Methods

Study site

The study took place over three consecutive non-breeding periods, hereby referred to as year1 (November 2017 – April 2018), year2 (September 2018 – April 2019), and year3 (November 2019 – March 2020) at a guinea savannah site on the Jos Plateau, Nigeria in West Africa (09°52’N, 08°58’E). This region experiences single pronounced wet and dry seasons lasting six months each, from May to October and November to April, respectively. Sites were primarily open scrubland with different and varying degrees of anthropogenic activities, e .g . farming, livestock grazing, tin mining, and fires (Hulme & Cresswell, 2012). These sites represent typical African dynamic habitats, where anthropogenic activities are constant and continuously changing throughout the year.

Mist-netting and resightings

Birds were captured using 9m, 12m, and 18m x 2.5m 5-shelf (16 x 16mm mesh) mist nets and conspecific playback. During year1, nets were set up in the morning between mid-November 2017 and mid-February 2018 (mean of four nets per day, open for 2h 50m), totalling 70 visits. In year2 nets were set up in the morning and/or evening from late October 2018 to mid-April 2019 (mean of 4.5 nets per day for 3h 24m), totalling 69 visits. Few additional birds were caught in year3 between mid-November 2019 and mid-February 2020 but were excluded from return rates and between-years site fidelity analyses. All individuals were sexed as either female, male or unknown, and aged as either first-year, adult, or unknown (Svensson, 1992). Each individual was given a unique combination of coloured leg rings (three colour rings and a metal ring). In total, 212 individuals were colour-ringed in year1, 115 individuals in year2, and 10 in year3. This work was conducted under the ethical guidelines of the AP Leventis Ornithological Research Institute Scientific Committee and all methods were approved by the School of Biology Ethics Committee of the University of St. Andrews (SEC17028).
Resightings were carried out at least once a week between sunrise and ~1030 hrs and/or between ~1500 hrs and sunset throughout the fieldwork period. Two observers undertook all observations. We interspersed starting points to avoid biases as a product of the time of day and air temperature. Resightings were not carried out during days of heavy rain. Once an individual was detected we proceeded to identify its complete colour combination using 10 x 40 binoculars. GPS points were recorded with a Garmin eTrex10™ GPS where individuals were first detected and/or captured. Due to the skittish and shy behaviour of Whitethroats, conspecific playback was used. In some cases, individuals were first detected and playback was then used to help reveal the complete colour-combination. In most cases, however, when there were no signs of activity, playback was used before detection. This did not seem to induce any significant movement in individuals, and we believe that most recorded GPS points reflect unbiased locations where the individuals would be without any interaction with observers. We tried to spend the same effort resighting all individuals, but we acknowledge that this may not have been always the case. 135 individuals were seen at least once after capture. Because of the high resighting effort, we are confident that departure months and site persistence were determined accurately.

Radio tag deployment

Between 25 October 2019 and 28 November 2018, 11 individuals were fitted with “LifeTags™”, a 0.45 g solar-powered and battery-free radio transmitter from Cellular Tracking Technologies™. Tags were attached to birds’ backs using an elastic leg-loop harness (Rappole & Tipton, 1991). Devices weighed approximately 0.51 g with the harness, corresponding to 3.4% (3.2 – 3.8%) of an individual’s body mass. As individuals were fitted with radio tags, an effort was made to seek them at least twice a week after tag deployment until 8 December 2019. All birds were observed for at least three days after tag deployment. When individuals were detected, efforts were made to observe and corroborate its ring combination. GPS coordinates were recorded where individuals were first seen or heard or when detection was strong. To determine whether radio tags had any negative effect on individuals, the residency period (number of days between when an individual was caught and the last time it was detected) and return rates (proportion of individuals that returned the following non-breeding period) was compared between 11 radio-tagged individuals and 11 randomly selected control birds, ringed during the same period. No significant differences were found regarding residency periods (F(1,20) = 0.05, p = .82) or return rates (χ² = 0.26, df = 1, p = .61) between radio-tagged individuals and controls.

Detection probabilities

The probability of detecting an individual directly affects how we calculate and categorise site persistence. Therefore, to estimate detection probability we used multiple datasets and methods.
  1. Manually : Detection probabilities were calculated by dividing the number of times a bird was detected (number of encounters) by the total visits to its home range between its first detection (excluding the date it was ringed) until its last detection for each year. We used data obtained from individuals that we knew were present at the study site during each visit (i .e . obvious long-term winter residents, see below) to be certain that their non-detection was due to detectability factors and not due to absence or death. This assumes that birds did not leave their home range at any time and that all birds, if present, had the same probability of detection. We used information collected from 20, 16, and 15 individuals during years 1, 2, and 3, respectively. All data were analysed separately by year and returning individuals were included in every year they were detected: excluding them would otherwise bias estimates by preferentially sampling first winter birds.
  2. MARK : With the same data, we proceeded to calculate detection probabilities using Cormack–Jolly–Seber (CJS) models in MARK software (White & Burnham, 2009). CJS models estimate both apparent survival (φ ) and detection probability (p ), where the former is the probability that an individual survives from one sampling occasion to the next, and the latter is the probability that, given that the individual is alive and in the sample, it is encountered (Hammond, 2018). Given that we used capture histories from individuals which we knew were present and alive (φ = 1), we were only interested in obtaining the detection probability for each year. We assumed that detection was constant throughout all encounters (φ (.)p (.)).
  3. Radio tags : Detection probabilities were calculated for three radio-tagged individuals that were detected at least during three visits in year2. Every time a radio-tagged individual was detected with the antenna, we proceeded to find it in the same manner that we would normally do during resightings. We then estimated detection probabilities by dividing the number of visits during which an individual was detected in ‘resighting’ conditions by the total number of visits that same individual was detected with the radio tag antenna.
The final overall detection probability was obtained by averaging all seven estimates: detections obtained manually and in MARK for all three seasons (total of six detections), and a detection obtained through radio-tagged individuals.

Site persistence

Once established that individuals undertook different residency strategies (see Appendix 1), we estimated the number of days individuals spent in the study area (days between when individuals were first and last detected). To facilitate further comparisons, however, individuals were grouped into residency categories as seen in Table 1. Individuals detected across more than one year were categorised independently each year.

Between-years site fidelity

Return rates were estimated by dividing the number of individuals that were seen in year i +1 by the total number of individuals ringed in year i . To determine the degree of between-years site fidelity of individuals that returned for at least two non-breeding seasons – how far an individual moved from year i to year i +1 – we calculated the centroid coordinate for each individual in each year and estimated the distance between centroids using the “distHaversine ” function from the “geosphere ” package version 1.5.10 in R (Hijmans, 2019; Fig. 1). Individuals were grouped into group A, individuals detected in years 1 and 2, group B, individuals detected in years 2 and 3, and group C, individuals detected in years 1 and 3 but not in year2. Individuals that were seen during all three seasons were not excluded from the analysis and were added to groups A and B.

Departure dates

We tested departure date repeatability of individuals seen for at least two non-breeding periods. Year3 birds were excluded from this analysis because resightings that year ended earlier, and final resightings were not likely to reflect true departure dates. We excluded records of all birds that were seen after 25 February (three weeks before the end of observations) to exclude birds that were highly likely to have not left before our last resighting effort of that year. We estimated repeatability using the “rpt” function in the “rptR ” package (Stoffel et al., 2017). This uses a linear mixed model framework where the groups compared for repeatability are specified by a random effect (i .e . individuals). Confidence intervals were estimated by running 1000 bootstraps. We calculated repeatability for adults and first-year birds, as well as for each residency category (i .e . long-term, short-term, and passage birds).

Statistical analyses

All data were analysed using R version 3.6.3 and RStudio version 1.1.456 (R Core Team, 2020) and a statistical significance level of p< .05 was chosen to reject the null hypotheses.

Detection probabilities

To compare whether detection probabilities were constant between non-breeding periods and methods, General Linear Models (GLMs) were performed.

Site persistence

We performed GLMs to understand whether site persistence, defined as the number of days an individual was present and detected in the area, varied across years, age, and sex. Birds that could not be aged or sexed were excluded from models that included these variables as predictors. First-year Whitethroats are difficult to sex accurately (Waldenström & Ottosson, 2000), so models using sex as an independent variable only include adults. Because of this, modelling for the effects of age and sex in residency periods was undertaken separately. Data from year3 were excluded from these analyses as well as those individuals whose age and sex were unknown.
We used a model averaging approach for models that had the same sample size using the “dredge ” and “model.avg ” function from the “MuMin ” package in R (Barton, 2020). This procedure entails carrying out all possible models from a base model (i .e . ‘days ~ age + year’ and ‘days ~ sex + year’), and calculating a weighted average of parameter estimates, such that parameter estimates from models that contribute little information about the variance in the response variable are given little weight (Grueber et al., 2011).

Between-years site fidelity

Chi-squared tests (χ2) were performed to detetermine the effects of year, age at year i (‘previous age’), sex, and residency at year i (‘previous residency’) on return rates. A model averaging approach was also undertaken to explore whether the distance moved from one year to another was dependent previous age, sex, and previous residency (base model: ‘dist ~ preage + sex + group + preres + preage*preres + preage*group’). All birds that could not be aged were excluded from models that included age as a predictor.

Residency repeatability

To explore whether individuals remained for similar periods across different years, or whether they repeated residency categories the following years, we estimated the percentage of individuals that remained (or changed) in each residency category. We carried out a linear model and estimated the correlation between the number of days spent in year i with the number of days spent in year i +1.

Departure dates

To describe population variation regarding departure dates, we pooled all observed dates across the first two years from individuals that left after January in a respective year. We then calculated the difference between each date and the date of earliest sighting and calculated the mean, standard error (se), and range. To describe intra-individual variation, we used data from individuals that were detected for at least two years. We calculated the difference between the two values for each individual observed in two years and calculated the mean, se, minimum and maximum values across all individuals. GLMs were performed to test for differences between individuals categorised by previous residency and previous age.