Introduction & Objectives

Accounting for availability bias, the time animals are unavailable to be detected by visual surveys, is a critical component for accurately estimating animal abundance and distribution in density spatial models. These models are used for conservation and environmental compliance purposes by the U.S. Navy, other federal agencies, and regional stakeholders. For air breathing animals at sea, such as sea turtles and marine mammals, the proportion of time spent below the sea surface can range from 5-90%, depending on species, season, and animal behavior. If availability bias estimates are not applied to spatial density models, animal abundance may be underestimated by as much as an order of magnitude in some cases, hindering conservation efforts, and significantly underestimating the potential impacts of human activities on these protected species. Applying robust availability bias estimates to density spatial models should be considered ‘best available science’ and actively pursued for the newer generations of density spatial models being produced.

Availability bias is often applied as a single, static number such as the mean proportion of time spent below the water based on depth profiles of tagged animals, or other similar metrics. However, animal dive behavior can vary widely by season, habitat, and life stage. More complex approaches that represent animal availability as spatially varying surfaces can now be developed and applied to density spatial models. Several frameworks have been implemented for sea turtles, including spatiotemporal regression models and Generalized Additive Models that relate dive behavior to environmental covariates. These models can estimate animal availability bias over broad spatial extents, diverse environmental conditions, and seasonal or finer temporal scales, assuming adequate sampling by tagged animals.

On the east coast of the United States, there is a critical data gap for dive data appropriate for availability bias estimates for endangered Kemp’s ridley (Lepidochelys kempii) and threatened green (Chelonia mydas) sea turtles, in part because these species are smaller and difficult to tag with depth recorders. Recent advances in tag technology have made tagging these species in the numbers required to generate robust availability bias estimates possible. Both species occur within nearshore and continental shelf waters ranging from New England to North Carolina on a seasonal basis, and from North Carolina to Florida year-round, and are found in multiple Navy training and testing ranges and other areas of interest to regional stakeholders such as windfarms and commercial ports.

Technical Approach

This project aims to tag 75 green and 75 Kemp’s ridley turtles over 3 years, in multiple seasons and locations along the East Coast, in order to robustly sample these species’ seasonal and year-round habitats with the end goal of producing spatial models of availability bias for both species at a seasonal or lower spatial resolution. Wild caught animals will be tagged at the Archie Carr National Wildlife Refuge, Indian River Lagoon, St. Lucie Nuclear Power Plant, and Trident Submarine Basin, all located on the east coast of Florida. Additional rehabilitated and wild caught animals will be tagged and released in partnership with several aquarium and rehabilitation centers north of Florida, the exact mix to be determined. In addition to the availability bias data products, the tracking and dive data collected as a part of this project will provide a valuable ecological baseline for other studies, as these species are underrepresented in the tagging record in the region.

The selected tags, a new, smaller model of Wildlife Computers SPLASH10 tags, are appropriate for tagging juvenile through adult specimens of both species, allowing maximum flexibility for tagging permits and available animals, as well as a range of turtle sizes to best represent size-based physiological differences in behavior. No animals smaller than 30cm straight carapace length (SCL) will be tagged as this is the smallest size generally considered to be visible from aerial survey platforms. The Wildlife computer dive data products are appropriate for calculating the dive statistics required for availability bias surface modeling, such as percentage time below a dive threshold and dive and surface intervals. Tags will not exceed 2-5% of the turtle's weight in air, and transmitter designs and attachment methods will be selected to minimize drag per federal and state tagging requirements for sea turtles. No turtles that are injured/ill or Fibropapilloma positive will be selected for this telemetry study.

Tags will be deployed over the course of the year, with the bulk deployed between spring and fall each year. Some tags may be reserved for deployments as late as March in the first two years of tagging to allow for maximum flexibility with the species and age classes tagged, and to sample southern overwintering habitat. In year 3, tag deployments will be completed by late fall to allow time for deployments to be completed prior to the capstone analysis in Year 4.

In spring, prior to tagging years 2 and 3, an evaluation of tag deployments and acquired telemetry data will be undertaken to ensure the data collected is robust and to assess whether any changes to tag attachment methodology, tag programming, or data collection is required. Additionally, the mix of species tagged, and the geographic coverage of tag data will be assessed, adjusting planned deployment locations and species mix as appropriate to tag roughly even numbers of both species across the three years of tagging, and sampling as many relevant environments as possible.
Year 4 will be the capstone analysis, fitting a Generalized Additive Model (GAM), or other appropriate statistical framework, relating environmental covariates to dive behavior. This approach has already been undertaken for species in the Gulf of Mexico and represents a proven approach ready for application and testing in other regions.

GAMs fit flexible relationships between a response variable (animal availability metrics) and explanatory covariates including remotely sensed environmental conditions, temporal covariates, and categorial considerations such as sex or size class. This allows for complex relationships to be modeled both spatially and temporally, as we would expect animal behavior to change as conditions and habitats change. Selected environmental covariates will plausibly influence animal dive behavior and hence, availability. Example candidate covariates include temperature, habitat structure, prey fields, and productivity metrics. Several embellishments to GAMs may be considered, including hierarchical models which would allow flexibility in the model for animals of different size classes or in different habitats.

Progress & Results

Initial funding for this project was provided in early 2024 and we plan to deploy our first tags early summer 2024 in Florida during the start of the green turtle nesting season, as well as on turtles captured from some coastal in-water sampling sites in the Indian River Lagoon, Trident Submarine Basin, as well as from the intake canal at the St. Lucie Power Plant in collaboration with project partner Inwater Research Group. Stay tuned for tag deployment and tracking updates!