New Statistical Methods Reveal Associations Between Loggerhead Turtles and Oceanic Eddies

PIFSC scientist Don Kobayashi, in the Center's Ecosystem and Oceanography Division, collaborated with staff in the Protected Species Division and colleagues in Taiwan and Japan to develop a new statistical technique to examine use of oceanic habitat features by loggerhead sea turtles in the North Pacific.

The new method was developed in a study of loggerhead turtles (Caretta caretta) in waters off Taiwan, where this species of sea turtle is commonly captured as bycatch in the coastal pond net fishery. The fishery is quite large, involving over one hundred individual moored nets along the coastline. Accordingly, the fishery is capable of exerting a substantial negative impact on the loggerhead turtles. Taiwan is not known to have any loggerhead nesting sites, and genetic analysis indicates that these turtles are likely from the Japan nesting stock. This North Pacific stock is a vital component of the worldwide loggerhead turtle population, hence any additional sources of mortality on the stock need to be carefully examined, monitored, and minimized.

The purpose of the collaborative study was to analyze the movement patterns of loggerhead turtles taken as pond net fishery bycatch in Taiwan by way of satellite tags attached to loggerheads caught in the fishery and released. The Argos geolocation data transmitted by the tags via satellite were used to infer the tagged turtles' patterns of habitat use in relation to regional oceanographic features such as large oceanic eddies.

Thirty five non-reproductive loggerhead turtles were tagged with satellite transmitters after being captured as bycatch in the coastal pond net fishery off the Pacific coast of Taiwan. Captures occurred from 2002-2008 and the turtles ranged in size from 64-92 cm SCL (69-95 cm CCL). Several different types of tags were used over the course of the study: Telonics ST-18 (used on 10 turtles), ST-14 (2 turtles), ST-20 (14 turtles), ST-24 (3 turtles), and Wildlife Computers SPLASH tags (6 turtles). Tags were affixed to the carapace of each turtle using polyester resin and fiberglass cloth. Tag transmission failure was very rare in this study. One ST-24 tag malfunctioned, so its data were not used in the analysis. The remaining 34 tags continued to transmit an average of 172 days (range of 6-503 days), providing tracking information for 5860 individual turtle-days. Positional data were downloaded from Argos and archived locally for processing.

Tracks of tagged loggerheads determined from Argos satellite data. Circles denote release locations 
            and stars the last transmitted positions.
Tracks of tagged loggerheads determined from Argos satellite data. Circles denote release locations and stars the last transmitted positions.

After all tags ceased transmitting, the raw Argos positional data were processed using a Bayesian state-space model (SSM). The SSM produced the most likely trajectory of each tagged turtle (graph above), taking into account Argos data quality codes and temporally adjacent positions. The SSM also recast the tracks into daily streams of points, thereby removing effects of variability in duty cycles of the tags.

Turtle tracks were merged with a suite of available oceanographic, bathymetric, and magnetic field data products. These include NOAA Pathfinder sea surface temperature (SST), AVISO altimetry products (sea surface height (SSH), geostrophic u-component, and geostrophic v-component), SeaWiFS ocean color, Smith and Sandwell bathymetry, and earth magnetic field data from the IGRF-10 model (total force, declination, and inclination). The ancillary data were examined on a daily basis and also integrated temporally over the entire track duration by averaging across the daily exposures per individual turtle. The SSM tracks were then merged with a new oceanographic data product which uses a time-series of remotely-sensed altimetry fields to quantify individual eddies. These energetic mesoscale features are one of the primary dynamic features in the ocean, after large oceanic currents and gyre circulation. Eddy shapes were reconstructed as circles using eddy-specific parameters, and the daily SSM turtle positions were compared to points along the circumference of the eddy. Radii at intervals of 5° of arc originating from the eddy central locations were used for the circle construction and the radii endpoint locations used for comparison to the SSM data. The central locations of all eddies were also compared to the SSM data. Eddies were classified as either cyclonic or anticyclonic by the nature of their SSH anomaly (negative SSH anomaly=cyclonic, positive SSH anomaly=anticyclonic), and further classified by eddy strength, as indicated by their vertical amplitude. Twelve distance measure metrics were calculated from this merging of datasets, reflecting a nested ordering based on eddy type (cyclonic or anticyclonic), eddy strength (any strength or strong), and the feature of interest (eddy edge or eddy center).

To model the relationship between turtle location and oceanographic features, a novel approach called eddy field randomization (EFR) was proposed. The method statistically evaluates the match/mismatch of turtle positions with specific eddy features. EFR infers attraction and/or aversion of the turtles to such features, and was used to demonstrate differential responses of the pelagic loggerheads to eddy characteristics (for example, eddy orientation, i.e., whether the eddy was cyclonic or anticyclonic), and to specific portions of the eddy (edges vs. centers).

The EFR approach was tested by examining how it performed on location data from 3 non-sentient sources in addition to the SSM daily turtle positions. First, EFR was applied to a set of random-walk tracks. Second, EFR was applied to a set of simulated passive particle tracks. Third, EFR was applied to location data from subsurface drifter buoys (drogued at 15 m depth) deployed within the spatial and temporal domain of this analysis (32,963 individual locations from 1291 buoys). Comparison of test results for the SSM turtle data and the non-sentient sources yields insights into eddy utilization by the loggerhead turtles and the underlying mechanisms, as well as the role of passive versus active orientation. The EFR analysis did not discern any associations between eddy features and the non-sentient location data evaluated, indicating no significant tendency for either attraction or aversion. The EFR analysis applied to the loggerhead turtle tag data, however, indicated that there was an attraction of turtles to cyclonic eddy features. This was most pronounced for the edges the eddy, and occurred regardless of eddy strength.

In addition to their attraction to cyclonic eddies, loggerhead turtles in the East China Sea near Taiwan also appear to utilize the continental shelf adjacent to the Yangtze River as a foraging area. This region is where the Yangtze River plume meets the Kuroshio Current intrusion. The complex dynamics make this region very productive. The sea floor here is shallow enough for benthic foraging by the turtles, yet also contains much eddy activity. This area is also intensively fished, primarily by boats from China. The extent of incidental or targeted take of loggerhead turtles by these and other fisheries over the continental shelf is largely unknown and needs further investigation. Likewise, loggerhead turtle diet and the community structure of both benthic and pelagic habitats are not well understood in this region and need further study.