Rpackage for improved analysis of marine animal movement data using hidden Markov models
Camrin D. Braun1,2*, Benjamin Galuardi3,4, Simon R. Thorrold2
Electronic tagging of marine animals is common throughout the world oceans. Many of these studies have deployed archival tags that rely on light levels and sea-surface temperatures to retrospectively track movements of tagged animals. However, methodological issues associated with light-level geolocation have constrained meaningful inference to species where it is possible to accurately estimate time of sunrise and sunset. Most studies have largely disregarded the oceanographic profiles collected by the tag as a potential way of refining light-level geolocation estimates provided by electronic tags.
Open-source oceanographic measurements and outputs from high-resolution models are increasingly available and accessible. We integrated temperature and depth profiles recorded by electronic tags, with empirical data and model outputs, to construct likelihoods and improve geolocation estimates for marine animals using an existing, but modified, state-space hidden Markov model (HMM). Our model (
HMMoce) exhibited as much as 6-fold improvement in pointwise error as compared to traditional light-level geolocation approaches and produced the lowest mean error in 3 of 4 cases when compared to the state-of-the-art tag manufacturer's HMM (GPE3).
HMMoce contained behavior state-switching capability not found in other comparable methods. The use of profile-based likelihood estimates proved useful when we removed data to emulate data returned from species that yield poor quality light data. The results demonstrated the general applicability of the
HMMoce model to marine animals, particularly those that do not frequent surface waters during crepuscular periods. Our model is available as an open-source
HMMoce, that uses a state-space HMM approach and leverages available tag and oceanographic data to improve position estimates derived from electronic tags.
The package is structured as follows: * Load the relevant tag data and establish a study area of interest. * Get the environmental data to base the likelihood calculations on. * Calculate the desired likelihoods (e.g. depth-temperature profiles, SST, etc) * Estimate parameters and run the model. Results are written out along the way. * Perform model checking and choose a final model.
HMMoce can be installed from CRAN from within
install.packages('HMMoce'). To get the latest developments, get it from GitHub using
For an example use of the package, please see the vignette using
Any scripts or data that you put into this service are public.
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.