Discussion

We present a novel tool that allows users to predict reproductive behaviour from GPS tracking data of red kites, and thus extract important reproductive parameters that can inform the population dynamics of this species (Katzenberger et al. 2021, Sergio et al. 2021, Pfeiffer and Schaub 2023). We developed these models with a large tracking dataset in Switzerland, but show that they can be transferred to different populations in Germany and successfully predict reproductive behaviour. We encourage the use of NestTool in other populations and species to estimate demographic parameters and assess the broader transferability of this tool to other species and populations.
Although the movement patterns of birds repeatedly visiting a nest site to incubate eggs or feed chicks can be highly diagnostic and allow an accurate identification of a bird’s breeding status (Picardi et al. 2020, Schreven et al. 2021, Bowgen et al. 2022, Overton et al. 2022, Ozsanlav-Harris et al. 2022, Eisaguirre et al. 2023), there are limitations to automatic identification (van der Wal et al. 2015). Red kites, in particular, can have individually variable home range behaviour, and also display profound differences in movements both between males and females and in different landscapes (Aebischer and Scherler 2021, García-Macía et al. 2022b, Spatz et al. 2022). Our tool provides probabilistic predictions and facilitates easy manual inspection of cases that fall below a user’s tolerance level of certainty. This tool is therefore not designed to work as a ‘black box’, but rather to assist researchers in the otherwise time-consuming manual inspection of events during the breeding season of tracked animals. By specifying the level of uncertainty that is tolerable, the user retains full control over the trade-off between accepting automated predictions and manually annotating the reproductive behaviour of many individuals.
NestTool provides accurate predictions for red kites using the movement metrics that we chose, but we would welcome further development and improvement for this tool to potentially become useful for other species. We selected 42 variables (Table S1) that are primarily focused on durations, distances, and home range areas. However, many alternative formulations are conceivable, including for example environmental features and historical context to put the movements of an individual into perspective of previous movements (Williams et al. 2020, Overton et al. 2022). The extraction of environmental features from remote-sensing data may be highly beneficial for some species, but may increase the computational cost of the tool, and make the transferability of models to other populations more problematic (Hothorn et al. 2011, Yates et al. 2018).
Because we did not use environmental or landscape variables, the variables that explained most of the variation between birds with and without a home range mostly described the amount of time individuals spent in a certain radius around the centre of the home range, either during the day, at night, or in total (Fig. S2). Territorial birds are expected to spend most of their time within their territory to be able to defend the territory against intruders and raise their brood (Hinde 1956, Kaufmann 1983), and the importance of time as a suitable predictor is therefore not surprising. Similarly, the prediction of nesting behaviour relied primarily on the amount of time spent in the vicinity of a potential nest site, and the median distance between diurnal locations and where birds roosted at night (Fig. S3), as most nesting birds generally sleep either on (females) or very close to (males) the nest (Aebischer and Scherler 2021).
By contrast, the prediction of breeding success relied primarily on movement metrics in the late chick stages (Fig. S4), including how often individuals revisited the nest during that phase, how much time they spent during the later chick stages, and when they last visited the nest site. Successful breeding requires chicks to be fed for 46 days, thus requiring multiple daily revisits to the nest for the entire duration of the breeding season, and regular presence at the nest until the chicks fledge (Aebischer and Scherler 2021, Scherler et al. 2023a). Although even unsuccessful breeders frequently remain in their territory and at their nest site, the number of revisits to the nest decreases when there are no more chicks that need to be fed (Scherler et al. 2023a). Moreover, in red kites brood loss primarily occurs during incubation (Nägeli et al. 2022), which then leads to a distinctive relaxation of the repetitive visits to the nest and an increase in the distance from the nest already during incubation (Fig. 2). These revisitation patterns are broadly similar to what has been used to predict breeding success in other altricial raptor species (Picardi et al. 2020, Eisaguirre et al. 2023, Murgatroyd et al. 2023), and we are therefore optimistic that this tool could be adapted to identify reproductive behaviour for other species as well.
Using an automated tool to classify and predict the existence and outcome of reproductive behaviour has clear benefits when comparing the population dynamics across several populations of a species. One key advantage of using tracking data at a resolution that is available across a range of studies is the reduced influence of survey effort, which can affect the assessment of reproductive success (Harris et al. 2020, Eisaguirre et al. 2023, Murgatroyd et al. 2023). By using a consistent algorithm across many populations, effects of survey effort or other regional adaptations can be overcome when understanding the range-wide population dynamics of species (Mattsson et al. 2022). This is particularly important for demographic parameters that are difficult to estimate from field observations, such as whether reproduction occurs or not. Especially in long-lived species, and populations approaching the carrying capacity of the environment, an increasing proportion of adult birds may defer breeding or fail at early stages of the nesting cycle, which would be very difficult to detect using field surveys (López‐Sepulcre and Kokko 2005, Catlin et al. 2019). Our tool offers a consistent classification approach that is independent of field survey effort and could therefore overcome some of the challenges in quantifying the proportion of a population that is attempting reproduction.
The age when the first reproduction attempt occurs is another important driver in the population dynamics of long-lived species (Saether and Bakke 2000, Sergio et al. 2021), and may change in increasing populations that approach carrying capacity (Ferrer et al. 2004, Margalida et al. 2020, Katzenberger et al. 2021). Because the pre-reproductive years of long-lived animals can be difficult to observe and can be affected by emigration, our tool offers a consistent approach to quantify at what age young individuals recruit into a population and attempt reproduction. Together with reproductive success, and the use of telemetry data to estimate survival (Sergio et al. 2019, Swift et al. 2020, Buechley et al. 2021), most of the key demographic parameters that are required for understanding population dynamics can therefore be inferred from telemetry data. Our tool shows that demographic parameters can be estimated consistently for different populations using the same approach, which could facilitate opportunities to assess population dynamics at larger spatial scales without the problem of inconsistencies in estimates between local or regional studies (Mattsson et al. 2022).
In summary, NestTool has the potential to facilitate the extraction of important demographic parameters from widely available tracking data for red kites in Europe. The availability of demographic information about when young birds recruit into the breeding population, and how successful they breed is useful to understand the population dynamics of the red kite, which may have changed in recent decades as populations have recovered (Katzenberger et al. 2019, 2021, Pfeiffer and Schaub 2023). We encourage red kite researchers to use this tool to assess the reproductive performance of their tracked individuals, and other researchers to adopt this tool for different species, as has been successfully done for species breeding in remote arctic areas (Eisaguirre et al. 2023, Ozsanlav-Harris et al. 2023).