Linear infrastructure drives biotic homogenization among bird species of a tropical dry forest

Linear infrastructures (LIs) such as roads, railroads, and powerlines are expanding rapidly around the globe. While most future developments are projected to take place in tropical regions, available information on impacts of LIs is biased towards single species studies of solely road impacts in temperate regions. Therefore, we investigated impacts of three types of LIs (road, railroad, and powerline) on the bird community of a tropical dry forest. Point-count surveys to record avian richness and abundance were conducted at 80 plots that were spatially stratified to include sites proximate to all possible LI combinations. Five measures of vegetation structure were collected at each plot as well. We then assessed the relationship between the bird community (i.e., richness, abundance, composition) and distance to each LI type while accounting for variation in vegetation structure. Species richness and abundance both declined significantly (25% and 20%, respectively) from edge habitat next to railroad to interior forest plots, while community composition was significantly altered by the distance to all three LIs. Road and railroad (both forms of dynamic infrastructure with moving vehicles) had similar effects on the bird community that contrasted with those of powerline (a type of static infrastructure). The resulting ordination reveled that Sri Lankan endemics are significantly disfavored by LI proximity, while the three species most often found proximal to LIs all have naturalized populations across the world. Our results emphasize that LI drive biotic homogenization by favoring generalist species at the expense of unique elements of the biota.


INTRODUCTION Types of linear infrastructure and their effects
Continuous expansion of the already extensive linear infrastructure (LI) network affect ecosystems around the world.More than 40% of the earth's total terrestrial surface, including numerous ecosystems and species, is located within 5 km of a road (Ibisch et al., 2017) and yet proposals for continued expansion proliferate.In the contiguous United States, more than 80% of forest land is already located within 1 km of a road (Riitters and Wickham, 2003).Moreover, scientists predict a 60% global increase in road and rail networks by 2050 (Dulac, 2013).Utility easements and industrial corridors, such as powerlines, pipelines, and seismic exploration lines, are also expanding at a rapid rate (van der Ree et al., 2015).All these different LI types are often used in tandem to provide access and energy, supporting a region's economic growth and development while traversing long distances through natural landscapes.
Due to their unique designs and functions, different LIs have the potential to differentially affect wildlife (Borda-de-Água et al., 2017;Forman, 2000).For instance, transport corridors, such as roads and railroads, exert a dynamic disturbance on the environment due to vehicular movement resulting in animalvehicle collisions and emission of pollutants at the site.Even among LIs that exert dynamic impacts, effects may differ based on frequency and speed of vehicular movement (Kušta et al., 2015;Waller and Servheen, 2005).Comparatively, the impact of LIs such as powerlines and pipelines are more static (no moving objects along the linear track).In this case, it is the presence of the infrastructure that affects wildlife by introducing artificial structures to the natural environment that could interfere with ecological interactions or animal behavior such as predation, perching, nesting, etc., (Avian Powerline Committee (APLIC), 2006; Infante and Peris, 2003).
Infrastructures that differ in their dynamic vs. static nature may still alter the physical environment in similar ways.For example, railroads and powerlines both create narrower corridors than multi-lane roads (Borda-de-Água et al., 2017).This can favor certain species, such as cottontail rabbits (Sylvilagus transitionalis), an early successional habitat specialist of conservation concern, that use such narrow corridors for movement among their habitat patches (Fenderson et al., 2014).Additionally, all linear corridors cause habitat fragmentation and thus create edge effects (Morelli et al., 2014;van der Ree et al., 2015).For example, roads can affect fauna, flora, and the abiotic environment as far as 1 -1.5 km (Forman and Deblinger, 2000;Shanley and Pyare, 2011).However, we know little about the extent of edge effects of different types of LIs.Understanding the varying types and scales of impact from LI enables informed decision making, leading towards sustainable development.

Gaps in LI research
There are several considerable gaps in information related to impacts of LI.These include biases with regard to the types of LI, the types of impacts, and the geographic regions that are typically studied.First, roads are the most extensively studied LI type around the world (Borda- de-Água et al., 2017;van der Ree et al., 2015).Roads cause direct mortality, habitat alteration, habitat fragmentation, and create barriers (Forman and Alexander, 1998).Although other LI types could potentially cause similar impacts to roads only a limited number of studies have investigated these impacts (Forman et al., 2003;van der Ree et al., 2015).Second, available literature on impacts of LI is skewed toward reporting direct mortality, while fewer studies report LI affecting biodiversity via other pathways such as changing animal behavior (Denoël et al., 2010;Halle and Stenseth, 2000;Mazerolle et al., 2005), spreading invasive species (Brown et al., 2006;Flory andClay, 2006, 2009) and altering species interactions (Davies et al., 2012;Ibáñezálamo et al., 2015;Planillo et al., 2018).Third, the majority of scientific work conducted on LI has occurred in temperate ecosystems (Monge-Nájera, 2018;van der Ree et al., 2015).However, given the substantial differences in faunal and floral assemblages between tropical and temperate regions, studies from temperate systems may not accurately predict LI impacts in all circumstances.For instance, tropical regions harbour a great deal of endemic species, which are prone to replacement by generalist species driving functional-homogenization (Clavel et al., 2011).Conservation and safety issues encountered due to LI also differ between the two climatic zones.For example, in temperate regions, colliding with large mammals is a greater issue for both humans and animals (Langbein et al., 2011;Shilling, 2016), whereas in tropical regions roadkills mostly consist of amphibians and reptiles (Karunarathna et al., 2013;Teixeira et al., 2013).In addition, since many tropical regions are in developing nations, there is a pressing need for infrastructure development.The infrastructure expansion mentioned earlier expects to add 25 million kilometers of new paved roads globally by 2050, and 90% of this new infrastructure is going to be built in tropical regions (Laurance et al., 2015).Therefore, it is crucial to expand knowledge of the impacts of LI in tropical regions.

Birds as an indicator taxon
All four terrestrial vertebrate classes are directly affected by habitat loss and disturbance due to construction and operation of LI (Forman et al., 2003;Forman and Deblinger, 2000).However, due to greater mobility (and therefore the ability to respond promptly to a given disturbance), as well as relatively high detectability, birds are ideal indicators to study the effects of LI (Marzluff et al., 2001).For a given group of animals to be used as an indicator, they first should respond to the impact of interest.
Birds are vocal animals and hence are likely to be disturbed by the noise created by both roads and railroads (Brumm, 2004;R Reijnen et al., 1995;Reijnen et al., 1996).In addition, certain bird species prefer to travel under forest cover rather than cross open areas, even when the forested route is substantially longer than the short cut in the open (Belisle and Desrochers, 2002).Therefore, gaps created in their habitat are likely to affect their movement patterns, possibly acting as a barrier for certain species (Gillies et al., 2011;Harris and Reed, 2002).Second, the indicator group should be common enough at the study site to provide a robust sample size.At the site selected for the present study (Figure 1), birds are the most species-rich vertebrate group (DWC, 2008).
Figure 1 Map of the study area showing the locations of the 80 survey points.The buffer region from which the first 70 survey points were selected is outlined in black, with the road (red line), railroad (orange line), and powerline (blue line) depicted within.The ten remaining survey points are to the south located in interior forest.
Therefore, to fill this crucial gap in information on impacts of different LI types in tropical regions, the present study attempts to quantify the spatial scale and relative magnitude of impacts from roads, railroads, and powerlines on the bird community in a tropical dry forest in Sri Lanka.The impact is quantified by looking at changes in bird species richness, abundance, and community composition in relation to distance from the selected LIs.

Study site
The study site was located in the Polonnaruwa District, North Central Province, Sri Lanka (8.07°N, 80.88°E).We selected this site for three main reasons.First, according to the National Physical Plan (National Physical Planning Department, 2019), this area will become the hub for four major urban centers by 2050.As a result, the area is scheduled for substantial expansions of its infrastructure network.
Second, the forest ecosystems present in this region are part of the largest network of dry zone protected areas in the country.This provides an opportunity to investigate the current impact of LI networks on dry forest biodiversity so that these relationships can be considered during the planned infrastructure network expansion.Third, the selected site has all three LIs of interest (i.e., road, railroad, and powerline) in close proximity to each other within the same forested landscape.
The landscape consists of two nationally protected areas, Thalapathkanda Forest Reserve and Minneriya National Park.The main road traversing the site (AA011-Maradankadawela-Habarana-Thirukkondaiadimadu Road) connects two cities, Habarana and Polonnaruwa.The study road carries an average of ~10,800 vehicles per 24 hr, of which 28% are heavy vehicles (buses or trucks), 55% are passenger cars and vans, 14% are two-wheeled motor vehicles, and 3% are safari jeeps (Hewavithana et al., unpublished data).The railroad runs from Gal Oya to Polonnaruwa with a daily traffic volume of six trains (from station records).The powerline (132kV) runs from Habarana to the Polonnaruwa 132/33 kV Grid substation.These LIs have all been in place for >20 years.Hence, we assume that the fauna and flora at the site have had adequate time to respond to most of the infrastructure's effects.
The major habitat type found in this area is Tropical Dry-Mixed Evergreen Forest, the largest natural terrestrial habitat type in the country (Gunatilleke et al., 2008).This habitat experiences a tropical dry hot monsoon climate characterized by a bimodal pattern of rainfall with relatively uniform high temperatures throughout the year (mean temperature of 28 o C).The main rainy season occurs between October and January coinciding with the north-east monsoon.The area experience desiccating winds between May and August during the long, hot south-west monsoon.The average altitude of the monitored locations is 187 m above sea level.

Bird surveys
To determine the distribution of avian biodiversity across the landscape, we selected 80 survey points.
The locations for these survey points were selected by marking the footprint of each LI through the forested landscape using ArcGIS 10.5.1 and then generating a 1 km buffer around all three LIs and merging them into a single region.We then selected 70 random points at least 100 m apart (mean distance to nearest neighboring survey point = 386 m) (Ralph et al., 1995) within the resulting buffered region such that ten points fell within each of the seven distance categories outlined in Table 1.Ten additional survey points were selected within the study area >1 km from all of the LIs.We used the point-count method to determine the abundance of birds at each survey site (Bibby and Burgess, 1993).All 80 survey points were surveyed three times over the course of the study: dry season (Jul-Aug 2019), wet season (Dec 2019-Jan 2020), and dry season (July-Aug 2020).shotgun microphone and a TascamDR-40) and later played back to confirm the recorded species.

Habitat variables
A total of five habitat variables (listed in

Variable Method
Canopy cover 5 photographs pointed directly overhead were taken at 1 m height.Images were analyzed for percent canopy cover using ImageJ software (Schneider et al., 2012) and the mean of the five measurements was recorded.
Shrub cover 5 photographs were taken at 1 m height and oriented horizontally with a 1 x 1 m white board in the background starting from ground level.Images were processed using ImageJ software (Schneider et al., 2012) to determine percent of the white board hidden by shrub cover.The mean of the five measurements was recorded.

Herb cover
Five 1 x 1 m quadrats were randomly placed within a 10 m radius from each survey point.The number of 20 x 20 cm sub-quadrats with herb cover was counted, and the mean of the five measurements was recorded as a percentage of total cover.

Diameter at Breast
Height (DBH) Number of trees with DBH >10 cm was counted within a 10 x 10 m plot centered on the survey point.

Plant species richness
The number of tree species with DBH >10 cm within the 10 x 10 m plot centered on the survey point was recorded.

Analysis:
1. Impact on species richness and abundance Model selection (AICc) was used to select the most parsimonious model relating bird species richness and abundance to distance from each infrastructure type.The full model included distance from each infrastructure type as well as all of the measured vegetation variables.

Impact on bird community composition
We employed a distance-based redundancy analysis (db-RDA) to examine the relationship between the measured vegetation variables, distance to each infrastructure type, and avian community composition at each survey point.Prior to analysis, we removed singleton species from the community matrix.We used Gower's distance to create the similarity/dissimilarity matrix between survey points.We then employed model selection (via the ordistep function in the vegan package; Oksanen et al., 2019) to identify which distance to infrastructure and which vegetation variables best predict avian community composition.
Using the resulting ordination, we created two summary axes: 1) an axis pointing toward survey points farthest from all three infrastructure types, and 2) an axis pointing toward survey points far from both road and railroad, but close to powerline.We then reflected each species score on both of these axes to determine their position along these important gradients.Species that fell more than halfway from the origin along either side of the first axis or the positive side of the second axis were identified as those most influenced by distance to LI or LI combinations.Further, a one-way ANOVA was conducted to determine if there was a difference between endemics and non-endemics in terms of their placement along the first summary axis generated above.All statistical analyses were performed in R ver 3.6.0(R Core Team, 2017) or JMP Pro 15 (SAS Institute Inc., 2019).Warakagoda et al., (2012) was consulted to categorize recorded species according to their relative forest dependency.The following keywords/phrases were used to determine if a given species prefers interior forest canopy: roosts high in trees, shy, deep forest, hidden in forest trees, usually nests at some height, and skulks in undergrowth.

RESULTS
Across the 80 survey points and three survey periods, we recorded a total of 95 bird species belonging to 42 families, including 11 migratory species (Table S1).Among these species, there were eight

Species richness and abundance are higher close to railroads
Both species richness and abundance were higher closer to railroads (p = 0.001 and p = 0.01, respectively).Mean species richness ranged from 43 species within 100 m of the railroad to 32 species >1 km from the railroad.Further, mean abundance decreased by 20% between the same distance categories.
Vegetation, road, and powerline had negligible impacts on species richness and abundance.

Linear infrastructure homogenizes avian communities
Avian community composition varied significantly with distance to all three LIs (road: p = 0.001, railroad: p = 0.001, powerline: p = 0.003).The five vegetation variables again did not have a significant effect on community composition.Of the 95 species recorded, we excluded three from the db-RDA because they were only encountered once during the entire survey period: Banded Bay Cuckoo (Cacomantis sonneratii), Brown Wood-owl (Strix leptogrammica), and Grey-breasted Prinia (Prinia hodgsonii).
The resulting ordination revealed that the road and railroad shape avian communities in similar ways (vectors largely parallel), in contrast to the powerline (vector orthogonal) (Fig. 2).Species' responses to LI fell into three main categories: 1) negatively affected by all three LIs, 2) positively affected by all three LIs, and 3) negatively affected by road and railroad while responding positively to the powerline (Table 3).

*endemic
Ten species were identified as negatively responding to all LIs whereas three species were noted as positively responding.Species that responded negatively to the presence of infrastructure tended to be those dependent on interior forest (7 out of 10), while those that responded positively to infrastructure tended to be generalists (3 out of 3).The ten species that were identified as notably avoiding all three LIs also included five endemics.Lanka Junglefowl (Gallus lafayetii), and Small Minivet (Pericrocotus cinnamomeus).In contrast, the three generalist species identified as most frequently occurring near all LI types were the Red-vented Bulbul (Pycnonotus cafer), White-rumped Munia (Lonchura striata), and Rose-ringed Parakeet (Psittacula krameri).The one-way ANOVA revealed that endemic and non-endemic species differed significantly in terms of their placement along the interior to edge axis (P = 0.002), with endemics being more likely to occur in interior habitat while non-endemics were more likely to be found in proximity to LI.One species, the White-browed Bulbul (Pycnonotus luteolus), was identified as preferring habitats close to the powerline while avoiding plots near the road and railroad.

DISCUSSION
Our findings demonstrate substantial variation in bird species' response to LIs.While overall species richness and abundance were highest close to LIs, particularly railroad, not all species exhibited this response.In particular, a set of species that preferred interior forest habitat, including five of the eight identified Sri Lankan endemics, were negatively affected by LIs.We also identified differential effects of roads and railroads (dynamic impacts) vs. powerline (static impact) on avian community composition.
Three previous studies have examined the impact of railroads on bird communities and all three found similar positive impacts on species richness and abundance.All three studies were conducted in Poland, two in a pine forest habitat (Wiacek et al., 2015;Wiącek et al., 2020) and the other in an agricultural landscape (Kajzer-bonk et al., 2019).It is interesting that the same pattern holds in a tropical dry-mixed evergreen forest as was observed in these temperate habitats.These observations could be attributed to the greater availability of additional resources at the habitat edge with a relatively low risk for accessing them (i.e., lower traffic at railroads than at roads, resulting in a lower collision risk and disturbance; Borda- de-Água et al., 2017;Rytwinski and Fahrig, 2012;Wiacek et al., 2015).Railroads attract birds because railroad-associated infrastructure act as excellent song, look-out, or resting posts safe from the attention of predators (Benítez-López et al., 2010;Morelli et al., 2014), and open terrain along the tracks provides good foraging habitats (Morelli et al., 2014;Terraube et al., 2016).
In contrast to the positive effects of railroads, previous studies of road effects on avian communities have found lower species richness and abundance in the vicinity of the road edge compared to interior forest plots (Arévalo and Newhard, 2011;Polak et al., 2013).While we did not see a negative effect of roads, we also did not observe a positive effect as we did for railroad.While some of the previous studies predict road noise to be the main driver for such negative trends (Arévalo and Newhard, 2011;Rien Reijnen et al., 1995), others point to road mortality as the main cause (Summers et al., 2011).Again, most of these studies have occurred in temperate regions (Hewavithana et al., unpublished data) and hence may not be the case for tropical regions.However, a comparable study conducted in a dry evergreen forest landscape in northeastern Thailand reports that both species richness and abundance increased with distance from the edge, but this relationship was only marginally significant for abundance (Khamcha et al., 2018).This variance in species response to LI development emphasizes the importance of not green lighting LI development activities in tropical regions just because data is limited.
We discovered a wide divergence in how the avian community responds to dynamic (roads and railroads) vs. static impacts (powerlines) as illustrated by their largely orthogonal vectors in the ordination.Regular vehicular movement on roads and railroads generate an array of dynamic impacts on the environment including animal-vehicle collisions, disturbance from noise and light, and pollution from vehicle emissions (Borda- de-Água et al., 2017;Benjamin Dorsey et al., 2015;van der Ree et al., 2015).
Therefore, roads and railroads affect animal populations by causing direct mortality (Gilhooly et al., 2019) in addition to triggering certain behavioral responses such as avoiding noisy or night-lit habitats (Jackson, 2000;Ruiz-capillas et al., 2013) and altering predation rate at edges (Pescador and Peris, 2007;Planillo et al., 2018).Further, the nature of the verge created by a road or railroad is different from that created by a powerline.For roads and railroads, it is necessary to keep the ground of the linear corridor either paved or completely cleared to ensure safe transportation.However, for powerlines it is only necessary to maintain the canopy at a certain height below the cables to ensure uninterrupted power transmission.Therefore, the linear corridor of a powerline is frequently left as a strip of grass or shrub.
These differential impacts may explain why the avian community present in the Sri Lankan dry forest responds differently to divergent LIs.
One species with a particularly distinct response to static vs. dynamic LI was the White-browed Bulbul.
During the present study, it was observed frequently in plots close to the powerline while avoiding plots near the road and railroad.This observation matches this species' habitat preference (i.e., scrub, forest understory, and shrubby gardens; Warakagoda et al., 2012), which are representative of the dense, short canopy habitat maintained in powerline corridors.Other studies that have investigated the impact of similar transmission corridors on birds report that these habitats have an immense conservation importance for scrub/shrub birds, as these corridors are maintained in an early stage of succession and hence serve as ideal scrub/shrub habitat (Hrouda and Brlik, 2021;King et al., 2009).The example of the White-browed Bulbul emphasizes the importance of considering differential impacts of LI types.
Overall response to LI showed wide variation between bird species.This could be because different guilds respond differently to the edges created by LIs.For instance, in a study conducted on edge effects in remnant rainforest patches embedded in two matrix types (mining vs. agricultural) on birds in southwest Ghana, forest specialists were the most negatively affected group (Deikumah et al., 2014).A study conducted in Thailand's dry-mixed evergreen forest revealed that the avian guilds most negatively affected by road induced edge effects include understory insectivores, arboreal frugivore-insectivores, and raptors (Khamcha et al., 2018).In addition, an experimental study showed that ground nesting birds with low-frequency calls are most prone to negative effects of vehicular movement (Polak et al., 2013).
Further, a meta-analysis conducted to understand the link between life history traits and population responses to roads asserts that more mobile birds and birds with larger territories are more susceptible to negative road and/or traffic effects than their counterparts (Rytwinski and Fahrig, 2012).It is thus unsurprising that forest specialists ( Our most alarming finding was that five out of the eight endemics recorded during our surveys were in the group of species most negatively affected by LIs, and that overall endemics were significantly more likely than non-endemics to occur in habitat distant from LIs.The occurrence of LI also seems to be creating favorable habitats for three generalist species that are globally known to be outcompeting native species with their naturalized populations (Eguchi and Amano, 2004;Klug et al., 2019;Nowakowski and Dulisz, 2019).Moreover, one of these three generalists, the White-rumped Munia, is known to be expanding its habitat even within its native range (Joshi, 2019).This suggests that further LI development would facilitate the spread of generalists and exert a greater pressure on endemics than on other species.
This might contribute to reducing species diversity at regional and global scales (Smart et al., 2006) resulting in biotic homogenization (Olden, 2008).Several studies report similar trends in biotic homogenization-replacement of endemic species (often forest specialist species) by other ubiquitous species as a result of human development.For instance, Blair (2004) demonstrated greater taxonomic overlap between bird communities in urban areas than in rural ones, a pattern evident in both oak woodlands of northern California and eastern broadleaf forests of Ohio.Similarly, Crooks et al. (2004) found that avian assemblages in southern California were progressively more similar to those in northern California and Ohio as sites became more urban.As LI is a major component of urbanization, the pattern we observed in our research parallels those found in these other studies.This indicates a negative overall effect of LI on beta diversity, and therefore the 'infrastructure tsunami' predicted to occur in the near future (Laurance 2018) must be tackled with great caution.
Our study highlights the importance of considering infrastructure-specific and species-specific responses to LIs rather than focusing solely on overall species richness or abundance when making LI development decisions.While overall species richness and abundance respond positively to LI (at least concerning railroad), which might be used to greenlight future development projects, those species that are the most unique components of Sri Lanka's biota (i.e., endemics) responded negatively, suggesting caution regarding how future LIs are placed.One solution could be to bundle LIs (Dorsey et al., 2015), as this will maximize interior forest habitat distant from all LIs, which our study indicates is favorable for a large set of forest-dependent species.Another option is to place LIs at forest edges during the design stage to avoid fragmentation.These options should be sought after first considering alternative sites, especially to avoid constructing LI through sensitive habitats (e.g., habitats occupied by endemics, or protected areas).
An understanding of how different biological communities respond to different types of LIs can not only help guide better decisions about placement of LIs, but it can also inform a reassessment of the total amount of LI development that is compatible with biodiversity conservation goals.

Figure 2
Figure 2 Ordination plot representing species response to distance from road, railroad, and powerline.

Table 1 :
Distance categories that were used to spatially stratify the points at which bird surveys were conducted.Powerline Plots within 1 km of all three infrastructure types Each survey lasted 20 minutes.The same individual conducted all surveys and all birds either seen or heard were recorded.Birds flying over the plot, above canopy level, were disregarded.All surveys were conducted between 0530 -0800 h.To avoid any observational bias due to difficulty in hearing birds above the noise of passing vehicles, each survey period was recorded (using a Sennheiser MKE 600

Table 2
(Rotenberry and Wiens, 1980;Seavy and Alexander, 2011)he vegetation structure, as this is known to have a role in structuring the avian community(Rotenberry and Wiens, 1980;Seavy and Alexander, 2011).These habitat variables were measured in February 2020.This study was conducted according to University of Miami Institutional Animal Care and Use Committee (IACUC) protocol 18-160, the Department of Wildlife Conservation in Sri Lanka (WL/3/2/56/17), and the Forest Department in Sri Lanka (R&E/RES/NFSRCM/2018/-02).

Table 2 :
Vegetation variables measured at each survey point

Table 3 :
Bird species that are notably associated with distance to linear infrastructure.

Table 3 )
, insectivores (Tickell's Blue Flycatcher, Black-naped Monarch, Black-capped Bulbul, White-rumped Shama, Brown-capped Babbler, and Small Minivet), ground-nesting birds (Sri Lanka Junglefowl), birds that are more mobile and with larger territories (Sri Lanka Grey Hornbill), and birds with low-frequency calls (Small Minivet, Black-capped Bulbul, Long Billed Sunbird, and Black-naped Monarch) are those most negatively affected in this Sri Lankan drymixed evergreen forest.