2.1 Study sites

We studied cavity-nesting bird communities in Northwest Ecuador, around the Mache-Chindul Ecological Reserve (0.3900, −79.6840; elevation 390–510 m near our study site) in the Esmeraldas Province (Figure 1). The area has been delineated as a BirdLife International “Important Bird Area” (“BirdLife Data Zone”, 2019) within the Chocó biogeographic zone, characterized by high biodiversity and severe rates of deforestation (Orme et al., 2005; Sierra, 1999). The Mache-Chindul Ecological Reserve was created in 1996 and consists of pristine forest, secondary forest fragments, and agricultural lands. The area has experienced rapid agricultural expansion in the last 50 years. About 6500 people live in the Reserve, typically on farms of 20–50 ha with agriculture including cacao, pasture for cattle, corn, beans, rice, plantain, and oil palm (Carrasco et al., 2013). Dominant forest types are humid evergreen and sub-humid evergreen forests, with canopy heights in primary forest spanning 30–40 m. Average monthly temperatures range from 26 to 28°C, and total annual precipitation ranges from 2 to 3.5 m, with the majority of the rainfall occurring between January and May. To study cavity limitation, we established fourteen 0.5-ha rectangular plots (100 m × 50 m), half within forest and half in agriculture, the latter consisting mostly of pasture but also with occasional cacao or plantain plants. Minimum distance between plots was 137 m, maximum distance was 1532 m, and mean distance was 650 m.

2.2 Bird surveys

To survey bird communities, FC and LC conducted 10-min point counts in four periods (February 2020, July 2020, May 2021, and July 2021) at the center of five agriculture plots and five forest plots (Figure 1). There are two peaks of avian breeding activity in the region, from February to May and October to November (Carrasco et al., 2013). During surveys, we recorded all birds seen or heard within 50 m from the stationary observer. We used circular plots for point count surveys so that the probability of detecting individuals was consistent in all directions. Within each of the four survey periods, each site was sampled three times within a week. Point counts took place between 6:00 and 11:30 a.m. (97% took place between 6:00 and 10:00 a.m.). We recorded the survey time, wind conditions (qualitative 0–3 scale, where 0 = no movement of branches, 1 = light breeze that moves leaves and small branches, 2 = moderate wind that moves shrubs and medium branches, and 3 = strong wind that moves large branches in the canopy), and fog conditions (qualitative 0–3 scale where 0 = no fog, 1 = fog present, visibility up to 100 m, 2 = visibility up to 50 m due to fog, and 3 = visibility up to 30 m due to fog) to help account for variation in detection between species, weather conditions, and habitats. No surveys were conducted in the rain. Finally, we used an existing database to identify cavity-nesting species (del Hoyo, 2015).

2.3 Modeling bird communities

To estimate the abundance of each bird species at each site while accounting for imperfect detection, we implemented a N-mixture model, where the number of individuals is counted during spatially and temporally replicated surveys (Kéry, 2018; Royle, 2004). We implemented the N-mixture model within a community modeling framework such that species-specific parameters were estimated using community-wide hyperparameters, allowing us to share information among species in the community and estimate parameters of rarer species (Dorazio & Royle, 2005; Kéry & Royle, 2015).

Here, “time” represents time of day, “wind” is the level of wind from 0 to 3, “fog” is the level of fog from 0 to 3, and “date” is the Julian day of the year. All variables, including the 0–3 scales, were scaled and centered prior to analysis. $\gamma 0$ is a land-use specific intercept (for forest versus agriculture), $\gamma 1$ is a species intercept, and $\gamma 3$ and $\gamma 4$ are fixed effects for wind and fog, respectively. Wind and fog were treated as fixed effects because we believed that each would have a consistent negative effect on the detectability of all species. $\mathrm{\gamma }5$ and $\mathrm{\gamma }6$ are species-specific slopes for date and date squared, as each species' activity (and thus detectability) could peak at different times of the year.

We implemented the model in R Version 4.0.0 using the package R2jags, which runs Markov chain Monte Carlo (MCMC) algorithms (R Core Team, 2013; Su & Yajima, 2012). We ran three chains starting at random initial values and 50,000 burn-in iterations. We included 50,000 post burn-in iterations, thinned at a rate of 50. We considered the chains to converge if the Gelman-Rubin statistics of the chains of every parameter were ≤1.1 (Gelman et al., 2004).

We expected the bird community on average to be positively associated with forest habitat (i.e., α1 > 0). We determined if cavity-nesting birds had a stronger positive association with forest than non-cavity-nesting birds by assessing whether α2 was positive and if its 95% Bayesian Credible Interval (BCI) did not include 0. We also determined if each species had a significant relationship with land-use by assessing whether the 90% BCI for the species slope (α3) crossed zero. We used a 90% BCI for species-level analyses because each parameter was estimated with fewer data than the community means. 90% BCIs are recommended when effective sample sizes are low (Kruschke, 2014; McElreath, 2020).

2.4 Natural cavity availability and use

In 2019, we scanned each tree within four of the agricultural and four of the forest plots with binoculars to identify natural cavities (entrance hole diameter >2 cm, >2.5 m high above ground level) (Figure 1; Cockle et al., 2010). We noted the height and orientation (in degrees) of each cavity, as well as the tree's height (using a hypsometer) and diameter at breast height (DBH). We conducted Welch's two-sample t-tests to determine if cavities in forest versus agriculture differed in height above ground, tree height above ground, and DBH. We conducted Rayleigh tests (Landler et al., 2018) to determine whether cavity entrances tended to face the same direction. Finally, to test whether natural cavity counts differed between land-use types, we conducted a Welch's two-sample t-test between the number of natural cavities found in agriculture versus forest sites.

We also quantified natural cavity use in each plot. Specifically, we conducted weekly observations of each natural cavity for 15 min from September 2019 to April 2020, noting any the activity of any cavity-nesting bird that we observed. We considered an activity event to be when a bird was observed inside, perching at the entrance, or excavating a cavity. This resulted in 19 observation periods per cavity and 980 observation periods total (multiple cavities were often visible simultaneously so they could be observed in the same 15-min period). To test whether natural cavity activity differed between land-use types, we calculated the total number of bird activity events observed per natural cavity (across all visits to a given plot) and then conducted a Welch's two-sample t-test to compare the number of activity events in forest versus agriculture.

2.5 Nest box addition experiment

To quantify use and success of artificial nest boxes, we constructed and placed 10 nest boxes in each of the five agricultural and five forest plots (100 nest boxes total; Figure 1). Specifically, we placed nest boxes on trees in six of the plots in September 2019 and the remaining four plots in February 2020. Boxes within a plot were located 10–20 m apart and 5 m off the ground, facing SW to minimize sun exposure. Half of the nest boxes in each plot were small (12-cm width × 12-cm depth × 24-cm height, with a 4-cm diameter entrance hole), and the other half were large (18-cm width × 18-cm depth × 38-cm height, with a 10-cm diameter entrance hole). Because nest box studies are rare in tropical ecosystems, small nest boxes were based on standard dimensions for attracting Western Bluebirds (Sialia mexicana) (Jedlicka et al., 2011). Large boxes were ~1.5× the dimensions of small nest box to attract larger species. We put 5 cm of sawdust in each nest box to imitate the conditions of natural cavities. We replaced 31 nest boxes throughout the study period when the boxes decayed or fell.

Nest boxes were monitored on a weekly basis from September 2019 to June 2022 (i.e., 2.5 years), resulting in 10,837 nest box visits. We used a telescoping pole with an endoscope to observe active nests and quantify the proportions of eggs that hatch, chicks that fledge, and nestlings/eggs predated. We noted any new activity from birds or mammals. We defined new activity as evidence of new materials in a nest box after being inactive for several weeks or different materials from what was previously present. Bird nesting materials were often sticks, bark, and leaves. Mammal nesting materials were often leaves (opossum) and dried grasses (squirrel). We considered an activity event to be an avian nesting attempt if any eggs were laid.

To quantify how land-use type and nest-box size affect nest-box activity, we implemented a Poisson mixed-effects model with a log link. The response variable was the total number of bird activity events at each box across the 2.5-year period (including nesting attempts). Explanatory variables included land-use type (binary) and nest-box size (large or small; binary), as well as the interaction between land-use and nest-box size. We included a random effect of “site” to account for spatial autocorrelation (i.e., multiple nest boxes at the same site) and an offset of “months” to account for the fact that some sites had nest boxes active for slightly longer than others (i.e., 28 vs. 33 months). We then repeated this analysis using mammal activity events. We excluded the interaction between land-use type and nest-box size in the mammal model so that the model could be identifiable. Finally, to understand how land-use type affected nest success, we implemented a binomial mixed-effects model with the same explanatory variables and a binary response variable, indicating whether at least one chick fledged during each avian nesting attempt.

3.1 Bird abundances across land-use types

Across 120 point counts, we detected 1093 individuals and 111 species, 26 (23.4%) of which are known to nest in cavities (Table S1). Out of 4440 abundance estimates (N), 38 did not converge (3%; all had R-hat ≤1.25, 88% of which had R-hat <1.2). $\gamma 0_{\mathit{LU}\left[2\right]}$ had an R-hat of 1.12, and all other parameters converged, including the core parameters among our log linear predictors. The 95% BCI for the community mean effect of land use on abundance was negative but crossed zero (Figure 2), meaning that, on average, species abundances did not differ between agriculture and forests. Eight cavity-nesting birds were more abundant in agriculture (House Wren; Troglodytes aedon, Dusky-capped Flycatcher; Myiarchus tuberculifer, Pacific Parrotlet; Forpus coelestis; Social Flycatcher; Myiozetetes similis; Red-bellied Macaw; Orthopsittaca manilatus, Red-billed Scythebill; Campylorhamphus trochilirostris; Masked Tityra, Tityra semifasciata; Bronze-winged Parrot, Pionus chalcopterus). One cavity-nesting bird was more abundant in forest (Collared Aracari; Pteroglossus torquatus). There was no significant interaction between being a cavity-nesting species and the effect of land-use on abundance (effect size −0.10, 95% BCI [−1.39, 1.21]). This suggests that cavity-nesting species did not differ from non-cavity nesting species in their responses to land use.

3.2 Natural cavity availability and use

As expected, there were significantly more natural cavities in forest than agriculture (mean 34 cavities/ha vs. 7.25/ha; t = −5.85, p-value = .0014; Figure 3). The mean cavity height above ground level in forest and agriculture were similar (13.8 [range 3–43 m] and 13.9 m [range 4–34 m], respectively; t = −0.04, p-value = .97) and natural cavities did not tend to be oriented in any cardinal direction (Figure S1; Rayleigh test statistic in forest = 0.11, p-value = .21; Rayleigh test statistic in agriculture = 0.06, p-value = .89). Mean DBH of trees with cavities was larger in agriculture than in forest (agriculture: 42.9 cm, range 7–200 cm; forest: 33.7 cm, range 10–63 cm; t = 1.96, p-value = .05). However, mean height above ground level of trees with cavities in forest and agriculture did not differ (agriculture: 22.0 m, range 3–46 m; forest: 21.6 m, range 6–35 m; t = 0.26, p-value = .80).

Across the nearly 1000 observation periods, we observed surprisingly low bird activity around natural cavities, with only seven instances of birds inside cavities in agriculture and two in forest (Table 1). Five of the seven observations of a bird inside a cavity in agriculture were of the Collared Aracari using the same cavity. In contrast, there was one observation of a Collared Aracari inside a cavity in forest. Other species observed inside natural cavities included Golden-olive Woodpecker (Colaptes rubiginosus; in forest and agriculture), Blue-and-white Swallow (Notiochelidon cyanoleuca; in agriculture), Black-cheeked Woodpecker (Melanerpes pucherani; in forest and agriculture), and Olivaceous Woodcreeper (Sittasomus griseicapillus; in forest). This low rate of observation precluded a statistical comparison of cavity activity events between forest versus agriculture.

3.3 Nest box addition experiment

Avian activity and nesting attempts (i.e., where an egg was laid) were much more common in experimentally placed nest boxes within agriculture compared to forest (Tables 2 and 3). There were 109 total avian activity events in agriculture and 5 in forest. Of those activity events, there were 52 nesting attempts in agriculture and 5 in forest. Correspondingly, our Poisson model indicated agriculture had significantly more avian activity events in boxes compared to boxes in forest (Figure 4, Table S2). Smaller nest boxes also had significantly more avian activity events than large boxes (Table 3). Finally, a significant negative interaction was detected between habitat and nest box size, such that the difference in avian activity between forest and agriculture was greater for small nest boxes (Figure 4, Table S2).

Seven bird species laid eggs in nest boxes (six in agriculture and two in forest; Tables 2 and 3). Five nesting attempts could not be attributed to a particular species because eggs disappeared before we could ascertain the species (all in agriculture). No species that used nest boxes were of conservation concern based on the IUCN red list classification. On average, avian nest success (i.e., successfully fledging at least one young) was 44% in agriculture (out of 52 nesting attempts) and 20% in forest (out of 5 nesting attempts). Out of the 57 total nesting attempts, we presumed that 13 nests had eggs predated and 12 nests had chicks predated because the eggs or chicks disappeared before an appropriate fledging age when chicks develop feathers. In our binomial model, neither box size nor habitat had significant effects on nest success, likely due to the low number of nesting attempts in forest (Table S2).

Unlike birds, mammal activity in experimentally placed nest boxes was more common in forest than in agriculture. In agriculture, there was one instance of mammal activity within a nest box (the brown four-eyed possum; Metachirus nudicaudatus) vs. 28 instances in forest (i.e., 22 brown four-eyed possum and 6 red-tailed squirrel (Sciurus granatensis) activity events across all sites; Table 2). Poisson models indicated that mammal activity was significantly higher in forest than agriculture, but no effects of nest box size were observed (Table S2). Finally, 10 nest boxes were occupied by wasps and bees, half in agriculture and half in forest.

4.1 Trends in bird abundance

On average, birds were not more abundant in forests than agriculture. Cavity-nesting species also did not differ in abundance between habitats, even though natural nest site availability was low in agriculture. One possible explanation is that agriculture may act as a sink habitat, drawing animals out from better habitat but not enabling sufficient reproduction for populations to persist in the absence of continued immigration (Gilroy & Edwards, 2017). If agriculture is a sink habitat, then biologists may be overestimating the ability of cavity-nesting birds to persist in agricultural landscapes and there could be unexpected crashes in populations of cavity-nesting birds if forest loss continues (Delibes et al., 2001). However, we found higher nesting success in agriculture than in forest (though there were few observations in forest), which does not support this hypothesis. Alternatively, the high bird abundances in agriculture may be explained by birds simply moving through agricultural systems and/or using multiple habitats to complete their life cycles; for example, foraging in agriculture and reproducing in forest (Frishkoff et al., 2019). This could mean that agriculture provides certain resources such as food and that increasing nesting habitat would further increase the use of agriculture. Looking forward, behavioral analyses could be used to identify the habitats that species preferentially move through, forage in, and reproduce in Ke et al. (2022), providing a clearer picture of the degree to which cavity-nesting birds rely on forests.

Of the species we observed, 23.4% of them are known to nest in cavities, which is high compared to the percentage of cavity-nesting birds globally (ranging from 10.9% to 19.5% depending on the global realm) (van der Hoek et al., 2017). Eight cavity-nesting species were more abundant in agriculture. All eight species have “Least Concern” conservation status from IUCN and mostly have wide global distributions. Thus, the species that are more abundant in agriculture could generally be more ecologically flexible and could be considered “winners” when forest is converted to agriculture (Järvinen & Ulfstrand, 1980). House Wrens had the most nesting attempts within nest boxes in agriculture (60% of the attempts in agriculture, no nesting attempts in forest) and are also known to readily use nest boxes in temperate agriculture (Baldwin & Bowen, 1928; Willner et al., 1983). Dusky-capped Flycatchers and Pacific Parrotlets each nested once in agriculture, and both have been previously documented using nest boxes (Bock et al., 1992; Collar et al., 2020). On the contrary, Collared Aracaris were significantly more abundant in forest, though they were observed using natural cavities in both forest and agriculture.

4.2 Natural cavities: Density and frequency of use

We found over four times the density of natural cavities in forest than in agriculture. If anything, this is likely a significant underestimate of the difference because it is difficult to find cavities that are high in the canopy and/or in dense foliage (Koch, 2008). Cockle et al. (2010) recorded over nine times the density of suitable cavities (at least 13 cm deep and 2.5 m high above ground level) in primary forest compared to logged forest in the Atlantic forest of Argentina. They also found that cavities were much more likely to be in large trees (>100 cm DBH). While the height of trees containing cavities was higher in agriculture than in forest, we did not find significant differences in the cavity height, DBH, or orientation of cavities in forest versus agriculture. Nonetheless, our estimates could be biased if we tended to overlook cavities high up in forest canopies. This is in contrast to studies such as White Jr et al. (2006) for example, which found that birds prefer cavities with entrances oriented westward.

Within the natural cavities that we did observe, bird activity tended to be quite low. Many of these cavities may not have been suitable for birds. Cavities that are too low in the canopy might be subject to higher predation risk and thus avoided (Nilsson, 1984). Additionally, cavities must also be large enough to satisfy species' needs. For example, Cockle et al. (2010) found that, among 86 cavities found through ground surveys, only 19% of them exceeded minimal size requirements for birds (at least 13 cm deep and 2.5 m high above ground level), and birds only occupied 25% of those suitable cavities. Similarly, Lima and Garcia (2016) found that birds occupied 26% of natural cavities in Mexico. Our natural cavity occupancy rate was even lower (6.6% with observations of a bird at the entrance or inside) and thus more similar to values reported from the Peruvian Amazon forest (2%; Brightsmith, 2005). This low activity rate made it challenging to achieve an adequate sample size to compare cavity occupancy between land-use types. Across the board, it seems that occupancy rates in tropical forests may be much lower than in northern temperate forests (e.g., 67% in Ingold and Ingold (1984); 57% in Peterson and Gauthier (1985)).

4.3 Nest boxes occupancy and reproductive success

Nest boxes were used much more often by birds in agriculture and by mammals in forest. Mammals were likely not outcompeting birds for forest nest boxes because nest box occupancy in forest was relatively low (32 out of 100 nest boxes had activity). Prior studies have also observed low avian activity in tropical forest nest boxes. For example, in Mexico, no nest boxes were occupied in mature forest, whereas 14% of 80 nest boxes were occupied in young forests (Lima & Garcia, 2016). Similarly, only 2% of nest boxes were occupied in an undisturbed Peruvian forest (Brightsmith, 2005), and only four of 16 nest boxes were used by Yellow-shouldered Parrots (Amazona barbadensis) over 3 years in Venezuela (Sanz et al., 2003). These numbers stand in stark contrast to studies conducted in temperate forests. In temperate regions, many nest box addition experiments have been shown to increase cavity-nesting bird and mammal populations (Aitken & Martin, 2012; Lindenmayer et al., 2009; Norris et al., 2018). This may be because tropical forests have much higher cavity density than temperate forests (e.g., mean 34 cavities/ha in our study vs. mean 2.4 cavities/ha in Swedish boreal forest (Andersson et al., 2018); mean 12.5 cavities/ha in Polish coniferous forest (Walankiewicz et al., 2014). Low nest box occupancy rates may occur if natural cavities are both ubiquitous and advantageous. For example, unlike artificial boxes, natural cavities can be found higher in the canopy and in live wood (Cockle et al., 2015; Nilsson, 1984; Wesołowski, 2002). Together, these studies and our work suggest that cavity-nesting birds may not be limited by nesting habitat in tropical forests, with other factors like competition, food resources, or predation constraining population sizes (Dhondt, 2012; Nilsson, 1984; Wiebe, 2011).

On the contrary, avian nest box activity was high in agriculture, and six species were observed using experimentally placed nest boxes in agriculture compared to two in forests (though sample size was low in forest). As a comparison, only two species were found occupying nest boxes in primary and logged forests in Argentina (Cockle et al., 2010). Similarly, only two species were found in nest boxes distributed across old growth forest, secondary forest, and coffee plantations in Costa Rica (Saker, 2015). Perhaps because tropical pastures have so few trees, populations of cavity-nesting birds may be especially limited by nesting habitat as compared to populations in selectively logged forests or coffee plantations. It is possible that bird activity was higher in agriculture than forest due to less mammal activity and thus less predation pressure from mammals. Smaller nest boxes also had significantly more avian activity events than large boxes, potentially because smaller bird species are more willing to use artificial nest boxes that are also at a lower height above ground. Importantly, there are not more small species compared to large cavity-nesting species in the community—out of 86 species that could nest in cavities, 42 were 20 cm or less in body length. The larger species may simply be more unlikely to use artificial nest boxes as a substrate and/or might prefer nesting higher in the canopy (e.g., trogons and parrots).

Importantly, we found that nest success was relatively high in artificial nest boxes (~50% fledging at least one young) compared to rates reported from other tropical studies. For example, nesting success of birds nesting in natural cavities has been estimated to be ~25% in lowland Panamanian forests (Brawn et al., 2011; Robinson et al., 2000). 55% of the 62 Puerto Rican parrot fledglings monitored survived to fledging, but this rate excludes nest failure in the egg stage (White Jr et al., 2014). Moreover, many studies of non-cavity nesting birds report lower nest success in fragmented versus natural land uses (Borges & Marini, 2010; Newmark & Stanley, 2011; Rangel-Salazar et al., 2008; Young et al., 2008). Other studies report increased nest success in areas with less canopy connectivity (Britt et al., 2014) and in older cavities (Brightsmith, 2005). If anything, we found that birds nesting in agricultural nest boxes were more successful than in forests; however, the difference was not significant, likely due to the very few nesting attempts we observed in forest. That said, it is possible that we could have slightly overestimated nest success if we missed nest failures near fledging. Specifically, we assumed a bird had fledged if we observed nestlings near fledging age on one visit and then found an empty nest on the next visit. Nevertheless, the nest success that we observed was sufficiently high that it seems that artificial nest boxes in agriculture provide suitable conditions to successfully rear tropical birds.