Unconventional Data

[Jonathon Kuntz]

Hi all,

I’m exploring whether the ctmm package can be used to estimate "isotopic home ranges" for sharks using AKDE, based on a time series of stable isotope data (δ13C and δ15N) instead of traditional telemetry data. My data is definitely unconventional for this purpose, but it might work since isotopic values vary across localities with distinct environmental baselines, which could act like "spatial movements" in isotopic space (conceptual figure attached).

Background on My Data:

• I have stable isotope data (δ13C and δ15N) collected over time for individual sharks.
• Each shark has repeated isotope records defined by age, similar to telemetry datasets where movement data is autocorrelated.
• Different foraging locations have unique isotopic signatures, making isotope data a good proxy for spatial ecology.
• The data is nested—multiple samples per shark, and sharks are grouped by populations or regions—so I need to account for individual-level autocorrelation while still looking at broader patterns, like population-level home ranges.

My limited understanding of the akde() function in ctmm is designed for telemetry data and typically requires latitude and longitude as inputs to calculate autocorrelated home ranges. While I don’t have spatial coordinates, I want to know if it is possible to treat δ13C and δ15N as "coordinates" in isotopic space and use time as the temporal dimension? Each shark would be treated as an individual with repeated measures, similar to how movement data is structured in telemetry datasets.

Thanks so much for any advice!

[Jonathon Kuntz]

Also, not sure if something similar has been asked before (I couldn't find anything scrolling through conversations), but I could also imagine this may be useful in terms of non isotope variables (e.g., altitude/depth, temperature, salinity, etc.)! Thanks again!

[Christen Fleming]

Hi Jonathon,

It sounds like you might be in luck, because ctmm is very hard coded to work on 1, 2, and 2+1 dimensional problems. From the broader perspective of timeseries modeling , the other model-structure assumption that is made in ctmm is that the 2 dimensions are very similar with the same autocorrelation timescales in every direction (to within a linear transformation of the two dimensions). You could certainly get your data into a timestamp, x, y  format which will import as fake UTM data, run the analysis, and then see how well the autocorrelation models look w.r.t. the variogram of the data and the correlogram of the residuals.

I'm curious as to what ontogenetic shift looks like in the data and how that would be modeled. If there's a simple drift model that I could add to ctmm to facilitate that, I'd be potentially interested.

Best,

Chris