}). Orchid community had low turnover at high elevation in transformed habitats (1150– ~1700 m), whereas in natural habitats, elevation had a consistent effect on turnover across the whole gradient (steady slope across the entire gradient at ζ₂; Figure 4c,d; Pearson's R² = 0.05; Figure S13). MS-GDM models showed good deviance explained across taxa (Figure S13).

Second, in natural habitats, local-scale canopy cover was the second most important determinant of bird and dung beetle communities (but first in widespread dung beetles ζ₄), causing a sharp rise in turnover between 10% and 50% cover, but minimal impact on community turnover above 50% canopy cover (Figure 4a,c,e,g). By contrast, in transformed habitats, geographical distance and landscape-scale forest cover emerged as strong forces of community turnover after elevation (high distance ~10 km, and mid-to-high forest cover 70%–100% at 1 km resolution).

3.2 Land-Use Change Has Diverse Effects on Montane Community Structures

We found patterns of biotic homogenization on birds and subtractive biotic heterogenization of dung beetles driven by changing patterns of widespread species post-deforestation, whilst we found a rapid decline in zeta towards zero in orchid communities demonstrates strongly localised species in both habitats (Table S3). Zeta-decline models show that neighbouring sites (sampling of nearest-neighbours -NON) were more likely to share species compared to random sites, suggesting that the structure of ecological communities is driven by isolation by distance (sampling across all-combinations -ALL; Figure 5, Table S3).

Our models revealed a larger set of widespread bird species in transformed habitats showing signs of additive homogenization (natural ~10% vs. transformed~20% of the communities, by ζ₁₀ in NON sampling, Figure 5a), with concurrent loss of smaller-ranged species from natural habitats (subtractive homogenization). In contrast, widespread dung beetles dominate natural habitats, but habitat transformation increases dung beetle community turnover towards a lower proportion of widespread species revealing subtractive heterogenization (natural ~20% vs. transformed ~5% NON, Figure 5b). This suggests that closer communities in natural habitats are more similar to each other (isolation by distance) in dung beetle communities than across bird and orchid communities. However, land-use change disrupted the effects of this spatial structure by reducing the prevalence of widespread dung beetles in natural habitats. Finally, orchid communities were mostly composed of rare species without spatial structure (ζ₂, Figure 5c) and changes in composition were due to the loss of species richness (Figures 2c and 5).

4.1 Land-Use Change Alters Drivers and Patterns of Community Turnover

Once natural habitats are converted to pasturelands, entire communities are reshaped as many species are unable to tolerate conditions in the transformed habitat. Our results align with the idea that land-use change can impact the effects of biogeographical forces such as elevation that govern community composition and structure (Peters et al. 2019). However, the novelty of our study lies in detecting specific elevations where land-use change exerts profound impacts on ecological patterns across the rare-to-widespread species spectrum.

Specifically, at low elevations in pasturelands (1100— ~1600 m), the rapid decline in bird species underscores the difficulty many species face in coping with transformed habitats. In the Andes, the highest bird diversity occurs at lower elevations, where land-use transformation intensity is high (Clark et al. 2012). Conversely, at mid-to-high elevations (~2500–3700 m), the low turnover of widespread bird species signals a group of dominant species can persist in transformed environments. This stands in stark contrast to natural habitats, where widespread bird species show high turnover in the opposite direction at higher elevations (~2500–3700 m). Our findings align with the idea that high-elevation bird communities might be preadapted or filtered for species that can survive when levels of forest cover in the landscape are naturally low (Betts et al. 2019; Mills et al. 2022). High-elevation landscapes in natural habitats present high turnover of widespread species, which might indicate a degree of species specialization to conditions such as sparse tree distribution, with encroached growth, and big canopy gaps (Bader et al. 2007; Betts et al. 2019). Our results contrast with patterns in the Australian avifauna where the highest shift from rare to widespread species occurs at low elevations. Instead, in the Andes, elevation influenced community turnover along the whole gradient (McGeoch et al. 2019). These differences may stem from several biogeographic factors such as the far larger elevational range in our study area (0- ~ 500 m Australia vs. 1100–3700 m this study) and the richer endemism of Andean bird communities (Myers et al. 2000).

We found that natural habitats along the whole elevation gradient retain more dung beetles and more diverse communities, which are largely lost post-deforestation. This might be explained by the high dependency of dung beetles to good habitat conditions such as low sun-exposure and high dung availability in natural habitats (Davis et al. 2001; Horgan 2005; Nichols et al. 2009). For instance, canopy cover is high in natural habitats which influences dung beetle community assembly by modulating local microclimates on which widespread dung beetles might rely (Williamson et al. 2022). Experimental gaps reduced the abundance of a key forest species (Anoplotrupes stercorosus), without compensatory recruitment of open land species (Staab et al. 2022). In contrast, transformed habitats present lower species richness, with fewer widespread species. This pattern might be explained by the availability and quality of dungs which are either present but poisonous for dung beetles or scarce. When cattle are present, farmers often use ivermectin that intoxicate dung beetle species potentially provoking a cascading effect in turnover (Verdú et al. 2015). Furthermore, many pasturelands have been abandoned in high Andean elevations (Clark et al. 2012), suggesting the availability of dung might be low or dependent on (often overhunted) mammals from nearby natural habitats (Nichols et al. 2009). To deepen our understanding of whether deterministic processes are occurring, traits can be used to reveal the mechanisms involved in community turnover (Edwards et al. 2021).

Orchid communities are composed of highly specialized species that become locally extinct after habitat transformation and orchid conservation should therefore focus on species richness (no signal of widespread species; Parra-Sanchez et al. 2023a). Thus, land-use drives species loss and alters the patterns of orchid turnover. Orchid communities have an inherent high turnover across elevation due to species rarity, but transformation of habitats at high elevations (> 3000 m) leads to complete local extirpation of species (Parra-Sanchez et al. 2024). Our results underscore the dependency of epiphytic orchid species (~92% of orchids in this study) upon good-quality habitat structure (Gentry and Dodson 1987; Zotz 2016; Reid et al. 2016), and align with the notion that orchid communities might require connectivity across landscapes (Janzen et al. 2020).

Following elevation, the second largest effect we found is that transforming natural habitats into pastures appears to change the scale at which different biophysical factors determine community structure, shifting from local to landscape-scale. In natural habitats, local-scale canopy cover is the second most important driver of bird and dung beetle turnover for rare and widespread species (but the first one in widespread dung beetles ζ₄, Figure 3). By contrast, in transformed habitats, geographical distance and landscape-scale forest cover emerge as a strong force of community turnover after elevation. As a result, anthropogenic habitat transformation restructures biodiversity in such a way that communities may increasingly depend on dispersal ability to move across the landscape. One potential strategy to counter this trend is the retention and expansion of high forest cover (> 40%) to ensure sufficient landscape connectivity, enabling species dispersal and reducing extinction risks, especially in response to the hostile conditions of pasturelands (Banks-Leite et al. 2014; Arroyo-Rodríguez et al. 2020). Thus, the reduced habitat quality and connectivity in transformed landscapes make dispersal highly relevant for communities.

4.2 Land-Use Drives Both Biotic Homogenization and Heterogenization

The zeta diversity framework allows the decoupling of land-use change effects on the spectrum of rare-to-widespread species to detect co-occurring signals of both biotic homogenization (bird communities) and biotic heterogenization (dung beetles) as a product of habitat transformation (McKinney and Lockwood 1999; Newbold et al. 2018). However, orchid communities lack any set of ‘winner’ species or increased differentiation after deforestation. These patterns are obscured when considering only beta-diversity, underscoring the value of using zeta diversity to understand community-level responses to habitat changes.

Biotic homogenization of bird communities in transformed habitats might be explained by the loss of sensitive species, with spatially proximal bird communities having twice the amount of more widespread species than natural habitats potentially driven by environmental filtering increasing more disturbance-tolerant ‘widespread’ species (McKinney and Lockwood 1999; Tabarelli et al. 2012; Püttker et al. 2015; Newbold et al. 2018). This is also supported by the decrease in explanatory power of MS-GDM at high zeta orders in transformed habitats, suggesting that communities of widespread bird species are more stochastically structured. Whilst some studies have concluded that homogenization in bird communities was driven by the changing distributional patterns of species with large ranges (Gilroy et al. 2014b; Newbold et al. 2018), we detected concurrent decrease in α diversity. At low elevations, local extirpation of rare species led to overall homogenisation, and this is likely to be of interest to conservationists, given the loss of species that were locally rare and did not colonize new areas within the locale to offset extirpations. Decreased dissimilarity contributes to large-scale homogenization, which poses the idea that some species are colonizing new elevations (e.g., eastern meadowlark Sturnella magna, Andean lapwing Vanellus resplendens) and potentially displacing natural occurring widespread species (Gilroy et al. 2014b). This highlights the importance of locally widespread species in driving regional patterns of community structure (Lennon et al. 2004; McGeoch et al. 2019).

In contrast, land-use change in dung beetles provokes biotic heterogenization, primarily driven by the loss (subtraction) of widespread beetle species in natural habitats. Our findings show that natural habitats retain four times more widespread dung beetle species than pasturelands. Likewise, variance explained declines more steeply in transformed habitats, suggesting highly fragmented communities, with fewer species consistently shared across sites. Dung beetles exhibit rapid species turnover at lower zeta orders in pastures, indicating a more immediate response to habitat loss. However, dung beetles display a progressive loss of community structure rather than a sharp threshold effect. This pattern is likely a product of the historical connectivity of the montane biomes in Colombia's eastern Cordillera which facilitated species dispersal (Escobar et al. 2005). However, widespread species (e.g., Deltochilum hypponum, a telecoprid, copronecrophagous beetle found across 36 plots in natural but absent in transformed habitats) are highly sensitive to forest loss due to soil compaction in pastures and can quickly experience range declines (González FA et al. 2009; Sweeney and Jarzyna 2022). In comparison, pasture species such as Andinocopris achamas, Uroxys coarctatus, and Onthophagus curvicornis are relatively rare in natural habitats. This may result from species adapted to exploit spatially contiguous natural conditions in which connectivity is disrupted by human disturbance (Noriega et al. 2021). For instance, the loss of canopy cover increases temperatures that reduce dung beetle diversity (Halffter and Arellano 2002). Additionally, dung and carrion, critical resources for dung beetles, are more abundant in connected natural habitats where mammal movement is less restricted (Horgan 2005; Nichols et al. 2009).

Finally, we provide evidence of orchid communities' high sensitivity to land-use change as they are the most negatively impacted taxon in our study. We also support the notion of a unique set of species structuring Andean orchid communities, with spatial rarity at their core (Parra-Sanchez et al. 2023b), which explains the similarity in patterns of species turnover across the whole range of rare to widespread species. Although turnover is high in both natural and transformed habitats, pastures are hostile for over 90% of species likely due to the lack of large old-growth trees for forest-dependant epiphytes impacting stochastic processes of orchid establishment (Parra-Sanchez and Banks-Leite 2022; Ospina-Calderón et al. 2023). The transformed habitat might act as a barrier due to altered microclimatic conditions, reduced tree density, and fragmented landscapes, which may limit the dispersal of epiphytic orchids (Parra-Sanchez et al. 2023a). A niche-based process may also be acting at a finer spatial scale. For instance, orchid seeds cannot tolerate drought in pastures (Olaya-Arenas et al. 2011; Mondragon et al. 2014; Ospina-Calderón et al. 2023) and orchids have very specific interactions with pollinators (Kindlmann et al. 2014; Ackerman et al. 2023), and the mycorrhizal fungi necessary for germination and survival that might be extirpated in isolated trees in pasturelands (Mccormick and Jacquemyn 2014; Romero-Salazar et al. 2022).