The six datasets collectively contain 6.0 million GPS locations and 164 million accelerometer measurements of 202 individuals of the nominate subspecies of the Eurasian oystercatcher Haematopus ostralegus ostralegus, collected using UvA-BiTS (Fig. 1).

Figure 1. 

Collecting GPS data of Eurasian oystercatchers using the UvA-BiTS system. a mistnets at a high tide roost site on Vlieland, with which birds were trapped b Eurasian oystercatcher equipped with colour-rings and UvA-BiTS GPS tracker c installation of a relay station via which data from trackers could be retrieved d a high tide roost site of Eurasian oystercatchers, showing a relay station in the background.

The datasets contain data from breeding and wintering individuals. A total of 98 breeding individuals were tagged on the Dutch Wadden islands, specifically on the saltmarshes of Schiermonnikoog (O_SCHIERMONNIKOOG), the polder meadows on Ameland (O_AMELAND) and on sandflats on Vlieland (O_VLIELAND). A total of 104 wintering individuals were tagged in the Dutch Wadden Sea at Balgzand (O_BALGZAND) and on sandflats on Vlieland (O_VLIELAND). Inland populations were studied in the Dutch city Assen (O_ASSEN) and in urban and agricultural areas near Antwerp in Belgium (O_WESTERSCHELDE) (Table 1; Fig. 2a). The dataset O_BALGZAND also contains one bird (animal-id: 5331220) that was captured on the nest on the saltmarsh of Schiermonnikoog. The number of tracked individuals per dataset can be found in Table 1. Since many oystercatchers migrate between their breeding and wintering sites, the data coverage extends beyond the sites where birds were tagged. Specifically, the breeding sites of tagged individuals ranged from Antwerp to Scandinavia and Russia, and wintering sites spanned from northern France to the Dutch Wadden Sea (Fig. 2b).

Figure 2. 

Maps of GPS positions collected from the six different datasets a map of the Netherlands and bordering areas of Belgium and Germany showing locations of study sites (indicated by bird symbols) and GPS locations b map of northwest Europe showing the full extent of the GPS locations. Maps show GPS locations with hourly intervals; higher frequencies in between GPS locations are omitted in this visualisation.

The datasets collectively cover a time period from 2008 until 2021 (Table 1, Fig. 3).

Figure 3. 

Number of GPS records per dataset per year.

Oystercatchers were trapped either in summer on the nest or in winter on their feeding grounds and roost sites. Oystercatchers in breeding populations were always adults that were caught on the nest using walk-in cages. Oystercatchers in wintering populations were caught using mistnets at night, either at low tide (O_BALGZAND) or at high tide (O_VLIELAND). The age of captured birds in winter was classified as either juvenile (1^st winter), subadult (2^nd winter) or adult (>2^nd winter) based on morphology (Cramp et al. 1983). At study sites on Schiermonnikoog and Vlieland, the sex of most birds was determined by DNA analysis of a small blood sample taken from the wing vein and, if available, sex is included in the datasets. Biometrics of trapped birds were taken and included in the datasets, including wing length, tarsus-toe length, bill length, bill tip height, bill tip width (all in mm) and bill tip shape (B = chisel-shaped, H = blunt or hammer-shaped, P = pointed, combined letters indicate intermediate bill tip shapes; van de Pol et al. 2009). Body mass of trapped birds was measured at all study sites and provided for all birds. All birds were equipped with colour rings and with an UvA-BiTS GPS-tracker (Bouten et al. 2013), attached on the back with a harness of Teflon tape that looped around the neck and wings.

A total of 202 individuals were equipped with a GPS tracker (Table 1), but for several individuals no data were registered, possibly due to the tracker malfunctioning. A minimum of 100 GPS records are available for 172 birds. Information on the end of a tracking session is included in the datasets when malfunctioning trackers were removed or when birds were found dead. In the project O_ASSEN, on one bird (animal-id: 5515867) a malfunctioning tracker was replaced by a new tracker.

The UvA-BiTS (Bouten et al. 2013) trackers used for these studies were all solar powered and had a weight of 18.0 g (O_SCHIERMONNIKOOG), 15.0 g (O_SCHIERMONNIKOOG, O_AMELAND, O_BALGZAND) or 13.5 g (O_SCHIERMONNIKOOG, O_VLIELAND, O_ASSEN, O_WESTERSCHELDE). The trackers record 3D GPS positions and include a tri-axial accelerometer that measures surge X, sway Y and heave Z. Accelerometer measurements were collected in samples of up to 10 s with a frequency of 20 Hz (i.e., a 2.00 s sample consists of 40 consecutive accelerometer measurements; Bouten et al. 2013).

All data collected by the GPS trackers were stored in the internal memory. The data were transmitted remotely to a base station, sometimes via in-between relay stations (Fig. 1c–d). A network of base and relay stations was set up around nesting sites (during the breeding season) or covering high tide roosts (during the non-breeding season). Mobile base stations were occasionally used to download data of birds that were found by colour-ring sightings and resided outside the station network. Data that were downloaded were automatically removed from trackers, thereby freeing up storage for new data. Due to the design of the GPS tracking system, no data could be downloaded from birds that left the study areas with station networks and never returned, except when birds were located based on colour-ring sightings for mobile download, or when trackers were retrieved from dead birds and data was subsequently downloaded from the tracker.

The settings of the GPS trackers, i.e., the intervals between successive GPS fixes, intervals between successive accelerometer samples and length of accelerometer samples, were flexible and could be changed anytime a GPS tracker connected to a base station. Accelerometer samples could follow directly upon a GPS fix or be taken in between GPS fixes. Different settings were used in different seasons and projects. In general, more data were collected when the memory of GPS trackers was empty, i.e., when birds resided within the area covered by receiving stations and data were frequently transmitted to base stations, and when the battery of the GPS trackers was fully charged, i.e., in summer when there is more sunlight. In winter, the battery of the trackers often drained, pausing data collection and consequently, there were data gaps for many birds each winter from November to January. When trackers were collecting data, GPS fixes were recorded at least once per hour, and often at higher frequencies (i.e., every 5, 10 or 15 mins). Sometimes, GPS trackers were set to record bursts with high frequency GPS fixes (i.e., every 16 s) for one or two hours per day during daytime. A total of 6.0 million GPS fixes were collected between 2008 and 2021 (Fig. 3). Data received by the base stations were automatically extracted, post-processed, and stored in a central PostgreSQL database which is part of UvA-BiTS, and accessible to participating researchers only.

The accelerometer samples (i.e., a burst of consecutive accelerometer measurements) can be used to derive movement and behaviour. Typically, behaviour was classified based on summary characteristics (e.g., mean X, standard deviation of Z, etc.) of the accelerometer samples, using a machine learning program that was calibrated with a training dataset. Within these projects, training datasets were acquired by annotating accelerometer samples based on detailed field observations (Shamoun-Baranes et al. 2012) or based on videos that were taken from birds with GPS trackers (van der Kolk et al. 2020a). For O_VLIELAND, a Random Forest model was trained to distinguish five behaviours (flying, walking, foraging, preening, and inactive) and had a prediction accuracy of 94.6% (van der Kolk et al. 2020a). A total of 7.0 million behavioural classifications based on the random forest model were included in the dataset O_VLIELAND, enabling the study of individual variation in behaviour and time budgets (Fig. 4). Note that the classification models were based on annotated behavioural data obtained in intertidal areas mainly in the non-breeding season, and that some behaviours are therefore not distinguished (e.g., no territorial display behaviour was included and incubating behaviour was grouped with inactive behaviour).

Figure 4. 

Example of how behavioural classifications included in dataset O_VLIELAND can be used to study time budgets. Time budgets are shown for five individuals in September 2018, which is the non-breeding season when birds were present in intertidal areas on or near Vlieland. The time when birds are foraging shifts every day by 0.5-1.0 hours, as the moments of low tide shifts with the lunar cycle.

GPS fixes that were likely incorrect (i.e., outliers) were marked in two ways: manually by the researcher in the UvA-BiTS database (indicated as TRUE in manually-marked-outlier) and automatically (in https://github.com/inbo/bird-tracking) before uploading to Movebank for GPS-fixes with speed above 45 ms^-1 or GPS-fixes with an angle below 30° and speed above 15 ms^-1 (indicated as TRUE in import-marked-outlier). The outlier count for each study is provided in Table 1. The rationale for these criteria is that migrating oystercatchers can travel at speeds up to around 30 ms^-1 but then move in a more or less straight direction (translating into a large angle). Outliers are typically characterised by high speed and a sharp angle (i.e., a movement towards the outlier is followed by a movement back to the original location). These criteria target mostly the largest outliers, and depending on the goal of data use, stricter filtering criteria may need to be applied.

To make the data openly available, all data were uploaded to Movebank (https://www.movebank.org), an online platform and database specialized in storing animal tracking and bio-logging data. The Movebank data model enables the description of animals, tags, deployments, detections, and other measurements recorded by or derived from animal-borne sensors, such as acceleration data (Kays et al. 2022). For the six datasets, reference data containing information about the animals, tags and deployments, as well as GPS and acceleration data were downloaded from the UvA-BiTS database and transformed to the Movebank data format (Movebank 2021) using SQL queries and R scripts (https://github.com/inbo/bird-tracking). This guarantees a consistent approach for all datasets and allows for repeating the process when new data become available for active studies. These data (and for O_VLIELAND additional behavioural classifications) were then uploaded to the Movebank database, with one study-identifier for each dataset (Table 1), documented with metadata and made available under an open Creative Commons Zero waiver.

To enable long-term and low-tech data preservation, data were also deposited as CSV files on the research repository Zenodo (https://zenodo.org). GPS, acceleration and behavioural data were split into separate files per year and compressed, making it easier to download data in manageable chunks. A datapackage.json file was included for each deposit, making it a Frictionless Data Package (https://specs.frictionlessdata.io/data-package/), a simple container format for tabular data. This file references all CSV files, organizes them into resources (reference-data, gps, acceleration, and biometric-measurements) and describes each of their fields, including data type, format and definition according to the Movebank Attribute Dictionary (Movebank 2021). The datapackage.json file also facilitates programmatic access to the dataset, such as with the R package ‘frictionless’ (Desmet and Oldoni 2022). Each deposited version on Zenodo is assigned a DOI upon publication, as well as a versionless DOI that always points to the latest version of the deposit (see Table 1 for their Zenodo identifiers).

Movement data can be used as general-purpose occurrence data. To enable wider discoverability and use, we reformatted our datasets to incorporate them in the Global Biodiversity Information Facility (GBIF, https://www.gbif.org) and the Ocean Biodiversity Information System (OBIS, https://obis.org). Reference and GPS data (excluding outliers; including fields informing on accuracy, e.g., coordinateUncertaintyInMeters) were transformed to Darwin Core (Wieczorek et al. 2012), and down-sampled to the first record per hour, to not needlessly flood GBIF and OBIS with high-frequency movement data. Metadata were transformed to the Ecological Metadata Language (EML) and included the same authors, keywords and DOI, and explained that data are down-sampled. The transformation process to Darwin Core and EML was automated with the custom developed R package ‘movepub’ (Desmet 2022). This automated approach includes fields that are not available for the datasets described in this study (organismName and reproductiveCondition), but could be included for future datasets that are transferred using this process. The datasets on Zenodo, GBIF and OBIS cross reference each other as well as the datasets on Movebank.