Effects of the environmental conditions and seasonality on a population survey of the Andean condor Vultur gryphus in the tropical Andes

Studied species and data collection

The Andean condor is the largest scavenger in South America (Winkler, Billerman & Lovette, 2020). The behavior of aggregating in communal roosting is widespread among vultures and has been particularly useful for monitoring this species (Perrig et al., 2019). Environmental conditions can influence this behavior (Beauchamp, 1999; Mazumdar, Ghose & Saha, 2017), resulting in thermoregulation benefits and low energetic costs to obtain partners and available resources (i.e., food and mates; Blanco & Tella, 1999; Beauchamp, 1999; Moleón, Bautista & Madero, 2011). Particularly for the Andean condor, researchers commonly find the communal roosts in areas with high avian diversity (Lambertucci, Ruggiero & Nascimento, 2013) and have been associated with philopatric behavior (Padró et al., 2019) and age-class segregation (Lambertucci, 2013). However, most of the existing knowledge about the Andean condor’s biology is limited to studies in the subtropical Andes. At the same time, tropical areas (e.g., Peru) remain poorly studied (Santangeli et al., 2022).

The study site is located at the Pampa Galeras-Bárbara D’Achille National Reserve (RNPG-BA)—SERNANP Ayacucho and is already recognized as a priority area for the conservation of Andean condors (Wallace et al., 2020). It is located in the highland of the Andean Puna ecoregion, in the central Andes of Peru, at 4,100 m (Fig. 1). This area presents hills from 50 to 150 m and slopes from 40° to 65°. In this cold-dry area, the minimum temperature is −8 °C, and the maximum is 16 °C, with an annual rainfall of 450 mm usually limited to September to March (SERNANP, 2015). Winds can reach speeds between 2–40 km/h (SERNANP, 2015). The study site represents the unique communal roost known nearby the RNPG-BA. The other two nearest communal roosts are in the locations of “Sondondo” and “Punta Gallinazo” (Reserva Nacional San Fernando), at 60 km north-northwest and 130 km southeast from the RNPG-BA, respectively. These nearest communal roosts have distinct climatic conditions and represent, for the Andean condor, a long journey from one roost to another despite its high mobility (the mean daily flight is 64 km, with a maximum range of 197 to 350 km; Lambertucci et al., 2014; Perrig et al., 2020; Padró et al., 2018).

The Andean condor surveys were conducted from November 2015 to March 2016, three times a week (=109 days), using binoculars (10 × 50 mm). We performed three surveys daily: 06:00 a.m., 12:00 p.m., and 06:00 p.m. (modified from Lambertucci, Jácome & Trejo, 2008). The early morning and late afternoon survey hours corresponded to the period most individuals were roosting (i.e., 06:00 a.m. and 06:00 p.m.). Also, we used the midday survey (12 p.m.), when Andean condors seldom use roosts, to compare their roost abundance with the other two time periods of the day (Sarno, Franklin & Prexl, 2000; Kusch, 2004). The roosted individuals were counted and separated into adults, with black and white plumage, and immatures, with brownish-gray plumage (see Perrig et al., 2019; Houston et al., 2020). Because distinguishing between males and females was difficult at certain distances and during the sunset, we decided to avoid distinguishing males and females in this study. We used surveys to estimate: (1) the minimum population size in the communal roost, calculated as the maximum number of adults added to the maximum number of immatures recorded, and (2) the immature/adult ratio.

Parallel to the surveys, we obtained the temperature and rainfall data from the Servicio Nacional de Meteorologia e Hidrologia (SENAMHI) weather station. This station is located in the facilities of the RNPG-BA checkpoint, 7 km southwest of the communal roost studied. Climatic data were available at 7:00 a.m. and are provided in the Supplemental Material.

Statistical analyses

We used a two-way ANOVA to compare the number of Andean condors surveyed across seasons and time of the day. Fitted values were plotted against the residuals to check for unequal error variances and outliers (see Supplemental Materials S1, S2). In addition, we assessed the extent to which temperature, rainfall, and seasonality predict the number of condors in the communal roosts using generalized linear models. Because the Andean condor survey resulted in a time series, we compared competing autoregressive generalized linear models (Autoregressive GLMs hereafter). Since we had no prior hypotheses about how climatic variables and seasonality would affect surveyed condors, we built Autoregressive GLMs representing all possible combinations of effects on the climatic variables to the Andean condor survey. We compared these competing models using the small sample corrected Akaike Information Criterion (AICc). We ranked the models according to the differences (∆AICc) of the model with the lowest AICc, which was assumed to be the most plausible. We considered models with ∆AICc < 2 as equally plausible. The Akaike’s weight (w) shows the weight of evidence for a particular model. Procedures to classify the dry and wet seasons and the rationale of the Autoregressive GLM are better explained next. All analyses were conducted using software R 4.1.0 (R Core Team, 2021).

We evaluated the influence of environmental variables on condor surveys using an Autoregressive GLM with a negative binomial distribution from the tscount package to model the total condors surveyed (version 1.4.3) (Liboschik, Fokianos & Fried, 2017). It is preferable to use autoregressive models when dealing with time series, such as the Andean condors surveys. A time series implies that the number of individuals y observed at time t, (y_t) is connected to the number of individuals observed in previous i observations (y_t-i), where i is the time lag included in the autoregressive model. The first step to modeling an autoregressive GLM is defining the number of time lags in the model, informed by a preliminary partial autocorrelation function (PACF). Although the number of time-lags parameters of the model in the autoregressive GLM must be prior informed, the model will estimate the magnitude to which each time-lag determines the observed values in y_t. The number of time lags included in the model is necessary to remove the temporal autocorrelation between the residuals. Therefore, we evaluated the partial autocorrelation using the forecast package (Hyndman & Khandakar, 2008; Hyndman et al., 2022), which suggests that observation y_t-1 is strongly related with one to four observations back in time i, i = 1, 2, or 4 (see Supplemental Material S1). Using i = 1 in the models as our first attempt was sufficient to remove the autocorrelation between the residuals (see Fig. S1.7), so we did not make other attempts. Because our first attempts using Poisson distributed models showed overdispersion, we built all models with a negative binomial distribution for a better fit (see Fig. S1.8).

Visual inspection of the number of surveyed condors unveils a gradual decrease over the year from about October, coinciding with an increase in the rainfall and an increase in the temperature, which no longer achieves negative values (Fig. S1.1). Thus, to define a breakpoint between the dry and wet seasons, we have used the function cpt.meanvar as default from the changepoint package (Killick & Eckley, 2014). We used visual inspection and a change-point analysis to define the dry and wet seasons. Using cpt.meanvar, we detected shifts in the number of condors, temperature, and rainfall based on changes between mean and variance. This approach allows us to establish an arbitrary breakpoint between the dry and wet seasons with statistical support. The breakpoints detected by the changepoint package were on 22 October for surveyed condors, on 27 August for temperature, and on 29 August for rainfall. Since we expected multiple factors influencing the use of the communal roost over the year by the Andean condors, including some not assessed in this study (i.e., food availability and mating), we assumed we could not define a breakpoint between the seasons strictly based on climatic variables. Then, we used the mean value of the three breakpoints (i.e., temperature, rainfall, and surveyed condors) to determine the end of the dry season and the beginning of the wet season, represented by mid-September (15 September). Therefore, the dry season in our study ranged from June to mid-September and the wet season ranged from mid-September to March. Despite the use of a broader sense of classification regarding the seasons, our delimitation is supported by other studies in the tropical Andes (e.g., wet season: Oct-Mar. (Jomelli et al., 2012); and Sep-Apr, (Silva, Takahashi & Chávez, 2008)).

Updates about the biology of the Andean condor

This study was the first to describe a remarked seasonality for the Andean condors using communal roost in the tropical Andes. So far, the only roost survey available for the tropics (on the arid coast of Peru, 60 km distant from our study site) showed regular use of the communal roost throughout the year (Vásquez, 2015). The difference between our study site and the communal roosts studied by Vásquez (2015) might be due to the low rainfall on the arid coast of Peru (approx. 8–80 mm/year; Olivares, Taype & Castro, 1994). Thus, in terms of seasonality, our results are more similar to other studies in subtropical areas of the Andean condor occurrence, pointing out a higher number of individuals in the wet season (i.e., Sarno, Franklin & Prexl, 2000; Pavez, 2020). This pattern, however, seems not to be unique to the Andean condor distribution (e.g., Kusch, 2004; Lambertucci, Jácome & Trejo, 2008). Therefore, it is likely that Andean condor populations in tropical areas present distinct displacement behaviors throughout the year on the coast and in the Andes. Despite the limited insight offered by our results on this matter, given that we only sampled one communal roost, Pennycuick & Scholey (1984) have already suggested that the predominant direction of the thermals might limit the connection between the coast and the highland Andes (but see Lambertucci et al., 2018).

The reduced surveyed condors in our study site during the wet season may reflect constraints imposed by the rainfall. While the Andean condors are likely to fly better and disperse more in the summer due to the thermals (De Martino et al., 2011), our study site may be temporarily unattractive due to the higher incidence of rainfall. An alternative explanation is the reduction of individuals due to higher dispersal during the breeding time (e.g., Lambertucci, Jácome & Trejo, 2008). Despite limited knowledge about the breeding season of the Andean condor in the tropics, there is evidence of egg-laying and incubation occurring between February and June (Wallace & Temple, 1988; Ríos-Uzeda & Wallace, 2007; Restrepo-Cardona et al., 2018). Nevertheless, the absence of immature condors in the study site is the main reason for the difference in Andean condors surveyed during dry and wet seasons. This pattern suggests that breeding time alone cannot explain the differences in the number of surveyed condors we observed over time.

The adult:immature ratio found at our study site is also remarkable and sheds light on important insights about the biology and conservation of the Andean condor. The minimum population size estimated suggests an adult:immature ratio of 1:3.37, which is highly disparate compared to previous studies that have reported a ratio close to 1:1 (range, 1:0.52–1:1.50; i.e., Sarno, Franklin & Prexl, 2000; Ríos-Uzeda & Wallace, 2007; Lambertucci, 2010; Escobar et al., 2015; Pavez, 2020). For long-lived and slow-reproductive species, the adult:immature ratio usually tends to be biased towards adults, which have small annual clutch sizes (e.g., Tella et al., 2013; Margalida et al., 2020).

Because the study site represents what we consider a relatively isolated communal roost (see methods), we identified at least two potential non-competing hypotheses to explain this adult:immature bias. First, biases towards a high number of immatures may suggest good recruitment in recent years. If this is the case, our results might suggest that the population of Andean condors in RNPG-BA is increasing. Indeed, more recent observations on this site have reported up to 45 individuals (S. Paredes-Guerrero, 2022 unpublished data). Alternatively, the social hierarchy is also a potential explanation. It is well-known that adults at the top of the social hierarchy have access to the best resources, leaving lower-quality resources for the individuals at the bottom of the social hierarchy, usually the youngest ones (Kirk & Houston, 1995; Donázar et al., 1999). The communal roost conservation priority must therefore consider a population spatially separated with different communal roosts providing shelter primarily for different phases of the Andean condor life cycle if the social hierarchy is that important. Finally, we recognize both hypotheses as possible because we investigated only one communal roost; nonetheless, this must be considered in the national census in Peru or other Andean countries to clarify these pressing questions further. Such results also chorus with other studies highlighting potential healthy populations based on the adult:immature ratio in Bolivia (Ríos-Uzeda & Wallace, 2007) and Ecuador (Naveda-Rodríguez et al., 2016).

Guidance for national surveys and conservation of the Andean condor

Our results support multiple concomitant surveys of potential communal roosts at each season. These surveys should be conducted mainly in the morning or late evening to avoid underestimation; different results are found in subtropical areas (Kusch, 2004). Surveys should also consider daily climatic conditions (e.g., Perrig et al., 2020). Particularly for the study site and potentially for other sites in the tropical Andes with marked climatic seasonality, surveys on colder and dryer days are more prone to capture the full potential (i.e., more individuals) of the Andean condor’s communal roosts. On the other hand, seasonality might not be a concern for areas with more stable climatic conditions. Finally, as expected, given that Andean condors can travel long distances (see Piana & Vargas, 2018; Perrig et al., 2020), protected areas could be an essential conservation tool for this species, but not alone (see Guido et al., 2020; Wallace et al., 2020). Instead, a network of protected areas, including communal roosts with different environmental characteristics, is important to provide adequate protection for Andean condors (Wallace et al., 2020). In the subtropical areas, significant progress has been made, particularly in the last decade, in mapping priority areas for the Andean condor (e.g., Perrig et al., 2019) and unveiling the spatial genetic structure of the population among multiple communal roosts (Padró et al., 2018, 2019). These topics, however, remain scarcely explored in the tropical Andes, which might be particularly important to clarify the extent to which the differential use of communal roosts along the year may reflect on priority areas selection and genetic structure of the condor populations. In brief, these areas might require particular strategies to promote an efficient national survey.