Acoustic and archival technologies join forces: A combination tag

Technological advances are key to maximizing the information potential in electronic tagging studies. Acoustic tags inform on the location of tagged animals when they are in the range of an acoustic receiver, whereas archival tags render continuous time series of logged sensor measurements, from which trajectories can be inferred. We applied a newly developed acoustic data storage tag (ADST) on 154 animals of three fish species to investigate the potential of this combination tag. Fish trajectories were reconstructed from logged depth and temperature histories using an existing geolocation modelling approach, adapted to include a likelihood for acoustic detections. Out of 126 detected fish (accounting for over 700,000 detections) and 25 tag recoveries, eight ADSTs rendered both acoustic and archival data. These combined data could validate that the original geolocation model performed adequately in locating the fish trajectories in space. The acoustic data improved the timing of the daily position estimates. Acoustic and archival tagging technologies provided highly complementary information on fish movement patterns and could partly overcome the limitations of either technique. Furthermore, the ongoing developments to acoustically transmit summary statistics of logged data would further increase the information potential of combination tags when tracking aquatic species.


| INTRODUC TI ON
Electronic tagging enables the spatiotemporal analysis of aquatic animal movements and vastly contributes to our understanding of these animals' behavioural and spatial ecology (Brownscombe et al., 2022;Hussey et al., 2015;Lennox et al., 2017). Over the past decade, technological advances have led to tag miniaturization and longer battery life, diverse attachment methods and increased data resolution (Hussey et al., 2015). Tags have been fitted with sensors (measuring e.g. pressure, acceleration, temperature and predation) to log or transmit information on behaviour, physiology and the physical environment (Brownscombe et al., 2019). These technological advances have allowed addressing a wider range of questions on a greater diversity of species (Brownscombe et al., 2022).
Two common electronic tagging technologies for aquatic animals are acoustic telemetry and archival tags. In acoustic telemetry, a tag transmits an acoustic signal, coded with a unique ID and optionally a sensor measurement. An acoustic receiver can detect this transmitted signal when the tagged individual is within the receiver's detection range. Detection data are accessed through the receiver.
Archival tags, on the other hand, store sensor measurements at a predefined time interval in the tag memory. These tags must therefore be recovered or send their information through satellites to access the logged data. The resulting time series can provide fine-scale information on vertical movement behaviour (Heerah et al., 2017), and environmental preferences (Righton et al., 2010), and can be used to reconstruct migration trajectories with geolocation modelling (Pedersen et al., 2008;Woillez et al., 2016).
Double-tagging, that is, tagging an animal with two tags, has been used to benefit from specificities and complementarity of different tag types (Gatti et al., 2021;Strøm et al., 2017). Aside from providing complementary information on ecology and/or physiology, the combined use of distinct technologies allows us to evaluate the interpretation of one technology's results and ground-truth modelled outcomes (e.g. geolocation models, as reviewed by Gatti et al., 2021).
In addition, double-tagging enables to assess tag retention and effect (Brownscombe et al., 2019;Verhelst et al., 2022). Although limitedly studied , double-tagging comes with reasonable concern over an increased impact of the tagged fish' welfare and movement behaviour. Combining technologies in one physical tag allows us to avoid the longer handling time in a more complex procedure and the added effect of the second tag. In this study, we report on the first utilization of a novel type of electronic tag that combines the technologies of acoustic telemetry and archival tagging.

| Tag specifications
We used the acoustic data storage tag (ADST; Figure 1), developed by Innovasea Ltd., in two sizes: ADST-V9TP (diameter 13 mm, length 65 mm, weight in air 8.5 g, transmitting power output 151 dB) and ADST-V13TP (diameter 16 mm, length 75 mm, weight in air 14.2 g, transmitting power output 154 dB). The ADST was equipped with a pressure sensor (maximum depth 68 m, accuracy ±1.0 m, resolution 0.3 m) and a temperature sensor (range −5 to 35°C, accuracy ±0.5°C, resolution 0.15°C). Tags were coloured bright red and fitted with a sticker with the contact details of the principal investigator and the mentioning of a reward (€25 or a T-shirt), to increase the probability of tag recoveries. The built-in floatation enabled tags to drift ashore when they got separated from the fish (e.g. due to predation, fishing or natural death).
Sensor data were stored as continuous time series on the tag itself. Sensor information at the time of transmission was also transmitted acoustically (69 kHz, MAP114, protocol A69-9006). When selecting the transmit ratio of temperature versus pressure measurements, we favoured depth use for its information potential on vertical movement behaviour. The transmitting and logging intervals were selected in consideration of the study species, the study objectives and the trade-off with battery lifetime (Table 1, more details in Supporting Information). Tag settings had to be selected at the time of ordering the tags, as the programming of settings had to be performed by the manufacturer. Because the ADST lacked an internal clock, the time of activation of the tag (i.e. by removing a magnet) had to be registered to the second. Upon retrieval of an ADST, the physical tag was mailed to the manufacturer to download the data.

| Data management
Acoustic detections could be registered on the permanent Belgian acoustic receiver network (Reubens et al., 2019), with the detection range distance (where the probability of detecting a tagged animal within a day exceeded 0.5) averaging from 500 to 700 m . The data management was facilitated through the European Tracking Network (ETN) database (https://lifew atch.be/ etn) (Reubens et al., 2019), archiving the data and metadata for both the acoustic and logged data.

| Analysis
For the recovered tags, trajectories were reconstructed with geolocation modelling, using a hidden Markov model (HMM). The hidden state (daily fish position) was estimated with an observation model, relating sensor measurements to environmental reference fields, and a movement model, describing the time dynamics of the state sequence as a Brownian random walk model (Pedersen et al., 2008).
Full details on the geolocation approach were outlined in previous publications (de Pontual et al., 2022;Woillez et al., 2016), but we describe below how this HMM was adapted for the application on ADST data in our study area.
The reference fields of bathymetry and temperature at depth for the observation model were drawn from the 3D Dutch continental shelf model in flexible mesh, 3D DCSM-FM (Zijl et al., 2021).
Building on an existing HMM, we decided to maintain an approach with a regular grid, rather than using the original irregular grid of the 3D DCSM-FM output (Liu et al., 2017). The depth and temperature irregular grids were rasterized to a regular grid (48.8°N -53.0°N, 3.2°W -5.0°E) with the field's finest resolution of 0.5′ x 0.75′ (latitude × longitude). The original 3D DCSM-FM output for the English Channel offshore area was at a coarser resolution of 1′ × 1.5′. Pixels in this area were resampled to the values of the nearest neighbouring cell to retain the highest resolution in the main area of interest (southern North Sea). The raster fields were transformed into a metric grid of a resolution of 1 km × 1 km. The temperature likelihood was estimated using a multivariate normal probability density function at the different depth layers (0, 5, 10, 15, 20, 25, 30, 50 and 100 m). This temperature likelihood was then multiplied by the depth likelihood (de Pontual et al., 2022).
Using the acoustic detection data, we implemented a detection likelihood. This likelihood layer was calculated differently for days with and without acoustic detections. If a fish was detected, the likelihood was set to 1 for the grid cell with the receiver location and 0 for the rest of the area. For days without detections, the grid cells with active receivers were assigned a detection likelihood of zero, with the rest of the field having an equal non-null value.
For European seabass, a behavioural switch was implemented (de Pontual et al., 2022) to discern two (daily) behavioural states: low versus high activity. As the behavioural pattern segmentation used here (Heerah et al., 2017) was developed specifically for seabass, we did not apply the behavioural switch for Atlantic cod and starry smooth-hound. Hence, the diffusion coefficient D (the mean daily distance covered by a fish, in km 2 /day) of the movement model was estimated with a maximum likelihood estimation for two behavioural states for seabass and one state for the other species. From the daily posterior probability distributions of the observation and movement model combined, we calculated the most probable sequence of positions (Viterbi track).
Model performance was evaluated using the information on acoustic detections. We defined positional accuracy as the distance between the known receiver location and the trajectory as estimated by the geolocation model without including the acoustic detections (detailed explanation in Supporting Information). Track sensitivity was defined as the distance between the entire trajectories reconstructed with and without implementing the detection likelihood. To account for potential errors in the timing of the estimated track, both metrics were calculated as timed (distance to the TA B L E 1 Tag settings applied for different species. Temperature (T) and pressure (P) sensor measurements were logged continuously at a fixed interval and were transmitted at a fixed ratio (more details in Supporting Information) estimated position on the exact day) and non-timed (minimum distance to the estimated positions on all days).

| RE SULTS
Up until June 2022, 25 tags were retrieved (16.2%): four tagged seabass were caught with rods and 21 tags were found washed ashore.
Plotting the depth and temperature histories of the tags, we could visually determine that two seabass and one cod died in the week after tagging; these datasets were omitted from the analysis. Eight ADSTs provided both acoustic and archival data ( Table 2).
The complementarity of the two electronic tagging data types was visualized in Figure

| DISCUSS ION
The unique value of combination tags consisted of the possibility to understand residency and habitat use in a specific area with a receiver array, in addition to studying migration behaviour and trajectories during the period animals were not detected. As illustrated by TA B L E 2 Overview of tags resulting in both acoustic and archival data with the number of days of archived data and days detected per fish, in addition to the archived depth (m) and temperature (°C) measurements. Performance metrics were computed in km: timed (TPA) and non-timed positional accuracy (NPA), timed (TTS) and non-timed track sensitivity (NTS the seabass in the port area, the bathymetry and temperature variability of (secluded) inshore areas might not reliably be accounted for in environmental reference fields. Acoustic data were vital to recognize the fish presence in this specific habitat. The inclusion of acoustic data could thus overcome the limited performance of geolocation models in coastal areas (due to an insufficient resolution of environmental reference fields), where the deployment of acoustic arrays would be relatively convenient. The vast contribution of the archival component was illustrated in the starry smooth-hound example. A solely acoustic tag would have only informed on site fidelity and some residency in the estuary, whereas with the archival data we were able to reconstruct its southward migration trajectory.

F I G U R E 2
Examples of tagging results for an Atlantic cod (left), European seabass (middle) and starry smooth-hound (right), shown with only the acoustic detection data (top), only the archival data (middle) and the combination of both in the ADST (bottom). White dots represent the locations of the active receivers with the locations of detections in blue (square: offshore wind farm; diamond: estuarine station; circle: harbour station). Archival depth and temperature histories were plotted over time and the modelled trajectories were visualized on the map in the timeline's colouring. Combining acoustic and archival data, trajectories were estimated with the inclusion of acoustic detection data in the geolocation model.
Acoustic detections informed on presence at specific locations, whereas the archival data contributed large-scale modelled trajectories on a low resolution and fine-scale information on behaviour and temperature experience on a high resolution.
The acoustic detections enabled the validation of the geolocation model, which was shown to perform in line with expectations for demersal and pelagic fish (Gatti et al., 2021). As illustrated by the smaller distances of the non-timed performance metrics, the geolocation model would adequately position the trajectory in space but would often err in the timing of the daily position estimates along the track. Building on an assumption of Brownian motion (Pedersen et al., 2008), the movement model of the geolocation assumed a fish to move to an area of high likelihood rather gradually. The acoustic detections, however, showed that fish movement could be abrupt in distinct periods of time.
To fully benefit from this information potential, combination tags should be highly modular. We regarded the floatability option as an important asset, as we retrieved the majority of recovered tags after washing ashore. Depending on the study species, researchers might opt for pressure and temperature sensors with a different range and resolution. Since the fish' temperature experience could have been drawn from existing temperature data series in the study area, the acoustic transmission of depth use information was preferred. The ability to (re-)program transmitting and logging settings and performing the data offload of recovered tags yourself, as well as the inclusion of an internal clock, would highly increase user-friendliness.
Other acoustic and archival tags on the market do entail these fea-

CO N FLI C T O F I NTE R E S T
The authors declare that they have no conflict of interest.

PEER R E V I E W
The peer review history for this article is available at https://publo ns.com/publo n/10.1111/2041-210X.14045.

S U PP O RTI N G I N FO R M ATI O N
Additional supporting information can be found online in the Supporting Information section at the end of this article.

Supporting Information S1
Table S1 Tag settings applied for different species. Temperature (T) and pressure (P) sensor measurements were logged continuously at a fixed interval and were transmitted at a fixed ratio. Signals were transmitted at a random delay between a minimum and maximum interval for a fixed period of time.

Supporting Information S2
Table S2 Definition of geolocation model performance metrics.

Figure S1
Visual explanation of performance metrics for evaluating the geolocation model. In this situation, the tagged fish was detected at day T3 (red dot). The trajectory was reconstructed without the information of the acoustic detection (dotted line) and with using the detection likelihood (undashed line). Blue arrows indicated which distance was used to calculate each metric.
Positional accuracy was calculated as the distance between the receiver location and the daily positon estimate for the day of the detection (timed) or the closest daily position estimate of the track (non-timed). Track sensitivity was calculated as the distance between daily position estimates of the same dates (timed) and as the minimum distance between daily position estimates of all dates (non-timed).

Figure S2
Performance metrics positional accuracy and track sensitivity, timed (purple) and non-timed (red), over time for the shark example (tag SN1293308).