Out of the forest: the movements and habitat selection of mixed-species bird flocks in agricultural landscapes adjacent to protected areas in Sri Lanka

Abstract

Background

The conservation of tropical biodiversity will require not only protected areas, but also strategies to retain animals in degraded secondary forests and countryside agricultural landscapes. This study aims to identify key landscape elements and land-use types that support the movement of mixed-species bird flocks in areas outside, but adjacent to, rainforests in lowland Sri Lanka.

Methods

We recorded flock movement pathways at five sites using handheld GPS, and documented flock composition, land-use, and habitat variables at waypoints every 15 min. Land cover was assessed through high-resolution satellite images, and classified into land-use categories based on dominant vegetation. To analyze habitat selection, we employed a resource selection function using generalized linear mixed models.

Results

A total of 112 flocks were observed, yielding 754 location records (average 7 ± 3 [SD] points per flock) across ten 3-km transects over two years. Overall, a diverse array of species (n = 66) was recorded in flocks. An interaction between land-use and time of day affected flock habitat selection, with preference for water-associated habitat such as marsh and riparian corridors, and mixed cultivation, particularly high after morning, while tea plantations were generally avoided throughout the day. Flock speed was inversely related with land-use preference, and increased when the leading species Orange-Billed Babbler (Argya rufescens) was present. Flock size and species richness strongly declined with increasing distance to the forest, with the exception that marsh, riparian corridors and mixed cultivation retained large flocks.

Conclusions

The findings emphasize how mixed-species flocks can move through human modified landscapes and how their organization (e.g., leading species) influences that process. To sustain avian diversity in agroecosystems, conservation strategies must be spatiotemporally explicit. The study highlights the importance of planning heterogeneous landscapes, including water-associated habitat and mixed cultivations, in order for flocks to persist in this agricultural landscape.

Supplementary Information

The online version contains supplementary material available at 10.1186/s40462-026-00627-z.

Background

Agricultural expansion, driven by human population increases, has been a major factor in the decline of biodiversity during the past century, and is likely to continue to adversely affect biodiversity going forward [1–3]. The major outcomes of this agricultural expansion have been habitat loss, fragmentation, and degradation [4, 5]. Most human-dominated landscapes restrict the activities of animals by increasing predation pressure [6], and providing sparse and low-quality resources, rendering them unsuitable for many species [7–9]. From a landscape perspective, agricultural land matrices often surround remnant forest patches, and these matrices can vary significantly in their permeability to biodiversity [10–12]. Some agricultural landscapes are quite ancient and continue to be traditionally managed, and contain natural elements such as habitat fragments or riverine corridors that allow quite high levels of biodiversity to persist [13, 14]. It is critical to identify which land-use types or landscape elements retain biodiversity, and how these should be arranged spatially [15]. As global efforts progress to protect 30% of terrestrial land [16], it is especially important to design agricultural areas near forests to provide both productivity and conservation [17, 18].

Movement is a particularly crucial aspect of biodiversity to investigate in agricultural landscapes. Nathan et al. [19] defined movement as the change in location of individuals over time, and provided a conceptual framework for the where, how and why questions that can be asked about it. Pyke [20] described the importance of speed and direction as fundamental properties of movement. With the development of tracking and statistical technologies, the field of movement ecology has grown rapidly in the last decades and it provides many implications for conservation management planning [21–26]. For any given species, movement is vital to sustain the genetic variability of established populations and colonize new ones [27, 28]. For multiple species, movement is important in meta-community assembly and in preserving species interactions and the ecosystem functions and services they support [29–31]. Movement studies have contrasted forest interior species and generalists as they move on their natural home ranges [32] and by translocating them in and across agricultural matrices [25, 33, 34]. They have documented how different landscape elements are important conduits for, or barriers to, animal movement [35, 36], as well as affecting the speed and linearity of the movement [37]. Ultimately, this research allows us to measure the functional connectivity of different parts of the landscape, whether they be human land-types, landscape elements or matrices [38, 39].

While many studies have investigated the movement of individual species, or even a suite of different species, movement patterns of mixed-species groups have been less studied (but see Rutt et al. [40] and Mokross [41]). Mixed-species groups are common among vertebrates, and a major type of social organization of forest animals [42]. They have interesting dynamics for the study of movement because different species can play different roles, including some species being leaders (sometimes called “nuclear species”), whereas others mostly follow [43, 44]. Therefore, there is the possibility of species interdependencies, in which the habitat selection of the leading species can influence that of the followers [45, 46]. For birds, studies of mixed-species flocks’ (hereafter, MSFs) movements have been mostly conducted in Amazonia, focused on road crossings [47], and the relationship of movement to fragmentation and subsequent reforestation [40, 41]. While Asian MSFs are known to come out of the forest in some areas [48], it is not well understood what land-use types allow them to do so, and how they change as they move away from the forest edge.

MSFs have been well studied in Sri Lanka [49–52], but as yet studies have not focused on their persistence within agricultural lands adjacent to forests or their movement patterns. In this project, we aimed to combine spatial tracking of MSFs with GIS categorization of land-uses, to identify the habitat selection of MSFs in human-modified landscapes, using resource selection functions (hereafter RSF; these functions model the probability of an animal selecting a specific resource or habitat type relative to its availability in the environment [53]). Our first hypothesis (H₁) was that MSFs would show a strong preference for forest fragments over other available habitats due to the continuous cover, structural complexity, and foraging resources [54]. In contrast, we expected monoculture tea plantations to be the least preferred land-use type, due to their simpler vegetative structure and lower food abundance [49]. We also hypothesized a temporal shift in habitat use, with MSFs increasingly selecting shaded, densely vegetated land-use types during the mid-day to mitigate heat stress and predation risk. Indeed, we have earlier noticed that MSFs even within forest slow their movement and stay in habitats near water during the middle of the day (EG, personal observation).

Our next hypotheses were related to speed. The second hypothesis (H₂) was related to H₁, as we expected MSFs to move slower in land-use types that they preferred. But, our interest in speed was particularly related to the question of how leader species influence MSF movements. We hypothesized (H₃) that the Orange-billed Babbler (Argya rufescens, OBBA) the most dominant leading species of MSFs in this lowland region [45], would facilitate the movement of MSFs, resulting in them moving at greater speeds, and consequently larger distances. In earlier studies, we have found that this species is fairly resistant to leaving the forest, at least in comparison to the leader of high-elevation MSFs in Sri Lanka, the Sri Lankan White-eye [48], which readily enters agricultural areas, and that the OBBA’s avoidance of strongly human-modified areas might mean that the whole lowland MSF system is less likely to leave the forest [45]. However, in the specific region of this study, we have observed that OBBAs do sometimes lead MSFs into adjacent non-forest and agricultural landscapes, and that the species can be seen in home gardens at the border of forests and occasionally even in the center of villages. Also, we have observed MSFs lacking this species in the same region to seem directionless, suggesting that the presence of OBBA is critical to MSF movement patterns [55].

Our fourth hypothesis (H₄) was that the size of MSFs would decline with increasing distance from the forest edge, and that the number of forest interior species would in particular plummet. We expected this because MSFs are primarily a forest phenomenon [42], and earlier work in Sri Lanka showed that they become infrequent in buffer and especially agricultural areas [48]. However, in this study we were especially interested to see whether some land-types might differ from others in retaining large and diverse MSFs.

Methods

Study area

Fieldwork was conducted in five study sites in the lowland southwestern wet zone of Sri Lanka (Fig. 1), with each site more than 10 km apart from the others and adjacent to lowland rainforests, with elevation below 500 m (Table 1; Fig. 1). Two 3-km transects were established at each site, running parallel to the forest edge and generally laid along foot paths. To maintain close association with the edge habitat, the transect pathway was consistently kept within 5 m of the forest edge. The starting point and direction of each transect were selected during a pre-survey to ensure the path was accessible and followed a clear, continuous forest edge.

Fig. 1.

The study locations, showing the study sites in the inset of Sri Lanka (with the study site circles not to scale), the transects (two 3-km transects per site) and the different land-use types found there. All colored areas were within 500 m of an MSF observation

Table 1.

Descriptions of the five study sites

Site no	Site name	Transect	Specific area	Latitude	Longitude	asl (m)	Adjacent forest reserves
1	Lankagama (L)	A	Wathugala	6.3698°N	80.4766°E	~ 323	Sinharaja Forest Reserve
		B	Pitadeniya	6.3644°N	80.4691°E	~ 295	Dellawa Proposed Forest Reserve
2	Kudawa (S)	C	Kosgulana	6.4141°N	80.4633°E	~ 472	Sinharaja Forest Reserve
			Ketalapaddala	6.4246°N	80.4464°E	~ 500
		D	Kudawa	6.4400°N	80.4165°E	~ 335	Sinharaja Forest Reserve
			Pitakele	6.4356°N	80.4037°E	~ 418
3	Erathne (E)	E	Dehigahahena	6.8093°N	80.4331°E	~ 500	Peak Wilderness Sanctuary
		F	Weniwelketiya	6.7956°N	80.4300°E	~ 426	Induruwa Erathne Mukalana (Part of Gilimale Forest Reserve)
4	Maliboda (M)	G	Bambaragala	6.8856°N	80.4606°E	~ 500	Peak Wilderness Sanctuary
		H	Orupilagama	6.9047°N	80.4450°E	~ 400
5	Kanneliya (K)	I	Panagala	6.2705°N	80.3288°E	~ 152	Kanneliya Forest Reserve
		J	Koralegama	6.2443°N	80.3323°E	~ 89

Tracking mixed species bird MSFs

Surveys were conducted by two observers (IW and DV) from 06:00 h until 18:00 h in the evening. When we encountered an MSF emerging from the forest edge, we recorded its movement path at 15-min intervals using a handheld GPS (Garmin eTrex 10 with approximately 5–10 m accuracy), following the MSF until it came back to the forest, moved into inaccessible habitat, or broke up into pieces. When following, the observers moved 10–15 m behind the MSF, with the purpose of minimizing interruption to MSF movement; the birds rapidly habituate to human followers.

In the fieldwork, an MSF was defined as more than two bird species moving in the same direction for at least 5 min [56]. Data on the composition of the MSF was taken approximately every 15 min, synchronous with the GPS location. All bird species moving in the direction of the MSF within ~ 20 m of another bird were recorded as being in the MSF; usually forest studies use a 10 m cutoff due to poor visibility [57], but because we were working outside the forest, visibility was high.

After we finished recording the movement of an MSF, we would restart our walk along the transect where we initially stopped to observe the last MSF, that day, or on earlier days. We walked transects only once to avoid repeatedly following the same individuals. New MSFs encountered were considered to be independent, although because birds were not marked, we cannot be sure different MSFs contained different individuals. Sampling was conducted during the relatively dry period (December to May) of each of two consecutive years (2023 and 2024); precipitation was lower in 2024 (Figure S1), but we did not notice that influencing the movement patterns of MSFs.

Landscape and habitat characteristics

At each point where we took a GPS location and assessed MSF composition, we also collected habitat information. Because birds were moving as a group in an MSF, and the MSF could be 20–30 m across at any one time, we concentrated on the land-use type of patches of habitat, and not on smaller-scale landscape elements like fencerows or single trees. The data at each point location included land-use type, according to a pre-defined classification scheme (see below), canopy density and canopy height. Canopy density was measured averaging four canopy images that were taken 5 m in the cardinal directions away from the GPS point representing the MSF’s center, using the fish-eye option of the “GoPro HERO8” camera (www.azandisresearch.com). Later these full HD images (1920 × 1080 progressively displayed pixels) were analyzed with “ImageJ 1.48” software [58] following the “hemispherical 2.0 manual” [59]. Canopy heights were measured using the Laser Rangefinder (YUKON Extend LRS-1000), averaging measurements at three locations near the MSF center at least 10 m apart to get a representative measurement of the environment through which the MSF was moving.

Landscapes outside the forests in this region have been mostly converted into agriculture and agroforestry, as well as residential property and infrastructure like roads (but built-up areas consisted of only 2.8% of the areas analyzed in the study). The remaining forest is often confined to riparian areas or scattered, degraded forest patches. Other land-types included agroforest, shrub, marsh, mixed-cultivation, and tea (see definitions in Table 2). Most tea had scattered shade trees present, so we did not differentiate between monoculture tea and shaded tea. The average canopy heights and canopy densities in these different land-use types are shown in Table S1.

Table 2.

Land-use categorization used for the study

Land use category	Forest Patches	Riparian Forest	Agroforest	Shrub	Marsh	Mixed cultivation	Tea	Built up
Description	Degraded forest patches and forest corridors consisting mostly of native trees and maximum canopy heights over 15 m that arose from natural succession, separated from the contiguous forest	Vegetation located up to 100 m from river or stream banks, separated from the contiguous forest.	Monoculture plantations (e.g., Pine, Eucalyptus, and Araucaria) established after the removal of natural forest cover from specific sites.	Areas dominated by woody plants that are typically smaller than trees (canopy heights less than ~ 10 m), often dominated by invasive and non-native species.	Areas often subject to frequent flooding and wet conditions throughout the day, with vegetation not above the shrub layer, and sometimes on the borders of paddy fields.	Cultivated landscapes that include home gardens and various economically important plants, excluding tea plantations.	All types of tea plantations, including monoculture tea plantations, sometimes with rows of other tree species planted for the purpose of shade or as windbreaks.	All man-made constructions including buildings, roads and infrastructure.
Re-categorization	Dense vegetation (characterized by high structural complexity and minimal light penetration)		Moderate vegetation (offers partial canopy cover and moderately dense vegetation)				Sparse vegetation (Minimal vertical structure, open canopy, and limited habitat complexity)

Land cover outside the forest edges was assessed through high-resolution satellite imagery (Maxar WorldView-3: +/- 0.31 m; Airbus Pleiades: +/- 0.30 m) via high-definition (8192 × 4639 pixels after cropping) Google Earth Pro images taken after 2024 (https://earth.google.com). We used ArcGIS 10.7 (https://www.esri.com) to classify areas into the seven pre-defined land-use categories; image digitization was guided by ground-truth data on the land-types collected while following the MSFs. All the map outputs (~ 1:5000 scale) were prepared using QGIS software 3.24.1 (https://www.qgis.org).

Statistical analysis

Land-use analysis by resource selection function (RSF)

We tested H₁, about what habitats MSFs prefer to move through, using RSFs. Resource selection refers to a suite of methods to determine resource or habitat use by animals that can be measured in multiple levels or scales [60]. There are several commonly used measures of habitat use or resource selection: Compositional analysis [61], Mahalanobis distance [62], Selection ratios [53, 63], Resource selection functions [64], and Resource selection functions with weighted distributions [65]. We selected RSFs wherein all variables are categorical, as they measure the relative probability of selecting habitat [65], are intuitive to interpret, and work well with generalized linear mixed models (GLMMs). We used non-weighted distributions because we were limited to using available habitat, rather than having confirmed avoidance data. Furthermore, the weighted distribution method works only if more than one of the covariates is continuous [66].

RSF compares the qualities of habitats where animals were present to what habitat was available in the area. To understand the characteristics of the available habitat, we generated random points using the ‘Create Random Points’ tool in ArcGIS (https://www.esri.com), with 10 random points records generated for every presence record. Random points were generated within buffers of three different distances: 100 m, 250 m, and 500 m from the MSF pathway, following the ‘assessment corridor’ method suggested by Selonen and Hanski [67] for measuring the habitat preference of moving animals. We conducted separate models for the three different scales. This approach allowed us to evaluate how scale might affect the measurement of habitat preference. Further, because there appeared to be a predictable diurnal cycle, we divided the day into three categories (06:00–10:00 hrs, Morning; 10:00–14:00 hrs, Midday; 14:00–18:00 hrs, Afternoon), and used time of day as a categorical predictor variable in the analysis.

All statistics were analyzed using the R statistical analysis version 4.4.2 (R Core Team 2024). A RSF was generated through logistic regression that can assess which variables affect MSF habitat or resource selectivity. The response variable represents the probability that the MSF used the habitat (used/available, requiring the use of a binomial error distribution), and this was influenced by land-use type, time of day and the land-use: time interaction, in a GLMM with the random factor of transect (10 types), nested in site (5 reserves). Thus, the model could be summarized in this equation:

A further elaboration of this model was used with a dataset that included all the data for all three buffer scales (100 m, 250 m, and 500 m), and pairwise interactions, to result in:

See below for our general modeling techniques and model simplification strategy.

Speed analysis

This analysis was developed to be complementary to the RSF one, and specifically to test H₂ that habitat preference and speed would be inversely related, with MSFs moving faster through land-use types that they less preferred. The distance traveled within each land-use type, and the time it took based on the time stamps of the waypoints, was calculated for each segment of time for each MSF pathway. Given the positive continuous data with non-zero values, we used the Gamma distribution. The analysis was similar to the RSF one, with the full model being:

To secondarily examine the relationship between movement speed and habitat selection, we calculated the correlation between a land-use types’ RSF preference (the probability of a MSF being present) and its average speed, and tested this with Spearman’s rank correlation test, as the data were not normally distributed.

In the GLMM above, land-use was found to be the only significant variable (see Results). We also observed that the presence of the leading species OBBA (“OBBA_P”, coded as 1 and 0) appeared to affect speed. We therefore developed a further model to test H₃:

OBBA were not present at one site. To test if speed, and the distances that MSFs traveled, different between sites, we used Welch’s ANOVA, due to unequal sample sizes between sites, and Games-Howell comparison tests, using the ‘rstatix’ package.

Distance to the forest edge analyses

These models were aimed to test H₄ about how the size and composition of MSFs changed as they moved away from the forest edge. In these models, characteristics of an MSF (total species, total number of individuals and number of forest interior species) were treated as separate response variables, and because these were counts, we used Poisson error distributions. Predictor variables included distance to the forest, land-use types and the interaction between distance to the forest and land-use. The straight-line distance to the forest edge was measured for each 15-min waypoint. The full model was thus:

When over dispersion was detected in these models, we used a negative binomial error family.

General modeling techniques

We analyzed the above models using generalized linear mixed models (GLMMs) in R using the ‘glmmTMB’ package for mixed modeling and parameter estimation. Our modeling approach followed Burnham & Anderson’s [68] information-theoretic framework. If the model was singular or failed to converge, we simplified the random effects structure from transect nested in site, to simply transect (but we did not simplify further, always retaining a random factor and thus a mixed model, as transect was part of our sampling design).

Model simplification was based on likelihood ratio tests, with non-significant interactions eliminated first, followed by non-significant variables. Residual diagnostics were performed using the ‘DHARMa’ package [69]. The differences between the levels of categorical variables were tested separately using Tukey’s Honestly Significant Difference (Tukey HSD) multiple comparison tests, using the ‘emmeans’ package. All mean values are shown ± standard deviations.

Results

Description of the dataset and MSF movement

A total of 112 (25 in 2023 and 87 in 2024) MSFs were observed, with 754 presence records (average presence records per MSF, 6.75 ± 2.71) in the five different study sites. MSFs averaged a total movement of 500.4 ± 280.4 m during an observation, summing the straight-line distance between each point location made 15 min apart, in an average time of 101.35 ± 40.63 min per MSF, which means the average speed was 0.23 ± 0.18 km/h. The mean maximum distance away from the forest the MSFs traveled was 133.68 ± 82.27 m. See Fig. 2 for an illustration of all MSF movements at the site level, with some examples of individual MSF movements as insets. Overall, of the 112 MSFs, 64 were seen to go out of the forest and return, whereas 43 broke apart, and 5 were impossible to track. As to the presence of the OBBA, this species was not present in MSFs in Kanneliya Forest Reserve (0 of 83 point locations), but was usually present at most of the other sites (633 of 672 point locations). As discussed below, this difference was associated with low speeds and distances traveled in Kanneliya compared to the other sites (see Fig. 2).

Fig. 2.

Examples of MSF movements. The land-use types for two sites, (A) Erathna, and (C) Kanneliya, are shown as in Fig. 1, with the map of MSF movements next to them (B and D, respectively; the waypoint pathways are in black with each waypoint made after 15 min shown as a pink dot, on the top panels, and a red dot in the bottom panels). The longer pathways of Erathna are apparent here, including one pathway that crossed between contiguous forest and a large forest patch. The lower panels (E, F) show two representative MSF movement pathways, from the start (yellow) to the end (blue) of the observation. The MSF in Erathna is typical of one that included OBBA and moved a long distance, as measured both by its full pathway and by its furthest distance from the forest, while the MSF in Kanneliya (notice the different scale) is characteristic of ones near that reserve that were missing OBBA and generally did not move very far or at fast speeds

RSF models for MSF habitat selection

RSF was conducted under three buffer scales (100 m, 250 m and 500 m) to test H₁. When scale was included in the model, the interactions with other variables were not significant (Scale: Time, χ²₂ = 0.025, p = 0.99, and Scale: Land-use, χ²₆ = 5.65, p = 0.46). When the model was simplified to remove the non-significant interactions, scale itself was not significant (χ²₁ = 0.014, p = 0.91). The lack of a scale effect is also seen in Fig. 3, in which relative differences between land-use types are not affected by scale. We selected the 500 m buffer for the subsequent analyses, as the marginal R² was highest among the three buffer models (0.082).

Fig. 3.

Predicted probabilities of an MSF being present in different land-use types and times of day across three buffers (100, 250 & 500 m). The probabilities range from 0.00 to 0.30, indicating the likelihood of a MSF being present in each category under the given conditions. Bars that share the same letter are not significantly (P < 0.05) different from each other

The land-use type: time of day interaction was significant (for the 500 m scale, χ²₁₂= 36.17, p < 0.001; this was true for the other scales as well), indicating that the effect of land-use on resource selection differed meaningfully across diurnal periods (see Fig. 3; Figure S2 shows pairwise comparisons between land-types). For example, during morning hours (0600–1000 h), forest was among the most preferred land-types, along with riparian habitats, shrub and mixed cultivation. However, by afternoon forest was the least-used land-type, and marsh was now among the highest. One thing that was consistent throughout the day, however, was an avoidance of tea, with it always being among the least preferred land-types (see Fig. 3).

MSF speed

Among the variables used to try to explain speed in testing H₂, including land-use, time and the land-use: time interaction, only land-use was significant (χ²₆ = 37.17, p < 0.001), with pairwise means showing that marsh and forest were transversed more slowly than agroforest, mixed-cultivation and tea (see letters representing multiple comparisons on the top of Figures S3 and S4). In general, the pattern illustrated by speed was the inverse of habitat selection: the Spearman’s rank correlation test revealed a statistically significant negative relationship between speed and habitat selection probability (ρ = -0.79, S = 100, p = 0.048, Figure S5).

Testing H₃, which investigated the influence of OBBAs, the interaction between OBBA and land-use was not significant (χ²₆ = 11.28, p = 0.080). But, the presence of OBBAs was significant in the reduced model (χ²₁ = 4.41, p = 0.035) and speed was uniformly higher when OBBAs were present than when they were absent (see Fig. 4).

Fig. 4.

MSF speed (km/h) across different land-use types, stratified by presence (left) and absence (right) of OBBA. Land-use categories include Forest (the reference in the models), Agroforest, Marsh, Mixed Cultivation, Riparian, Shrub, and Tea. Error bars represent 95% confidence intervals. Bars that share the same letter are not significantly (P < 0.05) different from each other

Comparing among the different sites, the speeds of MSFs, and the total distances traveled, were particularly low in Kanneliya Reserve, where OBBAs were absent (Table 3). ANOVA results revealed site was a significant factor influencing speed, with significant differences of MSF speeds in different sites (F₄ = 18.67, p < 0.001). Pairwise comparisons between study sites are shown in Figure S4; speeds in Kanneliya were significantly different than those in Lankagama, the reserve with the fastest speeds.

Table 3.

Average full path distance the flocks moved and their speed, and the average number of OBBA individuals

Study site	Average full path distance of MSF (m)	Average speed (km\h)	Average OBBA
Lankagama	548.99 ± 318.48	0.29 ± 0.15	10 ± 6
Kudawa	555.69 ± 300.28	0.24 ± 1.18	9 ± 3
Erathne	566.73 ± 319.88	0.25 ± 0.18	10 ± 6
Maliboda	434.51 ± 170.08	0.20 ± 0.16	7 ± 1
Kanneliya	343.84 ± 161.86	0.18 ± 0.14	0

MSF composition and distance to the forest

In tests of H₄, model simplification revealed that the complex model (with interaction between land-use and distance to the forest), performed best for the models in which the number of individuals and species were the response variable (for total individuals, the interaction was significant [χ²₆ = 14.291, p < 0.026], as it was for total species [χ²₆ = 26.295, p < 0.001]). For the model with forest species as the response variable, the interaction term was not significant (χ²₆ = 3.269, p = 0.77); but distance to forest (χ²₁ = 27.125, p < 0.001) and land-use (χ²₁ = 76.519, p < 0.001) were significant and retained in the model.

The marginal plots (Fig. 5) illustrate the predicted relationship between three metrics of MSF composition (total individuals, total species and forest interior species) and distance to continuous forest edge, segregated by land-type. While most land-types show declining number of species and individuals with distance, a few land-uses allowed these numbers to persist: particularly marsh and mixed-cultivation for numbers of individuals, and these two land-uses and riparian forest corridor for numbers of species. The number of forest species, however, falls with distance in all land-types. Pairwise comparisons between land-types are shown in Figures S6-S8.

Fig. 5.

Relationship between the distance to the forest and the number of total individuals, total species and forest species in the MSF in different land-use types. The significance (p-values for regression lines) for each land-use type are indicated, showing variations in MSF composition as distance from the forest increases

Discussion

MSFs are a forest phenomenon: they change composition and lose members as one moves from forest to buffer (agroforestry, degraded forest) to agriculture [70]. In Sri Lanka, earlier literature has shown that the lowland rainforest system led by OBBA is particularly reluctant to move away from forest; whereas the upland system is much more adaptable to buffer and agricultural environments [45]. Yet even though this is a lowland forest MSF system, in this study we were able to track MSFs as they moved outside of the edge of contiguous forests to adjacent agricultural areas, moving an average of ~ 130 m away from the forest. We found, as we hypothesized (H₁), that small patches of forests, especially riparian forests, were preferred habitats for these MSFs. But, more surprisingly, habitat preference changed over the day, and habitats such as the edges of marshes, and areas of mixed cultivation were also preferred at some time by MSFs. The preferred land-use types of MSFs were all dense and moderately dense vegetation, and MSFs avoided monoculture tea and its sparse vegetation with a lack of resources and protection. In addition to investigating land-use type, we also explored factors that influenced MSF speed and composition. Confirming H₂, speed was shown to have an inverse relationship to land-use preference. Related to H₃, for all land-use types, we found that the presence of the OBBA as the leader species made the MSF move farther and at greater speeds. Distance to forest was a major driver of MSF composition, with MSF typically losing individuals, species and forest species as they moved away from the forest, providing evidence for H₄. Yet we found some land-use types were able to retain high numbers of individuals and generalist species. Below we look in detail at our three major analyses in turn, and then continue to discuss the conservation implications of the study.

Land-use selectivity and temporal dynamics of MSFs

We evaluated habitat preference at the MSF level with RSF models to test H₁. This flock-centered approach obscures differences between species and individuals stemming from variation in their habitat use and preferences. RSF often demonstrates that animals’ habitat preferences often vary independent of habitat abundance. This functional response [71] to habitat selection can also vary with the spatial configuration of the habitat or can be temporally dynamic, which we observed in this study.

Land-use preferences obtained in this study varied temporally; for example, shrub was most preferred in the morning, riparian forests during midday and marsh in the afternoon. This suggests that MSFs adjust their habitat use in response to microclimatic conditions (e.g., heat, shade, resting and bathing opportunities). Typically, most of the MSFs we observed emerged from the forest edge during morning hours (usually before 10:00 h, depending on sunrise) and dispersed into surrounding landscapes. Later in the day, they either scattered or returned to the forest as an MSF, although often having a different group composition than earlier. If the MSFs continued to stay outside forest during the midday and afternoon hours, they mostly preferred to stay in shaded or moist habitats (such as riparian forests; Fig. 6). Ultimately, this time-dependent preference for different habitats means that heterogeneous landscapes will have the highest capacity for retaining MSFs.

Fig. 6.

Movement pathways of MSFs through riparian forest corridors that are situated near tea plantations and mixed cultivation landscapes. Dotted lines represent the center of the MSF movement paths and yellow arrows indicate the direction of the MSF movement. Red arrows represent the approximate width of the MSF as it traveled. The satellite images were obtained through Google Earth Pro software. Camera symbols in the top panels represent where the photographs in the bottom panels were taken and the direction of the photograph

MSFs showed a strong habitat preference for areas of mixed cultivation. This kind of habitat could also be called “home gardens” and they contain a diverse set of crops reflecting a system of indigenous agriculture common throughout South and Southeast Asia that is ancient but still evolving [72]. These habitats are particularly important for fruit-eating birds and species adapted to both forests and open habitats. Some other studies have confirmed the high bird species diversity in mixed cultivation, due to higher availability of food and stratification [73–76], as compared to tea plantations [77, 78].

Other land-use types were less preferred. Agroforests (Pine, Araucaria, Eucalyptus plantations) have been noted to be one of the least biodiverse land-use types [50, 78, 79] in the region due to the absence of structural components of native habitats such as understory vegetation [80], and we found them to have both low habitat preference and high speeds. Tea plantations also had generally low preferences throughout the day. Tea plantations in this landscape mostly exist on steep slopes alongside the contiguous forest. Usually, these tea plantations have been identified as one of the land-use types in the region with lowest bird diversity [73, 79, 81]. Birds in the MSF were very reluctant to go into the tea bushes themselves, with sometimes only a few babbler species entering them and the other species moving widely around the open area above the tea, especially if there were no shade trees. Shade trees were clearly beneficial for certain canopy bird species to move through tea to the adjacent vegetation.

MSF speed and pivotal role of nuclear species

MSFs moved across faster through less-preferred habitats like agroforests or monoculture tea plantations, and slower in preferred habitats such as forest patches, and marshes, supporting H₂, although the relationship was not perfect (for example, speeds in mixed cultivation and riparian forest were fairly high despite these land-uses being quite preferred). MSFs spend longer time intervals in these preferred vegetation types, engaged in foraging or resting. This phenomenon supports the “landscape resistance” theory of movement, which explains how species’ movement across a landscape is not uniform but is influenced by spatially heterogeneous environmental factors [82]. The behavior mirrors observations in fragmented landscapes globally [25], where birds minimize time in high-predation or low-resource areas such as open landscapes.

The other clear result from the speed analysis, testing H₃, was that in every kind of land-type, OBBA-led MSFs were faster than other MSFs (OBBA-led mean; 0.25 ± 0.42 km\h, OBBA-less mean; 0.18 ± 0.14 km\h). Speed here was collinear with total distance traveled: OBBA-led MSFs moved much further in terms of total pathway distance (OBBA-led mean; 524.8 ± 286.9 m, OBBA-less mean; 343.84 ± 161.9 m), although the furthest distance from the forest edge was not so different between MSF types (OBBA-led mean, 132.2 ± 79.0 m; OBBA-less mean, 130.0 ± 96.5 m). OBBA is well-known for being important to this system, being highly gregarious (averaging 16 individuals per MSF in the forest) and being the first species to cross a gap in 60% of observations [49]. It is constantly vocal, and because of its many eyes and kinship groups is an information-producing species that other species may follow [52]. Slow movement of OBBA-less MSFs has earlier been noted by Perera et al. [55] in a forest (Hiyare) close to Kanneliya, the reserve adjacent to our field site where we found OBBA completely absent. In these MSFs, other species appear to take over the role of leader, such as Sri Lanka Drongo (Dicrurus lophorhinus) and Scimitar Babbler (Pomatorhinus horsfieldii), but the MSFs have little forward motion, often seeming to move around in circles.

Our results emphasize the critical role of this gregarious species in the navigation of MSFs in suboptimal habitats, aligned with the predictions of H₃. Even though leading (or “nuclear”) species like OBBA can be quite common, they are important conservation targets if their habitat selection influences the habitat selection of endangered species in MSFs [49]. Thus, leadership has critical implications for conservation: protecting leader species like OBBA could enhance the resilience of MSFs in human-dominated landscapes [70]. Similarly, it is worthwhile to further study which species play the leading role in MSFs in lowland wet zone forest when the OBBA is absent.

MSF composition and proximity to the contiguous forests

Proximity to contiguous forest edge has a major impact on MSF composition, consistent with H₄. The proximity to forest has been reported as the key factor regulating biodiversity within different land-use types such as forest, buffer and agriculture [49, 83]. But what was unexpected from this study was the modulating impact of land-use type. Indeed, some of the most preferred land-use types outside of forest, such as riparian corridors, marshes and mixed cultivation, showed a pattern wherein the number of species and individuals did not decline with distance to the forest edge. Nevertheless, forest-specialist species declined uniformly in all habitat types. This means that MSFs in these preferred habitats outside of contiguous forest include a lot of generalists, which can use both forest and more open lands [32] and indeed we often saw these types of species join MSFs as they came out of the forest. Overall, the positive association between species richness and forested habitats such as riparian forest patches, even when distant from the edge of contiguous forest, highlights the conservation value of these “stepping-stone” habitats [37].

Conservation implications and recommendations

As MSFs prefer several different kinds of land-use types depending on the time of day, a heterogeneous landscape will be most protective of birdlife and the various ecosystem services that they provide. Forest-interior flocking species are predominantly insectivorous [84] offering valuable pest control services [85]. Meanwhile, facultative flocking species in agricultural areas, which include omnivores and frugivores, contribute additional ecosystem services such as seed dispersal [86]. The importance of heterogeneous landscapes for biodiversity and ecosystem services has been studied in various parts of the world such as landscapes that are dominated by shade coffee plantations [75, 81, 87], cacao [88] or cardamom [79]. These studies have identified certain forms of cultivation, such as shade coffee or cinnamon, to be more conducive to biodiversity conservation compared with unshaded or monoculture plantations [88, 89]. At the same time, many findings highlight the importance of native plants, rather than introduced or invasive species, in providing habitat for species survival [12, 79, 81, 90]. We strongly recommend maintenance of these heterogeneous landscapes, as their conversion into poor, more open habitat (tea, agroforests) will clearly negatively affect MSFs.

Tea is a major cash crop internationally, with global cultivation exceeding 6.5 million hectares, and its cultivation is expected to expand [91]. Sri Lanka plays a remarkable role in the tea production in the world, with tea cultivation having been established in the wet zone of the country since the 1830s [92]. In this study we found that birds in MSFs generally avoided tea, and our findings suggest that tea plantations require a significant alteration in order to be changed into sustainable and preferred habitats. We recommend maintaining tea plantations with shade trees or integrating multi-crop systems (e.g., cinnamon, pepper) that support biodiversity while remaining economically viable for landowners. Our preliminary observations suggest that such habitats may be more permeable to MSFs than tea itself. Such sustainable practices can provide critical refuge for wildlife while balancing agricultural productivity, supportive of the ‘land sharing’ approach [18]. Further efforts will need to understand how forest corridors, riparian corridors, and even narrow fencerows [37, 93] can improve the functional connectivity of tea-dominated landscapes.

The value of riparian forest corridors has been highlighted in several studies that have focused on the movement ecology of animals [36, 37, 83]. We found that the size of riparian corridors ~ 20–30 m (canopy width) wide was a good match to the breadth of MSFs, providing more width than just a fencerow, and they were clearly used by MSFs to move through highly-modified landscapes (see Fig. 6). These riparian corridors have recently been proposed as conservation targets globally [94] and we hope that this study adds another piece of evidence for their importance for conservation. In this part of the southwestern lowlands of Sri Lanka, small rivers nearby forests can flood suddenly and caused damage to the villagers and agricultural fields [94, 95]. It is therefore critical to leave some buffer areas around the river basin and this can serve a conservation function, as well as protect human safety.

Although modified land uses host many generalist birds that join MSFs, the number of forest-dependent species gradually declines across all land-use types, and most of the forest interior species have no prospects of survival outside of forest [96]. Importantly, these forest-dependent birds are unique and ecologically irreplaceable. Although generalist species and modified habitats can, to some extent, offset losses in phylogenetic and functional diversity, they cannot compensate for the specific ecological roles of forest specialists [97, 98]. Therefore, preventing further deforestation and connecting forest patches are essential strategies to safeguard forest-dependent birds and preserve their vital ecological functions. In this study, we occasionally observed MSFs move from one forest to another (see example in Fig. 2B). Because these birds were not marked, definitive evidence is still lacking on whether individual birds traveled the entire distance with the MSF. If birds are able to cross human-dominated areas and connect forests, it could be an important mechanism for genetic transfer across populations and the colonization of new populations [99].

Limitations of the study

MSFs are complex to observe because they have many individuals, are spread across a sizable area, and move rapidly, so that detailed records of exactly which bird was where were difficult for us to record without losing the MSF as it moved forward. We depended on observers following the MSFs, and we assumed that observer speed is similar to MSF speed, and that the movement of the observers did not influence that of the MSFs. In practice, at the beginning of the observation we would sometimes observe the birds detect the human observers, but they quickly habituated, so we do not think the presence of the observers biased the data. Nevertheless, future studies using banded and radio tagged birds could capture much more detailed data at the individual level and therefore see a different set of patterns.

Some limitations of our study are related to the modeling. Because our study did not rely on telemetry data, we also have limited presence data points for each MSF to analyze the habitat selectivity. For example, RSPF (Resource Selection Probability Function) and SSPF (Step Selection Probability Function) use telemetry data with a considerably higher number of point records with directions; in contrast, we generated random points at a 10:1 ratio to the observed data. Because we had only categorical variables for resources and did not have exact absence data and therefore absolute probabilities, we had to depend on RSF, rather than the more accurate analyses, such as weighted distribution RSPFs [67].

Other limitations were related to the image data. Due to insufficient resolution land-use data, distinguishing between some land-use boundaries proved challenging. For example, saying that MSFs use marshes does not mean they were in the middle of a rice paddy – actually MSFs do not use rice paddies, but they do use the vegetation on the sides of them. These areas are dominated by woody, early successional species, primarily the dense, fast-growing shrub Dillenia suffruticosa, which provides protective cover for the species during daytime. Anyway, we did not have enough spatial resolution to delineate different microhabitats within marshes. Likewise, we did not distinguish between monoculture tea plantations and those with shade trees, or mixed with other crops. Integrating fine-scale resource and land-use data, such as fine scale aerial images or LIDAR images, would strengthen habitat selection models. Further, such image data could identify the vertical structure of the vegetation or spatial configuration of the land-use types, all of which could be pursued as a part of future studies.

Conclusions

By integrating movement ecology with landscape-scale land-use analysis, we demonstrate that MSFs dynamically respond to human-modified environments, with land-use type, time of day, and leading species collectively shaping their movements. These insights underscore the need for spatially and temporally explicit conservation strategies to sustain avian communities in tropical countryside agroecosystems. Although they cannot replace forests, agroecosystems can be optimized for their diversity, in this case ensuring a variety of land-use types, like riverine forests, marshes and mixed cultivations provide the requisite resources for MSF persistence throughout the diurnal cycle.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

MSF

Mixed-species flock

OBBA

Orange-billed Babbler

RSF

Resource Selection Function

RSPF

Resource Selection Probability Function

SSPF

Step Selection Probability Function

Author contributions

IW: Conceptualization, Methodology, Investigation, Formal analysis, Visualization, Writing – Original Draft, Writing – Review & Editing. DV: Investigation, Validation. SK: Conceptualization, Supervision, Resources, Administration, Writing – Review & Editing. AJ: Supervision, Resources, Administration, Visualization, Writing – Review & Editing. EG: Conceptualization, Supervision, Resources, Methodology, Formal Analysis, Writing – Original Draft, Writing – Review & Editing.

Funding

IW is grateful for support for her PhD studies from the Chinese Scholarship Council (CSC-2021SLJ009558), as well as to the Idea Wild team for their assistance for field equipment.

Data availability

The datasets used and/or analyzed during the current study are available from the co-corresponding authors on request.

Declarations

Ethics approval and consent to participate

We confirm that our research was conducted under the law and ethical standards of Sri Lanka and China. Approval for the fieldwork was obtained from Department of Wildlife Conservation (Permit no: W/L/3/2/24/2023).

Consent of publication

Consent form attached.

Competing interests

The authors declare no competing interests.