Formicivora acutirostris, a threatened species in Brazil (Ordinance MMA 148, of 7 June 2022), is a territorial and socially monogamous bird that forms long-term pairs, sharing reproductive and defensive activities throughout the year (Sobotka, 2011; Reinert et al., 2012; Bornschein et al., 2015). The breeding season extends from August to February (Reinert et al., 2012); the species builds cup-shaped nests attached to vegetation, at a height between 30 and 220 cm above the ground (Reinert et al., 2012), and may renest up to eight times during a breeding season (Reinert, 2008). Formicivora acutirostris lays two eggs (Reinert et al., 2012), incubates them for approximately 16 days (Reinert, 2008), and the nestling stage extends for approximately 10 days (Reinert, 2008).

The study was conducted in the Guaratuba Bay, Guaratuba Ramsar Site (see https://rsis.ramsar.org/ris/2317), municipality of Guaratuba, on the southern coast of Paraná, southern Brazil ( Figure 3 ). We worked at the confluence of the Claro and São João rivers (“Continente”; c. 25.873°S, 48.761°W; 8.7 ha) and on Jundiaquara Island (“Jundiaquara”; c. 25.873°S, 48.759°W; 11.5 ha), located upstream on the São João river, and in part of the Folharada Island (“Folharada”; c. 25.866°S, 48.723°W; 16.3 ha), located downstream on this river. The studied areas contain “estuarine marshes” (sensu Doody, 2001), “tidal marshes” (Reinert et al., 2007), “subtropical salt marshes” (Bornschein et al., 2017), or pioneer formation of fluviomarine influence (sensu Veloso et al., 1991; IBGE, 1992). The environment experiences two daily floodings according to a mixed semi-diurnal tide (Lee and Chang, 2019).

The dominant plant species are crinum lily (Crinum americanum) and California bulrush (Schoenoplectus californicus). Locally, the following species can be abundant: the herbaceous southern cattail (Typha domingensis), swamp sawgrass (Cladium mariscus), giant leather fern (Acrostichum danaeifolium), the bush Talipariti pernambucensis, pond apple (Annona glabra), Brazil beauty-leaf (Calophyllum brasiliense), and white mangrove (Laguncularia racemosa) (Reinert et al., 2007; Bornschein et al., 2017; Favretto et al., 2022).

Fieldwork was conducted between January 2006 and March 2023. Until May 2008, it was conducted daily during the breeding season and for 10 days per month outside of it. From June 2008 to March 2023, fieldwork was conducted for 3–8 days per month throughout the year. We worked in the field as a group of two to five people, usually three, from dawn to approximately 1 p.m. and for a further two to three-and-a-half hours in the afternoon, before dusk.

All territorial birds were banded (with a metal numbered band and three colored plastic bands) in unique color codes for individual recognition with binoculars in order to determine the owners of the nests and the nest sequences. During banding, each individual was assigned with an identity code (referred to as “ID number”) and its minimum age was estimated based on the plumage [for details on juvenile and adult plumage patterns, see Reinert (2008)]. To reduce data variability, we estimated the birth date as either October 1 (a period with many early breeding season births) or February 1 (a period with the latest births in the season), in accordance with Bornschein et al. (2015). We assigned the closest birth date that best matched the observed plumage pattern at the time of the banding (Bornschein et al., 2015). Every year, we calculated the resulting age of the individuals by summing a year.

Target nests were systematically searched in 14–41 territories per year during the breeding seasons (August to February), making a total of 521 monitored territories. The nests were located mainly by observing the behavior of individual birds (e.g., carrying plant material or food and alarm vocalizations) that we followed up to discover nests. When finding nests at different stages, we calculated their initiation, as the date of nest construction completion and the start of egg laying, based on mean reproductive times studied by Reinert (2008). The nests were numbered and marked with a ribbon tied on the vegetation to facilitate the correct identification in the returns. We took the geographic coordinates of each nest and information on its height (as lip to the ground), thickness, and environment type. The thickness, which represents the robustness of the nest, may be a factor related to losses due to rips and holes (Palomino et al., 1998; Suárez et al., 2005) and was estimated as the average of the differences between (1) internal and external diameters and (2) nest height and depth of the incubation chamber.

We classified the environment type in which the nest was constructed, based on the dominant plant species. We named them by adding the suffix “etum” to the genus of the dominant species, following Braun-Blanquet (1979), as follows: (1) “Crinetum” (for the domain of the herbaceous C. americanum), (2) “Schoenoplectetum” (for the domain of the herbaceous S. californicus), (3) “Cladietum” (for the domain of the herbaceous C. mariscus), (4) “Acrostichetum” (for the domain of the herbaceous A. danaeifolium), (5) “Typhetum” (for the domain of the herbaceous T. domingensis), (6) “Taliparitetum” (for the domain of the bush T. pernambucensis), and (7) “Laguncularietum” (for the domain of the tree L. racemosa).

We calculated nest altitude based on an altimetric mapping, made with the Real Time Kinematic (RTK) technique in 2014 (when more than 4,000 altitude points were taken with 1 mm precision). The altitude points were distributed in grids of 20 m × 20 m or 5 m × 5 m, the latter for regions with terrain slope. A map was generated with contour lines for every 10 cm of altitude. The altitude of each nest was taken to be that of the altitude of the point that were closest to the nest, but within the same contour line as the nest.

Monitored nests were considered as having success or failure (= nest fate), and failure was attributed to predation, flooding, or tipping. We consider the nest to be successful if at least one fledgling leaves it. We attributed failure to predation when observed or inferred attacks on nests that led to loss of eggs, hatchlings, or the entire nest. We inferred predation if we observed evidence in the nests such as bottom or side holes, eggshell bottoms, or completely ripped nest material, with no signs of friction in the plant fibers. We attributed nest failure to flooding if we observed flooded nests, floating eggs around the nests, drowned chicks inside the nests, wet nests (on non-rainy days), and the presence of debris inside the nests. Finally, we attributed nest failure to tipping if we observed or inferred fallen nests due to the observation of friction in the plant support. We observed many empty nests that should have been active, but it was not possible to attribute the cause of the loss, such as nests that may have been lost due to strong winds or predator approach. In these cases, we attributed the cause of failure as undetermined and removed these data from the analyses.

After determining success or failure, we only counted subsequent nest attempts by the same pair (renesting case) that were carried out to completion (i.e., a complete cup), since there are changes in the height and thickness of nests during construction. We concluded that we were dealing with a case of renesting only when we were certain that no third nest had been built between the previous and subsequent nests.

We compared the characteristics of the previous and the subsequent nests (renesting) to evaluate the changes adopted by the individuals. We calculated the differences in nest attributes by subtracting the height (change in nest height), altitude (change in nest altitude), and thickness (change in nest thickness) of the nests (change in nest attribute = second nest attribute − previous nest attribute). Additionally, considering the position and environment of the nests, we calculated the distance between subsequent nests and assessed whether there was change in environment type.

We conducted three different analyses to assess the adoption of the WSLS strategy by F. acutirostris, as well as its adaptive significance. Initially, we tested the relationship between the nest attributes and their fates (analysis 1). Next, we evaluated the relationship between changes in nest characteristics in renesting cases and the fate of the previous nest (analysis 2) to determine whether the species makes systematic changes ( Figure 2 ). Finally, we tested the relationship between the changes in nest characteristics and the success or failure of the second nest (analysis 3) to assess the adaptive significance of these changes. In addition to the previously mentioned nest attributes (analysis 1) and changes in these attributes (analyses 2 and 3) to test our main hypotheses, we also included potential factors associated with nest fate or nesting behavior, such as seasonality (indicated by month of nesting initiation), environment type (simplified according to the sample size and objective), and parental age ( Table 1 ) in the models. We did not include these potential factors in analysis 3 due to the low sample size resulting from the model’s specificity (see below). For the same reason of sample size, thickness and altitude were not included in the models of this analysis. We used generalized linear mixed models (GLMMs) with different functions (see below) from the lme4 1.1-35.1 package (Bates et al., 2024) in R (version 4.3.3; R Core Team, 2024).

For analysis 1, we constructed three models, one for each nest attribute (height, altitude, and thickness), analyzing each one for the different nest fates ( Table 1 ). For analysis 2, we constructed 15 models, one model for each of the five changes in nest attributes (nest height, altitude, thickness, distance between nests, and environment type) across the three comparisons between success (“win”) or failure (“lose” = flooded, predated, or tipped) of the previous nest ( Table 1 ). For analysis 3, we built four models, one for each fate of the first nest (flooded, predated, tipped, and successful nets), comparing what led to the success or failure of the subsequent nest ( Table 1 ).

We included random variables in all models. Territory and year were treated as structural random predictors, meaning they were retained in all models (i.e., not removed during model selection; see below) to account for the identifier and time of repeated measures or longitudinal samples. Additionally, in all models, female ID number and male ID number were also included as random predictors, but they could be removed during model selection (see below) if this improved the model fit. The complete model structure can be found in Table 1 .

We disregarded the models with a sample size smaller than twice the number of predictor variables included. To avoid over-testing due to multiple parallel analyses, we adjusted the p-values for each of the three analyses separately. For this, we used the p.adjust function from the stats 4.3.3 R package (R Core Team, 2024), applying the Bonferroni method.

After selecting the best complete model distribution based on AIC values and convergence success, we proceeded with variable selection. We evaluated the importance ranking of the predictors using the MERF function from the LongituRF 0.9 package (Capitaine, 2020). We designated territory as the identifier and year as the time trajectory—for this reason, they are treated as structural variables in our GLMMs, as they contributed to the longitudinal structure of the models and did not have their importance obtained. The model was configured without any stochastic process, two predictor variables per split, and with other default settings. For random-effect predictors, importance was measured by summing their estimates, and for fixed-effect predictors, we assessed importance based on the percent increase in mean squared error (%IncMSE; Supplementary Tables 1–3 ). We sequentially removed the least significant random-effect predictors until their best combination based on AIC is identified, considering equally adjusted models with ΔAIC ≤ 2 (Bolker, 2008). Next, we applied the same method of removing the least important fixed predictors to obtain the best model.

Model validation was conducted by verifying normality and homoscedasticity of residuals using the residual, fitted, bptest, and shapiro.test functions of the stats (R Core Team, 2024) and lmtest 0.9-40 (Hothorn et al., 2022) R packages for normal and log-normal models. For binomial models, we validated the results by analyzing the receiver operating characteristic (ROC) curve and calculating the area under the curve (AUC) using the predict and roc functions of the stats (R Core Team, 2024) and pROC 1.18.5 (Robin et al., 2023) R packages. If our best models did not validate, we removed outliers, tested alternative distributions, or selected more complete models of the best distribution. In order to verify the differences in nest fate for analysis 1, we applied pairwise comparison tests using the emmeans and pairs function from the emmeans 1.10.4 (Lenth, 2024) and multcomp 1.4.26 (Hothorn et al., 2024) R packages.

The attributes of the nests of F. acutirostris were related to their fate ( Figure 4 ; Table 2 ). The height of the nests varied according to nest fate ( Figure 4A ), with flooded nests being at lower heights than predated (pairwise, p < 0.001), tipped (pairwise, p < 0.001), and successful nests (pairwise, p < 0.001). However, there was no difference in the height of successful nests versus predated (pairwise, p = 0.986) and tipped nests (pairwise, p = 1.000), and in the height of predated versus tipped nests (pairwise, p = 0.996). Nest height also varied between environments, with nests in Cladietum being at higher heights than those in Crinetum (pairwise, p = 0.001), Taliparitetum, Typhetum, and Schoenoplectetum (“other”; pairwise, p < 0.001); at lower heights than nests in Acrostichetum (pairwise, p = 0.002); and at similar heights to nests in Laguncularietum (pairwise, p = 0.195). Nests in Acrostichetum were at higher heights than those nests in Crinetum (pairwise, p < 0.001), in Laguncularietum (pairwise, p < 0.001), and in Taliparitetum, Typhetum, and Schoenoplectetum (“other”; pairwise, p < 0.001). There was no difference in the height of nests in Crinetum versus nests in Laguncularietum (pairwise, p = 0.560); nests in Crinetum versus nests in Taliparitetum, Typhetum, and Schoenoplectetum (“other”; pairwise, p = 0.856); or nests in Laguncularietum versus nests in Taliparitetum, Typhetum, and Schoenoplectetum (“other”; pairwise, p = 0.116).

Nest thickness was related to nest fate ( Figure 4C ; Table 3 ), with predated and tipped being thicker than flooded (pairwise, p = 0.046 and p = 0.012), and tipped nests being thicker than successful nests (pairwise, p = 0.043). Predated nests were equally thicker than tipped (pairwise, p = 0.664) and successful nests (pairwise, p = 0.274); and flooded nests were also equally thicker than successful nests (pairwise, z = 0.251; p = 0.774). We did not observe a relationship between nest altitude and fate ( Figure 4B ; Table 3 ).

Changes in subsequent nest characteristics were related to the fate of the first nest ( Figure 5 ; Table 3 ). After losing the nests due to flooding, the species constructed the second nest higher, making significant changes in nest height compared to chances in height after experiencing success (GLMM, p = 0.020; Figure 5A ). The changes in nest height following a successful nest were similar to those after predation and tipping ( Figures 5B, C ; Table 3 ). Similarly, the changes in nest altitude after experiencing success were similar to those after experiencing flooding and predation of the nest ( Figures 5D, E ; Table 3 ), and changes in nest thickness after a successful nest was similar to those after a predated nest ( Figure 5H ; Table 3 ). The distances between nests after experiencing success were similar to those after experiencing flooding, predation, and tipping ( Figures 5J–L ; Table 3 ), as well as the changes in nest environment ( Figures 5M–O ; Table 3 ). No other factors significantly influenced the changes in nests’ attributes adopted by the species ( Table 3 ). Because of the small sample size, it was not possible to analyze changes in nest altitude after tipping (n = 16), as well as changes in nest thickness after flooding (n = 141) and tipping (n = 2.).

Changes in nest height, distance between nests, and in the environment type did not influence the success rate of subsequent nests following flooding, predation, or success of the previous nest ( Figure 6 ; Table 4 ). Analysis 3 could not be conducted following previous nest tipping due to the small sample size (n = 11).