Diversity and activity of bird fauna in ephemeral river-created habitats in Amazonia

ABSTRACT Seasonal changes in runoff and inundation create river-habitats that are considered key to the diversity of Amazonian bird species. However, studies on avifauna in different riverine habitats remain scarce. We used camera traps to survey the avifauna in river-created habitats along 39 km of river in the eastern Brazilian Amazon. We examined the number of bird species detected and species composition in islands and margin banks, evaluated species–area relationships in these two habitats, and examined if the most abundant bird species have differences in activity pattern between habitats. Cameras recorded 21 bird species from 13 families (Accipitridae, Ardeidae, Bucconidae, Caprimulgidae, Cathartidae, Columbidae, Cracidae, Hirundinidae, Icteridae, Scolopacidae, Thraupidae, Threskiornithidae, and Tyrannidae). Species composition did not differ between habitat types but we found a discernible area effect, explaining as much as 84% on the number of species detected across island sites. Overlaps of activity pattern of bird species were high between habitat types, but differences in activity peaks were found for Molothrus oryzivorus and Pitangus sulphuratus while examining islands and margin banks. River-created habitats are relevant to Amazonian bird diversity and studies on these habitats must be increased to better understand their value across the Amazon.


Introduction
The annual rise and drawdown of river waters in the Amazon basin alters the landscape each year, creating riparian areas that alternate between terrestrial and aquatic phases (i.e.floodplain) (Hamilton et al. 2002;Carvalho et al. 2017).These seasonal changes in river flows generate a variety of river-created habitats, which include beaches and sandbars, sandbar scrub, river edge forest, varzea forest, transitional forest, and wateredge habitats (Remsen & Parker 1983;Sioli 1984).Such a diversity of areas that are under water during seasonally high river levels can provide habitats used by a variety of bird species during low river level periods.River-created habitats are important for Amazonian bird species, with approximately 15% of the nonaquatic avifauna of the Amazon basin dependent on these habitats (Remsen & Parker 1983).Additionally, differences in avifauna diversity between seasonally flooded and adjacent unflooded 'terra firme' forests were found across the Amazon (Haugaasen & Peres 2008;Beja et al. 2010), highlighting that habitat heterogeneity created by flooding contributes to maintaining bird diversity in Amazonian forest landscapes (Beja et al. 2010).
River seasonality can be seriously affected by hydropower developments, submerging areas permanently in remote regions that would otherwise be well protected (Norris et al. 2018a).The lack of baseline data documenting patterns in bird distribution, abundances and activity limits the capacity to develop effective mitigation and minimization actions across Amazonian rivers increasingly threatened by large and small hydropower developments (Latrubesse et al. 2021).
Additionally, the majority of studies on birds are focused to assess species richness and abundances, with only a small fraction of studies focusing on bird activity, which could be related to the inherent sampling difficulties in the field (Fonturbel et al. 2020).The temporal aspect of activity is not only relevant from an ecological perspective, but it can also provide information on changes to species behaviors and interactions resulting from anthropogenic alterations, and their impacts on community structure and niche partitioning (Frey et al. 2017).For example, habitat disturbance, with changes in forest structure, has been proven to affect bird daily activity patterns (Fonturbel et al. 2021).
During recent decades, camera traps have been proved to provide a cost-effective method to remotely monitor wildlife with low levels of observer interference (Burton et al. 2015;Sollmann 2018).Camera traps have been widely used to monitor terrestrial mammals, but their use to study birds is still relatively underexplored (Murphy et al. 2018;Luo et al. 2019;Fonturbel et al. 2021).Although usually used and considered most appropriate for large, ground-dwelling birds (Michalski et al. 2015;Paredes et al. 2017), camera traps have also been shown to be sensitive enough to detect small birds (O'Brien & Kinnaird 2008).Moreover, camera traps can provide similar results to those obtained by point counts for understory birds (Fonturbel et al. 2020).Camera traps are particularly useful for behavioral studies including studies on bird activity patterns, as cameras can operate 24 hours a day throughout long periods of time (Kuhnen et al. 2013;Luo et al. 2019;Mere Roncal et al. 2019;Fonturbel et al. 2021).
We used camera traps to document the number of species detected, composition, and activity patterns of birds in river-created habitats (islands and margin banks) in the State of Amapá, eastern Brazilian Amazon.Our objectives were to: (1) examine the number of bird species and species composition in two types of river-created habitats (islands and margin banks), (2) evaluate species-area relationships in these two habitats, and (3) examine if the most abundant bird species have differences in activity pattern between the two habitat types studied.Finally, we explore relevant considerations for management and conservation strategies for river-created habitats in the Brazilian Amazon.

Study area
The study was conducted along 39 km of the Falsino River (81-107 m.a.s.l.), in the State of Amapá, Brazil (N 0.924722, W 51.595833; Figure 1).The Falsino River flows for approximately 125 km and drains an area of 4211 km 2 (Linke et al. 2019).The 39 km river segment is perennial and runs between two sustainable-use protected areas, the Amapá National Forest and the Amapá State Forest (hereafter 'FLONA ' and 'FLOTA,' respectively).This particular stretch of river is 61 km from the nearest town and suffers relatively little anthropogenic influence (Norris & Michalski 2013;de Oliveira et al. 2015;Norris et al. 2018b), with only six houses in the river segment during our study period.
The river is immediately bordered by continuous forest cover (i.e. a border of canopy trees typically starts 1-4 m from the river's edge) and the riverbank rises abruptly along the studied section (Figure S1) such that forests close to the margin (e.g.110-554 m) are never flooded (Caron et al. 2021).The Falsino river flows through broadleaf evergreen tropical forests (Uatumã-Trombetas and Guianan moist forest ecoregions) (Olson et al. 2001;Dinerstein et al. 2017).The forests surrounding our study area are part of the eastern Amazon Guianan forests (Ter Steege et al. 2001) and consist predominantly of never-flooded closed canopy tropical rainforest vegetation dominated by members of the Fabaceae, Sapotaceae, Lecythidaceae and Lauraceae (Batista et al. 2015).The regional climate is classified by Köppen-Geiger as 'Am' (Equatorial monsoon) (Kottek et al. 2006), with the driest months from September to November (total monthly rainfall <150 mm) and the wettest months from February to April (total monthly rainfall >300 mm) (Paredes et al. 2017).The level of the Falsino River is also strongly seasonal (Figure S2), increasing twofold from low water (long-term average 345 cm in November) to high water (long-term average 711 cm in May), with water level rising rapidly in rare cases at the end of the dry season (e.g.2.2 m over 16 days) (Norris et al. 2020).
Historically, the water of Amazonian rivers has been broadly grouped into three classes (black, white and clear) based on chemical and physical characteristics (Sioli 1984;Junk et al. 2011).Based on such a classification the Falsino River is a 'clear-water' (Junk et al. 2011), which usually includes rivers that drain Precambrian Guiana and Brazil Shield kaolinite clay soils (Leenheer 1980;Sioli 1984;Junk et al. 2011).More recent advances (Dallaire et al. 2019) developed standardized global classification of river reaches that integrates five categories of variables: (1) hydrology, (2) physiography and climate, (3) fluvial geomorphology, (4) water chemistry, and (5) aquatic biology.Based on this system, the Falsino River is classified as 'very hot, high moisture region, medium river, medium and high stream power,' and rivers with the same reach types represent approximately 3.15% of Amazonian rivers by length (Figure S3).

Bird data
In order to sample bird species, we used 22 infrared camera traps (Bushnell Trophy Cam, 8 MP, Overland Park, KS, USA) to conduct a three-month assessment in the study area (September-December 2019).Cameras were installed on 19 sites (eight islands surrounded by the river and 11 margin banks bordered by both forest and river) during the low river level season (de Oliveira et al. 2015), when islands and sand banks along the river are exposed.During the study period river width ranged from 59 to 166 m (mean = 83 m) at the study sites and islands were 21 to 64 m (mean = 40.3m) from the nearest margin.Islands (median size = 636 m 2 ) and margin banks (median size = 622 m 2 ) were characterized by the presence of open sand banks and forest vegetation, which included shrubs and trees (see Figure S1).We maximized the spatial independence between sites by establishing an average (±SD, range) distance along the river of 15.5 km (±10.4 km, min.-max.= 0.05-39.0km, n = 342 comparisons).
Cameras (min.-max.= 1-2 per site) were unbaited, installed at 30-40 cm above the ground and faced the largest open areas of the island or margin bank.The open areas sampled were ephemeral, i.e. all were 100% submerged with the rise in river level from December 2019 to January 2020 and remain submerged until river level drops from August/September.Cameras functioned continuously (24 hours a day) and were configured in hybrid mode (taking three photos followed by a 40 s video film post-activation), with intervals of 15 s between videos, and date-time stamp enabled.For all subsequent analyses, we considered only independent records, considering only photos/videos with over 30 min intervals in case the same species was recorded during the same day on the same camera (Michalski 2010;Michalski & Norris 2011).The species recorded by camera traps were identified with the aid of standard field guides for regional birds (Sigrist 2008(Sigrist , 2009)).We checked species identification twice, conducting two independent assessments (CM, and FM) and contrasting them.For this study, we only considered birds with a body size >15 cm, weighing on average 507 g (range = 26-2872 g), which can be classified as medium and large-bodied birds (Lindenmayer et al. 2018).Scientific and English names follow an available checklist for South American Birds (Pacheco & Agne 2019).
We classified each species into one of five diet guilds: fruit-nectar, invertebrate, omnivore, plant-seed, and vertebrate-fish-scavenger (Wilman et al. 2014).Body mass was obtained from the literature (Wilman et al. 2014).Bird status category follows the IUCN Red List of Threatened Species (IUCN 2020).

Environmental data
All sampling sites were classified as islands or margin banks taking in consideration the location of the areas and the presence/absence of connectivity with the adjacent mainland forest along the Falsino River during the sampling period.Sites were selected based on the following criteria: areas of >5 m 2 of exposed sand and/or fine gravel that were sufficiently raised above the river level not to be waterlogged at a depth of 15 cm (Quintana et al. 2019;Michalski et al. 2020).The area of each site was measured by mapping the total surface using a handheld GPS in situ.For margin banks, we measured the largest open continuous area that we could identify as being part of the sampling site delimited between the river and forest border (Figure S4).

Data analysis
All analyses were performed in R (R Development Core Team 2020).To determine the distance between each site surveyed along the river as well as total river length sampled we used functions available in the riverdist package (Tyers 2017).To assess whether the sampling effort was sufficient to record the majority of bird species, we constructed and compared cumulative species curves with the specaccum function of the Vegan package (Oksanen et al. 2019).To assess the overall representativeness of our sampling effort with camera traps in the study sites, we used the First order jackknife estimator to extrapolate the number of species (i.e.estimate the number of undetected species) based on the frequency of detected species (function specpool, package Vegan) (Oksanen et al. 2019).
To test for differences in species composition between islands and margin banks we performed nonmetric multidimensional scaling (NMDS) ordinations based on Jaccard dissimilarity matrix of bird species records of all sites surveyed (function metaMDS, package Vegan) (Oksanen et al. 2019).We also performed a NMDS ordination based on the Bray-Curtis dissimilarity matrix weighted by the number of independent records of bird species for all sites (function metaMDS, package Vegan) (Oksanen et al. 2019).
To explain patterns in the number of bird species detected across sites we used generalized linear models (GLMs, error distribution family = poisson).As explanatory variables we used habitat type (island or margin bank) and total area (squared meters, log x + 1 transformed) of islands and margin banks.The influence of these variables on number of species detected was tested with separate GLMs.To improve numerical stability of the GLMs the continuous variable was standardized (centered and scaled by its standard deviation).To identify the best model we compared the difference in second-order Akaike's information criterion corrected for small sample sizes (AIC c ) values in relation to the first-ranked model (∆AIC c ) (Burnham & Anderson 2002).A value of ∆AIC c ≤ 2 indicates equally plausible models (Burnham & Anderson 2002).We then plotted the number of bird species detected against islands and margin banks area (log x + 1) in order to evaluate area-effects.
To assess bird activity, we extracted date and time for each bird record identified and converted time figures to radians (function gettime, package activity) (Rowcliffe 2019).Using the converted data, we fitted activity density kernels for each site type (island and margin bank) for all species recorded, and selecting only species with at least 12 independent records in three or more sites surveyed.Then, we performed pairwise comparisons on those activity patterns using the ∆ s overlap coefficient (Ridout & Linkie 2009), which gives the degree of overlap between two activity density kernels (overlapEst function, package overlap) (Ridout & Linkie 2009).Finally, we tested the significance of those comparisons using 999 permutations (compareCkern function, package activity) (Rowcliffe 2019).
Although the accumulation curve did not reach an asymptote for birds registered in islands and margin banks, we obtained 74.1%, and 72.3% of the expected bird species for these two habitat types (Figure 2).When considering the trophic guild, 10 species were classified with a diet based on invertebrates, four based on vertebrate-fish-scavengers, four based on fruitnectar, two were omnivorous, and only one was based on plant-seed (Table 1).Ten species (47.6%) were present in both habitat types (island and margin bank).Two species were only recorded in islands (Brownchested Martin -Progne tapera and Blue-gray Tanager -Thraupis episcopus), while six were only recorded in margin banks (Striated Heron -Butorides striata, Boat-billed Heron -Cochlearius cochlearius, Rufescent Tiger-Heron -Tigrisoma lineatum, Black Vulture -Coragyps atratus, Pale-vented Pigeon -Patagioenas cayennensis, and Black Curassow -Crax alector) (Table 1).Among all bird species, the Swallowwinged Puffbird and the Great Black Hawk had the highest number of independent records registered (53 and 42, respectively).
There was a discernible species-area relationship for the number of species detected in island habitats (R 2 = 0.84), which was not observed for margin banks (Figure 3).Similarly, models containing area [(either in a single model or with an interaction with habitat type (margin bank)] were also ranked as equally plausible when predicting patterns in the number of species detected at our sites (Table 2).Overall, both plausible models significantly explained variation in the number of species detected (model R squared = 0.56 and 0.26), reinforcing that area was the most important variable of those included in the models (Table 2).All other models, including the null model, and those containing only habitat type (islands vs margin banks) and habitat type with the additive effect of area were ranked lower based on their AIC c values (Table S1).
When considering species composition, islands and margin banks presented similar species that could not be differed in a nonmetric multidimensional scaling (NMDS) using either Jaccard or Bray-Curtis dissimilarity matrices (0.133 and 0.134 stress values, respectively) (Figure 4).
Among the seven species for which we compared activity patterns between islands and margin banks, the Great Black Hawk had the highest temporal overlap (0.89; Table 1 and Figure 5), whereas the lowest temporal overlap between the two habitats was for the Great Kiskadee (0.18; Table 1 and Figure 5).We found contrasting bird activity patterns between islands and margin banks for two species (Molothrus oryzivorus and Pitangus sulphuratus) (Table 1).While M. oryzivorus has a narrower activity pattern concentrated in the middle of the afternoon in islands, it shows a wider activity along the entire daylight in margin banks (Figure 5).P. sulphuratus showed a single peak of activity around midday in islands and bimodal crepuscular activity (early morning and end of afternoon) in margin banks (Figure 5).

Discussion
This study conducted in Amazonian river-created habitats showed that (1) area was the strongest determinant of the number of bird species detected; (2) activity patterns at islands and margin banks differed for some bird species; and (3) camera traps can be helpful for studying bird behavioral/habitat use patterns, providing valuable data on diel activity patterns.These observations allow an improved understanding of the importance of ephemeral river-created habitats for Amazonian birds and also provided a detailed characterization of the diel activity of several bird species.
We found a similar species composition between islands and margin banks, which may be a reflection of all sites receiving the same regime of rise and drawdown of the river each year.While some studies found differences in bird species composition between floodplain and 'terra firme' forests (Haugaasen & Peres 2008;Beja et al. 2010), others could not find such differences (Mere Roncal et al. 2019).Differences between bird species composition between river islands and 'terra firme' forests were also found in the Amazon basin (Cintra et al. 2007a).We found in our study species that are considered associated with river-created habitats such as Butorides striata, Tigrisoma lineatum, Hydropsalis climacocerca, Mesembrinibis cayennensis (Laranjeiras et al. 2019), and species such as Pitangus sulphuratus, that was considered to have evolved in Amazonian river created habitats (Remsen & Parker 1983).
Our rarefaction curves and the difference between the observed and extrapolated number of bird species indicate that our sampling effort was representative of the expected species at river-created habitats using this technique in our study area.We suggest that camera trap surveys of river-created habitats can provide a useful standardized method to complement audio/visual sampling and inform development of additional studies.Including additional audio/visual surveys would certainly increase substantially the species recorded.Bird surveys often concentrate monitoring activities during early morning/evening (Moncrieff et al. 2021).The photos showed a different activity pattern, with a peak of records around midday at both islands and margin banks.This suggests that the timing of visual and audio based surveys (e.g. point counts) may need to be extended to include midday hours to represent the diversity of species using different river-based habitats.Yet, these river-created areas would also be challenging to survey.For example, as these ephemeral river-based habitats remain exposed to direct sunlight, mist nets would need to be intensively/ continuously monitored to reduce risk mortality due to heat stroke/dehydration/stress.
The 21 bird species recorded in our study is similar to a number that was recorded with camera traps in the lowland Peruvian Amazon, with 16 species of terrestrial  Our results showed that area was the strongest predictor of the number of bird species detected, and together with the habitat type interaction explained as much as 56% of the overall variation in the species detected across all 19 sites surveyed.Area effect was particularly relevant for island habitats, showing that larger islands have larger number of bird species.Similar results were also found for bird species in forest patches, which can behave like real islands in hostile matrix (Michalski & Peres 2017).Similar results, with smaller number of vertebrates (including birds) in smaller islands were also found for permanent islands created by a hydroelectric dam in the Amazon (Benchimol & Peres 2015).
Overall, the activity pattern of all bird species studied here did not differ between islands and margin banks, with birds showing an activity peak around mid day.This result is in line with what was observed in another Amazonian study (Griffiths et al. 2020), as well as a study conducted in Amazonian floodplain and 'terra firme' forest that showed birds with activity pattern characterized as either diurnal or mostly crepuscular with some diurnal activity (Mere Roncal et al. 2019).Although habitat disturbance has been shown to reflect differences in bird activity patterns (Fonturbel et al. 2021), there is no reason to see differences in our sites surveyed apart of their differences in habitat type, as our study region is largely undisturbed (Norris et al. 2018b;Quintana et al. 2019;Michalski et al. 2020).Thus differences in activity pattern found for the two bird species (Molothrus oryzivorus and Pitangus sulphuratus) can only be attributed to habitat type.However, because of our limited temporal replication, as we only sampled one season in one year, our results should be interpreted with caution.
Finally, our results concur with previous studies that found camera traps to be useful for documenting the presence of birds, identifying birds to species level and contributing relevant data on activity patterns (O'Brien & Kinnaird 2008;Ridout & Linkie 2009;Murphy et al. 2018;Mere Roncal et al. 2019;Griffiths et al. 2020).Along with the employment of almost noninvasive techniques such as camera traps, we believe the data on ephemeral river-created habitats provides valuable information on bird species levels, as well as highlighting the importance of these habitat types that are usually overlooked in the scientific literature.

Figure 1 .
Figure 1.Location of the study region between the Amapá National Forest (FLONA) and the Amapá State Forest (FLOTA), Amapá State, eastern Brazilian Amazon.(a) Amapá State in Brazil; (b) FLONA and FLOTA (polygons) in Amapá State; (c) linear dark gray representing Araguari and Falsino rivers (with white and dark gray circles representing islands and margin banks, respectively) where camera traps were installed.The star shows Porto Grande city, the nearest town in the study region.

Figure 2 .
Figure 2. Cumulative curves for bird species sampled with camera traps in islands and margin banks in the study area.Detection of species recorded in site randomized 1000 times and results used to derive mean (black line) and 95% confidence intervals of the mean (light gray polygon).(a) Cumulative curve for bird species in all sampled sites; (b) cumulative curve for bird species in islands; (c) cumulative curve of bird species in margin banks.Dashed horizontal lines represent extrapolated number of species pooled across sample sites.

Figure 3 .
Figure 3. Relationship between area (log 1 + x m 2 ) and number of bird species detected with camera trapping in islands and margin banks.Regression lines (mean) and shaded areas (±95% CI) were obtained from model predictions.

Figure 4 .
Figure 4. Nonmetric multidimensional scaling (NMDS) ordination for bird species sampled with camera traps in islands and margin banks in the study area.NMDSs were based on Bray-Curtis dissimilarity matrix weighted by the number of independent records of birds in all sites surveyed, and on Jaccard dissimilarity matrix of bird species records in all sites surveyed.Circles are scaled to the area (m 2 ) of islands and margin banks.Black and gray circles represent islands and margin banks, respectively.

Table 1 .
Status category, trophic guild, number of occupied sites, number of independent records in islands and margin banks, and coefficient of overlapping (∆) for activity density kernels between islands and margin banks for all the 21 bird species examined.
(Table1), occurring in over 50% of all sites surveyed

Table 2 .
GLM model results (slope coefficients with associated ± SE in parentheses) of predictors of number of bird species at islands and margin banks in the eastern Brazilian Amazon.Only coefficients from the 'best models' based on AIC c model selection (delta AIC c < 2) are shown.Significant P values are shown in bold.
(Wilman et al. 2014)20ncal et al. 2019).Similarly, another camera trap study in the Peruvian Amazon identified 13 bird species (weight >1 kg) in mineral licks(Griffiths et al. 2020).Although we included smaller birds (body size >15 cm) in our study, nearly half of all birds sampled (47.6%) have a body mass larger than 180 g(Wilman et al. 2014), which makes our results comparable with those previous studies.