Skip to main content

Anthropogenic noise reduces avian feeding efficiency and increases vigilance along an urban-rural gradient regardless of species’ tolerances to urbanisation

Cite this dataset

Merrall, E. S.; Evans, K. L. (2020). Anthropogenic noise reduces avian feeding efficiency and increases vigilance along an urban-rural gradient regardless of species’ tolerances to urbanisation [Dataset]. Dryad.


Anthropogenic noise can adversely impact urban bird populations by interfering with vocal communication. Less research has addressed if anthropogenic noise masks the adventitious sounds that birds use to aid predator detection, which may lead to increased vigilance and reduced feeding efficiency. We test this hypothesis using a controlled playback experiment along an urban-rural gradient in Sheffield (UK). We also test the related predictions that anthropogenic noise has the greatest impacts on vigilance and feeding efficiency in rural populations, and on species that are more sensitive to urbanisation. We focus on six passerines, in order from most to least urbanised (based on how urbanisation influences population densities): blue tit Cyanistes caeruleus, robin Erithacus rubeculla, great tit Parus major, chaffinch Fringilla coelebs, coal tit Periparus ater and nuthatch Sitta europaea. We used play-back of anthropogenic urban noise and a control treatment at 46 feeding stations located along the urban-rural gradient. We assess impacts on willingness to visit feeders, feeding and vigilance rates. Exposure to anthropogenic noise reduced visit rates to supplementary feeding stations, reduced feeding rates and increased vigilance. Birds at more urban sites exhibit less marked treatment induced reductions in feeding rates, suggesting that urban populations may be partially habituated or adapted to noisy environments. There was no evidence, however, that more urbanised species were less sensitive to the impacts of noise on any response variable. Our results support the adventitious sound masking hypothesis. Urban noise may thus interfere with the ability of birds to detect predators, reducing their willingness to use food rich environments and increase vigilance rates resulting in reduced feeding rates. These adverse impacts may compromise the quality of otherwise suitable foraging habitats in noisy urban areas. They are likely to be widespread as they arise in a range of species including common urban birds.


The methodology used to collect these data is described in Merrall, E. S. and Evans, K. L. 2020. Anthropogenic noise reduces avian feeding efficiency and increases vig-ilance along an urban-rural gradient regardless of species’ tolerances to urbanisation. – J. Avian Biol. 2020: e02341

Site selection and urbanisation metrics

Work was conducted in and around Sheffield (53°22′N, 1°20′W), which is England’s fifth largest city, with a population of c. 575,000.  We defined urban areas as 1km x 1km grid cells with more than 25% impervious surface (following Gaston et al. 2005) and selected 16 urban sites within 3 km of the city centre, 19 urban sites that were more than 3 km from the city centre and 11 rural sites that were between 1 and 3 kms away from the city’s outer limits (using the above definition of urban areas). This approach enabled us to haphazardly select sites that were spread across the urban to rural gradient (see below for quantification), with the fewest sites in rural areas due to their greater homogeneity in background noise levels. All sites within the urban area were public parks and green-spaces with some woody vegetation cover, and all rural sites were woodland (rather than the alternatives of moorland or farmland) to maximise the similarity of the focal bird community along the urban to rural gradient. 

Urbanisation intensity was quantified at each site using the ‘Urbanisation Index’ software developed by Seress et al., 2014 (available at: This software uses a semi-automated method where it takes a 1km2 area from google maps around the coordinates of each location, and then uses manually inputted training points to score each image for vegetation cover, forest, buildings and paved roads using 100x100m2 cells. It then uses principle component analysis (PCA) to calculate an urbanisation intensity score for each area.

Experimental design

Field work took place between the 12th February and 3nd April 2019. Our general approach was to assess avian feeding and vigilance rates at feeding stations under three experimental conditions: play back of urban noise and two control treatments (play back of natural noise and a silent control without playback).  The urban recording was created by splicing together 5-minute sound recordings from each of four locations in central urban Sheffield using a Zoom H4n sound recorder and Cubase LE AI Elements 10, with a mixture of traffic, pedestrian and construction noise. The urban noise treatment was played at approximately 80 decibels (dbc), i.e. the typical volume of anthropogenic noise in busy urban areas during the day (Maryland SHA, 2018). The natural control used a mixture of songs of summer migrants (barn swallow Hirundo rustica; whinchat Saxicola rubetra; redstart Phoenicurus phoenicurus and common whitethroat Sylvia communis). Recordings were obtained from Xeno-canto (2005-2019) and were vetted to ensure that they did not include alarm calls to ensure that this treatment did not include vocalisations to which birds using the feeders were likely to respond. The natural sound control treatment was played at approximately 40 decibels (dbc) that matched natural sound levels of avian vocalisations. The urban treatment lasted 40 minutes while each control lasted 20 minutes, with a 10-minute habituation period between equipment set-up and beginning playback, and between the urban treatment and the two control treatments (which each lasted 20 minutes).  We used two green, portable SONY SRS-XB10 Bluetooth controlled speakers (IPX rated 5) at each site located approximately a metre from the feeding station in a spatial configuration that created a surround sound effect.

Each site’s feeding station was set up four to seven days before conducting the experiment to enable birds to habituate to the presence of the food source. Each station consisted of two standardised hanging feeders, each with two feeding ports, filled with sunflower hearts. These are a nutritious food source that has a negligible handling time and is thus widely used by a wide range of species. To reduce the risk of disturbance from grey squirrels Sciurus carolinensis, which can discourage birds from using feeders (Bonnington et al. 2014), poles were greased and sunflower hearts were coated in chilli powder (which squirrels avoid, whilst birds exhibit no adverse response). Feeders were placed in relatively open locations to allow easy observation, but close to vegetation cover to encourage birds to approach and use the feeders. Feeding stations were located away from footpaths to minimise effects of human disturbance.

Treatments were applied in a randomised order at each site (with the three treatments being applied sequentially). Sites were visited in a haphazard manner with regard to the site’s urbanisation score. All data collection took place at least 1.5 hours after civil dawn and before civil dusk, to avoid spikes in bird activity early and late in the day. Data were not collected when it was snowing or raining (except occasional light drizzle), or at high wind speeds as such conditions interfered with activity levels and noise transmission.

Each treatment was filmed using a Panasonic (HC-X920) HD Camcorder and observed from an approximately 10m to 15m distance. Videos recorded birds feeding on the feeder and those feeding on spilt food beneath the feeder. If disturbance events occurred during the treatment, such as a human or dog passing close to the feeder, and interrupted birds’ feeding behaviour, data collection was paused until birds resumed normal activity. Videos were analysed, and for each visit we recorded the species, visit duration (seconds), number of pecks (as a measure of feeding rate) and the amount of time (seconds) spent performing vigilance behaviour, defined following Quinn et al. (2006), as when the bird raised its head and scanned. For each site, the temperature (oC) and wind speed (kn) were also recorded using data from the nearest weather observation site (Met Office, 2019). The seven weather stations used ranged from 220 m to 6.1 km away from the study site.

Statistical analysis

All statistical analyses were conducted using R Studio (RStudio Team, 2016). Three response metrics were calculated from the videos for each species per treatment per site: i) visit rate, i.e. the rate at which the species visited the feeder per hour of treatment, ii) the peck rate per hour of treatment, and iii) the vigilance rate (in minutes) per hour of time spent on the feeder. A total of 19 species were observed using the feeders or feeding on fallen seed on the ground below, of which six occurred at ten or more sites and were included in data analysis (great tit Parus major 46 sites; blue tit Cyanistes caeruleus 44 sites; coal tit Periparus ater 33 sites; Eurasian robin Erithacus rubeculla 35 sites; nuthatch Sitta europaea 14 sites; and chaffinch Fringilla coelebs 13 sites). These species vary substantially in their ability to maintain high population densities in urban environments. This is indicated by their urbanisation scores as calculated by Evans et al. (2011), i.e. the ratio of urban to rural population densities obtained from Breeding Bird Survey data from approximately 3,000 randomly selected 1 km x 1 km squares located across the UK. These scores, from most to least urbanised species, are: blue tit 1.46; robin 0.99; great tit 0.74; chaffinch 0.25; coal tit 0.23, nuthatch 0.17. Matched paired t-tests demonstrated that each of these six species’ visit, peck and vigilance rates did not differ between the two forms of control (P ranges from  0.110 to 0.877) and these data were thus merged to form a single control treatment. 

We modelled visit, peck and vigilance rates using mixed effect models (lme4 package; Bates et al, 2015). These models pool data across species and include the following main effects as predictors: species (fixed effect), treatment (fixed effect), the site’s urbanisation intensity, date (number of days from the 1st of January), time of day (hours since sunrise), temperature (°C), wind-speed (kn), and site (random effect). We also included the following interaction terms: species*treatment*urbanisation intensity (to test the predictions that treatment impacts vary with species and along the urbanisation gradient); species*treatment (to test the prediction that species vary in their responses to treatment); treatment*urbanisation intensity (to test the prediction that, regardless of species identity, urban populations are less sensitive to treatment effects); and species*urbanisation intensity (to take into account the potential that species’ vary in how their visit, peck and vigilance rates change along the urbanisation gradient, although such patterns are not associated with our objectives of assessing impacts of urban noise and are not explored further). A significant species*treatment interaction term would indicate that species varied in their responses to the urban noise treatment. Consideration of the correlations between each species’ parameter estimates from this interaction term and their urbanisation scores (from Evans et al. 2011; see above) enable us to assess if species with greater tolerance to urban development are less sensitive to treatment effects than species that are more impacted by urban development.

We start by building a full model that includes all main effects, the three-way interaction term and the three two-way interaction terms. We simplify this full model by removing non-significant interaction terms in a step-wise manner according to their P values, with significance being determined using the car package (Fox and Weisberg, 2019) to calculate Wald chi-square statistics (Type III). We use this approach as the alternative of using information theoretic approaches based on criteria such as change in Aikaike Information Criteria (AIC) is not recommended when models include interaction terms (Cade, 2015).



Bates, D., Maechler, M., Bolker, B., Walker, S. 2015. Fitting Linear Mixed-Effects Models Using lme4. J. Stat. Software, 67: 1-48.

Bonnington, C., Gaston, K. J. and Evans, K. L. 2014. Assessing the potential for Grey Squirrels Sciurus carolinensis to compete with birds at supplementary feeding stations. Ibis, 156: 220-226.

Cade, B.S., 2015. Model averaging and muddled multimodel inferences. Ecology., 96: 2370-2382.

Evans, K. L., Chamberlain, D. E., Hatchwell B. J., Gregory, R. D.  and Gaston, K. J. 2011. What makes an urban bird?

Glob. Change Biol. 17: 32-44.

Gaston, K. J., Warren, P. H., Thompson, K. and Smith, R. M., 2005. Urban domestic gardens (IV): the extent of the resource and its associated features – Biodivers. Conserv., 14: 3327-3349.

Fox, J., Weisberg, S., 2019. An R Companion to Applied Regression, Third edition. Sage, Thousand Oaks CA.Gaston, K. J., Warren, P. H., Thompson, K. and Smith, R. M., 2005. Urban domestic gardens (IV): the extent of the resource and its associated features. Biodivers. Conserv., 14: 3327-3349.

Maryland SHA. 2018. Sound Barriers Guidelines – Highway Traffic Noise [online]. State Highway Administration [Accessed 15 November 2018], Maryland Department of Transportation. Available at

Met Office. 2019. Weather Observations Website (WOW) [online] WOW Met Office [accessed 14 February 2019]. Available from:

Quinn, J. J. Whittingham, M. J. Butler, S. and Cresswell, W. 2006. Noise, predation risk compensation and vigilance in the chaffinch Fringilla coelebs.  J. Avian Biol. 37: 601-608.

RStudio Team. 2016. RStudio: Integrated Development for R. RStudio, Inc., Boston, MA URL

Seress, G., Lipovits, Á., Bókony, V. and Czúni, L., 2014. Quantifying the urban gradient: a practical method for broad measurements. Landscape. Urb. Plan., 131: 42-50.

Xeno-canto, 2005-2019. Sharing bird sounds from around the world [online]. Xeno-canto Foundation [Viewed 6 February 2019]. Available from:

Usage notes

Both files contain metadata sheets that explain each variable.