Pygmy blue whale telemetry and biologging data

Full description

As is commonly done for many animals where direct observation of foraging is difficult, foraging areas of pygmy blue whales have previously been defined from analysis of data obtained from satellite tracking tags deployed on pygmy blue whales. These studies (Double et al. 2014, Moller et al. 2020 and Thums et al. 2022) documented pygmy blue whale migration between Australia and Indonesia and identified areas where whales spent the most time and where they had movement behaviour (as calculated from surface location estimates from the tags) thought to be indicative of foraging (slower speed and lots of turns), such as the Bonny upwelling region of South Australia, Perth Canyon, NW Cape region and the Banda Sea. Although there is a theoretical basis for assigning putative foraging behaviour along the horizontal movement paths of animals based on such analyses of location estimates from satellite tags (Kareiva & Odell 1987, Zollner & Lima 1999), foraging actually occurs while the animals are diving and so having data from the vertical dimension is important for validation of foraging areas defined using these methods.


To characterise and track vertical movement behaviours and validate inferred foraging behaviour from movement models, eight pygmy blue whales were double tagged by AIMS and CWR (in partnership with Woodside) in the Perth Canyon, Western Australia with Fastloc GPS tags (Wildlife Computers LIMPET; Low Impact Minimally Percutaneous Electronic Transmitters, type: SPLASH10-F-333) and pop-up satellite-linked archival tags (PSATs) (Wildlife Computers MiniPAT) in 2021 – 2022, and another double tagged at Ningaloo in 2023. The GPS tags provided location estimates and the PSATs provided depth and summarised accelerometery time-series (Mobility and activity-time series, ATS) data from the whales for up to 40 d duration, encompassing presumed foraging grounds and migration areas. Four PSATs were recovered, providing access to the high-resolution (1 s sample rate of depth and Mobility) data archive on board the tag. For the remaining five tags, the summarised time-series data (75 s sample rate of depth and ATS) transmitted through the Argos satellite network (Argos), was used for the analysis. Given the difficulty of recovering tags from long (> a few days) deployments, both recovered and transmitted datasets were used to determine whether lower temporal resolution depth and accelerometry data transmitted via Argos (compared to the archived data on board recovered tags) can provide sufficient detail to characterise pygmy blue whale diving behaviour, especially foraging and feeding. Diving behaviour was characterised using a supervised Random Forests dive behaviour classification function to determine where and when pygmy blue whales forage. The locations where foraging and lunge feeding dives occurred was compared to areas of putative foraging inferred from a movement model (State-space model) and to important foraging areas previously defined from spatial analyses based on horizontal movement data only (Thums et al. 2022).


Transmitted depth time series (75s) was adequate for identifying foraging dives, but accelerometry metrics were key (error increased to 18% without it) to distinguishing lunge feeding dives from foraging dives without lunges.


Foraging and lunge feeding dives occurred in three main foraging areas: 1) Centred at the head of the Perth Canyon, extending from offshore of Cape Naturaliste to offshore of Jurien Bay, 2) offshore of Geraldton and the Abrolhos Islands and 3) offshore of Ningaloo, extending from approximately Coral Bay up to offshore of approximately the Montebello Islands (~19 °S). Foraging/ feeding was also detected in the Savu Sea (~8 °S), offshore of Bremer Bay and far off the shelf of the Kimberley region of Western Australia while migrating (~15 °S, ~120 °E).


Despite a weak temporal relationship between putative (inferred from a movement model) and actual foraging, there was generally good spatial overlap detected, but predominantly in high use areas with lower use and more opportunistic foraging areas being less likely to be detected by the model. More opportunistic foraging occurred off north-west Australia where foraging dives were shallower, horizontal travel rates faster, and there was an absence of a diurnal pattern in diving. This suggests a reliance on more ephemeral prey than off south-west Australia where whales have high residency.


Our test of movement models to define foraging areas is extremely useful given its common usage in ecology and our spatial delineation of foraging areas assists with conservation management.

Lineage

Maintenance and Update Frequency: biannually

Statement: Seven pygmy blue whales were double tagged (6 in the Perth Canyon, Western Australia and 1 at Ningaloo) with Fastloc GPS tags (Wildlife Computers LIMPET; Low Impact Minimally Percutaneous Electronic Transmitters, type: SPLASH10-F-333) and pop-up satellite-linked archival tags (PSATs) (Wildlife Computers MiniPAT) in 2021 – 2023. Two other whale were tagged with PSATs only. The LIMPET tags provided location estimates obtained from the GPS reciever on board the tag in addition to location estimates obtined through the Argos satellite network using the Doppler effect. The PSATs provided depth and summarised accelerometery time-series (Mobility and activity-time series, ATS) data from the whales for up to 40 d duration, encompassing presumed foraging grounds and migration areas. Four PSATs were recovered, providing access to the high-resolution (1 s sample rate of depth and Mobility) data archive on board the tag. For the remaining five tags, the summarised time-series data (75 s sample rate of depth and ATS) transmitted through the Argos satellite network (Argos), was used for the analysis. For whales where PSATs were deployed without LIMPETs (n = 2), the movement paths of the whales from the PSATs were determined using the Wildlife Computers geolocation processing software that makes two location estimates per day (at dawn and dusk). The software (GPE3) uses observations of twilight, temperature, and depth along with corresponding reference data on sea surface temperature and bathymetry to determine an animal’s trajectory. To account for location error and standardise the location estimates at set intervals (calculated from the average number of raw location estimates per day), a correlated random walk model was applied to all the location estimates received from the LIMPETs including Argos (location classes 3, 2, 1, 0, A, and B with estimated error of <250 m, 250 – 500 m, 500 – 1500 m, >1500 m, and unknown, respectively) and Fastloc GPS using the R (R Core Team 2022) package foieGras (Jonsen et al. 2020). Then, a move persistence model (mpm) (Jonsen et al. 2020) was applied to provide an objective behavioural index (g) along the track. The index, known as move persistence, is a continuum ranging between 0 (decrease in speed and directionality = low move persistence) and 1 (increase in speed and directionality = high move persistence). Segments of relatively low move persistence are generally indicative of foraging, but could also represent resting and/or breeding (Bailey et al. 2009), while segments of relatively high move persistence are related to migration or transit behaviour (Jonsen et al. 2019). Move persistence was also summarised into a binary measure, based on the threshold move persistence (g) of 0.8 developed by Thums et al. 2022. Putative foraging was inferred where satellite location points along the track had g<0.8 and transit/migration was inferred for those points with g ≥ 0.8 as Thums et al. (2022). The depth time–series were analysed in R (R Development Core Team 2022) using the library diveMove (Luque 2007) to define dives. After zero offset correction, dives were identified in the time-series according to a minimum depth threshold (we used 1.5 m) and a minimum dive duration (10s). Dive phases (descent, bottom period, and ascent) and summary statistics were calculated for each dive (Luque 2007). A suite of dive statistics were calculated for each dive in addition to those supplied by diveMove. We also calculated the number of feeding lunges and the number of wiggles in each dataset, using custom algorithms to detect spikes in Mobility and ATS as lunges and large undulations in depth as wiggles. These combined dive statistics (these and those produced by diveMove) were then used to develop the dive classification functions (see below). We developed a supervised diving classification. Here a training dataset must be provided, where the dive classes have been specified a priori on a subset of the data to train a classification function. The prior classifications were obtained by two scientists independently visualising each dive, classifying it as one of seven types which were agreed a-priori. The seven dive types were decided based on previously described dive types for pygmy blue whales (Davenport et al. 2022), eastern North Pacific blue whales (Oleson et al. 2007), fin whales (Fonseca et al. 2022) and other diving vertebrates (e.g. Shreer et al, 2001). Two training datasets were made; one for each of the transmitted and recovered data. The dives where the classification from both scientists agreed formed the training datasets where each dive type was subset to the sample size of the least common dive type for each whale. We used the Random Forests model (Breiman 2001) to develop the classification functions using the R package randomForest (Liaw and Wiener, 2002). The dive summary statistics calculated (see above) were collated to use as predictors in the classification functions. We used the Random Forests model to classify all dives in the record for the recovered dataset. The Random Forests error rate was high for foraging and lunge feeding dives in the transmitted dataset and in contrast, the error rates of the human classifiers were equal or better, thus we used the visual classifications for the transmitted dataset (the classifications of one of the scientists). The classified dives were merged (by timestamp) with the location estimates and plotted spatially and temporally to determine where and when important behaviours (such as foraging) occurred. The location estimates received from the LIMPETs had to be interpolated at the sampling interval of the PSAT data to facilitate the merge. In many cases the LIMPET duration was longer than the PSAT duration, thus we could only geolocate the dives for which we had corresponding location data and the 3D satellite tracks we present are then necessarily cropped to the duration of the diving data (PSAT deployment duration).

Notes

Credit
Thums M, Ferreira LC, Davenport A, Jenner M, Möller L, Russell G, McCauley RD, Jenner C (2025) Tracking pygmy blue whale diving behaviour and validation of foraging areas defined from horizontal movement data. Global Ecology and Conservation 57:e03362. This research was co-funded by Woodside Energy Ltd, with additional support provided for one of the field trips by the North West Shoals to Shore Research Programme supported by Santos.