Seeking insight on how to analyze long-term snake data

[aheffernan]

I'm currently working on a dataset with over 30 years of mark-recapture data for a threatened rattlesnake in a protected area. I'm fairly new to modeling in general, but I'd like to try to extract some population demographics from my data, particularly survivorship and approximate population numbers.

In the study, we conduct snake surveys throughout the active season to mark new captures and recapture previous individuals, and opportunistically capture snakes that are reported by park visitors, but we also conduct frequent road surveys to recover dead individuals, some of whom are previously marked.

I was thinking that a Burnham joint live-dead framework might be most appropriate for this dataset, but I would appreciate any insight into whether JS or CJS models would be better.

It may also be important to note that the areas being surveyed have varied over time, but all are within the same protected area and the cover the same metapopulation of snakes, although search effort has varied significantly over the course of the project. I'm not sure if these factors would influence my ability to analyze the data, but if they would, any suggestions on how to work around these issues would be great.

Overall, I'm hoping to get some input on how best to go about formatting my data, what model framework would be most compatible, and whether there are aspects of this study that would violate some key assumptions of the models. If anyone could recommend any R scripts and beginner resources for starting something like this, that would also be excellent!

Thanks!

[B.K. Sandercock]

Hi A. Heffernan,

A couple of suggestions on options for models for your rattlesnake data.

Take a look at how much information you have for live encounters versus dead recoveries. If the number of dead snakes is relatively low, it might not be enough information to estimate the extra parameters in the Burnham model. Also, sounds like most of your dead recoveries are within the same sampling area. Supplementary resighting and dead recovery data are often more useful if taken from a larger area.

JS models give estimates of abundance but often with large confidence intervals that still might be suitable for your needs. If you want better estimates of abundance, you might look into closed population models and consider if your sampling design is sufficient to meet assumptions of closure. Closed models also have more options for handling issues affecting probability of capture.

CJS models might be the best place to start for your dataset for some exploratory analyses and see what you can get for parameter estimates for apparent survival and detection. The Mark manual has detailed information on how to set up the encounter histories. Note that you can code the encounter histories to censor the individuals that are known to be dead. The convention in Mark/RMark is to have a negative sign in front of the count for a particular encounter history if an individual that turned up dead. The negative sign means not released at last encounter so that the trailing zeros in the individual encounter history are censored and do not go into the calculations of possibly alive but undetected. In practice, if a sample only includes a handful of dead individuals, the coding does not usually have a big impact on the parameter estimates.

Good luck, Brett

[aheffernan]

Hi Brett,

Thanks for the thoughtful reply! Looking at my dataset, I only have 85 dead recoveries where the individual ID was known, out of ~2000 captures. You're likely right that my sample size would be too small to estimate those extra parameters. I'll try a CJS with censoring of known mortalities and see how that goes!

Thanks again, Alex