Varying trap effort

[Prashant_mahajan]

Hello,

I am trying to include effort as a covariate to model capture probability (g0). However, due to high variability in effort across detectors (range: 11 m to 7500 m), I am encountering the following warning message:

Extreme 'beta' in 'fastsecrloglikfn' (try smaller stepmax in nlm Newton-Raphson?)

As a result, my density estimates appear to be unreliable. However, when I exclude effort from the model, the results seem reasonable.

My study includes 289 detectors, with detections recorded at 92 of them. I would like to incorporate effort as a covariate while ensuring stable model performance. Could you suggest the best approach to handle this issue?

Thank you

[murray.efford]

There is no need to model effort as a covariate - just include a 'usage' attribute in the detector object. Pehaps that is what you are doing? Also, it may be better to use a hazard detection function (HHN, HEX) with parameter lambda0, not g0.
Murray

[Prashant_mahajan]

Hi,

I have my summary of capthist as follows:

Object class capthist
Detector type count
Detector number 289
Average spacing 250 m
x-range 543232.1 556724 m
y-range 5261384 5286856 m

Usage range by occasion
1
min 11.81
max 7544.24

where usage 11.81 m is the min length of the transect in a detector and 7544.24 is the maximum length of the transect in a detector.

And my trap file looks like this:

X.TrapID X Y Effort
1 114 550982.1 5286142 246.74
2 115 551232.1 5286142 517.36
3 116 553232.1 5286142 76.02

The code for calling trap file is:

trapfile2021 <- read.traps(data = traps.2021, trapID = "trapID",
detector = "count",
binary.usage = FALSE)

And the model that I am using is this:

m.0 = secr.fit (coy.cap2021, model = list(D~1, lambda0~1, sigma~1),
mask = mask.2021, detectfn="HHN", method = "Nelder-Mead")

What should I put in lambda0 to make it as function of varying length of transects in each detector with a single occasion.

Best,
Prashant

[murray.efford]

Nothing. If your detectors have a usage attribute that is automatically used as a sort of offset, by default.

[Prashant_mahajan]

Okay, thank you. But how can I determine the effect of effort on lambda0 when the model output only provides the intercept?

[murray.efford]

We are assuming a known relationship: doubling effort doubles the expected number of detections. How could it be otherwise? See Methods in Ecology and Evolution 2013, 4, 629–636. If you want a more clunky approach you can fit a detector-level covariate, but remember to suppress 'usage' (details = list(ignoreusage = TRUE)).

[Prashant_mahajan]

Okay great. Thank you so much.