Livestock grazing impacts components of the breeding productivity of a common upland insectivorous passerine: Results from a long‐term experiment

Citation for pulished version (APA): Malm, L. E., Pearce-Higgins, J. W., Littlewood, N. A., Karley, A. J., Karaszewska, E., Jaques, R., Pakeman, R. J., Redpath, S. M., & Evans, D. M. (2020). Livestock grazing impacts components of the breeding productivity of a common upland insectivorous passerine: Results from a long-term experiment. Journal of Applied Ecology, 57(8), 1514-1523. https://doi.org/10.1111/1365-2664.13647


| INTRODUC TI ON
One third of farmland in the European Union (EU) consists of permanent grasslands. However, the proportion of livestock fed through natural grazing is decreasing in the majority of European countries, and in many countries outside Europe (van den Pol-van Dasselaar, de Vliegher, Hennessy, Isselstein, & Peyraud, 2015). As a result, more polarized management (i.e. intensification or abandonment) is anticipated in traditionally pastoral landscapes, of which many are of High Nature Value (Meiner & Bas, 2017). After decades of concerns about unsustainably high levels of grazing (Fuller & Gough, 1999), new concerns regarding under-grazing are emerging (IEEP, 2004). Fuller, Gillings, Lauder, and Crowe (2013) found that bird species in upland habitats in Britain and Ireland have shown the strongest range contractions, during a 40-year period, compared to birds in other habitat types. Some population declines occurred alongside increases in livestock densities since the mid-20th century (Fuller & Gough, 1999). However, many species, particularly ground-nesting birds (Sullivan, Newson, & Pearce-Higgins, 2015), are still declining in abundance, despite lower sheep densities in some parts of the British uplands in recent years (Hayhow et al., 2017).
Livestock grazing intensity can affect ground-nesting, insectivorous birds through a number of mechanisms. Firstly, livestock may have a direct impact on demographic parameters, for example by trampling or predating nests and chicks (Jarrett, Calladine, Wernham, & Wilson, 2017). Secondly, livestock may alter vegetation structure by their effect on sward height and density. This can not only alter the suitability of the habitat for different species, thus affecting bird settlement patterns (Loe et al., 2007), but may also alter the abundance and/or availability of their prey (Buchanan, Grant, Sanderson, & Pearce-Higgins, 2006;Dennis et al., 2008). Food availability is a function of both prey abundance and accessibility. For example, ground-foraging, insectivorous birds have been shown to prefer shorter vegetation with high arthropod abundance and accessibility, rather than simply high arthropod abundance (Douglas, Evans, & Redpath, 2008;Pearce-Higgins & Yalden, 2003). Thirdly, changes in vegetation structure may also affect the visibility and hence vulnerability of nests to predators (Homberger, Duplain, Jenny, & Jenni, 2017). Moreover, predators may increase as a result of improved habitat suitability and/or higher population densities of other prey, such as small mammals (Evans et al., 2015). In the longer term, changes in grazing pressure, or complete removal of livestock, may alter the composition of vegetation, particularly the ratio of shrubs to sedges and grasses (Fuller & Gough, 1999), with further impacts on the abundance of bird species , although detailed studies are lacking.
To gain a mechanistic understanding of how livestock grazing pressure affects upland birds, replicated experiments that manipulate stocking densities are necessary, but rare. Experimentally managed sheep grazing in Norway resulted in a higher abundance of birds with increasing sheep density (Loe et al., 2007) while Johnson, Kennedy, and Etterson (2012) found that breeding success of two ground-nesting passerines in the United States did not vary with cattle grazing pressure. However, both studies investigated shortterm bird responses in the first few years after the experiments had commenced. Land management change can take several decades, or longer, to reach their full effects on plant composition (Pakeman, Fielding, Everts, & Littlewood, 2019) and to have potential knock-on effects on other taxa across trophic levels. Yet to our knowledge, no experimentally managed grazing study has examined the long-term effects of grazing management on avian breeding success, largely due to the logistical and financial constraints of maintaining such experiments at large spatial scales.
Here, we use a replicated landscape-scale grazing experiment (the Glen Finglas experiment; Evans, Redpath, Evans, Elston, & Dennis, 2005) with 13 years of continuous manipulation to study the impacts of four livestock grazing treatments (i.e. sheep at two different stocking densities, mixed cattle and sheep grazing and no grazing) on meadow pipit Anthus pratensis, the most common passerine in the British uplands. Previous work using the Glen Finglas experiment has documented short-term effects on pipit breeding density (Evans et al., 2006(Evans et al., , 2015 and egg size , which were both enhanced under low intensity, mixed cattle and sheep grazing. Furthermore, the pipit offspring sex ratio was 5. Synthesis and applications. Livestock grazing management can have different outcomes for different upland birds. Our results showed that, with time, meadow pipit breeding productivity tended to be higher when sheep grazing intensity was reduced and/or mixed with cattle, and lower when livestock were removed, but not significantly so. Removal of grazing, however, can significantly increase bird species richness. The long-term experiment showed an overall decline in fledglings regardless of grazing treatments, potentially a result of increased predator numbers harboured by nearby developing woodland, highlighting the importance of considering wider landscape processes in grazing management decisions.

K E Y W O R D S
agriculture, avian biology, grasslands, meadow pipit, moorland, nest survival, predation, temporal change biased towards more male nestlings in plots with low-intensity sheep or mixed livestock grazing (Prior, Evans, Redpath, Thirgood, & Monaghan, 2011), while arthropod abundance and plant biomass increased with decreasing livestock densities (Dennis et al., 2008;Evans et al., 2015). However, there was no significant effect of grazing pressure on nest survival during the early years of the experiment . Here, after more than a decade of continuous grazing, we predict that long-term effects of grazing management will have significant effects on nest survival and fledgling output, unlike results from early in the experiment.
In particular, if meadow pipit breeding productivity is a function of vegetation structure and arthropod prey availability, we hypothesize that the effect of grazing pressure is enhanced in the later stage of the experiment, with lowest productivity in both the intensively grazed and ungrazed treatments. We investigate how grazing pressure and management duration affect the following Finally, although principally designed to understand the mechanisms by which grazing impacts pipit breeding productivity, we use the Glen Finglas experiment to investigate whether there are any long-term changes in the overall breeding bird community. were set up to both have low-intensity grazing pressures with the same annual vegetation biomass offtake and to maintain stocking at similar rates to those pre-experiment.

| Bird surveys, nest monitoring and vegetation sampling
The meadow pipit breeds in a range of grassland types and builds concealed nests on the ground. Incubation and nestling development each take approximately 13 days before chicks are ready to leave the nest.
Breeding bird surveys were carried out in 2003-2004 and 2015-2016 to study immediate and long-term effects of livestock grazing treatments on meadow pipit breeding density and output. Breeding territories were determined by mapping all breeding activity of meadow pipit and other bird species during six surveys of each plot from late April to early July, with at least 3 days between each survey. Following , a territory was defined as a cluster with at least two independent observations of breeding behaviour (i.e. singing, alarming, food carrying or encounters of active nests) and territories could be separated from adjacent ones by simultaneous observations of singing males. Each year, territories were assigned to plots by the same method and person (DME) on maps of accumulated observations from surveys. On the small number of occasions when territories spanned two or more plots (9.9% of all territories), these were assigned to the plot where the majority of observations occurred. With standardized sampling effort, nests were located by searching through plots every 2-5 days, depending on weather conditions, through the whole breeding season. Nests were found by flushing incubating or brooding females while walking or rope dragging, and occasionally by observing birds arriving at the nests.
Once found, each nest was checked every 3 days (weather permitting) while active, through each stage of the nest period (i.e. egg laying to hatching and hatching to fledging, hereafter referred to as the egg and nestling stages). The meadow pipit lays one egg per day until a clutch of two to 5 eggs is completed. Partial predation was not observed in active nests, so the clutch size recorded when no additional eggs were found on following visits was therefore not likely to have been altered by predators. Partial mortality did occur (in 29% of nests), but unhatched eggs were then found in the nest and dead nestlings were found in or just outside the nest. A nestling was considered as successfully fledged when recorded alive just before fledging, unless it was found dead on the post-fledge visit on day 15-17 after hatching. The number of territories per plot was used as a measure for breeding density, but it was not possible to successfully find every nest within the plots. We therefore calculated the

| Statistical analyses
All meadow pipit breeding parameters (see below) were analysed using GLMMs. Nests found after hatching were not used in proportional survival analyses for the incubation stage or total numbers of fledglings per nest. Within the models, treatment (categorical factor of the four grazing treatments) and sampling period (categories early or late) were the primary factors of interest, with the latter indicating either early (2003)(2004) or late stage (2015-2016) effects of the treatments. A significant interaction between treatment and sampling period would indicate that grazing effects had changed between the two sampling periods. A significant effect of treatment across both sampling periods would suggest that any effect of different grazing pressure was already apparent at the early stage of the experiment. An effect of sampling period across all treatments would indicate that changes affecting the whole study area occurred between the two sampling periods, and therefore were unlikely to be related to grazing treatments.
The GLMMs for meadow pipit breeding density and the total number of breeding bird species per plot were tested with treatment and sampling period both as an interaction and as separate fixed effects. For meadow pipit breeding density, the number of territories of other bird species were also included as a fixed effect to control for potential disturbance/competition within the plots. The GLMM for Julian hatch date also contained grazing treatment and sampling period as an interaction and separate fixed effects, and the number of meadow pipit territories as a fixed numerical effect. The GLMMs for clutch size, number of fledglings, egg-and nestling-stage nest survival and overall survival all had the same fixed effects: treatment; sampling period; the interaction of treatment and sampling period; Julian hatch date; Julian hatch date 2 and number of meadow pipit territories per plot. The observed and estimated number of fledglings per plot were analysed similarly, but excluded the number of territories, which is directly linked to the estimate of fledglings per plot. Julian hatch date and Julian hatch date 2 (both numerical) were included to account for a linear and quadratic effect, respectively, of seasonal variation. The number of territories was included to control for potential competition or positive effects by conspecifics. Fixed effects in all models were tested with likelihood ratio tests and removed sequentially if making the model worse in terms of model convergence and AIC score. All tests had the random effects: block (one of three experiment areas); replicate within block (which also takes into account altitude); plot; and calendar year, in line with Gelman and Hill (2006). Nest ID was also included in nest survival models as an observation level random effect (OLRE) to control for over-dispersion (Harrison, 2014). As the observed number of fledglings per plot may be affected by a potential difference between treatments in probability of finding nests, we first tested the effect of treatment on the ratio of nests found to territories per plot in a binomial GLMM. Details on selected models and probability distributions applied can be found in Table S2.
We further tested whether treatment effects on breeding productivity changed when taking initial vegetation communities into account, and whether potential effects of grazing treatment on productivity were mediated through a change in vegetation height.
The models for breeding density and fledgling output per nest were therefore run with and without vegetation variables to determine whether the grazing treatment effects were affected by the initial vegetation communities and/or if they were solely due to vegetation changes. To characterize the initial vegetation communities in each plot, cover data from 2002 were subject to detrended correspondence analysis (DCA; Hill & Gauch, 1980)  Since traditional r 2 values are not applicable to GLMMs we calculated marginal and conditional r 2 values, which provide estimates of variance explained by fixed effects only and by both fixed and mixed effects respectively (Nakagawa & Schielzeth, 2013). All models and graphs were analysed/produced in R version 3.5.2 (R Core Team, 2018). GLMMs were conducted using package lme4 (Bates, Maechler, Bolker, & Walker, 2015). Post hoc tests for pairwise comparisons were carried out with package lsmeans (Lenth, 2016) and p-values of the comparisons were adjusted with the Holm-Bonferroni method (Holm, 1979).

| RE SULTS
Across the four breeding seasons, a total of 295 meadow pipit nests were found of which 268 were monitored until breeding outcome was confirmed (Table S1).

| Breeding density
The meadow pipit breeding density was significantly affected by grazing treatment across all years (n = 96, χ 2 = 15.59, p = 0.001) with lowest density in Low (M ± SD = 2.88 ± 0.9 territories/plot) and highest density in High (3.96 ± 1.37 territories/plot) and Mixed (3.92 ± 1.25 territories/ plot). There was no interaction between grazing treatment and sampling period ( Figure 1; Table 1) but there was a significantly higher breeding density in the early sampling period (3.81 ± 1.38 territories/plot) than the later one (3.27 ± 0.96 territories/plot; χ 2 = 4.09, p = 0.043). Including vegetation variables in the model did not change the significance/nonsignificance of treatment and sampling period, but there was a significant effect of the initial vegetation community composition through the first DCA axis (χ 2 = 4.87, p = 0.027; Table S2). Higher breeding density was associated with lower scores on the first DCA axis, plots with a higher abundance of mire species such as Myrica gale and Narthecium ossifragum. Higher scores on axis 1 where instead plots with a higher representation of acid grassland species such as Agrotis capillaris and Anthoxanthum odoratum (see Table S4; Figure S5 for all species).

| Clutch size and hatch date
There was no significant effect of grazing treatment or the interaction of grazing treatment and sampling period on clutch size (Table 1)  there were no significant changes in treatment effects between sampling periods (Figure 2; Table 1).

| Fledgling output and nest survival
The number of fledglings per nest was highest in Low and Mixed treatments but there were no statistically significant effects of grazing treatment across all years, nor by treatment-sampling period interaction (n = 268; Figure 3a; Table 1) or in the model including vegetation variables (Table S2). There was no significant effect of treatment on the probability of detecting nests (Table S5) Table S2).
Nest survival was highest in Low and Mixed treatments (Figure 4) but there was no significant difference in overall nest survival be-  Table 1).
The nestling-stage nest survival was neither significantly affected by grazing treatment nor by the interaction of grazing treatment and sampling period (n = 213; Figure 4; Table 1 (33%, compared to 17% in the early sampling period). Other nests were abandoned or did not hatch/died for unknown reasons.

| Breeding bird species richness
There was a significant interaction of treatment and sampling period on the number of bird species per plot (n = 96, χ 2 = 12.10, p < 0.007). This was significantly higher in Ungrazed (2.33 ± 1.44 species/plot) than Mixed (1.33 ± 0.65 species/plot) in the later sampling period, while no significant differences between treatments were seen in the early sampling period ( Figure 5; Table 1; Table S3). Seven bird species, including black grouse Tetrao tetrix and common cuckoo Cuculus canorus, were only observed breeding in the later period, and mainly in Ungrazed (Table 2).

| D ISCUSS I ON
We provide the first experimental results of the long-term effects of livestock grazing intensity on the breeding performance of a common upland insectivorous passerine, as well as changes in the overall bird community. After a 12-to 13-year period, meadow pipit breeding density was significantly lower in the Low treatment. Conversely, the highest rates of egg-stage failure occurred in High and Ungrazed plots, where overall nest survival also tended to be lower, although not significantly so. There were no significant changes in grazing treatment effects over time but, across the experiment, both the egg-stage and overall nest survival declined with time. These results suggest that other processes at the wider landscape scale, such as changes in surrounding habitat and predator densities, are important for breeding meadow pipits and that these may be more apparent than long-term effects of variation in grazing treatment.

F I G U R E 5
Willow warbler Phylloscoupus trochilus

| Changes in treatment effects between sampling periods
Given the association of meadow pipits with habitat mosaics (Douglas et al., 2008;, we hypothesized that positive effects of the low-intensity treatments (Low and  (Tables 1;   Table S2). Therefore, other factors such as weather and predation pressure unrelated to grazing treatments may contribute to a larger proportion of the variation in breeding success (see 'Long-term regional changes' below), but see

| Overall treatment effects
The highest meadow pipit breeding density was found in the High and Mixed treatments. This supports results from a landscapescale sheep grazing study by Loe et al. (2007) in Norway, which found both meadow pipit and total bird abundances to be highest in intensively grazed plots compared to low intensity or ungrazed plots, at least in the short term. At Glen Finglas, the initial vegetation community composition, but not current vegetation height, had an effect on breeding density. Previous results from the early period of the experiment suggested that meadow pipit breeding density is mainly driven by availability of arthropods that are common in upland bird diets (see Buchanan et al., 2006), where vegetation heterogeneity is high (Evans et al., 2015). Evans et al. (2015) also showed that vegetation height heterogeneity was greater in  (Misenhelter & Rotenberry, 2000). At Glen Finglas, previous studies found that activity indices of foxes Vulpes vulpes were highest in Ungrazed plots and declined as a result of increasing grazing pressure (Villar, Lambin, Evans, Pakeman, & Redpath, 2013). The higher nest failure rate in High plots could instead be explained by an increased exposure to predators through lower vegetation biomass (Baines, 1990), or even by predation by sheep (Jarrett et al., 2017).

| Long-term regional changes
Several breeding parameters were affected by sampling period across all grazing treatments with smaller clutch sizes, lower overall nest survival, (near significantly) fewer fledglings per nest and lower egg-stage nest survival in the later sampling period (Table 1).
Considering that most nests failed due to predation, the change in nest survival is likely caused by a regional change in predation pressure, and meadow pipit breeding success (but not local breeding density) has been observed to increase under experimental predator removal (Fletcher, Aebischer, Baines, Foster, & Hoodless, 2010). The area of native woodland on the estate on which this study is located has increased during the course of the experiment. This could contribute to the higher nest predation in the later period by providing increased cover for predators (Söderström, Pärt, & Rydén, 1998), but more research is necessary.
The growing interest in natural woodlands, afforestation and rewilding will drive the need to find ways for such management to work effectively in parallel with traditional land use such as grazing (Pettorelli et al., 2018).

| CON CLUS IONS
Our results show that longer term effects of grazing intensity can affect the breeding density and egg-stage nest survival of the meadow pipit, with lowest survival in High or Ungrazed plots, but that over 12-13 years this has little effect on overall nest survival or number of fledglings produced. Treatment effects on fledgling output were not significantly stronger in the later period of the experiment.
Instead, there was lower nest survival in the late compared to early sampling period, mainly caused by predation across all grazing treatments. Grazing exclusion was associated with an increase in bird species richness in the later stage of the experiment, probably due to a gradual change in vegetation structure and composition. Further studies disentangling the effects of regional predator abundances and local management on both predator numbers and predator behaviour would be needed to identify the causes of observed predation pressure on breeding birds.

DATA AVA I L A B I L I T Y S TAT E M E N T
Data available from the Dryad Digital Repository https://doi.org/ 10.5061/dryad.9zw3r 22bd (Malm et al., 2020).