Unreasonable pop abundance estimates

[SoConfused]

Hello,

It's my first time trying to use POPAN in MARK. I have a 3-occasion dataset (which worries me - is that too short?) of sturgeon mark-recap. The setup is as follows:
1) in early 2000, thousands of juveniles were released in an upstream location. Over the years, they've been making their way into more downstream areas of the river.
2) In addition, a small wild population is known to exist in the area.
3) In 2010, 2012, and 2015, the downstream area was sampled using approximately equal effort between the years.
4) The following is a table of capture histories, by occasion and hatchery/wild:
Hatchery Wild
001 289 50
010 149 26
011 21 2
100 328 48
101 31 3
110 17 2
111 3 0
5) A few fish were entrained through a hydroelectric dam further downstream (and therefore are removed from the system); I entered these as loss on capture. All were Hatchery fish; 12 had "100" capture histories, and one had "010" capture history.

Ideally, I'd be able to estimate hatchery/wild populations separately. If not, I can just calculate a total pop abundance and use the proportion of wild fish out of the total to estimate the wild pop.

I tried combinations of the following parameters: Phi.1, Phi.g, Phi.t; p.1, p.t; N.1, N.HW; pent.1, pent.t.
Most of the models did not converge, which I assume is due to the short dataset. The ones that did converge, predicted a ridiculously low N (~850), and a very high p (~ 0.79) and fairly low (for sturgeon) Phis - ~0.4-0.6.

When I ran a CJS model just for comparison, I got the values I would've expected - Phis of ~ 0.8-0.9, and p of 0.1-0.2...

Here are the questions:
1) Can I even use a three-occasion dataset for POPAN?
2) Why the odd estimates in POPAN, and what can I do about it?
3) If I can't use POPAN, where do I go next? I'd really not want to go to closed-pop models at this point...

Thanks so much!

[Bryan Hamilton]

Three occasions is the minimum for open population models. I can't really comment on the odd estimates and what to do about it. Another year of data collection could really help tighten up the parameter estimates and the model convergence issues. You won't be able to use closed population models as its pretty clear the population is not closed.

Were all the juveniles marked in 2000 when they were released? You have a lot of individuals captured only once. These don't contribute much information to the models.

You could find the strongest model for p, then use this to model Phi and N. With capture probability so high, the abundance estimates won't be much higher than the number of captures.

What are you trying to model with N? The total population over the three years (super-population) or the group and time specific abundances?

[SoConfused]

Thanks for the reply! Ideally, I think that I would want the occasion-specific pop abundances.

Yes, all the hatchery fish were marked when they were released, but there is no db for the tag numbers (don't ask me why, I have no idea why someone would do that), so there's nothing I can do with that data.

My worry is that the p-estimates from the POPAN models are basically garbage. There's no way that recap rate is so high.

Is there some (even if weird) way to use CJS instead, and use the recap rate for some oddball occasion-specific pop estimate? It's probably stupid - but if, say, my recap rate at occasion 2 is 0.1, and I know how many fish I caught at occasion 2, could I calculate an occasion-specific coarse population estimate? And then repeat for occasion 3...

[Bryan Hamilton]

I ran your data in RMark and got some results. The estimates for p are much lower, around 0.16. Subsequently the estimates for N are higher, 3820 for hatchery and 596 for Wild. Survival was still high, 0.7.

The occasion specific estimates for N are also seem more reasonable, with fairly large confidence intervals.

[AdamC]

I think Bryan and I had the same thought. The estimated parameters you listed don't seem to match your capture history, although what went wrong is unclear. Here is how I ran your numbers in RMark:

Code: Select all

library(RMark)
ch <- rep(c("001","010","011","100","101","110","111"),each=2)
freq <- c(289,50,149,26,21,2,328,48,31,3,17,2,3,0)
group <- rep(c("Hatchery","Wild"),length.out=14)
data <- data.frame(ch,freq,group,stringsAsFactors = FALSE)
mod <- mark(data,model = "POPAN")
mod$results$real

And the output:

Code: Select all

> mod$results$real
                  estimate          se          lcl          ucl fixed    note
Phi g1 a0 t1     0.7396175   0.1571249    0.3646721    0.9335848             
p g1 a0 t1       0.1182818   0.0359291    0.0639219    0.2085697             
N g1 a0 t1    4373.7095000 534.2938700 3476.9100000 5591.1941000             
pent g1 a1 t2    0.1361216   0.0717578    0.0454806    0.3425741             
pent g1 a2 t3    0.1361216   0.0717578    0.0454806    0.3425741