Assessing the impacts of livestock grazing on upland bird breeding territories using drone surveys
Data files
Dec 10, 2024 version files 201.89 KB
-
Combined_Plot.csv
3.28 KB
-
Combined_Territories.csv
16.02 KB
-
Combined_VegPoints.csv
177.22 KB
-
README.md
5.37 KB
Abstract
Ground nesting birds are sensitive to habitat structure, so understanding this relationship is fundamental to managing habitat to maintain or enhance bird populations. We used an existing long-term, large-scale experiment with routine monitoring of meadow pipit territories to assess the capability of drone-based remote sensing as a means of capturing relevant habitat information. Normalised Difference Vegetation Index (NDVI) captured differences in stocking density between treatments and autumn measured NDVI was well correlated to field measurements of vegetation height and density. Spring and autumn NDVI were negatively correlated due to dominant tussock-forming species dying back over winter. Meadow pipit apparent territory size was positively correlated to autumn NDVI and to the difference between autumn and spring NDVI. Apparent territory size was larger where there was more tussocky vegetation that comprise the areas least preferred for foraging. The long-lived nature of tussocks likely constrains the effect of the grazing treatments on meadow pipit breeding apparent territory size. Drone-based remote sensing of habitat characteristics appears to be a powerful way forward to understand bird-habitat associations.
README: Assessing the impacts of livestock grazing on upland bird breeding territories using drone surveys
https://doi.org/10.5061/dryad.x69p8czv3
Description of the data and file structure
Files and variables
File: Combined_Plot.csv
Description:
Variables
- Block: Experimental block (A to F)
- Treatment: Coded treatment (I High 9 sheep per plot, 2.7 sheep/ha; II Low sheep 3 per plot, 0.9 sheep/ha, III Mixed sheep and cattle, grazing offtake equivalent to II, IV no livestock grazing)
- Plot_ID: Combination of Block and Treatment codes.
- SPR_NDVI_MEAN: Mean NDVI from spring image at the plot level. Unitless, scale -1 to 1.
- SPR_NDVI_STD: Standard deviation of NDVI across the plot in spring. Unitless, scale -1 to 1.
- AUT_NDVI_MEAN: Mean NDVI from autumn image at the plot level. Unitless, scale -1 to 1.
- AUT_NDVI_STD: Standard deviation of NDVI across the plot in autumn. Unitless, scale -1 to 1.
- 2020PinFrameHeight: Mean height of first contact of pin drops from the 2020 detailed vegetation sampling. Unit: centimetres (cm).
- 2020StdDevHeight: Standard deviation of height of first contact across the plot. Unit: centimetres (cm).
- 2021_Structure_MEANDensity: Mean density recorded during autumn structure measurements across the plot. Unit: centimetres (cm).
- 2021_Structure_STDDensity: Standard deviation of density recorded during autumn structure measurements. Unit: centimetres (cm).
- 2021_Structure_MEANHeight: Mean height recorded during autumn structure measurements across the plot. Unit: centimetres (cm).
- 2021_Structure_STDHeight: Standard deviation of height recorded during autumn structure measurements. Unit: centimetres (cm).
File: Combined_Territories.csv
Description:
Variables
- Territory: Territory label
- SPR_NDVI_MEAN: Mean NDVI from spring image for each territory. Unitless, scale -1 to 1.
- SPR_NDVI_STD: Standard deviation of NDVI across each territory in spring. Unitless, scale -1 to 1.
- AUT_NDVI_MEAN: Mean NDVI from autumn image for each territory. Unitless, scale -1 to 1.
- AUT_NDVI_STD: Standard deviation of NDVI across each territory in spring. Unitless, scale -1 to 1.
- Allocation: Plot allocation for each territory. NA Territory mainly outside the experimental area, so not allocated to plot.
- Block: Experimental block (A to F). NA Territory mainly outside the experimental area, so not allocated to a block.
- Trt: Coded treatment (I High 9 sheep per plot, 2.7 sheep/ha; II Low sheep 3 per plot, 0.9 sheep/ha, III Mixed sheep and cattle, grazing offtake equivalent to II, IV no livestock grazing). NA Territory mainly outside the experimental area, so not allocated to a treatment.
- Stocking: Sheep stocking rate (sheep/ha) equivalent. NA Territory mainly outside the experimental area, so not allocated to a grazing level.
- NDVI_DIFF: Difference between mean spring and mean autumn NDVI for each territory. Unitless, scale -1 to 1.
File: Combined_VegPoints.csv
Description:
Variables
- ID: Code for each recording point from the autumn structure survey. A-F = block, I-IV = treatment, 1-81 point number.
- SPR_NDVI_MEAN: Mean NDVI from spring image for circle of 2 m radius around each point. Unitless, scale -1 to 1.
- SPR_NDVI_STD: Standard deviation of NDVI from spring image for circle of 2 m radius around each point. Unitless, scale -1 to 1.
- AUT_NDVI_MEAN: Mean NDVI from autumn image for circle of 2 m radius around each point. Unitless, scale -1 to 1.
- AUT_NDVI_STD: Standard deviation of NDVI from autumn image for circle of 2 m radius around each point. Unitless, scale -1 to 1.
- Block: Experimental block (A-F)
- Plot: Coded treatment (I High 9 sheep per plot, 2.7 sheep/ha; II Low sheep 3 per plot, 0.9 sheep/ha, III Mixed sheep and cattle, grazing offtake equivalent to II, IV no livestock grazing)
- Point: Structure recording points within each plot (1 to 81)
- 2021StructureMEANDensity: Mean density recorded during autumn structure measurements at the point. Unit: centimetres (cm).
- 2021StrcutureSTDDensity: Standard deviation of density recorded during autumn structure measurements at the point. Unit: centimetres (cm).
- 2021StructureMEANHeight: Mean height recorded during autumn structure measurements at the point. Unit: centimetres (cm).
- 2021StructureSTDHeight: Standard deviation of height recorded during autumn structure measurements at the point. Unit: centimetres (cm).
- Comm: National Vegetation Classification community at each point: U5 Nardus stricta–G. saxatile grassland (Nardus grassland), M15 Scirpus cespitosus (now Trichophorum cespitosum)–Erica tetralix wet heath), the mire communities M6 Carex echinata–Sphagnum recurvum/auriculatum (Carex mire), M25 Molinia caerulea–Potentilla erecta (together with small areas of M23 Juncus effusus/acutiflorus-Galium palustre called Molinia mire in this paper), U4 Festuca ovina–Agrostis capillaris–Galium saxatile grassland (Agrostis-Festuca grassland), and U20 Pteridium aquilinum–G. saxatile (bracken). NA Could not be allocated to a community.
Code/software
Files are .csv files
Access information
Other publicly accessible locations of the data:
- N/A
Data was derived from the following sources:
- N/A
Methods
Study system
The long-term Glen Finglas grazing experiment (56°16’N, 4°24’W) was established in 2002 on a patchwork of upland, unimproved plant communities including upland wet heathland (according to the UK National Vegetation Classification, Rodwell 1991, 1992) mainly (in order of area) U5 Nardus stricta–G. saxatile grassland (Nardus grassland), M15 Scirpus cespitosus (now Trichophorum cespitosum)–Erica tetralix wet heath), the mire communities M6 Carex echinata–Sphagnum recurvum/auriculatum (Carex mire), M25 Molinia caerulea–Potentilla erecta (together with small areas of M23 Juncus effusus/acutiflorus-Galium palustre called Molinia mire in this paper), U4 Festuca ovina–Agrostis capillaris–Galium saxatile grassland (Agrostis-Festuca grassland), and U20 Pteridium aquilinum–G. saxatile (bracken). The climate is cool and humid (mean temperatures for January and July were 3.04 and 14.12°C, respectively, for 1981–2010 and the mean annual rainfall 2,230 mm (Perry and Hollis, 2005), and the soils are humus-iron podzols with peaty gleys (Pakeman et al. 2019). Shrubs have recently spread in plots in one block (see below) and affected the composition of the bird community (Malm et al. 2020).
The experiment used a randomised block design made up of six blocks (A-F) in three groups of two, with each block formed of four plots, each plot being 3.3 ha. Plots were fenced in the winter of 2002-03 and the plots in each block were allocated randomly to one of four grazing treatments in spring 2003. The treatments are High - a tripling of sheep numbers to 2.7 ewe ha-1, Continued - grazing at the same low intensity to the management prior to the establishment of the experiment at three ewes per plot (0.9 ewe ha–1) to act as the control for the experiment, Mixed – a partial substitution of sheep by cattle (0.6 sheep ha-1) and two cows each with a single, suckling calf grazed for four 4 weeks in autumn (September-October) to give the same year-round grazing intensity as the Continued treatment, and None - no livestock present to represent agricultural abandonment. Sheep were removed during the winter (November to April) and for short periods for routine farming operations. All experimental blocks were mainly surrounded by similar mosaics of the vegetation communities represented in the experiment and managed similarly to the Continued treatment.
Vegetation height data have been collected in two ways. Firstly, as part of detailed vegetation composition recording 25 fixed points were established per plot on the intersections of a square grid at 40 m intervals (Pakeman et al. 2019). At each fixed point, a vertical pin-frame of five pins, orientated consistently (1 m east) from the fixed point, was used to measure the cover of individual plant species and litter and the height of first contact (total 125 pins per plot). Sampling has been carried out every three years, including in 2020 when strict movement restrictions were in place due to Covid-19, in late July/early August prior to cattle being introduced to the plots. Secondly, an annual estimate of vegetation height and density was taken in late August/early September at 81 points per plot (Evans et al. 2015), the 25 fixed points plus intermediates on a 20 m grid. Three measures of maximum vegetation height were made at each point using a stick marked at 5-cm intervals (at arm’s length, to the front and either side of the observer). The visibility of white marks at 5-cm intervals on the vertically held stick was recorded to measure vegetation density.
Bird census
Within each plot, meadow pipit territories were mapped between April and June each year (since 2002) using an adaptation of the standard Breeding Bird Survey methodology (BBS, Heywood et al. 2023). In contrast to the BBS methodology, transects were spaced at much shorter distances, generally 25-45 m depending on the size and shape of each block, and were walked more or less parallel to the outer boundaries of each block. Bird activity was recorded on A3 field maps using the Common Bird Census symbology (Gilbert et al., 1998) in a single band which stretched as far as reliable identification was possible, typically within 100 m, but generally took place within the block being walked, paying particular attention to bird breeding behaviour, such as song flight, alarm calls, food or faecal sac carrying, territorial encounters and using nest locations where detected (Dennis et al. 2005).
Detailed work at the start of the experiment found that observed pipit breeding behaviour centres around nest locations (Evans et al. 2005) and that they stay close to the nest when foraging (~75% observations within 60m; Douglas et al. 2008), with average distances reducing to 29.3 m ±2.89 (1SE) in treatment (i) when provisioning nestlings (Vandenberghe et al. 2009). Thus, we are confident that observed activity is indicative of a territory even if the nest location has not been discovered. Efforts are made to avoid double-recording, given the short distances between adjacent transects. All bird species are recorded but meadow pipit being by far the most abundant species at the study site, the latter only was used in the present analyses.
Bird surveys in 2021 were carried out in blocks A/B on 28th April, 5th, 11th, 25th May and 1st June; in blocks C/D on 29th April, 7th, 13th, 27th May and 2nd June; and in blocks E/F on 26th April, 2nd, 9th, 24th and 30th May. Surveys took place only in suitable weather conditions, i.e., not in winds exceeding 8 m s-1 and not in heavy rain. During the initial weeks, the higher areas of the plots could still be covered in snow; such days were equally avoided or surveyed later in the morning after the snow had melted. Following the completion of fieldwork, the registrations from all visits to a block were transcribed onto territory maps and grouped into apparent territories on the basis of clusters taking note especially of birds that had been recorded simultaneously and likely associated with different apparent territories as well as territorial encounters. We use the phrase “apparent territory” as they are based on a limited set of observations and subsequent interpretation. We concede that pipits occasionally use areas outside of these apparent territories and that territorial boundaries may be fluid. However, our adapted BBS methodology is suitable for determining territory size for our purposes as the frequency of visits is higher and spacing of transects is closer, allowing a more detailed mapping of territories across the experimental plots. Furthermore, birds were all surveyed and identified by the same team, with territory maps finalised by the same observer (DME) since 2002, and the geo-spatial spread of observations used to delineate the apparent territory area. Thus, we are confident that comparing territories based on the spread of breeding bird observations is robust within the context of this analysis. The transcribed maps from 2021 showing apparent territories were scanned and saved as jpegs. These were then georeferenced in ArcGIS Pro 3.3 (ESRI, Redlands, USA) and subsequently digitised.
NDVI data capture
On the 1st, 2nd and 22nd of April 2021, and again on the 2nd and 8th of September 2021 aerial imaging surveys were conducted at the experimental site. The flight planning software Ground Station Pro 2.0.17 (DJI, Shenzhen, Guangdong, China) was used to demarcate each experimental site and conditions for the survey set as follows: height above ground level: 120 m, resulting in a ground sample distance of c. 7 cm (approximate, as terrain height varied within survey areas); image overlap: 80% forward, 70% side overlap. The unmanned aerial vehicle (UAV) deployed was a Phantom 4 Multispectral RTK (DJI, Shenzhen, Guangdong, China) linked to RTK (real-time kinematic positioning) base station during data collection, which carries a payload of one RGB and five monochromatic sensors (Filters: Blue (B): 450 nm ± 16 nm; Green (G): 560 nm ± 16 nm; Red (R): 650 nm ± 16 nm; Red edge (RE): 730 nm ± 16 nm; Near-infrared (NIR): 840 nm ± 26 nm) and has an upward facing irradiance sensor for radiometric normalisation purposes. Flight lines were adjusted to be perpendicular to the angle of the sun (course angle option in Ground Station Pro). Pix4Dmapper, Education Licence (Pix4D, SA, United States) was used to create Digital Surface models (DSM) prior to orthomosaic generation (structure from motion algorithms). Pix4Dmapper was also used to generate reflectance maps (incorporating the irradiance data) from the monochromatic images, that were then used to calculate and create NDVI maps, which were exported as GEOTIFF files. The 120 m flight altitude was more than twice the threshold (50 m) where disturbance affects have been detected on nesting birds (Cantu de Leija et al. 2023). Since 2008 there has been little between year variability in vegetation height, though heights have slowly increased (Pakeman et al. 2019). Consequently, the NDVI survey should be representative of adjacent years.
Data processing
All data processing was completed using ArcGIS software version 10.8.1 and all spatial datasets were projected using the WGS1984_UTM_Zone_30N coordinate system to match that of the drone imagery. Each meadow pipit apparent territory was assigned a unique ID and the area of each apparent territory calculated. Subsequently, the tabulate intersection tool was used to calculate the area of each territory within each of the experimental plots. As some territories crossed the boundaries between experimental plots, the percentage of each territory contained in each plot was calculated using the ArcGIS field calculator. The number of territories in each plot were calculated using the frequency function.
Spatial layers containing the experiment plots, apparent meadow pipit territories and the locations of the 81 vegetation sampling points in each plot were overlayed on the NDVI maps produced from the drone imagery. The zonal statistics function in ArcGIS was used to produce tables containing the mean NDVI and standard deviation at the plot and territory scale as well as within a 2m radius of each of the vegetation sampling points in each plot. This was performed using the Spring and Autumn NDVI maps.
Vegetation height and density values from the 2021 annual structural measurements and height measurements from the 2020 pin frame survey were later joined to the NDVI tables at the respective scale of plot, meadow pipit apparent territory and vegetation sample point. Where meadow pipit apparent territories crossed the boundary between plots, they were assigned the plot which contained more than half the apparent territory.