Study area

The study was conducted in the Sakaerat Environmental Research Station (SERS), which is part of the Phuluang Non-hunting Area, with a total forest area of 160 km². SERS was declared a UNESCO Biosphere Reserve in 1977. It is located in northeastern Thailand (14°30ʹN, 101°55ʹE), with an elevation of 280–762 m asl. The average annual rainfall is approximately 1200 mm, with the dry season (average rainfall 220 mm) generally occurring from November to April and the wet season (average rainfall 920 mm) from May to October. The average temperature is 27 °C, ranging from 19 to 36 °C over the year. SERS has a five-lane highway to the south, villages and agricultural areas to the east and west, and the northeast is connected to the Lam Phra Phloeng Dam. It includes two major forest types, dry evergreen forest covering 70% of the area and dry dipterocarp forest covering 20%, with the rest consisting of small patches of bamboo, plantations, and grassland. The study focused on an area within dry evergreen forest ranging in elevation from 355 to 523 m asl within a kilometre of the forest edge (Fig. 1). This edge is “hard”, in the sense that the forest ends abruptly at a five-lane highway (Route 304). The road has been in place for over 60 years but there has been recent widening along part of its length with consequent loss of native vegetation on both sides.

Bird surveys

We conducted bird surveys using point-counts during the breeding season February–August 2014 and March–August 2015. Our bird sampling units consisted of survey points (16 points during 2014, 24 points in 2015, 24 in total) set up along parallel transects (eight points/transect). Transects were 700 m apart. The total area sampled was approximately 140 ha of dry evergreen forest and was adjacent to the highway (Fig. 1). The points were arranged to sample a gradient of distances 0, 120, 240, 360, 480, 600, 720, and 840 m from the edge; thus, in total our design resulted in three points at each distance from the edge. Survey points were > 700 m from other forest types (Fig. 1). Each point was surveyed for 10 min during each survey (Gale et al. 2009). To reduce the bias of time of the day, the starting point was reversed each month. From February to May 2014 we conducted surveys at 8 points (one point at each distance from the edge) twice per month, and from June to August 2014 we surveyed 16 points (two points at each distance) once per month. During 2015 from March to August we conducted monthly surveys at 24 points (three points at each distance) and to increase the number of surveys in July and August 2015 we repeated the surveys (a day after completing the initial 24 points) on eight of these points (one point at each distance), with the aim of accounting for every species present in our study area regardless of its detectability (Sliwinski et al. 2016). The detections were recorded as either seen or heard or both. Birds in flight were not counted. Points were surveyed in the early morning, starting at approximately 06:00 a.m. (sunrise) and finishing before 09:00 a.m., as this time has the highest singing rates (Gale et al. 2009). All point-count surveys were conducted by one observer to minimize potential variance among observers (Ralph et al. 1993).

We classified birds into 11 guilds following Johns (1986) and Lambert and Collar (2002): arboreal frugivore-insectivore (AFI), arboreal nectarivore-insectivore (ANI), bark-gleaning insectivore (BGI), foliage-gleaning insectivore (FGI), granivore-insectivore (GI), raptor (R), sallying insectivore (SaI), terrestrial insectivore (TI), terrestrial insectivore-faunivore (TIV), terrestrial insectivore-frugivore (TIF), and understory insectivore (UI) (Additional file 1: Table S1).

Vegetation measurements

Vegetation structure is expected to change along a forest edge, which has the potential to be a mechanism to influence the composition of the bird community. To quantify vegetation structure, we measured the potentially influential vegetation variables; 10-m radius circular plots were established at the 16 bird survey points which were surveyed in 2014. Within each plot we counted the number of stems in three height classes (0.5–3 m, > 3–5 m and > 5 m) in order to estimate stem density (stems/ha) and measured the DBH (diameter breast height) of all trees with DBH > 5 cm in order to estimate the basal area (m²/ha). The percentage vegetation cover was estimated for five plant height classes: < 0.5 m, > 0.5–3 m, > 3–5 m, > 5–10 m and > 10 m.

Traffic noise

Ambient noise is expected to be high close to the road edge, which could affect the bird community close to the road edge. To investigate the potential effects of traffic noise on avian guilds, ambient noise measurements were conducted during bird surveys at the 24 bird survey points during March–August 2015. At each of the bird survey points we measured ambient noise in decibels (dB) for 10 min using the application Sound Meter version 1.6.1 on a smartphone (Lenovo A680) held 1.2 m above the ground pointed towards the road. We recorded the average, minimum and maximum ambient noise for each sample period.

Data analysis

To assess the effect of the forest edge on the richness (number of species observed) and abundance (counts) of the avian community sampled from the 24 survey points, we categorized our avian data into guilds and constructed generalized linear mixed models (GLMM) in R using package lme4 (Bolker et al. 2011). GLMM provides a flexible approach for analysing data with a variety of sampling units with uneven and small sample sizes and repeated samples that allows for random effects. We modelled abundance and richness using a Poisson distribution. Our models included fixed effects of vegetation (vegetation density, stem density of vegetation with heights between 0.5–3 m, 3–5 m, 5–10 m, and stem density of trees with DBH > 5 cm, vegetation cover—percentage cover of vegetation at < 0.5 m, 0.5–3 m, 3–5 m, 5–10 m and > 10 m, and basal area), ambient noise, and distance from forest edge. Although we collected vegetation structure data for only 16 of the 24 bird survey points, our preliminary surveys and recent measurements of vegetation structure in the same study area found that the vegetation structure followed a consistent pattern along our interior-edge distance gradient (R. Angkeaw, unpubl. data; W. Petersen, unpubl. data). For the data analysis we used averaged values to represent the vegetation at each of the 8 distances sampled from the road. We used avian guilds as a random effect to allow for different responses to the edge among guilds. We used maximum ambient noise in this analysis, as this is likely to have the most effect on the avian community relative to other noise measurements (Ware et al. 2015). We did not conduct ambient noise measurements in 2014, but a traffic survey by the Bureau of Highway Safety, Department of Highways, on Route 304 along our study area in 2014 (Department of Highway 2014) found a similar traffic volume to our study in 2015 (678 vs. 609 vehicles/h). Since traffic noise is determined by traffic volume (Arevalo and Newhard 2011), we used our 2015 data to represent the ambient noise in both years.

We excluded year and spatial variables (transect and elevation) from this analysis as we found no support for either of them affecting counts or richness. One variable of any pair that was highly correlated (r > 0.7) was removed from the analysis. We expected that ambient noise and vegetation variables would change with distance from the edge, so when these factors were correlated with distance, we regarded distance from edge as the best predictor of changes in species richness and abundance of avian guilds.

Bird species richness and abundance

A total of 272 independent surveys were conducted during our two-year study. We recorded 2781 detections and identified 70 species of birds (Additional file 1: Table S1). Some species were observed for relatively short periods, such as migratory species; e.g. a group of warblers, cuckoos and the Siberian Blue Robin (Larvivora cyane). Most species (69%) could be found at every distance from the edge into the interior, but some species were recorded at particular distances from the edge, including three species of woodpeckers, Orange-breasted Trogon (Harpactes oreskios), Oriental Pied Hornbill (Anthracoceros albirostris), and Thick-billed Green Pigeon (Treron curvirostra), which were found only in the interior. On the other hand, there were some species found only near the edge, such as prinias (Prinia sp.), Asian Koel (Eudynamys scolopacea), Scaly-breasted Munia (Lonchura punctulata), and Olive-backed Sunbird (Cinnyris jugularis). Total bird species richness and abundance (including both resident and migratory birds) both increased with distance from the edge, but this relationship was only marginally significant for abundance (richness: R² = 0.95, p < 0.001; abundance: R² = 0.69, p = 0.056) (Fig. 2).

Vegetation structure

Hopea ferrea (Dipterocarpaceae), Lagerstroemia duperreana (Lythraceae), and Shorea henryana (Dipterocarpaceae) were the dominant species in the canopy layer (> 30 m). The dominant species in the understory layer (< 20 m) were Hydnocarpus ilicifolia (Flacourtiaceae), Walsura trichostemon (Meliaceae), Memecylon ovatum (Melastomataceae), and Memecylon geddesianum (Melastomataceae). The total basal area for trees with DBH > 5 cm was 28.18 m²/ha. The stem density was 15,524 stems/ha for stems 0.5–3 m in height, 642 stems/ha for stems 3–5 m, and 1176 stems/ha for stems > 5 m. The stem density for trees with DBH > 5 cm was 1202 stems/ha. Estimated mean vegetation cover by height layers was 44% < 0.5 m; 72%  0.5–3 m; 55%  3–5 m; 63%  5–10 m; and 81% > 10 m.

When we considered the vegetation structure from the forest edge into the interior, we found that the total basal area was larger further from the edge (Fig. 3). The stem density of trees was also higher at the interior. The stem density of saplings was higher at both edge and deeper interior areas, relative to intermediate distances (Fig. 3). The percentage cover of the vegetation from the ground to the middle layer was higher closer to the edge in contrast to the percentage cover of vegetation at the canopy layer which was higher further from the edge (Fig. 3).

Traffic noise

From March–August 2015 the mean ambient noise was 45 dB (range: 29–82 dB). The mean maximum ambient noise within 100 m of the forest edge was 75 dB (range: 67–82 dB) and in the interior (at 800 m) was 58 dB (range: 42–69 dB) (Fig. 3). The mean traffic noise on low traffic days was 44 dB (range: 29–78 dB) and on high traffic days was 45 dB (range: 33–82 dB). Total traffic volume on Route 304 was 553–700 vehicles/h and average traffic volume was 609 vehicles/h. Average traffic volume on the low traffic days was 564 vehicles/h (range: 553–582 vehicles/h) and average traffic volume on the high traffic days was 642 vehicles/h (range: 586–700 vehicles/h). Ambient noise was significantly higher closer to the forest edge, especially within 100 m of the edge, and gradually decreased towards the interior (R² = 0.90, p = 0.002).

Avian guild responses to forest edge

We removed four variables that were correlated (r > 0.7) with distance from edge: basal area and the percentage coverage of vegetation at > 10 m height, which were positively correlated with distance from edge, and maximum ambient noise and percentage coverage of vegetation at 3–5 m height, which were negatively correlated with distance from edge. We retained distance from edge as a fixed effect to represent effects of these four variables (Fig. 3).

For guild-specific predictions, the models suggested that there was an effect of the distance from edge and vegetation structure on richness and abundance of avian guilds, and distance to edge also represented ambient noise and several characteristics of vegetation structure because of correlations (Fig. 3). However, each guild responded differently to the edge (Fig. 4). We found some evidence to support negative effects of forest edge on richness of bark-gleaning, sallying, terrestrial, and understory insectivores and we also found some evidence of negative edge effects on arboreal frugivore-insectivores (Table 3; Fig. 4a). The richness of these guilds were higher at greater distances from the edge, where the density of saplings, basal area, and percentage cover of higher vegetation layers were higher, but percentage cover of lower layers (0.5–5 m) and ambient noise were lower (Fig. 3). Edge effects seemed to have positive effects on richness of arboreal nectarivore-insectivores (Table 3; Fig. 4a). We found only limited evidence to suggest there were negative effects of forest edge on the richness of foliage gleaning insectivores and raptors, although their richness were higher further from the edge (Table 3; Fig. 4a). We found no evidence to support effects of forest edge on the richness of granivore-insectivores, terrestrial insectivore-faunivores, and terrestrial insectivore-frugivores (Table 3).

Guildsa	Richness				Abundance
	Coefficients	SE	Lower 95% CI	Upper 95% CI	Coefficients	SE	Lower 95% CI	Upper 95% CI
AFI	0.17	0.07	0.01	0.32	0.12	0.06	− 0.01	0.25
ANI	− 1.25	0.28	− 1.81	− 0.68	− 1.38	0.28	− 1.94	− 0.81
BGI	2.92	0.56	1.78	4.05	2.95	0.58	1.78	4.11
FGI	0.11	0.07	− 0.03	0.25	0.10	0.06	− 0.03	0.23
GI	− 1.17	0.91	− 3.01	0.66	-0.33	0.91	− 2.15	1.49
R	0.83	0.45	− 0.08	1.74	0.81	0.44	− 0.07	1.69
SaI	0.54	0.11	0.31	0.76	0.53	0.10	0.32	0.73
TI	0.51	0.13	0.24	0.77	0.64	0.11	0.40	0.87
TIF	0.20	0.32	− 0.45	0.85	0.24	0.32	− 0.41	0.89
TIV	− 1.16	0.68	− 2.53	0.21	− 0.66	0.61	− 1.88	0.56
UI	0.62	0.14	0.32	0.91	0.68	0.13	0.41	0.94

Positive coefficients indicate guilds that were more likely to be found away from the edge; negative coefficients indicate guilds that were more likely found near the edge

^aArboreal frugivore-insectivores (AFI), arboreal nectarivore-insectivores (ANI), bark-gleaning insectivores (BGI), foliage-gleaning insectivores (FGI), granivore-insectivores (GI), raptors (R), sallying insectivores (SaI), terrestrial insectivores (TI), terrestrial insectivore-frugivores (TIF), terrestrial insectivore-faunivores (TIV), and understory insectivores (UI)

Similar to richness, we found some evidence of negative effects of edge on the abundance of bark-gleaning, sallying, terrestrial, and understory insectivores (Table 3; Fig. 4b).

We also found some evidence suggesting positive effects of edges on abundance of arboreal nectarivore-insectivores (Table 3; Fig. 4b). The abundances of bark-gleaning, sallying, terrestrial, and understory insectivores were higher further away from the edge where the density of saplings, basal area, and percentage cover of vegetation in higher layers were higher, but percentage cover of vegetation in lower layers and ambient noise were lower (Fig. 3). There was very limited evidence to suggest there were negative edge effects on the abundance of arboreal frugivore-insectivores, foliage gleaning insectivores, and raptors, although their abundances were higher further from the edge (Table 3; Fig. 4b). We found no evidence to support edge effects on the abundance of granivore-insectivores, terrestrial insectivore-faunivores, and terrestrial insectivore-frugivores (Table 3).

For most avian guilds, there were gradual shifts in abundance and richness from the edge into the interior; in contrast, the bark-gleaning insectivores were only observed at distances > 600 m from the edge (Fig. 4), which appeared to be related to basal area and vegetation cover at the canopy level (Fig. 3). The effect of traffic/ambient noise seemed to occur mostly within 100 m (Fig. 3) (see the results on the effect of traffic/ambient noise below). The vegetation structure near the forest edge (< 350 m) was significantly different from the forest interior. The difference was most notable within 100 m from the edge (Fig. 3).

We generated 15 models to evaluate the influence of the potential variables on richness and abundance of avian guilds that were correlated with distance from edge. The models with strong support (ΔAICc < 2) included maximum ambient noise and largely the same variables (stem density of saplings, percentage cover of vegetation at < 0.5 m and  5–10 m) as the models with strong support (ΔAICc < 2) from the analysis which included “distance from edge” without ambient noise. Maximum ambient noise was negatively associated with species richness (β = − 0.04) and abundance (β = − 0.06). Other variables in the top-ranked models and models with strong support that had influence on species richness included stem density of saplings (β = 0.22) and percentage cover at  5–10 m (β = − 14). For abundance, stem density of saplings (β = 0.25), the interaction between maximum ambient noise and stem density of saplings (β = − 0.13), and percentage cover at  5–10 m (β = − 12) were the important factors, with influence in the same direction as the models with distance from edge. Some bird guilds (sallying, terrestrial, and understory insectivores) responded negatively to ambient noise opposite to the responses to distance to edge; there was a strong negative correlation (r = − 0.8) between ambient noise and distance from edge.

Edge effects on avian guilds

Overall, species richness and abundance of most avian guilds were reduced close to the edge, similar to findings of other tropical studies (Watson et al. 2004; Deikumah et al. 2014). Distance from the edge, density of saplings, basal area, and percentage cover of vegetation > 10 m had positive effects on most guilds. In contrast, increased vegetation cover in low and middle layers and ambient noise seemed to have negative effects. The vegetation structure near the forest edge was different from the forest interior. Differences extended to 350 m from the edge, but the clearest differences were within the first 100 m. Similar to previous studies, forest edge areas had a higher density of small trees and saplings, resulting in higher vegetation cover near the ground but lower cover in the canopy, as well as lower total basal area and lower density of larger trees (Dale et al. 2000; Watson et al. 2004). However, the density of saplings in our study area was higher at both edge and deeper interior areas, but relatively lower at intermediate distances. Furthermore, the edge in our study was a busy (approximately 950 cars/h) five-lane highway with substantial traffic noise, especially within 100 m of the edge, which may have further negatively impacted at least some guilds as demonstrated elsewhere (Arevalo and Newhard 2011; Ware et al. 2015), and is further explained below.

Influence of vegetation structure on avian community

The lower richness and abundance of insectivores near the edge in our study was similar to other tropical studies (Laurance et al. 2004; Mammides et al. 2015), and may be explained by the relatively unsuitable microhabitats created by the increased understory density and reduced canopy cover, which may lead to higher light intensity, higher temperatures, and lower humidity (Pollock et al. 2015). Vegetation structure had the strongest effect on species richness and abundance of insectivorous birds, with higher complexity of vegetation structure positively associated with species richness and abundance of most insectivores, especially bark-gleaning, sallying, terrestrial, and understory insectivores (Ferger et al. 2014; Mammides et al. 2015). In our study area, vegetation structure within 100 m from the edge was generally simpler, with only one or two layers of particularly dense small trees and saplings and a greater cover of vegetation near ground level. These structural changes may have negatively affected the foraging microhabitats for understory insectivorous birds (Pollock et al. 2015). The reduced layers of small trees may limit shelter and foraging substrates for sallying and understory insectivores, such as the Black-naped Monarch (Hypothymis azurea) and Orange-breasted Trogon (Harpactes oreskios); whereas the greater cover at ground level may be an obstacle to terrestrial insectivores such as the Puff-throated Babbler (Pellorneum ruficeps). In contrast, the forest interior had a more complex vegetation structure, with a higher basal area, higher density of larger trees, more vegetation layers, and less ground cover, which probably provided more diverse arthropod resources and foraging habitats (Ferger et al. 2014).

Bark-gleaning woodpeckers and understory insectivores (trogons and kingfishers) also were negatively affected by the edge. Their richness and abundance were higher in the interior and positively related to the basal area, density of large trees and canopy cover, which has also been reported by Mahmoudi et al. (2016), and Whelan and Maina (2005). This group of birds may also avoid edge areas due to the rarity of particular foraging and nesting substrates, including larger trees and stumps (Newton 1994; Lindell et al. 2004).

There was a small, but significantly negative response to edges by arboreal frugivore-insectivores, similar to some other studies in the tropics (Watson et al. 2004; Mammides et al. 2015). Slightly higher richness and abundance of frugivores in the forest interior may be associated with a higher basal area and density of large trees (Deikumah et al. 2014) and more potential food resources (Ferger et al. 2014). Higher basal area, density of large trees and vegetation cover at the canopy level of our forest interior could provide greater fruit availability (Lindenmayer et al. 2012). Some other studies have found frugivores to be tolerant to edges (Menke et al. 2012), but the tolerant species were mostly generalists, and included small-bodied frugivores such as the Yellow White-eye (Zosterops senegalensis) and Common Bulbul (Pycnonotus barbatus) that occur at soft edges with agriculture areas. On the other hand, richness and abundance of forest specialists and large-bodied frugivores, such as hornbills and pigeons, decreased at the edges (Deikumah et al. 2014; Mammides et al. 2015). Sixty percent of our birds in this guild were forest interior species (Round et al. 2011) and some had relatively larger body sizes, such as the Asian Fairy-bluebird (Irena puella; 70 g), Oriental Pied Hornbill (Anthracoceros albirostris; 750 g), and Thick-billed Green-pigeon (Treron curvirostra; 150 g). Moreover, the roadside habitat adjacent to our forest edge was unlikely to attract frugivores in the way that agricultural habitats often do.

The nectarivore-insectivore guild, such as sunbirds, was the only one to show a positive response to forest edge, similar to other studies (Laurance 2004; Watson et al. 2004). This guild depends greatly on nectar which is likely more abundant closer to the edge where higher light levels have been suggested to cause understory shrubs and lianas to produce more flowers, especially lianas which are also denser closer to the edge at all height levels (Levey 1988; Barber and Marquis 2011).

Influence of traffic noise on avian community

A decline in richness and abundance of birds close to road edges has been observed in many studies and noise may have significant effects on avian communities in general (Arevalo and Newhard 2011; Polak et al. 2013; Ware et al. 2015). Traffic noise from Route 304 was high (average = 67 dB) within 100 m from the edge and then rapidly decreased into the interior. Noise disturbance from this busy five-lane highway could therefore be the primary explanation for the decline in richness and abundance for most avian guilds at the edge (Arevalo and Newhard 2011), although this effect was confounded with vegetation changes in our study. Traffic noise distracts birds, making them more vulnerable to predation, and disrupts the singing for pairing during the breeding season, which may have resulted in edge avoidance by some avian guilds (Arevalo and Newhard 2011; Polak et al. 2013), such as the terrestrial insectivores, Green-legged Partridge (Arborophila chloropus) and Puff-throated Babbler (Pellorneum ruficeps), that frequently vocally communicate especially during the breeding season. Because our assessment is correlative, we encourage future studies to further evaluate noise at edges in an experimental fashion to confirm this mechanism (e.g. Ware et al. 2015).

Finally, the lack of an edge effect on raptors may be explained by their wide-ranging behaviour, with movements likely larger than the distance we surveyed from the forest edge. We reported few raptors (0.02 individuals/km²) and they typically occur at low densities (Andersen 2007). Minimal edge effects on foliage-gleaning insectivores may also be a result of life history traits of members in this guild, of which around half were small, non-forest specialists, including the Common Tailorbird (Orthotomus sutorius), Common Iora (Aegithina tiphia), and Plain Prinia (Prinia inornata) (Round et al. 2011). In the case of granivore-insectivores, terrestrial insectivore-faunivores, and terrestrial insectivore-frugivores, we found no evidence of edge effects on their richness and abundance. However, each of these guilds had only one species and their sample sizes were small.