Evidence of large‐scale range shift in the distribution of a Palaearctic migrant in Africa

Long‐distance Palaearctic migrant birds are declining at a faster rate than short‐distance migrant or resident species. This is often attributed to changes on their non‐breeding grounds and along their migratory routes. The European honey buzzard (Pernis apivorus) is a scarce migrant in southern Africa that is declining globally. This study assessed the distribution and abundance of honey buzzards in southern Africa over the past four decades and compared it to trends in the East African population to examine possible drivers of population expansion in southern Africa.

Previous research has shown that wind patterns en route and habitat quality affect where migrating birds establish non-breeding territories (Gunnarsson et al., 2005;Norris et al., 2003;Richardson, 1990;Saino et al., 2004;Thorup et al., 2003;Vansteelant, Kekkonen, & Byholm, 2017). Adult birds are more able to account and correct for the effects of wind but inexperienced juveniles, who are likely not navigating to a specific location, are not as able to rectify their direction (Thorup et al., 2003). The direction and strength of the wind are likely to have a large effect on where young birds initially arrive in Sub-Saharan Africa (Vansteelant et al., 2017). However, it is not clear how this may affect where they eventually settle.
Habitat quality in the non-breeding season has also been shown to have strong effect on both survival and breeding success in Neotropical and Palaearctic migrants (Gunnarsson et al., 2005;Norris et al., 2003;Saino et al., 2004). For example, more productive non-breeding territories have been shown to correlate with earlier arrival and better breeding territories in American Redstart (Setophaga ruticilla; Norris et al., 2003). The same is true for Icelandic Black-tailed Godwit (Limosa limosa); individuals that spent the non-breeding season in more productive coastal habitats had higher quality breeding sites and improved breeding success than those that selected less productive inland non-breeding locations (Gunnarsson et al., 2005). Once a productive site is located, many adult migratory birds, particularly long-lived species such as raptors and waterbirds, return to the same non-breeding site year after year (Cresswell, 2014). It is believed that returning to known suitable sites benefits long-term survival by reducing mortality risk during the non-breeding season (Cresswell, 2014).
These sites are likely to be located during juvenile exploratory movements earlier in life, which allow the bird to find more suitable locations (Cresswell, 2014).
Long-distance Palaearctic migrant bird species are declining at a faster rate than both short-distance migrant and resident species (Sanderson, Donald, Pain, Burfield, & van Bommel, 2006;Thaxter, Joys, Gregory, Baillie, & Noble, 2010;Vickery et al., 2014). This is believed to be due to change in climate and habitat degradation on both the breeding and wintering grounds as well as mortality during migration. A study of British migrant birds reflected this trend, with Afro-tropical migrants declining more rapidly than species spending their winters in the United Kingdom or other parts of Europe (e.g., residents and short-distance migrants; Thaxter et al., 2010). From 1986 to 2006, species that spent the non-breeding season in the humid West African forest and savanna declined at a greater rate than those in the more arid northern zone of Africa (Thaxter et al., 2010). This may be due to large-scale habitat loss and fragmentation from forest clearing and intensification of agriculture in this region (Thaxter et al., 2010).

The European honey buzzard (Pernis apivorus Linnaeus 1758)
is a Palaearctic raptor that migrates annually from its breeding grounds in Europe to its non-breeding grounds in Africa (Cramp & Simmons, 1979). This dietary specialist breeds in dense forest where it feeds mostly on wasps and bees (Hymenoptera) with frogs (Anura) as an important alternate prey (Gamauf, 1999;Itämies & Mikkola, 1972). Following breeding (or fledging in the case of juvenile birds), honey buzzards migrate south en masse (Cramp & Simmons, 1979).
It is thought that the vast majority of birds spend the Northern Hemisphere winter in the tropical rainforests of west and central Africa with a small proportion migrating to East and southern Africa (Agostini et al., 2007;Bruderer, Blitzbau, & Peter, 1994;.

While the European honey buzzard is listed as a species of Least
Concern by the IUCN, it has declined across Western Europe and Scandinavia over the past 30 years (Benusan, Garcia, & Cortes, 2007;IUCN, 2019;Kostrzewa, 1987;Löhmus, 2005). This may be due to habitat transformation and loss on both the breeding grounds in Europe as well as the non-breeding grounds in Africa (Benusan et al., 2007;Kostrzewa, 1987;Löhmus, 2005). Persecution during migration also plays a role in the decline (Brochet et al., 2016). Migration counts at Gibraltar in Spain and Batumi in Georgia show stable or declining numbers of honey buzzards passing annually (Batumi Raptor Count, 2017;Benusan et al., 2007).
The European honey buzzard is generally considered a scarce migrant to southern Africa (Cramp & Simmons, 1979;Hockey & Ryan, 2005). Over the past three decades, the number of reported records of this species has appeared to increase. This study aimed to examine the drivers of the apparent increase in the European honey buzzard population in southern Africa using citizen science data. We assessed two hypotheses: (a) the growth in records was due to an intensification of birdwatching and citizen science in the southern African subregion and (b) the changes further north in the honey buzzards range (e.g., habitat loss) have driven the range shift in the European honey buzzard and have consequently caused an expansion in the southern African population.

| Southern African rarity record collection
European honey buzzard records in southern and East Africa were collected from a variety of sources. East African records were selected for comparison based on tracking data which indicate that most southern African honey buzzards migrate from the eastern portion of their breeding range and travel through East Africa on their way to and from southern Africa (C. Howes, P. Byholm, & C. T. Symes, unpublished data).
To begin, eight museums were surveyed to determine if they had European honey buzzard specimens from southern and East Africa. In addition to European honey buzzard records, southern African rarity records for all subregion rarities were also collected from SARBN, the NRC and the Zest for Birds website. Pelagic rarities were excluded. The total number of rarity records provided a measure of observer effort over the study period. The ten most regularly occurring migrant rarities were assessed to examine whether records of all vagrant species were increasing due to increased observer skill and effort in the subregion. Each of the ten occurred at a rate of above one record per season. They included one migrant raptor species, Western Marsh-harrier (Circus aeruginosus) and nine migrant shorebird species, that is American Golden Plover (Pluvialis dominica),

| Southern African Bird Atlas data collection
The Southern African Bird Atlas Project (SABAP) is a collection of bird distribution data collected by citizen scientists (SABAP 2, 2018).
It is divided into two parts, SABAP 1 and SABAP 2, which use different protocols and spatial scales. SABAP 1 occurred from 1987-1991 while SABAP 2 is ongoing since 2007. SABAP 1 was collected by quarter degree grid cell (QDGC), and SABAP 2 is collected by pentad (Bonnevie, 2011). A pentad is one ninth of a QDGC and is a 15 min by 15 min grid (~9 km by 9 km). SABAP 1 had sampling periods of two to seven days in a QDGC during which all bird species heard or seen were recorded (Bonnevie, 2011). All SABAP 1 data were used. SABAP 2 has three classes of data of which we used only full protocol data. A full protocol atlas card for SABAP 2 is a minimum of two hours and a maximum of five days in a given pentad. Every bird species seen or heard during this period is recorded. Cards are then submitted, and the reporting rate of each species is calculated using the total number of species records divided by the total number of full protocol cards for each pentad. The data for European honey buzzard were downloaded by species as SABAP 1 versus SABAP 2 reporting rates at a QDGC level in August 2017.
In addition, SABAP 1 versus SABAP 2 data for Forest Buzzard (Buteo trizonatus) were also downloaded for the same period. This species appears to have similar habitat requirements to European honey buzzard as well as presenting a challenging identification to many birdwatchers. Therefore, it is the ideal species to compare with European honey buzzard in terms of range expansion in order to provide a quality check of the atlas data.

| European honey buzzard record analysis
For all records of European honey buzzard from southern Africa and East Africa, the date and location were recorded and catalogued.
Repeat records, based on date and location, were removed from analysis. The records were then plotted spatially in ArcGIS 10.3. A QDGC grid was plotted over the records, and the records were extracted to the grid to understand the distribution of honey buzzards spatially and temporally. The data were grouped seasonally from July to June to include the full non-breeding season of the European honey buzzard. Total records per season were compared for each subregion over time.
To examine the effect of observer effort on the number of southern African records, the total number of European honey buzzard records per season was divided by the total number of rarity records per season. This was then plotted to assess whether observer effort accounted changes in the number of records over time.

| Southern African rarity record analysis
Records of the other ten southern African vagrant species examined were similarly plotted over seasons (July-June), a period covering an entire Austral summer period. In addition, the annual number of records for each species as well as European honey buzzard was compared before and after SARBN using a Wilcoxon rank sum test to determine the effect of increased birdwatching effort and communication on reporting rate.
In order to assess whether trends in European honey buzzard numbers differed significantly from other vagrant species, a generalized linear mixed model for number of annual vagrant records was created. Species was the random variable. The fixed variables were year, perapi, a binomial variable indicating whether or not the species was European honey buzzard, and an interaction term between year and perapi. The model was created using the "glmer" function (with a Poisson distribution) from the package "lme4" in R. The function "confint" was used to determine 95% confidence intervals for all fixed variables. Confidence intervals that did not overlap with zero were taken to be significant.

For both European honey buzzard and Forest Buzzard in South
Africa, the number of QDGCs in each reporting rate change category (SABAP 1 only, decrease in SABAP 2, increase in SABAP 2, and SABAP 2 only) was compared using a χ 2 test to determine whether the two species showed similar changes in range and reporting rate between the two periods. Only South Africa was assessed as Forest Buzzard does not occur in the other southern African countries.  (Hansen et al., 2013). The two datasets included in our analyses were the forest cover for the year 2000 and the year of gross forest loss. Pixels (30 m by 30 m) were defined as forest if there was over 30% cover, as was defined in the forest loss layer (Hansen et al., 2013). The total area of forest for the subregion was calculated, and the amount of forest loss annually was subtracted from previous

| Global forest change data collection and analysis
year's total forest area.
Rolling correlations were created for the relationship between the annual number of honey buzzard records in southern Africa in relation to the annual number of honey buzzard records in Tanzania to assess the relationship between the two populations over time.  African rarity records that were European honey buzzards were computed in order to assess how observer effort affected the relationship. A rolling correlation is a time series analysis which uses a moving window (in this case time) to assess the correlation of two variables during that window. This was used to determine how the relationships between the two populations as well as the two populations and habitat availability may have changed over time. The rolling correlations were calculated, using the R package "roll," for seven different intervals ranging from three years (the minimum number of points for a correlation) to nine years. Different intervals were used because a species response to altered habitat, particularly that of a habitat specialists, is not usually immediate (Uezu & Metzger, 2016).
For all time intervals, a minimum of three points was required to include a correlation value. The R 2 values for each window were plotted for the last year of that window.

| RE SULTS
Nine southern African European honey buzzard specimens were found in three of the eight museums (see Supporting Information Table S1). These specimens were collected as early as 1894. Six specimens were from South Africa and three were from Zimbabwe.
Eight specimens from three museums were from East Africa. The first specimen was collected between 1895 and 1897. Six specimens were from Kenya and three were from Tanzania.  A total of 2,517 subregion rarity records of 87 species were assessed (Supporting Information Table S2). The three most common species were European honey buzzard (n = 1,094), Green Sandpiper (n = 179) and Pectoral Sandpiper (n = 96). The annual number of records has consistently increased over time with the greatest number of records documented in the 2016-2017 season (n = 433; Figure 3a). The proportion of European honey buzzard records has also increased over time with a similar pattern to the total number of honey buzzard records (Figure 3b).

| Comparison of European honey buzzard and other rare bird species in southern Africa
Of the rare species examined, all 11 showed a significant increase in southern Africa between the annual number of records before and after the introduction of SARBN ( Green Sandpiper showed a longer and less drastic increase over time than European honey buzzard over the same period as is indicated by the interaction term ( Figure 4).

| Comparison of European honey buzzard and Forest Buzzard SABAP data
The European honey buzzard had a significantly greater proportion of pentads with an increase between SABAP 1 and SABAP 2 than the Forest Buzzard (χ 2 (3) = 155.16, p < 0.01). The majority (80%) of QDGCs with European honey buzzard were new in SABAP 2 while Forest Buzzard had smaller percentage (25%) of new QDGCs (Supporting Information Figure S2).

| Correlations of European honey buzzard records in southern Africa and Tanzania
The five-year interval rolling correlation between southern African and Tanzanian records showed the strongest negative correlation during the 1980s and the strongest positive correlation during the 2010s ( Figure 5). The 1990s and 2000s showed very weak correlations. The trends were similar for rolling correlation intervals ranging from five to nine years (Supporting Information Figure S3).
Over the 16 years assessed, there was an 11.6% (46,367.0 km 2 ) loss of forest area in East Africa (Table 3) The trend was similar for intervals ranging from five to nine years (Supporting Information Figure S6).

| D ISCUSS I ON
From the records collected for European honey buzzard from multiple sources, it is unquestionable that there has been an increase in both the abundance and distribution of the species in southern Africa over the 34 years for which data were analysed. This massive increase more likely reflects an actual increase in the number of birds migrating to the subregion, rather than an artefact of increased birdwatcher effort in the area. Over the period of largest expansion in southern African honey buzzards, there was a small decline in Tanzania.

| Explanations for the population increase
Based on the evidence detailed above, the European honey buzzard has truly increased in abundance and distribution in southern Africa and this is not merely a reflection of increased reporting rate.
While other species have also increased, none have increased to the same extent and the number of honey buzzard records appears to have increased disproportionately to the overall birdwatching effort. ). This may indicate that European honey buzzards have been driven further south due to the loss of habitat in East Africa (Hansen et al., 2013). The extreme drought may have exacerbated the effect of forest loss and caused a reduction in honey buzzard prey availability (Awange et al., 2016;Hassan et al., 2014). This hypothesis is further supported by the slight decline in Tanzanian honey buzzard records over the same period as well as the small contraction of its range.
TA B L E 1 Eleven "vagrant" southern African bird species as well as European honey buzzard (Pernis apivorus) in Tanzania with the total number of accepted records, the average number of annual records (±standard error) before (August 1982-July 2007) and after (August 2007-July 2017) the Southern African Rare Bird Newsletter and the increase factor (e.g., how many times the average annual number of records has increased post-SARBN) is as a result of plasticity within individuals but plasticity within individual migrants has rarely been observed (Gill et al., 2014). It has been shown that new recruits to the population are more likely to arrive earlier, perhaps due to earlier hatching subsequently facilitating earlier migrations to and from the breeding grounds (Gill et al., 2014). These recruits are likely to be driving changes in population-level migration patterns (Gill et al., 2014). In the case of the European honey buzzard, it seems likely that the juvenile birds would be the drivers of the southwards overwintering range shift based on tracking studies of the species. Adults have high site fidelity and are unlikely to change their non-breeding grounds while juvenile birds wander widely in Africa for at least three years, presumably to find the most suitable non-breeding territory (Strandberg, Hake, Klaassen, & Alerstam, 2012). This pattern strongly supports the serial residency hypothesis which predicts that a proportion of juvenile birds will find suitable non-breeding habitat that can be used as an adult, allowing for overall population resilience to large-scale change in the species' environment (Cresswell, 2014).
This study documents the response of one migrant species, the European honey buzzard, to changes in its non-breeding grounds. We show that the apparent increase in southern Africa is a real expansion of the population in the subregion and hypothesize that it is new recruits that are driving this southwards shift.
This shows remarkable flexibility for a specialist species but may have negative, long-term consequences for the species due to decreased breeding success of late-arriving individuals. A southwards shift is in contrast to many previous studies of Palaearctic migrants which have predicted or documented a northwards shift (closer to the breeding grounds) in response to climate change (Ambrosini et al., 2011;Barbet-Massin, Walther, Thuiller, Rahbek, & Jiguet, 2009;Visser, Perdeck, Van Balen, & Both, 2009). This may be because the focus of past studies has also been on climatological factors rather than habitat variables. With a likely increase in the collection of bird distribution data across Africa due to new and repeat atlas projects in Kenya and Nigeria as well as throughout southern Africa, we will be able to better answer questions regardsing how both climate and habitat changes are affecting migrant birds on the non-breeding grounds at a finer scale. With much of the current understanding of Palaearctic migrant birds' response to global change based on European data, we emphasize the importance of better using these African distribution data resources to increase our understanding of shifts in non-breeding ranges, phenology and more for migrant bird species.

ACK N OWLED G EM ENTS
We would like to thank BirdLife South Africa for access to the archives of the National Rarity Committee and other important literature, as well as access to their membership numbers, Trevor Hardaker for the Southern African Rare Bird News and the Zest for Birds archives, the Southern African Bird Atlas Project and all of its citizen scientists for their contributions, and Neil Baker and the atlassers of the Tanzania Bird Atlas for their records. We also acknowledge Andreas Lindén for his assistance with statistics. Lastly, we recognize the National Research Foundation for funding.

DATA ACCE SS I B I LIT Y
All species records have been uploaded onto Dryad under DOI number https://doi.org/10.5061/dryad.8m36rq2.