Breeding phenology can have strong demographic consequences. Yet, the reasons why coexisting animal species differ in breeding phenology have received little attention. We tested selection pressures that underlie life history variation, such as age-specific mortality, diet, and body size to explain the breeding phenology of 16 coexisting songbird species in arid shrubland in South Africa. The average start and end dates for the earliest and latest species differed by 1.5 months, with a gradient among the remaining species. Nest predation risk generally increased through the season, although species differed in seasonal patterns. Species with lower annual adult mortality, greater seasonal increases in nest predation, and greater nest predation rates had earlier start and stop dates, thereby reducing demographic risks. Species with higher adult mortality had larger brood sizes that required more food and they bred later when food was more abundant. Evolved timing of breeding thereby reflecting risk management and food availability related to longevity and brood size. These factors may place unrecognized constraints on within-species responses to climate change. Given the importance of phenology for fitness, phenology should be integrated as a core life history trait in future theory, and evolutionary constraints need to be considered in responses to climate change.

We studied the breeding birds in the coastal shrubland of Koeberg Nature Reserve, South Africa during 7 breeding seasons: 2000-2004, 2014, and 2016. This site and its habitat were described in a study by Nalwanga et al. (2004). Adult mortality of banded birds was studied from 2001-2007.

Nests were found using parental behaviors, where nest-building activity in particular was readily observed, following Martin & Geupel (1993). Large numbers of nests were monitored each year using long-term protocols to estimate nest predation rates and dates of nest initiation. We located and monitored a total of 6754 nests for the 16 most common species over the seven years (Table 1).

We used the date that the first egg was laid in a nest as the initiation date of a nest. In cases where we found nests after the clutch was complete or had nestlings, we back-dated based on the average clutch size, incubation, and nestling periods. We defined the starting and stopping dates each year as the initiation dates of the 5% earliest and 5% last nests found for each species. We used the averages of these 5% dates across the seven years as the average first and last initiation dates for each species.

Replication Statement – The authors wished to examine the causes of differences in the timing of breeding among coexisting bird species, based on 87 to 1520 nests per species studied over 7 years.

Nests were generally checked every other day, but sometimes up to four days following work breaks, to determine status and predation. Nests were checked daily or twice daily near hatching and fledging to obtain exact transition timing. Nest predation was assumed when all nestlings disappeared more than two days prior to average fledging age, and parents could not be found feeding fledglings following Martin & Geupel (1993).

Daily nest predation rates were estimated using the logistic exposure method (Shaffer, 2004) based on R x64 v4.1.l0 for Windows (R Development Core Team, Vienna, Austria). This approach considers the number of days a nest was observed with eggs or nestlings that were exposed to possible predation between each nest check to minimize any potential bias from successful nests found late in their nesting cycle (Mayfield, 1961; Shaffer, 2004). The Mayfield Maximum Likelihood method provides a single estimate of nest predation rate, whereas the logistic exposure approach allows examination of changes in nest predation with covariates, such as date. A recent study showed that these estimates are highly correlated between these two methods (Şahin Arslan & Martin, 2024). Using the logistic exposure method, we estimated the rate of change (slope) of daily nest predation rates relative to the ordinal date for each species separately. We pooled all data across years to obtain the largest and most robust sample possible for each species. We used the ordinal date of each nest check, nest fate at the check, and the exposure interval since the previous check, or half the interval if predation occurred, to model nest predation as a function of ordinal date under the logistic exposure method for each species (Shaffer, 2004). Our data included 33,339 nest exposure intervals encompassing 56,047.5 exposure days across the 16 species. We modelled nest predation from the average first 5% of initiation dates through 6 days after the average last 5% of initiation dates for each species as a general estimate of nest activity over the season.

Species generally differ in nest predation rates based on nesting habits (Martin, 1995), which may affect starting and stopping dates of breeding.  Assuming seasonal changes in daily nest predation risk, then differing timing can yield differing nest predation rates that may not reflect differences in the nesting habits of species. For example, a species that breeds early may have a different predation rate than one that breeds later due to seasonal changes in predation rates and not necessarily reflecting the relative risk of their nesting habits. Consequently, we estimated a late-season date-controlled daily nest predation rate for each species to isolate the relative difference in risk among species controlled for seasonal changes in predation. We used the latest average initiation date (ordinal date = 285, or 12 October) in which all species were still initiating nests on average to obtain a late-season date-controlled daily predation rate for each species. We used the logistic exposure relationship for each species to predict their predation rate on this date. This date was constrained by Cape Robin-chats (Dessonornis caffra) which stopped breeding at the earliest. We used nest predation late in the season because species with higher risk may try to avoid this high nest predation late in the season by nesting earlier and we tested this possibility in our models for explaining start and stop dates.

Adult mortality was estimated using netting, banding, and resighting of birds during the breeding seasons from 2001 to 2007 by Lloyd et al. (2014). Estimates were obtained using Cormack-Jolly-Seber modelling in a Bayesian framework and taken from Lloyd et al., (2014).

We accounted for phylogenetic effects (Felsenstein 1985) in all analyses using the Caper package (Orme et al., 2018) in R x64 v4.1.0 for Windows (R Development Core Team, Vienna, Austria) based on phylogenetic generalized least squares analyses (PGLS). The phylogenetic tree was obtained from www.birdtree.org (Jetz et al., 2012) using the Hackett et al. (Hackett et al., 2008) backbone and imported into the program Mesquite (Maddison & Maddison, 2011) where a majority rules consensus tree was constructed from 1000 trees (see Figure S1 in Supporting Information). This consensus tree was then used in phylogenetically controlled analyses (PGLS) that incorporated Pagel’s lambda (Pagel, 1992) to transform branch lengths and reduce over-correction for phylogenetic effects (Symonds & Blomberg, 2014). We standardized all variables to z-scores for analyses. We tested average start and stop dates as separate dependent variables and began with a global model of possible explanatory covariates that included: annual adult mortality rate, seasonal change in nest predation rate, late season nest predation rate, diet (insects vs fruit/seeds), nest type (enclosed vs open), and body mass. We used a standard backward, stepwise elimination modeling approach in the PGLS models, where we removed the most insignificant variable at each step to arrive at a final model of significant variables. Insignificant variables and their statistics at their last step are reported to clarify the full model that was tested.

We further tested the possible importance of collinearity among covariates in influencing breeding phenology (see Figure 1) using a phylogenetic path analysis based on the R program phylopath (van der Bijl, 2018). In the end, we only report the path analyses in the supplementary material because they added little additional information beyond the PGLS results since inter-correlations among variables were largely lacking except for one between nest type and body mass and one between adult mortality probability and the rate of change in nest predation over the season, as discussed in results.