How to model mixed cohort + individual capture histories?

[Jmackay]

Hello!

I'm looking for advice on how to best model data from 1 season of mark-recapture of juvenile coho salmon. My primary goal is to estimate abundance, and hopefully survival, to explore some hypotheses. My problem is that I have a mix of individual and cohort-based capture histories:

Fish were sampled in one closed mark-recapture event per month (2 secondary events) for each of 5 months (5 primary occasions) over the 2025 season. Individual fish received cohort-based VIE tags until/unless they grew large enough to be PIT tagged (and thus, gain an individual capture history). As such, my dataset is a mix of VIE tagging (both secondary events have the same mark; cannot discern individual capture history) and individual PIT tagged fish (which may have VIE tags from previous occasions). I've appended a small example of capture histories for clarity at the end.

I was initially hoping to conduct a robust-design model to estimate abundance and survival, though at this point I'm unable to incorporate the VIE data. I could use only the PIT data (individual histories), but I fear the sparse early-season PIT data (smaller fish = high proportion of VIE marks) and low overall recapture rates would result in poor estimates.

A couple of alternative options have occurred to me: 1. Collapse the secondary occasions and form pseudo capture histories for a Jolly-Seber model, or 2. Use two separate models - a robust model using only the PIT data to estimate survival/capture probability, and multiple closed capture models using PIT and VIE data for each secondary occasion to estimate abundance. However, I worry that this would be compromised by the low recapture rates.

Is there a way to format the data to use both VIE and PIT data for a robust model? Is it be better to run separate open/closed models? Is there another approach I could consider?

Advice is greatly appreciated! I'm happy to provide more information if needed.

Many thanks,

Jamie


P.S. some example capture histories:

___________|---P1---|---P2---|---P3---|---P4---|---P5---|
___________|-S1-S2-|-S1-S2--|-S1-S2-|-S1-S2-|-S1-S2--|
fish 1 (PIT): ---0--1-----0---0-----1---1-----0---1----0---0---
fish 2 (VIE): ----1----------0---------1----------1--------0-----
fish 3 (both):---0----------1-------1---0-----0---1----1---1---

[jhines]

I think you can use robust-design with these data, but you would need to add a 3rd secondary occasion for VIE.

Code: Select all

fish 1 (PIT) : -01 -00 -11 -01 -00
fish 2 (VIE) : 1-- 0-- 1-- 1-- 0--
fish 3 (both): 0-- 1-- -10 -01 -11
Fish4 (PIT*) : 0-- 0-- -01 -11 -10

Fish 1 was captured in primary period 1 with PIT tag, so it will not be available for VIE captures (1st secondary period of each primary period.

Fish 2 never got a PIT tag, so it was not available for PIT capture occasions (2nd and 3rd secondary periods).

Fish 3 was captured via VIE in the 1st 2 primary occasions, then captured via PIT in primary occasions 3,4 and 5.

Fish4 was not captured via VIE, then captured via PIT, so it would have zero’s for occasions before the first PIT capture, indicating it was not captured via VIE in the 1st 2 primary occasions.

This should allow you to estimate detection probabilities for PIT occasions, but not for VIE occasions (unless you add constraints on them).

I’m assuming here that the VIE marking allows you to identify which primary occasions fish were captured.

I think the bigger problem will be grouping the small and larger fish together such that they have the same survival and capture probabilities. Perhaps a better approach would be to run a robust design, multi-state model with small and large fish, where they could transition from small to large and you could estimate survival for each group.