The endangered Rice’s whale (Balaenoptera ricei) is endemic to the Gulf of Mexico (GOM) and has a small population of fewer than 100 individuals. Little is known about their foraging ecology. Suction cup tags were attached to two Rice’s whales to collect information on their diving kinematics and foraging behavior. The tagged whales primarily exhibited lunge-feeding near the sea bottom and to a lesser extent in the water column and at the sea surface. During foraging dives, the whales typically circled the prey before executing one or two feeding lunges. Longer duration dives and dives with more feeding lunges were followed by an increase in the breathing rate. The average lunge rate of both animals was much lower than expected based on comparative research on other lunge-feeding baleen whales, which may be related to differences in prey or inter-specific differences, or may be an indication of different foraging conditions. Overall, these data show that Rice’s whale foraging behavior differs from other lunge-feeding rorqual species and may be a significant factor in shaping our understanding of their foraging ecology. Efforts to mitigate threats to the Rice’s whale will benefit from improved understanding of patterns in habitat use and fine-scale ecology within the population.

Field methods

Rice’s whales were tagged with a suction-cup-attached Acousonde tag (Greeneridge Sciences, Inc.) using the NOAA Ship Gordon Gunter to search for whales and the vessel’s 7-m Rigid Hull Inflatable Boat (RHIB) R3 to approach the free-ranging animals. The Acousonde tag was attached to the whale using a hand-held pole deployment method. After tag attachment, the whales were tracked by observers on the Gordon Gunter when possible, during daylight hours. A VHF receiver on the tag allowed it to be detected when it was above the water surface and provided direction for recovery after the tag detached from the whale. Milky Way was tagged in 2015 between 20 September 09:44 and 23 September 01:10 (CDT) at 29.261^o N and 86.268^o W, and in 2018, Edna was tagged on 3 July 17:16 at 28.757^o N and 85.705^o W with the tag remaining attached until 4 July 18:25 (CDT). The water depth at the location of tagging was determined post hoc from online bathymetry data³⁵. 

The Acousonde (Model B003B) tag includes temperature and pressure sensors, triaxial magnetometers and accelerometers, and a hydrophone. The data sampling differed between the two tag deployments. In 2015, all non-acoustic sensors were sampled at 5 Hz. In 2018, the sampling rates were: temperature at 5 Hz, pressure at 10 Hz, accelerometer at 800 Hz, and magnetometer at 40 Hz. Acoustic data were sampled at 9110 Hz during both deployments. For consistency in analyses, the accelerometer and magnetometer data from 2018 were down-sampled to 10 Hz. 

Tag data analysis

The accelerometer, magnetometer, and pressure data were calibrated and corrected for changes in tag placement using the tagtools package (animaltags.org) in Matlab (MATLAB 2016b, Mathworks, Natick, MA). Pitch and roll were calculated from the accelerometer data, and heading was calculated from the combination of the accelerometer and magnetometer data. The tag depth sensor was temperature corrected using periods when the whale was at the surface before and after deep dives when the tag was temperature equilibrating. The correction was performed using the fix_pressure function from the tagtools package. Dives were detected automatically using the find_dives function from the tagtools package, with a minimum dive depth set at 10 m.

The International Geomagnetic Reference Field (IGRF-13) was used to estimate the orientation of the earth's magnetic field at the location of the tag (2015: declination = -3.4986°; inclination = 58.2491°; 2018: declination = -3.1925°; inclination = 58.4539° ). A transformation applied to the tag magnetic data (x', y', z') yielded the cardinal directions north, east, and toward the earth’s center (x, y, z):

x' y' z' sinθcos∅ sinθsin∅ -cos∅ -sin∅ cos∅ 0 cosθcos∅ cosθsin∅ sin∅ =x y z

where theta is the earth's field declination and a phi = 90 degrees minus the earth's field inclination.  

Tag data were manually analyzed using the Matlab-based Triton software package³⁶ (version 1.0 2021 09 21, https://github.com/MarineBioAcousticsRC/Triton) with a customized add-on “remora” software module, MTViewer (inspired by Burgess MT Viewer, www.acousounde.com), that synchronizes displays of data from acoustic and kinematic (pressure, orientation) sensors and includes a tool for annotating events. The following events were manually annotated and occurrence times were extracted for further analysis: breaths, depth of neutral buoyancy, circling, foraging lunge, descent end, and ascent start. Breaths were identified by both minima in depth and the broadband sound of exhalation/inhalation on the hydrophone. We subdivided breaths in inhalation and exhalation, and only inhalation times were used in further analysis. 

We used a proxy for estimating neutral buoyancy by identifying the depth during both descent and ascent at which fluking stopped and gliding downward or upward (respectively) began. Neutral buoyancy was then calculated as the average of those points. Circling was identified by a >180° change in heading. Foraging lunges were identified as the point at which a sharp increase in speed occurred, followed by a rapid deceleration. Lunges typically also contained large changes in pitch and roll, which helped to identify them. Descent end and ascent start were the points between which the dive leveled to a relatively narrow depth range, near the maximum depth of the dive. Following annotation, we extracted timing information for each event for further analyses using custom Matlab-based routines. Events were assigned to day or night periods based on the time of "civil" twilight, when the sun was at 6^o below the horizon (http://users.softlab.ntua.gr). 

Estimation of energy expenditure

To get an estimate of the minimum energy required by Rice’s whale to sustain their foraging behavior, we calculated the energy expended as work done during foraging using the minimum specific acceleration of the animals. We estimated mass as an average body mass of 6000 kg, from a generally accepted length-to-weight conversion for baleen whales, using species-specific constants for the closely related Bryde’s whales (a = 0.012965, b = 2.74)³⁰. Length was calculated as the average length (9.2 m) from documented Rice’s whales¹. 

Acceleration was calculated as the minimum specific acceleration based on acceleration measured by the tag accelerometer. Measured acceleration is comprised of acceleration due to propulsion, body rotation, and gravity, where the acceleration due to propulsion by the animal is known as the specific acceleration. The minimum acceleration that is due to specific acceleration, the minimum specific acceleration³⁷ is calculated by combining the data streams from the magnetometer and accelerometer.  The magnetometer only measures body rotation, which makes it possible to separate the acceleration component that is due to body rotation from the component that is due to propulsion. We estimated velocity as the whale’s swimming speed from the tag jiggle (high-frequency vibrations of the tag on the animal that increase with speed), which was regressed against the tags’ accelerometers’ Orientation-Corrected Depth Rate (OCDR), following Cade et al.³⁸. Because the speed estimation was most reliable with a sampling rate of >5 Hz, speed was calculated for 2018 first, against which the speed values of 2015 were checked to see if they were reliable.

Statistics

To compare dive characteristics between individuals, we tested whether the measured variables were normally distributed using a Shapiro-Wilk test of normality. As none of the variables were normally distributed, we used a Wilcoxon rank test to compare measurements between individuals. We investigated the effect of activity (descent, circle, lunge, ascent) and individual on power and energy expenditure with linear models. For power, we used a Gaussian-distributed model with log transformation of power to correct for non-linearity in the data, while for energy expenditure, we used a generalized linear model (GLM) with a Gamma distribution. We investigated the influence of energy expenditure and individual on breathing rate using a GLM with quasi-Poisson distribution. Additionally, we tested for the differences in dive depth of lunges preceded by a circle vs lunges that were not preceded by a circle using a linear model. All models were checked against violations of the model assumptions with model diagnostics, and final models were obtained through dredging (MuMIn package), an automated way of selecting the most parsimonious model that corrects for order in variable selection. All statistical tests were performed in R 4.1.2 (R Core Team 2022).