Mask size and model fit issues with small dataset

questions concerning anlysis/theory using program DENSITY and R package secr. Focus on spatially-explicit analysis.

Mask size and model fit issues with small dataset

Postby pantheraz » Wed May 08, 2024 8:46 am

Dear Murray,

I am using SECR for my master thesis and it has been very helpful so far. I have however run into some problems and I was hoping that you (or someone else) could provide me with help. I am doing a camera trap study on African civets and I would like to estimate the density of African civets in the area. I am using secr 4.6.6 in R 4.2.3 on Windows 10. My dataset is unfortunately quite limited (44 observations of 3 individuals on 14 out of 20 camera traps).

When fitting the null model, there was too little data to determine the starting values, so I set these manually (D = 0.00001, g0 = 0.1, sigma = 1270.699). The density output I got for a mask with buffer width of 4*sigma seems reasonable. However, when I increase the mask size, the density becomes smaller. This seems odd to me, because making the mask too large should not noticeably affect the final estimate right?

When I try to fit the models for learned responses, I run into a problem. The learned response model (b) runs fine but it yields the exact same density as the null model. When trying to run the detector-specific learned response model (bk) I get an error:

Code: Select all
 civet.fit.bk <- secr.fit(civet_CH,
+                       mask = clippedmask.civet,
+                       start = list(D=0.00001, lambda0 = 0.1, sigma = 1270.699),
+                       model = g0 ~ bk)
Checking data
Preparing detection design matrices
Error in aperm.default(temp, c(2, 1, 3)) :
  'perm' is of wrong length 3 (!= 2)


Lastly, when displaying the AIC-values, I get values for AIC but I get NA for AICc, dAICc and AICcwt:
Code: Select all
model   detectfn npar    logLik     AIC AICc dAICc AICcwt
null D~1 g0~1 sigma~1 halfnormal    3 -77.53938 161.079   NA    NA     NA
b    D~1 g0~b sigma~1 halfnormal    4 -77.53938 163.079   NA    NA     NA


I wonder if there is an easy fix for these problems, or are some of them the result of my very small dataset? Any help would be greatly appreciated. I have already searched for answers in the forum but to no avail. If you have time I could also send you my script and data, just let me know. Thanks a lot in advance!

Kind regards,
Luuk
pantheraz
 
Posts: 1
Joined: Tue May 07, 2024 9:28 am

Re: Mask size and model fit issues with small dataset

Postby murray.efford » Wed May 08, 2024 5:48 pm

I'm sad to say I think a dataset with three detected individuals is woefully inadequate for capture-recapture methods of any sort, and that is likely to be the source of your particular problems. Certainly a model with 4 parameters won't fly, and it is unclear how you can know 'true' sigma to interpret the apparent buffer effect.
Murray
murray.efford
 
Posts: 712
Joined: Mon Sep 29, 2008 7:11 pm
Location: Dunedin, New Zealand


Return to analysis help

Who is online

Users browsing this forum: No registered users and 2 guests

cron