Burnham LD model with irregular sampling

[Urge4Sturg]

I have set up encounter histories in the LDLD format to estimate survival rates for an impounded white sturgeon population in the Columbia River.

Sturgeon were first individually marked with PIT tags beginning in 1994.

Stock assessments are conducted every third year starting in 1999 to estimate abundance.
Stock assessment sampling consists of a winter marking run and a summer recapture run.
Harvest (commercial and recreational) occurs winter-summer in some years

We are interested in using these data in addition to reported harvest data to estimate annual survival rates in this population.

I have given fish encountered in live sampling a 1 in the L column if they were encountered in either winter or summer sampling in a given year. Harvest occurs before, during, and after the two sampling periods in a stock assessment year. The live sampling period ends up being as long, or longer than the harvest period in years where live sampling occurs.

Is it appropriate to include the harvested fish in the year they were caught, given live sampling can occur before, during, or after harvest? Or should they be placed in the preceeding D column?

How does violoating the assumption of a longer dead recovery interval affect parameter estimates? Fidelity is high (an assumption of 1 would not be unreasonable as it is probably between 95 and 99%) as this population is in an impoundment.

Also, I have defined the first time interval as 5, and used annual encounters for the rest. When i get survival estimates in a S(t) model, the estimate of S1 is a 5-year survival rate, while the rest are annual, correct?

For off years(i.e., no live sampling), i have fixed p=0, but included harvest data. Does this affect survival rate estimates in the off years?

Do i need to rethink the encounter history structure i am considering or redifine sampling intervals? I have been RTFM (chapter 10) and took the intermediate workshop, i'm just a bit rusty and looking for some additional human input.


Thanks in advance!

[claudiapenaloza]

Everything sounds good as far as I can tell... I'll answer your questions as best I can and then I have a suggestion that might make your life easier.

Is it appropriate to include the harvested fish in the year they were caught, given live sampling can occur before, during, or after harvest? Or should they be placed in the preceeding D column?

I'll answer this with a suggestion at the end.

How does violating the assumption of a longer dead recovery interval affect parameter estimates?

A long dead recovery interval is not an issue... the issue is the "as long or longer" live sampling period. The first problem of having a very long live recovery sampling period is that you may be violating the assumption of closure within the sampling period. The second problem is less evident/talked about and affects your survival estimates (Kendall, comm. pers.)… it is not the same for an individual that was marked, say 1Jan1999, to survive to 1Jan2000 as it is for another individual marked at the same time to survive to 1July2000. Is there a way you can shorten the live sampling periods? Is there a peak sampling period?
Can you give more details as to how live samples are collected (e.g., for the winter sample we go out to the field to catch fish on 3-day trips every 2-weeks from Nov-Mar)?
I'm wondering if you could use a Robust Design…

When i get survival estimates in a S(t) model, the estimate of S1 is a 5-year survival rate, while the rest are annual, correct?

Yes

For off years(i.e., no live sampling), i have fixed p=0, but included harvest data. Does this affect survival rate estimates in the off years?

Yes, but not in a "bad" way. The harvest data feeds directly into the estimation of true survival, whether or not you have live recaptures (just don't set r=0 when you have harvest .

Do i need to rethink the encounter history structure i am considering or redifine sampling intervals?

My 2 cents… I think it would be a lot easier to have 6-month-long "years", i.e., separate winter and summer live sampling periods. Each of these 6 month periods would start with your first live capture sampling for that season; harvest for the season would also start the same day and continue to the last day of the season (whether or not it actually lasted that long). This would make it a lot easier to assign a dead recovery. If you capture an individual in a given season AND it is harvested thereafter, it would get a "1" in the "L" and a "1" in the "D" column for that season. If you did not capture an individual in a given season but it was harvested in that season, it would get a "0" in the "L" and a "1" in the "D" column for that season. Breaking the live sampling into two six month chunks also reduces the "skewed" survival issue I mentioned before.

Hope that helps for now… please tell us more.

[Urge4Sturg]

I rearranged the encounter histories to fit the burnham model more "cleanly" i think. I have an LD column for years with live sampling effort and lumped all harvest data into the D column for the interval between live sampling years. The i defined the time intervals when I imported the .inp to reflect the irregular sampling.

I built a general model and did the GOF testing. Everything seems alright, however results aren't very interesting. We would ultimately like to have survival and fidelity estimates for three different size classes representing juveniles, subadults and adults. A simple group variable will not work, however, because given the duration of our study fish could grow into a larger size class.

Would it be appropriate to use a Barker multi-state model with three strata defined by size? Transition probabilities would represent growth from one group to the next. Backwards transitions (e.g. Adult to subadult) and skipping a stage (e.g. juvenile to adult) would be fixed at 0.