ABSTRACT: Aim: A key goal in ecology is to develop more effective ways to understand species’ distributions in order to facilitate both their study and conservation. Many species distribution modeling analyses have been performed to date, using either structured survey data or unstructured citizen science data; these two pools of data have tradeoffs in terms of data density, spatiotemporal coverage, and accuracy. Recent studies have shown that combining structured and unstructured survey data can greatly improve the accuracy of species distribution models for birds, but most of this work has focused on north temperate bird species and uses bird atlas data that is much more common in the temperate zone than elsewhere. We sought to adapt a data pooling approach from the literature on north temperate bird biology to create distribution models for a selection of secretive suboscine bird species that occur in a highly diverse region of the southwestern Amazon. We adapted a data pooling methodology which typically uses structured point count data and eBird observations by substituting the former with acoustic detections collected with autonomous monitoring devices. The pooled dataset was used to model encounter rate, assess habitat variable importance, and predict site occupancy probabilities. Models produced with the pooled dataset performed better than those produced with the eBird-only dataset and yielded occupancy patterns that better reflect the major habitat gradients in this region. Our result supports prior work on eBird data pooling and shows that using acoustic rather than human survey data is a scalable approach to modeling the distributions of other species in Amazonia as well as more broadly across the global tropics.

KEYWORDS: birds, neotropics, acoustic monitoring, species distribution modeling, Amazon, eBird

INTRODUCTION: Quantifying the environmental factors that influence organismal occurrence patterns has helped address key research questions in 21st century biology (Guisan & Thuiller, 2005; Brotons, Herrando, & Pla, 2007; Shirley et al., 2013). Some of these include describing life history strategies (Brambilla & Ficetola, 2012; Segal, Massaro, Carlile, & Whitsed, 2021), estimating the relative contributions of ecological partitioning and competitive interactions in community structuring (Freeman, Strimas-Mackey, & Miller, 2022), and predicting threats due to anthropogenic change (Velásquez-Tibatá, Salaman, & Graham, 2013).​​ A key research tool in this work is species distribution modeling, which relates information on species occurrence (presence, presence-absence, and/or abundance) to environmental covariate sets (Peele, Marra, Sillett, & Sherry, 2016; Stowell, Wood, Stylianou, & Glotin, 2016; Ahmed, Atzberger, & Zewdie, 2020; Stowell 2022). Most of this work has focused on north temperate species due to longstanding historical biases in biological research (Stutchbury & Morton, 2008). While applying species distribution modeling techniques to vastly more speciose tropical communities is becoming more feasible as data steadily accumulates over time and more efficient surveying techniques are developed (Fernández‐Arellano et al. 2021; Rumelt, Basto, & Mere Roncal, 2021), data density shortcomings still represent significant research impediments.

The Amazon rainforest is one of the most speciose regions for birds globally (Amadon, 1973; Haffer, 1978). Although much theoretical ecology work in this region has examined rivers and other habitat discontinuities (Hayes & Swelal, 2004; Fernandes, Cohn-Haft, Hrbek, & Farias, 2014; Oliveira, Vasconcelos, & Santos, 2017) or interspecific competitive interactions (Pearson, 1977; Brooks, 1998) as drivers of diversity, it is clear that they have also been shaped by local-scale spatial gradients and habitat heterogeneity over evolutionary timescales (Stratford & Stouffer, 2013; Castilheiro, Santos-Filho, & Oliveira, 2017; Fluck, Cáceres, Hendges, Brum, & Dambros, 2020; Maximiano et al., 2020). Efficiently gathering data on how these gradients mediate bird occurrence patterns has the potential to answer important questions about community assembly processes and habitat partitioning within clades of Amazonian birds.

Acoustic monitoring is a key technique for surveying tropical rainforest birds as it requires limited person-hours in the field, is more effective at detecting species living in heavily vegetated understory than camera traps or human observers, and can be used to survey for long periods of time, providing a potentially rapid solution to current data limitations in tropical regions (Alvarez-Berríos et al., 2016; Leach, Burwell, Ashton, Jones, & Kitching, 2016; Rumelt et al., 2021). Due to the a priori nature of acoustic survey design, this technique has high location specificity, low spatiotemporal bias, and targeted coverage of local-scale habitat gradients thought to be important to the study species (Pacifici et al., 2017; Robinson, Ruiz‐Gutierrez, & Fink, 2018; Miller, Pacifici, Sanderlin, & Reich, 2019). By repeatedly sampling survey sites, researchers can also use these data in occupancy models to jointly predict occurrence and detection processes (Mackenzie et al., 2002). As with other types of field surveys, however, acoustic monitoring projects are limited by cost, logistics, and scheduling concerns, and are typically only used to gather data on specific sites or areas and at specific times of year. An important research priority is to find effective ways to combine acoustic survey data with other datasets to increase generalizability over larger spatial areas.

Citizen science datasets have shown real promise as a means of filling spatial gaps in ecological data (Dickinson, Zuckerberg, & Bonter, 2010; Bonney et al., 2014; Bradter et al., 2018; Gouraguine et al. 2019) as they have high data densities, are free to use, and cover broad spatial scales. However, they also typically contain higher levels of spatiotemporal survey bias and greater unevenness in observer quality relative to traditional survey methods (Crall et al., 2011; Aceves-Bueno et al., 2017). A variety of research has therefore examined how to mitigate sources of bias in citizen science data (Pacifici et al., 2017; Miller et al., 2019, Steen, Elphick, & Tingley, 2019; Johnston et al., 2020, Feldman et al., 2021; Van Eupen et al., 2021). The largest citizen science project for birds is the eBird project (Sullivan et al. 2009; Wood, Sullivan, Iliff, Fink, & Kelling, 2011; Sullivan et al., 2014), with 70.5 million bird observation records submitted by casual observers around the world (eBird, 2022), and eBird-based species distribution modeling methods, typically incorporating the use of 500 m spatial resolution MODIS landcover variables to describe important factors mediating occurrence, have been in development for over a decade (Fink et al., 2010; Fink et al., 2014; Kelling et al., 2015; Fletcher et al., 2019, Johnson et al., 2020; Fink et al., 2020). Several recent publications have demonstrated that combining eBird data with small amounts of structured survey data collected from the same geographic area and time period (referred to in the literature as “data pooling;” Fithian, Elith, Hastie, & Keith, 2015) is a simple yet powerful way to increase the predictive power of eBird species distribution models (SDMs) (Johnson et al., 2020). This approach has the potential to address long-standing regional biases in ornithology and rapidly scale up to enhance distributional knowledge of large numbers of bird species. However, prior integrated modeling approaches have combined eBird data with structured survey data from The Nature Conservancy and the Breeding Bird Atlas (Robinson et al., 2020; Johnston et al., 2021), and similar large bodies of survey data are largely unavailable in the global tropics. An important step in scaling up the data pooling approach is therefore a rigorous test of this methodology’s generalizability to other forms of structured survey data, including acoustic monitoring data, that can serve as a replacement for human survey data in regions where the latter does not exist.

Here, we apply a framework that integrates data from acoustic surveys with eBird observations to model the occurrence of several species of secretive interior forest suboscine songbirds in a highly diverse region of primary forest in southwestern Amazonia. This is the first time an integrated modeling framework combining acoustic monitoring and citizen science data has been applied to Amazonian birds. We aim to test the impact of this data pooling approach, along with a high resolution land cover dataset of the primary ecological gradients in this region (finer spatial scale than the 500 m grid typically used for eBird species distribution modeling) on the classification accuracy of eBird-based SDMs produced for bird species whose life history characteristics make them less likely to be detected by human observers. This is a first test of this methodology prior to its potential application to other study systems in the Neotropics as well as other regions of the global tropics.

MATERIALS AND METHODS:

1. Survey site and study species All acoustic survey data were collected at Los Amigos Biological Station, which contains approximately 145,000 hectares of lowland primary rainforest in the province of Madre de Dios in southeastern Peru (Pitman 2010), on the station’s existing transect grid (Tobler, Carrillo‐Percastegui, Leite Pitman, & Powell, 2008). Habitat structure in this region is heterogeneous, being structured by successional gradients associated with increasing distance from the channels of large rivers (Gentry, 1993), and this habitat diversity occurs alongside high avian alpha diversity (~600 species: eBird, 2022). The two high-level forest habitat types present at the site are floodplain forest (flooded for at least part of the year) and terra firme(never flooded); differences in time since disturbance, soil moisture, and edge effects are responsible for creating distinct understory vegetation characteristics in these two habitats as well as the ecotone between them (Gentry, 1993; Nobre et al., 2011). Members of several avian clades regularly segregate spatially by habitat type, suggesting that these high habitat and species diversity metrics are related (Rosenberg, 1990; Harvey, Aleixo, Ribas, & Brumfield et al., 2017; Mere Roncal et al.; 2019).

We focused our analyses on a set of five suboscine songbirds in Furnariida, an extremely speciose and ecologically varied clade in Amazonian forests (Oliveiros et al., 2019; eBird, 2022): Formicarius analis, F. colma, Myrmothera campanisona , Akletos goeldii, and Oneillornis salvini (Table 1). We chose these species because they had the most occurrences in the acoustic dataset among the Furnariida present at Los Amigos. Although these five species vary with respect to their utilization of the key ecological gradients in this region, all are terrestrial or semi-terrestrial, cryptically-colored, and have strong affinities for intact forest with heavy understory vegetation (Marcondes & Brumfield, 2019), which makes them important indicators of forest disturbance (Laurence, Stouffer, & Laurence, 2004; Thornton, Branch, & Sunquist, 2012). This set of life history traits makes detecting our target species using in-person survey methods challenging; however, they all have loud, distinctive vocalizations that are easy to record with acoustic monitors.2. Acoustic data collectionAcoustic data were collected as part of an automated monitoring project using 10 SWIFT autonomous recording units (ARUs; Kahl et al., 2019) cycled across 34 survey points (Figure 1a) spanning the main habitat gradients at this site. One field season ran from late July to early October 2019, during the dry season in southeastern Peru; the second ran during the wet season from early January to late February 2020 (Pitman 2010). Recorder sampling frequency was set to 16kHz to cut acoustic frequencies greater than 8kHz (Landau 1967); this allowed us to reduce recording file size while retaining acoustic fidelity in the frequency bands in which our target species’ vocalizations occur (<3.5 kHz). Microphone gain was left at the default -33 dB. Recorders were placed ~1m above the ground with the microphone facing towards the ground, which is optimal for recording terrestrial birds. Although this survey methodology was designed for tinamous, these parameters are also appropriate for capturing the vocalizations of terrestrial suboscines as the two clades broadly overlap in forest strata and acoustic frequency bands. Recording during the 2019 field season took place continuously between 5:00 and 7:30 hrs and 16:00 and 18:30 hrs to capture the dawn and dusk choruses, respectively. A change in the scope of the tinamou project during the 2020 field season led us to increase our recording windows to continuously from 3:30 to 7:30 hrs and from 14:30 to 18:30 hrs, as well as 15 minutes every half-hour between 8:30 hrs and 14:00 hrs and between 19:30 and 24:00 hrs. Additional details of recording site selection, placement, and recording parameters for the 2019 field season are described in [removed for double-blind review].

3. Processing of acoustic data Survey audio was captured as a series of 30 minute WAV files (or 15 minutes outside of peak recording times during the 2020 season) spanning each day’s recording period. Processing was performed using the BirdNET-Analyzer platform, which is pre-trained on a list of ~3000 bird species that includes our five target species and much of the avifauna of Amazonian Peru (Kahl et al., 2021). The classifier examines 3-second moving windows spanning the source audio and returns possible detections with associated detection probabilities (see Table S1 for full classification parameters). A characteristic of all probability-based moving window classifiers is the need to set a minimum threshold for accepted detections. We lumped overlapping and adjacent detections for each species into a single acoustic event and calculated maximum and mean detection probability (p_maxand p_min) for each event. All events with p_max > 0.80 were retained; these parameters were chosen by examining a random sample of 300 detections for each species (or all detections if the number was less than 300) and selecting a cutoff that eliminated the false positives in each sample. Positive detections for each species were plotted on a map and outliers manually checked for accuracy. Finally, these data were joined to a database of all recordings made during the two field seasons as well as a lookup table containing date-time and spatial metadata; this had the effect of creating a presence-absence dataset in which 30-minute recordings containing the target species are labeled “presences” and those without “absences” (we hereafter refer to presence-absence datasets formatted in this manner as “zero-filled” as this is the term used in the literature on eBird SDM; Strimas-Mackey et al., 2020).

4. eBird data and landcover covariates Occurrence data for the five target species were acquired from the eBird data portal (Sullivan et al., 2014). Data were pre-processed using the ‘auk’ package in R (Strimas-Mackey, Miller, & Hochachka, 2018) to select for checklists containing our five target species that were submitted in Madre de Dios between 2010-2022 using semi-structured protocols (Stationary or Traveling). This list was further refined to retain complete checklists (those where the observer specifies that they reported all species they were able to identify) with survey duration less than five hours, effort distance less than 5 km (for traveling checklists), and number of observers less than 10 in order to increase sampling effort evenness within the dataset. As with the acoustic survey data, eBird presence data were joined to the complete database of eBird checklists from this region with the same filter parameters and zero-filled; this had the effect of creating a presence-absence dataset in which checklists containing the target species were reported as “presences” and those without as “absences.” The canonical eBird species distribution modeling framework (Strimas-Mackey et al., 2020) uses a MODIS land-use/land-cover dataset (Freidl et al., 2002) as the basis for analyses, with elevation as an auxiliary continuous predictor variable. As the MODIS land cover dataset lumps most Amazonian forest types into an “evergreen broadleaf forest” class, and is relatively low resolution (500 m), we felt it necessary to develop a land cover dataset that is resolved to a finer spatial scale and contains more nuanced land cover classes that are relevant for analyses in the Amazon. The base layer for our land cover classification is the Sentinel-2 10m Land Use/Land Cover Time Series (Kara et al., 2021), which contains 11 land cover classes. We augmented class four of the Sentinel-2 dataset (“Flooded Vegetation”) by superimposing two land cover classes (“occasionally open water” and “always inundated”) from a synthetic aperture radar dataset of forest inundation (Sheng & Alsdorf, 2005). In order to better capture the natural gradient that exists between terra firme and floodplain forest types in lowland Madre de Dios, we chose to evaluate areas covered by class two in the Sentinel-2 dataset (“Trees”) with three types of continuous data: the SRTM+ 30m Digital Elevation Model (DEM) (Farr et al., 2007), height above nearest drainage (HAND) (Nobre et al., 2011) calculated from the same DEM, and a Sentinel-2-derived canopy height layer (Lang, Jetz, Schindler, & Wegner, 2022). Although we produced our land cover dataset at the spatial extent of the Amazon basin, we clipped it to the outline of the Madre de Dios province for the purposes of this study. All raster values were resampled to 30m for improved processing efficiency. GIS layer manipulation was performed using QGIS version 3.26.0 (QGIS.org 2022).

5. Modeling framework Our modeling approach follows that used by Robinson et al. (2020), Strimas-Mackey et al. (2020), and Johnson et al. (2021), unless otherwise noted. Although Strimas-Mackey et al. (2020) discusses how to use eBird data to create encounter rate, single-species occupancy models, and GAM-based abundance models, our acoustic data do not capture abundance, and we are therefore only able to perform the first two types of analyses here. All model processing was performed in R version 4.2.0 (R Core Team, 2022).5a. Encounter rate Encounter rate is a frequently-used metric in the literature on eBird SDM modeling that is proportional to occupancy and uses survey date-time, duration, distance covered, and number of observers as covariates in order to reduce between-site variability in detection (Guillera-Arroita et al., 2015; Strimas-Mackey et al., 2020). To produce encounter rate models for our species and evaluate the benefit of including structured survey data from the acoustic dataset, we used a bootstrapping analysis to generate submodels either from eBird data alone or from the pooled eBird+acoustic dataset (n=25 per analysis; Johnston et al., 2021). In each bootstrap iteration, the eBird dataset for our target species was spatiotemporally subsampled by overlaying a 3 km hexagonal grid created using the ‘dggridr’ package (Barnes & Stahr, 2017) over Madre de Dios and sampling one observation per grid cell per week; positive and negative detections were sampled separately to reduce class bias (Robinson et al., 2018). Spatiotemporal subsampling was not performed on the acoustic survey data because our strategy of placing recorders at set intervals along transects inherently reduced spatial bias. For the set of iterations using pooled data, we combined the two datasets by treating each acoustic sampling event as a pseudo-checklist submitted under the Stationary protocol with number of observers = 1 and duration = 30 or 15 min (depending on the length of the audio file); this is an established strategy in the literature for combining structured survey data with eBird data for occupancy modeling (Robinson et al., 2020). Land cover summary data were extracted for all spatial points by setting a 300m buffer radius around each point and calculating percent coverage of each land cover class and mean of the continuous predictor datasets within the buffer (“pland_XX_” and “X_median”, respectively). A class-balanced, GAM-calibrated (Robinson et al., 2018) random forest model was produced from the subsampled eBird data using the ‘scam’ (Pya, 2021) and ‘ranger’ (Wright & Ziegler, 2017) packages to model encounter rate. We used an 80%-20% data partition to train and evaluate each model, respectively; the validation dataset for the combined dataset contained both eBird and acoustic data. Models were evaluated based on: area under the curve (AUC), which measures the degrees to which the model correctly predicts presence localities and minimizes false presence predictions across all binary cutoff thresholds; sensitivity, the probability that a given presence prediction represents a true presence; specificity, the probability that a given absence prediction represents a true absence, and Mean-Squared Error (MSE), which measures the mean of the squared deviations between predicted probability of presence and true presence-absence state. These metrics were as applied to the hold-out set for the model produced in each bootstrap iteration and used to assess changes in accuracy between the two data pooling approaches. Package ‘ranger’ also calculates relative predictor importance for each model using the Gini Score (Gini, 1936), where value of importance is equal to the mean decrease in model accuracy when the predictor is excluded during the fitting process (Strimas-Mackey et al., 2020), and we extracted the importance values from each model to determine whether relative predictor importance was different between the eBird and pooled bootstrap runs. Finally, we made two prediction surfaces for Madre de Dios, one by taking the mean of the output rasters of the 25 bootstrap runs using just eBird data and the other from the mean of the output rasters from all 25 runs using eBird+acoustic data.

5b. Occupancy modeling Occupancy modeling was used to jointly model occurrence and detection probability as separate processes. This modeling approach involves partitioning predictors into site- and observation-level subsets and selecting between several candidate models using subsets of these predictors. eBird data can be used for occupancy modeling, but it requires special processing to select a subset of the data that represents repeat visits by the same observer to the same site within closure periods (Mackenzie et al., 2002) to meet the standard set of occupancy model assumptions (Strimas-Mackey et al., 2020). We used observation start time, checklist duration, day of year, number of observers, and distance traveled as detection covariates for both datasets (number of observers and kilometers traveled set equal to 1 and 0, respectively, for all acoustic records). All predictors were standardized by subtracting the mean and dividing by the standard deviation, and any correlated predictors (r ≥ 0.75) were removed. Spatiotemporal subsampling was again applied to eBird data: we discovered that the additional constraint of only including repeat visits necessitated less stringent subsampling to ensure adequate sample sizes for analysis, so we randomly selected 10 samples per site per 3 km grid cell rather than one. As with the encounter rate model we did not spatially subsample the acoustic data.

We first created single-species occupancy models using the R package ‘unmarked’ (Fiske & Chandler, 2011) with eBird data. We evaluated our global models for goodness-of-fit using SSE, χ2, and Freeman-Tukey tests (n=1000; MacKenzie & Bailey, 2004). We then performed AICc-based model selection using the “dredge” function from MuMIn(Bartoń, 2022), which evaluates all possible models. We retained all models with ΔAICc ≤ 2.5 and used model averaging to produce a final occupancy model (Supplement). While we used the same procedure to produce pooled models, it was critical to address the issue of differing detection/non-detection ratios between the two component datasets. The nature of our acoustic survey design yielded long segments of audio containing zero bird detections, mainly in the middle of the day and after dark. By contrast, it is exceedingly rare for eBird observers to submit “null” checklists where zero species are reported. As a result, the number of sampling events was much higher for the acoustic dataset than for the eBird dataset, with a correspondingly lower detection/non-detection ratio. As working with data where this ratio is closer to one seems to be key for producing this type of occupancy model (Strimas-Mackey et al., 2020), we subsampled the acoustic detections so that the total number of sampling events and the ratio of positive to negative detections was the same as in the eBird dataset before creating the pooled dataset. The end result of each modeling attempt was a raster layer for Madre de Dios consisting of occurrence probabilities for the given species. AUC, sensitivity, and specificity metrics were collected from each occupancy model by comparing site detection/non-detection status with predicted occurrence probability and choosing a cutoff threshold that maximizes Cohen’s Kappa (Johnston et al., 2021).

RESULTS:

Our total sampling effort amounted to 5792.5 hours of collected audio, 1216.5 from the 2019 field season and 4575 from the 2020 field season, of which 66% came from terra firme and 34% from floodplain. Relative time allocation within terra firme and floodplain was approximately equal to the number of recorder-deployments within each habitat (42 vs. 23; 65% vs. 35%). We recorded 222 independent detections for O. salvini , 159 for F. analis , 100 forF. colma , 99 for A. goeldii , and 45 for M. campanisona . Filtering the eBird dataset with the chosen effort metrics yielded 1789 records for F. analis , 624 for A. goeldii , 279 for F. colma, 232 for M. campanisona , and 217 forO. salvini . In the encounter rate analysis, AUC values were higher for all species using the pooled models than using the eBird-only models, with mean AUC increasing from 0.82 to 0.96 (Figure 2). This pattern appears to have mainly been driven by increases in specificity (all species had higher or equivalent specificity values in the pooled models, versus one out of five with higher sensitivity, Figure 2). The habitat variables that were most important for the encounter rate models for all species, irrespective of data source, were elevation_median ,HAND_median , and canopy_cover_median (Figure 3). In the pooled models however, the absolute importance of all habitat predictors was higher than in the eBird-only models, and the importance of detection covariates relative to habitat predictors generally decreased. The covariate time_observations_started , which was the most important predictor variable in the eBird-only models for all species except M. campanisona , dropped in relative importance in the pooled models to between 2nd and 5th place in all but one species. Similarly, duration_minutes , which was among the top five predictors for all species in the eBird-only models, decreased in relative importance for all five species in the pooled models. By contrast, elev_median increased or maintained its relative importance in the pooled models for all five species, andhand_median increased or maintained its relative importance in the pooled models for four of the five species.

Occupancy model metrics showed mixed results with respect to overall model performance. When comparing the pooled occupancy models to the eBird-only models, AUC values were higher in two species, showed no appreciable difference in one species, and shifted lower in the other two species. However, breaking AUC into its component pieces suggested that specificity was much higher (with concomitant decreases in sensitivity) in the floodplain species’ pooled models, and vice versa for two of the three terra firme species (Figure 4). Inspection of the parameter estimates showed that in all but one species, adding acoustic survey data increased the significance of one or more of the three habitat variables that were a priori though to have the most influence on occurrence (Table 2) (the exception, M. campanisona , was likely due to sample size limitations as this was the species with the lowest number of total detections in the combined dataset). Visual inspection of occupancy model outputs (Figure 5) revealed a similar story: eBird-only models for all species predict essentially constant rates of occupancy across forested sites in Madre de Dios, while the pooled occupancy models conform much more closely to the terra firme -floodplain gradient and exhibited broadly realistic occurrence patterns (i.e., occupancy for terra firme species was higher interra firme , and vice versa for floodplain species). The exception was unrealistic occupancy values observed for the threeterra firme species in montane areas of Madre de Dios, which is likely the result of extrapolating the positive relationship between elevation and occupancy far outside the parameter space of the models. We found little evidence for lack of fit in most species-model combinations, though the pooled model for F. colma, eBird-only model for Myrmothera campanisona and both models for O. salvini each failed one of the three statistical tests (Table S2).

DISCUSSION:

Our results broadly support previous work showing citizen science data can be combined with small amounts of structured survey data collected over ecological gradients of known importance to target species to better model species distribution patterns (Pacifici et al., 2017; Miller et al., 2019, Robinson et al., 2020, Johnston et al., 2021). Including acoustic survey data led to significantly improved model accuracy and error metrics for encounter rate models. Improvements to the occupancy models was more modest, but they nonetheless yielded far more realistic habitat parameter estimates and occurrence maps than the eBird-only occupancy models

A potential explanatory factor for these results is that eBird checklists in Madre de Dios are non-randomly distributed, being most common near the major rivers that serve as the main means of transportation in this region (Figure 1b). We included traveling checklists with lengths <5 km in this analysis, and although this ensured we had access to an adequate amount of data for modeling, this distance is large relative to the length of the terra firme -floodplain gradient. Many eBird records in this region are submitted to station or lodge “hotspots” that represent a network of trails extending across a larger area, leading to hotspots containing a mixture of terra firme and floodplain data. The coordinate point used for a hotspot is therefore often not representative of the habitat in which data collection occurred, which makes using eBird data for inference over short, steep ecological gradients more challenging. It is therefore likely that eBird data more accurately reflects the habitat affinities of floodplain species than terra firme species, and the overall effect of adding acoustic data to the occupancy models is an improvement in the models’ ability to predict where terra firmespecies do occur, and where floodplain species do notoccur. The fact that key habitat variables increased in importance relative to detection covariates with the addition of structured survey data in the encounter rate analysis provides further evidence for this assertion by showing that the observed improvements are being driven in large part by higher weighting of habitat variables relative to detection covariates.

That integrated modeling can be used to improve SDM models for Amazonian birds has important implications for understanding the distribution of tropical species, as does the finding that these improvements can be realized with acoustic data. Despite being vastly more speciose than the north temperate zone, the Tropics have significantly lower participation rates in eBird (La Sorte & Somveille, 2020), yielding lower overall data densities. Although we are aware of a handful of recent studies that have used eBird data to perform SDM analyses specific to tropical birds (Girish & Srinivasan, 2022; Ramesh, Gupte, Tingley, Robin, & DeFries, 2022), these studies were both conducted in regions of South Asia that have higher eBird data densities than the Tropics as a whole (La Sorte & Somveille, 2020), and neither used an integrated modeling approach similar to ours or those of Robinson et al. (2020). An important characteristic of eBird observations is that they are non-randomly distributed, with preference given to areas near roads and known public recreation areas (Zhang 2020). This makes using eBird data for modeling particularly challenging in regions such as the Amazon that have high levels of beta diversity and relatively few, narrow transportation corridors relative to their land area. Low availability of eBird data away from these corridors can give rise to unrealistic model predictions, particularly when there is a known association between habitat and transportation corridor placement, i.e., major rivers being associated with both boat traffic and floodplain forest (Figures 1b and 5). Our results suggest that including a small quantity of structured survey data away from human movement corridors alleviates some of these problems.

Another advancement is the use of acoustic monitoring rather than in-person field surveys as the structured dataset. Prior integrated modeling approaches using eBird data have incorporated data from TNC and the Breeding Bird Atlas (Robinson et al., 2020, Johnston et al., 2021). Few tropical countries have a significant history of conducting similar bird atlas surveys (Horns, Adler, & Şekercioğlu, 2018); by contrast, an increasing number of biological stations have available acoustic survey data (Alvarez-Berríos et al., 2016; Do Nascimento, Campos-Cerqueira, & Beard, 2020; Blake, 2021; Campos et al., 2021; Rumelt et al., 2021). We argue that acoustic surveying techniques are well suited to addressing this data gap; providing a similar quality of data for highly vocal species as human observations (Blake, 2021), they have the added benefits of being non-invasive and easier to conduct for extended periods of time. They also have a proven history of providing critical information on poorly-known species (Pérez-Granados & Schuchmann, 2021; Zhong et al., 2021; Pérez-Granados & Schuchmann, 2022). In the past, the use of acoustic survey data for landscape-scale analysis has been hindered significantly by the need for manual processing of audio by trained listeners with intimate familiarity of the avian soundscape being examined (Sugai, Silva, Ribeiro, & Llusia, 2019); however, this barrier is rapidly being overcome withy automated detection techniques, either project-specific (Rumelt et al., 2021) or publicly-available classifiers such as BirdNet (Kahl et al., 2021).

It is important to note that future work incorporating acoustic monitoring data for occupancy modeling should involve careful consideration of survey design and the meaning of model outputs. We have used acoustic monitoring data as a stand-in for human survey data in this analysis and integrated it with a larger dataset of human observations, but we would be remiss for claiming that they offer exactly the same patterns of inference. Although acoustic monitoring has efficiency benefits over human surveys (Blake, 2021), human observers are capable of detecting non-vocal individuals by sight, and it is probable that two models generated for a less vocal species, one using human survey data and the other using acoustic monitoring data, would generate different conclusions about that species’ patterns of occupancy. Our analysis preferentially looks at a small set of secretive but highly vocal species for which acoustic monitoring is a highly suitable survey approach, but for other species or clades the considerations could be quite different. Furthermore, some species’ rates of vocalization may vary with habitat quality across their home ranges (Wood & Peery, 2022): for instance, lekking species that have particular habitat constraints on lek site selection (Durães, Loiselle, & Blake, 2008). Therefore, depending on the question being asked, it is important to consider whether the chosen survey design will yield “acoustic occupancy” rates that are comparable to the species’ true occupancy. A study that directly compared the use of point-count surveys and acoustic monitoring data in an integrated modeling framework like the one explored in this analysis would be fascinating and much appreciated. Generating a more complete picture of occupancy patterns of Amazonian birds is a critical research priority given the extent of anthropogenic change in this region. The Amazon at large currently experiences threats from deforestation (Silva-Junior et al., 2021) and fire (Silveira, Silva‐Junior, Anderson, & Aragão, 2022), and the area bordering the Ruta Interoceánica, a newly-constructed east-west road connecting Madre de Dios and the Peruvian Andean provinces with the Brazilian state of Acre, is a major deforestation hotspot within Peru (Sánchez et al., 2021). Much of this deforestation is intended to clear land for agriculture and is rapidly changing the biological characteristics of this region. Madre de Dios has also become a hotspot of illegal artisanal gold mining in recent years: this activity denudes floodplain areas near rivers and oxbow lakes of vegetative cover and releases large quantities of mercury into the environment (Diringer et al., 2019; Gerson et al., 2022). Understanding which species’ habitat use characteristics make them most vulnerable to anthropogenic pressures will be crucial for informing conservation efforts and acquisition of land for the establishment of protected areas.

CONFLICT(S) OF INTEREST:

The authors declare no conflict of interest.

REFERENCES:

Aceves-Bueno, E., Adeleye, A. S., Feraud, M., Huang, Y., Tao, M., Yang, Y., & Anderson, S. E. (2017). The accuracy of citizen science data: a quantitative review. Bulletin of the Ecological Society of America , 98(4), 278-290.

Ahmed, N., Atzberger, C., & Zewdie, W. (2020). Integration of remote sensing and bioclimatic data for prediction of invasive species distribution in data-poor regions: a review on challenges and opportunities. Environmental Systems Research , 9(1), 1-18.

Alvarez-Berríos, N., Campos-Cerqueira, M., Hernández-Serna, A., Amanda Delgado, C. J., Román-Dañobeytia, F., & Aide, T. M. (2016). Impacts of small-scale gold mining on birds and anurans near the Tambopata Natural Reserve, Peru, assessed using passive acoustic monitoring.Tropical Conservation Science , 9(2), 832-851.

Amadon, D. (1973). Birds of the Congo and Amazon forests: a comparison. Tropical forest ecosystems in Africa and South America: a comparative review. Smithsonian Institution Press , Washington, DC, 267-277.

Austad, A. M. (2021). Birds on camera: assessing avifaunal detections by camera traps in the Amazon (Master’s thesis, Norwegian University of Life Sciences, Ås ).

Barnes, R. & Sahr, K. (2017). dggridR: Discrete Global Grids for R. R package version 2.0.4. https://github.com/r-barnes/dggridR/. doi:10.5281/zenodo.1322866

Bartoń, K. (2022). MuMIn: Multi-Model Inference. R package version 1.46.0. https://CRAN.R-project.org/package=MuMIn.

Blake, J. G. (2021). Acoustic monitors and direct observations provide similar but distinct perspectives on bird assemblages in a lowland forest of eastern Ecuador. PeerJ , 9, e10565.

Bonney, R., Shirk, J. L., Phillips, T. B., Wiggins, A., Ballard, H. L., Miller-Rushing, A. J., & Parrish, J. K. (2014). Next steps for citizen science. Science, 343(6178), 1436-1437.

Bradter, U., Mair, L., Jönsson, M., Knape, J., Singer, A., & Snäll, T. (2018). Can opportunistically collected Citizen Science data fill a data gap for habitat suitability models of less common species?. Methods in Ecology and Evolution, 9(7), 1667-1678.

Brambilla, M., & Ficetola, G. F. (2012). Species distribution models as a tool to estimate reproductive parameters: a case study with a passerine bird species. Journal of Animal Ecology, 81(4), 781-787.

Brooks, D. M. (1998). Competition and coexistence in Neotropical birds: a latitudinal comparison. Texas A&M University.

Brotons, L., Herrando, S., & Pla, M. (2007). Updating bird species distribution at large spatial scales: applications of habitat modeling to data from long‐term monitoring programs. Diversity and distributions, 13(3), 276-288.

Callaghan, C. T., Martin, J. M., Major, R. E., & Kingsford, R. T. (2018). Avian monitoring–comparing structured and unstructured citizen science. Wildlife Research, 45(2), 176-184.

Campos, I. B., Fewster, R., Truskinger, A., Towsey, M., Roe, P., Vasques Filho, D., … & Gaskett, A. (2021). Assessing the potential of acoustic indices for protected area monitoring in the Serra do Cipó National Park, Brazil. Ecological Indicators , 120 , 106953.

Castilheiro, W. F. F., Santos-Filho, M. D., & Oliveira, R. F. D. (2017). Beta diversity of birds (Passeriformes, Linnaeus, 1758) in Southern Amazon. Ciência animal brasileira, 18.

Cerqueira, M. C., Cohn‐Haft, M., Vargas, C. F., Nader, C. E., Andretti, C. B., Costa, T. V., … & Ferraz, G. (2013). Rare or elusive? A test of expert knowledge about rarity of Amazon forest birds. Diversity and Distributions, 19(7), 710-721.

Chaves, W. A., Sieving, K. E., & Fletcher Jr, R. J. (2017). Avian responses to reduced-impact logging in the southwestern Brazilian Amazon. Forest ecology and management, 384, 147-156.

Cintra, R., & Cancelli, J. (2008). Effects of forest heterogeneity on occurrence and abundance of the scale-backed antbird, Hylophylax poecilinotus (Aves: Thamnophilidae), in the Amazon forest. Revista Brasileira de Zoologia, 25, 630-639.

Condit, R., Pitman, N., Leigh Jr, E. G., Chave, J., Terborgh, J., Foster, R. B., … & Hubbell, S. P. (2002). Beta-diversity in tropical forest trees. Science, 295(5555), 666-669.

Cooper, C. B., Dickinson, J., Phillips, T., & Bonney, R. (2007). Citizen science as a tool for conservation in residential ecosystems. Ecology and society, 12(2).

Crall, A. W., Newman, G. J., Stohlgren, T. J., Holfelder, K. A., Graham, J., & Waller, D. M. (2011). Assessing citizen science data quality: an invasive species case study. Conservation Letters, 4(6), 433-442.

Dickinson, J. L., Zuckerberg, B., & Bonter, D. N. (2010). Citizen science as an ecological research tool: challenges and benefits. Annual review of ecology, evolution, and systematics, 149-172.

Diringer, S. E., Berky, A. J., Marani, M., Ortiz, E. J., Karatum, O., Plata, D. L., … & Hsu-Kim, H. (2019). Deforestation due to artisanal and small-scale gold mining exacerbates soil and mercury mobilization in Madre de Dios, Peru. Environmental science & technology, 54(1), 286-296.

Do Nascimento, L. A., Campos-Cerqueira, M., & Beard, K. H. (2020). Acoustic metrics predict habitat type and vegetation structure in the Amazon. Ecological Indicators , 117 , 106679.

Durães, R., Loiselle, B. A., & Blake, J. G. (2008). Spatial and temporal dynamics at manakin leks: reconciling lek traditionality with male turnover. Behavioral Ecology and Sociobiology, 62(12), 1947-1957.

eBird. 2022. eBird: An online database of bird distribution and abundance [web application]. eBird, Cornell Lab of Ornithology, Ithaca, New York. Available: http://www.ebird.org. (Accessed: Date [e.g., September 7, 2022]).

Farr, T. G., Rosen, P. A., Caro, E., Crippen, R., Duren, R., Hensley, S., … & Alsdorf, D. (2007). The shuttle radar topography mission.Reviews of geophysics , 45 (2).

Feldman, M. J., Imbeau, L., Marchand, P., Mazerolle, M. J., Darveau, M., & Fenton, N. J. (2021). Trends and gaps in the use of citizen science derived data as input for species distribution models: A quantitative review. PLoS One, 16(3), e0234587.

Fernandes, A. M., Cohn-Haft, M., Hrbek, T., & Farias, I. P. (2014). Rivers acting as barriers for bird dispersal in the Amazon. Revista Brasileira de Ornitologia, 22(4), 363-373.

Fernández‐Arellano, G. J., Teixido, A. L., Bernardon, B., Bueno, E. R., Ferreira, T. V., Goncalves, S. R., … & Pinho, J. B. (2021). Knowledge gaps and biases in the Pantanal indicate future directions for ornithological research in large wetlands. Ibis, 163(3), 784-797.

Finer, M., Yupanqui, O., Suarez, D., & Novoa, S. (2021). Using Satellites to Detect Illegal Logging in Peruvian Amazon. MAAP, 139, 7.

Fink, D., Hochachka, W. M., Zuckerberg, B., Winkler, D. W., Shaby, B., Munson, M. A., … & Kelling, S. (2010). Spatiotemporal exploratory models for broad‐scale survey data. Ecological Applications, 20(8), 2131-2147.

Fink, D., Damoulas, T., Bruns, N. E., La Sorte, F. A., Hochachka, W. M., Gomes, C. P., & Kelling, S. (2014). Crowdsourcing meets ecology: hemisphere-wide spatiotemporal species distribution models. AI magazine, 35(2), 19-30.

Fink, D., Auer, T., Johnston, A., Ruiz‐Gutierrez, V., Hochachka, W. M., & Kelling, S. (2020). Modeling avian full annual cycle distribution and population trends with citizen science data. Ecological Applications, 30(3), e02056.

Fiske, I., & Chandler, R. (2011). Unmarked: an R package for fitting hierarchical models of wildlife occurrence and abundance. Journal of statistical software, 43, 1-23.

Fithian, W., Elith, J., Hastie, T., & Keith, D. A. (2015). Bias correction in species distribution models: pooling survey and collection data for multiple species. Methods in Ecology and Evolution, 6(4), 424-438.

Fletcher Jr, R. J., Hefley, T. J., Robertson, E. P., Zuckerberg, B., McCleery, R. A., & Dorazio, R. M. (2019). A practical guide for combining data to model species distributions. Ecology, 100(6), e02710.

Fluck, I. E., Cáceres, N., Hendges, C. D., Brum, M. D. N., & Dambros, C. S. (2020). Climate and geographic distance are more influential than rivers on the beta diversity of passerine birds in Amazonia. Ecography, 43(6), 860-868.

Freeman, B. G., Scholer, M. N., Ruiz-Gutierrez, V., & Fitzpatrick, J. W. (2018). Climate change causes upslope shifts and mountaintop extirpations in a tropical bird community. Proceedings of the National Academy of Sciences, 115(47), 11982-11987.

Freeman, B. G., Strimas-Mackey, M., & Miller, E. T. (2022). Interspecific competition limits bird species’ ranges in tropical mountains. Science, 377(6604), 416-420.

Friedl, M. A., McIver, D. K., Hodges, J. C., Zhang, X. Y., Muchoney, D., Strahler, A. H., … & Schaaf, C. (2002). Global land cover mapping from MODIS: algorithms and early results. Remote sensing of Environment, 83(1-2), 287-302.

Gentry, A. H. (Ed.). (1993). Four neotropical rainforests. Yale University Press.

Gerson, J. R., Szponar, N., Zambrano, A. A., Bergquist, B., Broadbent, E., Driscoll, C. T., … & Bernhardt, E. S. (2022). Amazon forests capture high levels of atmospheric mercury pollution from artisanal gold mining. Nature communications, 13(1), 1-10.

Gini, C. (1936). On the measure of concentration with special reference to income and statistics. Colorado College Publication, General Series, 208(1), 73-79.

Girish, K. S., & Srinivasan, U. (2022). Community science data provide evidence for upward elevational range shifts by Eastern Himalayan birds. Biotropica.

Gouraguine, A., Moranta, J., Ruiz-Frau, A., Hinz, H., Reñones, O., Ferse, S. C., … & Smith, D. J. (2019). Citizen science in data and resource-limited areas: A tool to detect long-term ecosystem changes. PloS one, 14(1), e0210007.

Greeney, H. F. (2020). Thrush-like Antpitta (Myrmothera campanisona ), version 1.0. In Birds of the World (S. M. Billerman, B. K. Keeney, P. G. Rodewald, and T. S. Schulenberg, Editors). Cornell Lab of Ornithology, Ithaca, NY, USA. https://doi.org/10.2173/bow.thlant2.01

Guillera‐Arroita, G., Lahoz‐Monfort, J. J., Elith, J., Gordon, A., Kujala, H., Lentini, P. E., … & Wintle, B. A. (2015). Is my species distribution model fit for purpose? Matching data and models to applications. Global Ecology and Biogeography, 24(3), 276-292.

Guisan, A., & Thuiller, W. (2005). Predicting species distribution: offering more than simple habitat models. Ecology letters, 8(9), 993-1009.

Haffer, J. (1978) Distribution of Amazon forest birds. Bonn. Zool. Beitr, 29, 38–78.

Harvey, M. G., Aleixo, A., Ribas, C. C., & Brumfield, R. T. (2017). Habitat association predicts genetic diversity and population divergence in Amazonian birds. The American Naturalist, 190(5), 631-648.

Hayes, F. E., & Sewlal, J. A. N. (2004). The Amazon River as a dispersal barrier to passerine birds: effects of river width, habitat and taxonomy. Journal of Biogeography, 31(11), 1809-1818.

Hochachka, W. M., Fink, D., Hutchinson, R. A., Sheldon, D., Wong, W. K., & Kelling, S. (2012). Data-intensive science applied to broad-scale citizen science. Trends in ecology & evolution, 27(2), 130-137.

Horns, J. J., Adler, F. R., & Şekercioğlu, Ç. H. (2018). Using opportunistic citizen science data to estimate avian population trends. Biological conservation, 221, 151-159.

Householder, J. E., Janovec, J. P., Tobler, M. W., Page, S., & Lähteenoja, O. (2012). Peatlands of the Madre de Dios River of Peru: distribution, geomorphology, and habitat diversity. Wetlands, 32(2), 359-368.

Johnston, A., Auer, T., Fink, D., Strimas‐Mackey, M., Iliff, M., Rosenberg, K. V., … & Kelling, S. (2020). Comparing abundance distributions and range maps in spatial conservation planning for migratory species. Ecological Applications, 30(3), e02058.

Johnston, A., Moran, N., Musgrove, A., Fink, D., & Baillie, S. R. (2020). Estimating species distributions from spatially biased citizen science data. Ecological Modeling, 422, 108927.

Johnston, A., Hochachka, W. M., Strimas‐Mackey, M. E., Ruiz Gutierrez, V., Robinson, O. J., Miller, E. T., … & Fink, D. (2021). Analytical guidelines to increase the value of community science data: An example using eBird data to estimate species distributions. Diversity and Distributions, 27(7), 1265-1277.

Kahl, S., Stöter, F. R., Goëau, H., Glotin, H., Planque, R., Vellinga, W. P., & Joly, A. (2019, September). Overview of BirdCLEF 2019: large-scale bird recognition in soundscapes. In CLEF 2019-Conference and Labs of the Evaluation Forum (Vol. 2380, No. 256). CEUR.

Kahl, S., Wood, C. M., Eibl, M., & Klinck, H. (2021). BirdNET: A deep learning solution for avian diversity monitoring. Ecological Informatics, 61, 101236.

Karra, K., Kontgis, C., Statman-Weil, Z., Mazzariello, J. C., Mathis, M., & Brumby, S. P. (2021, July). Global land use/land cover with Sentinel 2 and deep learning. In 2021 IEEE international geoscience and remote sensing symposium IGARSS (pp. 4704-4707). IEEE.

Kelling, S., Johnston, A., Hochachka, W. M., Iliff, M., Fink, D., Gerbracht, J., … & Yu, J. (2015). Can observation skills of citizen scientists be estimated using species accumulation curves?. PloS one, 10(10), e0139600.

Krabbe, N. and T. S. Schulenberg (2020). Rufous-capped Antthrush (Formicarius colma), version 1.0. In Birds of the World (J. del Hoyo, A. Elliott, J. Sargatal, D. A. Christie, and E. de Juana, Editors). Cornell Lab of Ornithology, Ithaca, NY, USA. https://doi.org/10.2173/bow.rucant2.01

La Sorte, F. A., & Somveille, M. (2020). Survey completeness of a global citizen‐science database of bird occurrence. Ecography, 43(1), 34-43.

Landau, H. J. (1967). Sampling, data transmission, and the Nyquist rate. Proceedings of the IEEE, 55(10), 1701–1706.

Lang, N., Jetz, W., Schindler, K., & Wegner, J. D. (2022). A high-resolution canopy height model of the Earth.arXiv preprint arXiv:2204.08322 .

Laurance, S. G., Stouffer, P. C., & Laurance, W. F. (2004). Effects of road clearings on movement patterns of understory rainforest birds in central Amazonia. Conservation biology, 18(4), 1099-1109.

Leach, E. C., Burwell, C. J., Ashton, L. A., Jones, D. N., & Kitching, R. L. (2016). Comparison of point counts and automated acoustic monitoring: detecting birds in a rainforest biodiversity survey. Emu, 116(3), 305-309.

MacKenzie, D. I., Nichols, J. D., Lachman, G. B., Droege, S., Andrew Royle, J., & Langtimm, C. A. (2002). Estimating site occupancy rates when detection probabilities are less than one. Ecology, 83(8), 2248-2255.

MacKenzie, D. I., & Bailey, L. L. (2004). Assessing the fit of site-occupancy models. Journal of Agricultural, Biological, and Environmental Statistics , 9 (3), 300-318.

Marcondes, R. S., & Brumfield, R. T. (2019). Fifty shades of brown: Macroevolution of plumage brightness in the Furnariida, a large clade of drab Neotropical passerines. Evolution, 73(4), 704-719.

Marra, P. P., & Remsen Jr, J. V. (1997). Insights into the maintenance of high species diversity in the Neotropics: habitat selection and foraging behavior in understory birds of tropical and temperate forests.Ornithological Monographs , 445-483.

Matutini, F., Baudry, J., Pain, G., Sineau, M., & Pithon, J. (2021). How citizen science could improve species distribution models and their independent assessment. Ecology and evolution, 11(7), 3028-3039.

Maximiano, M. F. D. A., d’Horta, F. M., Tuomisto, H., Zuquim, G., Van Doninck, J., & Ribas, C. C. (2020). The relative role of rivers, environmental heterogeneity and species traits in driving compositional changes in southeastern Amazonian bird assemblages. Biotropica, 52(5), 946-962.

Melo, A. S., Rangel, T. F. L., & Diniz‐Filho, J. A. F. (2009). Environmental drivers of beta‐diversity patterns in New‐World birds and mammals. Ecography, 32(2), 226-236.

Mere Roncal, C., Middendorf, E., Forsyth, A., Cáceres, A., Blake, J. G., Almeyda Zambrano, A. M., & Broadbent, E. N. (2019). Assemblage structure and dynamics of terrestrial birds in the southwest Amazon: a camera‐trap case study. Journal of Field Ornithology, 90(3), 203-214.

Miller, D. A., Pacifici, K., Sanderlin, J. S., & Reich, B. J. (2019). The recent past and promising future for data integration methods to estimate species’ distributions. Methods in Ecology and Evolution, 10(1), 22-37.

Nobre, A. D., Cuartas, L. A., Hodnett, M., Rennó, C. D., Rodrigues, G., Silveira, A., & Saleska, S. (2011). Height Above the Nearest Drainage–a hydrologically relevant new terrain model. Journal of Hydrology , 404 (1-2), 13-29.

Oliveira, U., Vasconcelos, M. F., & Santos, A. J. (2017). Biogeography of Amazon birds: rivers limit species composition, but not areas of endemism. Scientific Reports, 7(1), 1-11.

Oliveros, C. H., Field, D. J., Ksepka, D. T., Barker, F. K., Aleixo, A., Andersen, M. J., … & Faircloth, B. C. (2019). Earth history and the passerine superradiation. Proceedings of the National Academy of Sciences, 116(16), 7916-7925.

Pacifici, K., Reich, B. J., Miller, D. A., Gardner, B., Stauffer, G., Singh, S., … & Collazo, J. A. (2017). Integrating multiple data sources in species distribution modeling: a framework for data fusion. Ecology, 98(3), 840-850.

Patten, M. A., J. van Dort, and P. F. D. Boesman (2021). Black-faced Antthrush (Formicarius analis), version 1.1. In Birds of the World (T. S. Schulenberg, Editor). Cornell Lab of Ornithology, Ithaca, NY, USA. https://doi.org/10.2173/bow.blfant1.01.1

Pearson, D. L. (1977). Ecological relationships of small antbirds in Amazonian bird communities. The Auk, 94(2), 283-292.

Peele, A. M., Marra, P. M., Sillett, T. S., & Sherry, T. W. (2015). Combining survey methods to estimate abundance and transience of migratory birds among tropical nonbreeding habitats. The Auk: Ornithological Advances, 132(4), 926-937.

Pérez-Granados, C., & Schuchmann, K. L. (2021). Passive acoustic monitoring of Chaco Chachalaca (Ortalis canicollis) over a year: vocal activity pattern and monitoring recommendations. Tropical Conservation Science, 14, 19400829211058295.

Pérez-Granados, C., & Schuchmann, K. L. (2022). Automated signal recognition as a useful tool for monitoring little-studied species: The case of the Band-tailed Nighthawk. Ecological Informatics, 101861.

Pitman, N. C. A. (2010). An overview of the Los Amigos watershed, Madre de Dios, southeastern Peru. Amazon Conservation Association, Washington, DC.ß

Pya N (2021). scam: Shape Constrained Additive Models . R package version 1.2-12. https://CRAN.R-project.org/package=scam.

QGIS.org, 2022. QGIS Geographic Information System. QGIS Association. http://www.qgis.org

R Core Team (2022). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. https://www.R-project.org/.

Ramesh, V., Gupte, P. R., Tingley, M. W., Robin, V. V., & DeFries, R. (2022). Using citizen science to parse climatic and land cover influences on bird occupancy in a tropical biodiversity hotspot. Ecography, 2022(9), e06075.

Robinson, S. K., & Terborgh, J. (1995). Interspecific aggression and habitat selection by Amazonian birds. Journal of Animal Ecology, 1-11.

Robinson, O. J., Ruiz‐Gutierrez, V., & Fink, D. (2018). Correcting for bias in distribution modeling for rare species using citizen science data. Diversity and Distributions, 24(4), 460-472.

Robinson, O. J., Ruiz‐Gutierrez, V., Reynolds, M. D., Golet, G. H., Strimas‐Mackey, M., & Fink, D. (2020). Integrating citizen science data with expert surveys increases accuracy and spatial extent of species distribution models. Diversity and Distributions, 26(8), 976-986.

Rosenberg, G. H. (1990). Habitat specialization and foraging behavior by birds of Amazonian river islands in northeastern Peru. The Condor, 92(2), 427-443.

Rumelt, R. B., Basto, A., & Mere Roncal, C. (2021). Automated audio recording as a means of surveying tinamous (Tinamidae) in the Peruvian Amazon. Ecology and Evolution , 11 (19), 13518-13531.

Sánchez, A. C., Bandopadhyay, S., Briceño, N. B. R., Banerjee, P., Guzmán, C. T., & Oliva, M. (2021). Peruvian Amazon disappearing: Transformation of protected areas during the last two decades (2001–2019) and potential future deforestation modelling using cloud computing and MaxEnt approach. Journal for Nature Conservation, 64, 126081.

Schulenberg, T. S., Stotz, D. F., Lane, D. F., O’Neill, J. P., & Parker, T. A. (2010). Birds of peru. In Birds of Peru . Princeton University Press.

Segal, R. D., Massaro, M., Carlile, N., & Whitsed, R. (2021). Small‐scale species distribution model identifies restricted breeding habitat for an endemic island bird. Animal Conservation, 24(6), 959-969.

Sheng, Y., & Alsdorf, D. E. (2005). Automated georeferencing and orthorectification of Amazon basin-wide SAR mosaics using SRTM DEM data. IEEE transactions on geoscience and remote sensing, 43(8), 1929-1940.

Shirley, S. M., Yang, Z., Hutchinson, R. A., Alexander, J. D., McGarigal, K., & Betts, M. G. (2013). Species distribution modeling for the people: unclassified landsat TM imagery predicts bird occurrence at fine resolutions. Diversity and Distributions, 19(7), 855-866.

Silva-Junior, C. H., Pessoa, A., Carvalho, N. S., Reis, J. B., Anderson, L. O., & Aragão, L. E. (2021). The Brazilian Amazon deforestation rate in 2020 is the greatest of the decade. Nature Ecology & Evolution, 5(2), 144-145.

Silveira, M. V., Silva‐Junior, C. H., Anderson, L. O., & Aragão, L. E. (2022). Amazon fires in the 21st century: The year of 2020 in evidence. Global Ecology and Biogeography, 31(10), 2026-2040.

Steen, V. A., Elphick, C. S., & Tingley, M. W. (2019). An evaluation of stringent filtering to improve species distribution models from citizen science data. Diversity and Distributions, 25(12), 1857-1869.

Stevens, H. C., Metz, E. M., Saboya Del Castillo, P., Alván, J. D., & Bowler, M. T. (2019). Use of autonomous audio recordings for the rapid inventory of birds in the white‐sand forests of the Peruvian Amazon. Journal of Field Ornithology, 90(1), 70-79.

Stouffer, P. C. (1997). Interspecific aggression in Formicarius antthrushes? The view from central Amazonian Brazil. The Auk, 114(4), 780-785.

Stowell, D., Wood, M., Stylianou, Y., & Glotin, H. (2016, September). Bird detection in audio: a survey and a challenge. In 2016 IEEE 26th International Workshop on Machine Learning for Signal Processing (MLSP) (pp. 1-6). IEEE.

Stowell, D. (2022). Computational bioacoustics with deep learning: a review and roadmap. PeerJ, 10, e13152.

Stratford, J. A., & Stouffer, P. C. (2013). Microhabitat associations of terrestrial insectivorous birds in Amazonian rainforest and second‐growth forests. Journal of Field Ornithology, 84(1), 1-12.

Strimas-Mackey, M., Miller, E., & Hochachka, W. (2018). auk: eBird Data Extraction and Processing with AWK. R package version 0.3. 0.

Strimas-Mackey, M., Hochachka, W. M., Ruiz-Gutierrez, V., Robinson, O. J., Miller, E. T., Auer, T., … & Johnston, A. (2020). Best practices for using eBird Data. Version 1.0. Ithaca, NY: Cornell Laboratory of Ornithology.

Stutchbury, B. J., & Morton, E. S. (2008). Recent advances in the behavioral ecology of tropical birds. The Wilson Journal of Ornithology, 120(1), 26-37.

Sugai, L. S. M., Silva, T. S. F., Ribeiro Jr, J. W., & Llusia, D. (2019). Terrestrial passive acoustic monitoring: review and perspectives. BioScience, 69(1), 15-25.studied species: The case of the Band-tailed Nighthawk. Ecological Informatics, 101861.

Sullivan, B. L., Wood, C. L., Iliff, M. J., Bonney, R. E., Fink, D., & Kelling, S. (2009). eBird: A citizen-based bird observation network in the biological sciences. Biological conservation, 142(10), 2282-2292.

Sullivan, B. L., Aycrigg, J. L., Barry, J. H., Bonney, R. E., Bruns, N., Cooper, C. B., … & Kelling, S. (2014). The eBird enterprise: An integrated approach to development and application of citizen science. Biological conservation, 169, 31-40.

Terborgh, J., Robinson, S. K., Parker III, T. A., Munn, C. A., & Pierpont, N. (1990). Structure and organization of an Amazonian forest bird community. Ecological Monographs, 60(2), 213-238.

Thornton, D. H., Branch, L. C., & Sunquist, M. E. (2012). Response of large galliforms and tinamous (Cracidae, Phasianidae, Tinamidae) to habitat loss and fragmentation in northern Guatemala. Oryx, 46(4), 567-576.

Tobler, M. W., Carrillo‐Percastegui, S. E., Leite Pitman, R., Mares, R., & Powell, G. (2008). An evaluation of camera traps for inventorying large‐and medium‐sized terrestrial rainforest mammals. Animal conservation, 11(3), 169-178.

Van Eupen, C., Maes, D., Herremans, M., Swinnen, K. R., Somers, B., & Luca, S. (2021). The impact of data quality filtering of opportunistic citizen science data on species distribution model performance. Ecological Modelling, 444, 109453.

Velásquez-Tibatá, J., Salaman, P., & Graham, C. H. (2013). Effects of climate change on species distribution, community structure, and conservation of birds in protected areas in Colombia. Regional Environmental Change, 13(2), 235-248.

Wilkinson, D. (2020). Fitting occupancy models with unmarked. https://doi90.github.io/lodestar/fitting-occupancy-models-with-unmarked.html

Winkler, D. W., S. M. Billerman, and I. J. Lovette (2020). Antthrushes (Formicariidae), version 1.0. In Birds of the World (S. M. Billerman, B. K. Keeney, P. G. Rodewald, and T. S. Schulenberg, Editors). Cornell Lab of Ornithology, Ithaca, NY, USA. https://doi.org/10.2173/bow.formic2.01

Wright, M.N., & Ziegler, A. (2017). ranger: A Fast Implementation of Random Forests for High Dimensional Data in C++ and R. Journal of Statistical Software, 77(1), 1-17. doi:10.18637/jss.v077.i01

Wood, C., Sullivan, B., Iliff, M., Fink, D., & Kelling, S. (2011). eBird: engaging birders in science and conservation. PLoS biology, 9(12), e1001220.

Wood, C.M. and Peery, M.Z. (2022), What does ‘occupancy’ mean in passive acoustic surveys?. Ibis, 164: 1295-1300. https://doi.org/10.1111/ibi.13092

Zhang, G. (2020). Spatial and temporal patterns in volunteer data contribution activities: A case study of eBird. ISPRS International Journal of Geo-Information, 9(10), 597.

Zhong, M., Taylor, R., Bates, N., Christey, D., Basnet, H., Flippin, J., … & Ferres, J. L. (2021). Acoustic detection of regionally rare bird species through deep convolutional neural networks. Ecological Informatics, 64, 101333.

Zimmer, K. and M.L. Isler (2020). Goeldi’s Antbird (Akletos goeldii), version 1.0. In Birds of the World (J. del Hoyo, A. Elliott, J. Sargatal, D. A. Christie, and E. de Juana, Editors). Cornell Lab of Ornithology, Ithaca, NY, USA. https://doi.org/10.2173/bow.goeant1.01

Zimmer, K. and M.L. Isler (2020). White-throated Antbird (Oneillornis salvini), version 1.0. In Birds of the World (J. del Hoyo, A. Elliott, J. Sargatal, D. A. Christie, and E. de Juana, Editors). Cornell Lab of Ornithology, Ithaca, NY, USA. https://doi.org/10.2173/bow.whtant1.01

Figure 1: (a) survey points at Los Amigos Biological Station and (b) eBird checklists within Madre de Dios. Most eBird checklist locations occur near riverine corridors, whereas structured survey points span the HAND gradient. Black and yellow box in (b) denotes extent of (a).