Colext Error: non-finite finite-difference value [14]

369 views
Skip to first unread message

Michele Chiacchio

unread,
Sep 30, 2021, 8:14:49 AM9/30/21
to unmarked

Dear all,

I am trying to run a multi-season model with Unmarked. There are a total of 16 sites, surveyed 9 days each for 4 years (some of the sites were not surveyed all the years and I inserted NAs for the corresponding dates).

It worked fine until I added the data for the last season (2021). Then, when I try to run a model selection for detection parameter

fmx1.1<-colext(~1,~year-1,~year- 1,~year+date+rain+air+hour+cow+ROCK+bush+log+veg,umf_RT) 

it returns an error message:  " Error in optim(starts, nll, method = method, hessian = getHessian, ...) :  non-finite finite-difference value [14]"

Could anyone provide any ideas?
Interestingly (or worringly), when I tried the same code for another species (same covariates, only different presence/absence values) it runs perfectly fine.

Best,
Michele Chiacchio

Ken Kellner

unread,
Sep 30, 2021, 11:05:32 AM9/30/21
to unmarked
Hi Michele,

How do the 2021 data for the species look compared to previous years? Are there a similar number of total detections, total sites with at least a couple detections, etc? Anything stand out? If there are fewer detections or the detections line up such that categorical covariates don't line up with any detections, that could cause unmarked to struggle to fit the model.

You are trying to fit a pretty complicated model with a a relatively small dataset. I would try starting with a very simple model for detection and then add covariates one at a time to see if you can identify if it's a particular covariate causing issues.

Ken

Michele Chiacchio

unread,
Oct 4, 2021, 8:29:49 AM10/4/21
to unmarked
Dear Ken,

yes, that's what I figured. What I did was to delete the variables one by one to see if the problem was remaining and apparently if I remove the "air" covariate it works just fine.

But what I don't understand is why with the other species (whom presence/absence data are not much different from this one) it works without any problem. With the exact same covariates combination.

Michele
Reply all
Reply to author
Forward
0 new messages