Autonomous PIT tag antenna use in MARK

[Snook]

I am new to the mark-recapture world, recently starting up a recapture study using autonomous PIT tag antenna. We have 4 study creeks, with 3 antenna in each creek (at a lower, middle, and upper stretch). The antenna do not span 100% of each location, but come close. Additionally, the antenna lays along the bottom of the creek, and only read the first 30 cm or so of the water column (the creeks being from .5-2.5m in depth). However, our study species typically swims along the creek bottoms, so this is a sufficient read range. In initial investigations, we found that the antenna pick up 40% of the fish we tag.

I am curious as to a few things. My focus is using these antenna to look at movement between creeks and within creeks, as well as get a measure of apparent survival (which I was going to do by using the 'recaptures only' option in MARK) in each creek by pooling the antenna data together for each creek.

I would like to know:

1) In terms of using a Jolly Seber estimate for apparent survival, what issues do you think may arise as far as heterogeneity, etc. Is the recaptures only option the right way to proceed?

2) Is it possible to get any information on abundance from these antenna? We do not need any sort of precise population estimate, only some measure of population size that we can use to compare between the four creeks. I reiterate, all we are concerned with is comparing between the creeks, not getting any true values we can compare to other systems.

3) What else might we be able to explore with these antenna? They are such a fascinating system and there seem to be so many possibilities, so I am always interested in hearing what others from other fields have to think.




Thanks for any input, and I look forward to reading everyone's responses,


Andrew

[jlaake]

You have released animals with marks and that puts you into the release-recapture(recovery) world which is CJS and not JS. The difference is that you are not modelling the initial capture process nor do you capture unmarked individuals at each resampling occasion. Many of the models in MARK are based on the CJS format in which the model is conditioned on the initial release. They include the Recaptures only, the Recovery, both, and various version of the multistate models. JS models the initial capture process. In mark they include JS, POPAN, Pradel models and others.

From what you described of your system you'll want to use the recaptures only or multistate models. You can't get abundance unless you sample something other than the animals you released.

--jeff

[Eric Janney]

Snook,

The use of remote underwater PIT tag antennas is relatively new in fisheries. Most people using this technology are analyzing the detection data in an observational way and not incorporating it into a capture-recapture analysis. There are a number of issues that arise when you incorporate remote detections into a capture-recapture analysis.

1) In terms of using a Jolly Seber estimate for apparent survival, what issues do you think may arise as far as heterogeneity, etc. Is the recaptures only option the right way to proceed?

Heterogeneity can certainly be a nasty problem when you incorporate remote detections into your study design because fish that move more have a much higher probability of being remotely detected than fish that don't move. Also, fish that reside in close vicinity to the antennas will have a much higher prob. of detection than individuals residing further from the antennas. Another issue that you need to think about is the assumption that your sampling period is instantaneous. In traditional capture-recapture studies you try to keep your sampling period as short as possible to meet this assumptions. The remote systems operate continuously which can cause major problems. There are C-R models that deal with this to some degree. A model developed by Richard Barker allows for continuous resightings between sampling periods.

2) Is it possible to get any information on abundance from these antenna? We do not need any sort of precise population estimate, only some measure of population size that we can use to compare between the four creeks. I reiterate, all we are concerned with is comparing between the creeks, not getting any true values we can compare to other systems.

No, I can't see how it is possible. The antennas only give you information about the tagged portion of the population. All capture-recapture models that estimate N need information concerning the unmarked portion of the population. Incorporating remote detections will also bias recruitment and lambda estimates using the Pradel models. The remote systems in essence create a nasty trap happy response which will bias recruitment parameters low. The trap happy response happens because the probability of initial capture is much lower than the probability of being resighted/detected on the remote systems.

3) What else might we be able to explore with these antenna? They are such a fascinating system and there seem to be so many possibilities, so I am always interested in hearing what others from other fields have to think.

I agree that remote PIT systems are making capture-recapture studies much more feasible for studying fish populations. I have been using remote underwater PIT systems since 2005 and have had tremendous success improving recapture rates. You may want to get in touch with Aaron Berger at Michigan St. University. He has a paper in review that incorporates remote antennas into a multi-state capture-recapture study design to study movement and survival of cutthroat trout. I Also recommend attending the MARK workshop if you haven't already. It is well worth the time and cost.

[mcmelnychuk]

3) What else might we be able to explore with these antenna? They are such a fascinating system and there seem to be so many possibilities, so I am always interested in hearing what others from other fields have to think.

Andrew,

I have a couple of other suggestions. First, if antennae are relatively inexpensive you could double them up close to one another but assume they are independent detection stations. This should result in better estimates of p for at least one of the pair (you could construct a model using a robust design-type approach, a CJS approach constraining Phi=1 for the short segment between adjacent stations, or a multi-state approach with the same Phi constraint and a Psi=0 constraint between these same stations). Doubling them up would also decrease the probability of a fish passing by undetected at both stations, p_both=1-(1-p_1)(1-p_2)). Alternatively, you could assume that adjacent antennae are part of the same detection station, and staggering them slightly down the creek from one another should increase p for the overall station.

Second, you could incorporate mobile sampling into your program. If the creeks are fairly narrow, you should be able to sweep most of the area while walking up or downstream with a hand-held reader. You could then treat these data as live resigtings that occur between the stationary stations for use in a Barker model. see:

Roussel, J. M., Haro, A. & Cunjak, R. A. (2000). Field test of a new method for tracking small fishes in shallow rivers using passive integrated transponder (PIT) technology. Canadian Journal of Fisheries and Aquatic Sciences 57, 1326-1329.

Keeler, R. A., Breton, A. R., Peterson, D. P. & Cunjak, R. A. (2007). Apparent survival and detection estimates for PIT-tagged slimy sculpin in five small new Brunswick streams. Transactions of the American Fisheries Society 136, 281-292.

Cheers, Mike

[Snook]

Thank you for the great replies, they have been immensely helpful.

I have a few additional questions:

1) In one reply, the difference between incorporating the capture process and those that do not is brought up.
"You have released animals with marks and that puts you into the release-recapture(recovery) world which is CJS and not JS. The difference is that you are not modelling the initial capture process nor do you capture unmarked individuals at each resampling occasion. Many of the models in MARK are based on the CJS format in which the model is conditioned on the initial release. They include the Recaptures only, the Recovery, both, and various version of the multistate models. JS models the initial capture process. In mark they include JS, POPAN, Pradel models and others. "

My question is one of why I should not incorporate the tagging event into my survival estimates? Why is there this delineation?

2) We would really like to examine abundance differences between our four creeks. I understand this cannot be done using the antenna due to their inability to see unmarked individuals. However, can we use the tagging process to come up with some rough metric to compare our creeks?

We seine for our fish, with occasional recaptures as we scan each fish we catch, looking for PIT tags. There are seemingly high rates of emigration and immigration for our study subject, our seine pulls are not standardized to a single area/type/length and cannot be due to our system. We try to sample as much of the creeks as possible, but only a small portion of each creek is actually sample-able, and much of our effort is repeated on the same dozen areas in each creek. Is there any sort of Jolly-Seber open model that can be used, or is our situation too biased to generate population size data?


Andrew

[jlaake]

In your initial post you didn't describe an initial capture process with seining and only described "recapture" via antenna. If you are trying to model capture/recapture process of seining and passive recapture as well then you have a big task in front of you. I'm not aware of a model that will work for the process you described. It seems similar to the Barker model but without recoveries and a "resighting" process via the antenna between capture (seining) events. However, that model is based on the CJS formulation that does not model the initial capture. What it sounds like you need is a multi-state POPAN formulation. I'm not aware of one in MARK. Maybe it is possible with Carl Schwarz's POPAN program.

You can estimate abundance with a CJS model using an ad-hoc approach of dividing the number caught by the estimated capture probability. See the book by Amstrup et al on capture-recapture.

regards --jeff

[PSchueller]

[quote="jlaake"]
You can estimate abundance with a CJS model using an ad-hoc approach of dividing the number caught by the estimated capture probability. See the book by Amstrup et al on capture-recapture.

Because CJS is conditioned on initial capture, doesn't this model give us a recapture probability, as it does not model the initial capture event. Doesn't this result in some bias if the recapture probability is used in the canonical abundance estimator of number counted/prob of detection. Isn't a closed capture model estimate of capture probability a much more appropriate parameter to be used in the canonical estimator?

[jlaake]

I was only saying that it was possible to estimate abundance within the CJS formulation and not that it was preferable. Most of the information about capture probabilities is in the "recapture probabilities". If you look at the original closed form estimators (not to be confused with closed population) for the fully time-dependent Jolly-Seber model, all of the information about p is derived from the subsequent recaptures of marked animals and it explicitly assumes that marked and unmarked animals have the same probability of capture for estimation of abundance. In that model, the number of new entrants at each occasion is time dependent, so I believe there is no information about p(t) in the initial capture events because the number of new entrants varies each time as well. If restrictions are added to the model (e.g., constant parameters, or death/emigration only model) then you can start to get information about p from the initial captures.

I'm not sure I follow your comment about use of a closed population estimator. If that is in reference to a robust design, then I agree with what you are saying as that was the reason for the development of a robust design. I'm not certain that was relevant to the original posting but I don't know the specifics of the particular sampling problem to address fully what is needed.