README.md

TwilightFree method of geolocation

A new method of geolocation from noisy light data using a hidden Markov model

For my Honours project I developed a new method of light-based geolocation that relies on the overall pattern of day and night rather than an explicit dependence on twilights as per the threshold, curve, or template-fit methods. For an introduction to geolocation using recorded light, please see https://en.wikipedia.org/wiki/Light_level_geolocator

This repository holds important data for my thesis and a paper which has been accepted detailing the 'twilight free' method.

Important Update: TwilightFree R package now available

This git is important as it provides a data archive including supplementary material for the published article. However, TwilightFree has been packaged for R and is available (with tutorials) from https://github.com/ABindoff/TwilightFree

Supplementary materials

twilight_free_algorith_example_ses.Rmd contains a script that replicates an example from the methods paper and calculates measures of accuracy and precision.

quick_start_guide.md provides a quick-start tutorial for users who want to use the (very small) set of functions to analyse their own light data from archival tags. A tutorial will be added for users who wish to improve their location estimates by including SST data

parameter_sensitivity_new_data.md extensively tests the sensitivity of the shading/noise and movement parameters using data from a single animal, then uses the optimal parameters from that example on three more datasets, finally performing another PSA on an animal displaying markedly different behaviours to the first. This analysis is useful because it is expected that users supply shading/noise and movement parameters from previous studies, preferably from double-tagging experiment. The analysis demonstrates that parameters determined from a double-tagging experiment can inform analyses of other tag data from the same species, but that optimal parameters are not necessary for useful results in biology and ecology.

Known bug:

• where the grid (spatial map) crosses the equator it is possible for the MAP estimate to occur (somewhat obviously) on the wrong side of the equator on any given day (the algorithm picks a mathematically plausible but ecologically implausible solution). One quick fix is to break the data into parts and provide a different grid for each part. Alternatively, calling essieRaster(fit_object) and finding the max in the appropriate hemisphere also works. A solution is currently in development and testing.