Adjusting for change in study area size necessary?

51 views
Skip to first unread message

jcl...@utk.edu

unread,
Jul 20, 2022, 10:01:58 AM7/20/22
to secr
I have a 14-year SCR data set on black bears based on DNA CMR in an area that was flooded during some years.  The flooding is reflected in the usage statistics for the traps for those years.  I am using openCR to estimate D and N, recruitment, phi etc. using a Pradel-model type.  My question is whether I should use the number of traps each year as a covariate on overall detection or will this be accounted for automatically by the usage data?  Previous analyses suggest that the majority of bears do not shift their home ranges because of flooding so I would not expect a big increase in density in the areas that were not flooded during flood years.  How should I estimate N from D for those years?
Thanks!
Joe

Murray Efford

unread,
Jul 20, 2022, 5:14:14 PM7/20/22
to secr
Hi Joe
Good to see you working on this dataset. When I looked at it a few years ago one of my hesitations was that I didn't have usage data for some spots that seemed not to have been sampled in some years (spillway 2007, 2011). Now you're onto it!

Usage data alone will take care of the differences in sampling - no need for a covariate. I'm not quite sure what you mean by a Pradel model type.

It's a puzzle how to estimate D (and N) when the extent of habitat changes between years because of flooding. My first reaction is that you could come really unstuck using an open model, because there are very specific unmodelled effects (limited temporary movement out of flooded areas, changing extent of habitat). If you are happy to argue/have good evidence that bears suirvive in situ then this is not an issue, and the open D-hat (N-hat) apply. I tend to place more trust in closed models for D (and N), anyway.
Cheers
Murray

jcl...@utk.edu

unread,
Nov 7, 2022, 8:50:42 AM11/7/22
to secr
Hi again.  I plan on using Murray's suggestion to use the open models to estimate lambda (my primary parameter of interest) and then use closed models to estimate D (and N).  Given that strategy, does it make sense to try to fit an open model with bk and h2 effects on lambda0 and sigma when those parameters mostly go toward estimating D, with estimates of lambda being pretty robust to biases in D?  I was considering testing some simpler models to estimate lambda (e.g., phi~t, lambda~t, lambda0~1, p~1) to decrease computing time and because the JSSA models are so finnicky when the become more complex.  Thanks!

Murray Efford

unread,
Nov 8, 2022, 3:07:51 PM11/8/22
to secr
I hesitate to prescribe, but... I think I agree. You can estimate the demographic parameters with a PLB model bypassing D altogether. Defining learned responses and heterogeneity in an open model raises all sorts of potential complexity (within- vs between-session effects, etc.). Heterogeneity does bias turnover (phi, f, lambda) estimates, but rather little. And h2 is just one stab at addressing it.
Murray
Reply all
Reply to author
Forward
0 new messages