Question about SECR DESIGN

[baptiste plana]

Hello, SECR forum,

I am currently working on SECR analyses regarding Timon lepidus. I have analyzed two years of camera trap data to determine the population parameters needed to run simulations. There is significant heterogeneity: 

$`session = mon_capthist2025, h2 = F` link estimate SE.estimate lcl ucl D log 1.24757584 0.340012498 0.73828565 2.1081887 g0 logit 0.09999775 0.009908712 0.08218375 0.1211633 sigma log 30.20771809 1.256025540 27.84457305 32.7714212 pmix logit 0.66709476 0.081079217 0.49487733 0.8038668 $`session = mon_capthist2025, h2 = M` link estimate SE.estimate lcl ucl D log 1.24757584 0.340012498 0.73828565 2.10818870 g0 logit 0.03491133 0.003607674 0.02849066 0.04271534 sigma log 100.76462322 5.849364751 89.93682867 112.89601204 pmix logit 0.33290524 0.081079217 0.19613315 0.50512267 $`session = mon_capthist2026, h2 = F` link estimate SE.estimate lcl ucl D log 1.2071715 0.29727525 0.7502944 1.9422552 g0 logit 0.1271644 0.01323387 0.1034046 0.1554371 sigma log 33.1805161 1.40168419 30.5450500 36.0433737 pmix logit 0.6670948 0.08107922 0.4948773 0.8038668 $`session = mon_capthist2026, h2 = M` link estimate SE.estimate lcl ucl D log 1.20717154 0.297275254 0.75029435 1.94225523 g0 logit 0.04528549 0.003942436 0.03815676 0.05367175 sigma log 110.68105830 6.290766106 99.02222169 123.71260164 pmix logit 0.33290524 0.081079217 0.19613315 0.50512267

So today I’m looking to run simulations with different parameters in order to answer two very specific questions.

First:

- Which parameters of the spatial trap system influence the accuracy of SECR estimates of Timon lepidus density?

Second, for management and field action purposes, I aim to develop a standardized method that can be replicated at sites similar to the original one. To this end, I am seeking to:

- Identify the minimal setup required to obtain a density estimate with sufficient precision, using as few resources as possible.

However, this is all rather unclear, and I am completely lost among the various studies when it comes to establishing a relevant workflow that aligns with SECR analysis methods.

I have therefore focused on setting up simulations with a group effect on the populations to model this heterogeneity. I am also varying the densities to ensure the results are representative of other existing populations.

sim_final <- run.scenarios(

nrepl = 100, trapset = trapset_valides, scenarios = sim_lo, maskset = maskset_valides,

fit = TRUE, extractfn = exfn_sex,

fit.args = list(model = list(g0 ~ h2, sigma ~ h2), hcov = “group”,

details = list(fastproximity = TRUE)),

ncores = max(1, ncores_dispo - 1), byscenario = FALSE, seed = 85

)

As for the trap parameters, since there is no reference area for an area that could contain a population of the species, I am testing the following range of trap area sizes:

area_sigmaM <- (0.5, 1, 1.5, 2, 3, 4) * sigmaM (since males have the highest sigma), and I set the number of observations to 70 because I do not wish to vary this figure, as the camera traps remain installed throughout the monitoring season.

For spacing, I’m varying:

(20, 30, 40, 50, 70, 90, 120) in meters

So my number of sensors is determined by these variables

I’d also like to test different sensor layout patterns: square, rectangle, cluster, and staggered.

I’m not sure whether I should conduct a full factorial design for all variables or whether I should first analyze spacing and coverage, and then examine the influence of geometry.

To reduce my computation time, I’ll apply an initial filter to eliminate outlier designs:

En_min <- 5

En2_min <- 1

rotRSE_max_prefilter <- 0.30

I’m not sure if I’m on the right track, and I’d appreciate your guidance on the best decisions and the ideal methodology to follow. I’d welcome any advice and recommendations to help me refine my approach.

I apologize that this message doesn’t really address issues directly related to a coding error, etc. I hope you can still help me.

Thank you in advance for your valuable help and all your work on researching and documenting the packages.

Have a great day

Baptiste Plana

[Murray Efford]

Hello Baptiste

I sympathise with you regarding the lack of a standard workflow for SECR study design. I have my own perspective that is mostly laid out in Ch 8 of Efford (2025) The SECR book. Maybe that makes it more complicated than it needs to be, and the language and emphasis vary from author to author. Basically, you need a representative design that delivers enough initial captures n and recaptures r while avoiding bias from an array that is far too small or far too widely spaced. Probably you got there already. If you haven't already found it I recommend playing with the app at https://www.stats.otago.ac.nz/secrdesignapp/.

Some other scraps:

Your initial example doesn't convince me there is significant heterogeneity: you have fitted one h2 model, and the pmix parameter seems close to the default start value (worrying).

You seem to expect the optimal design to depend on local characteristics - that may be overthinking it, given other uncertainties (detection parameters, density etc.).

I don't exactly follow your concern about population areas. Are we talking buffer width, array size or what?

I know people have had trouble wrangling group scenarios in secrdesign, so that is better left until you are fully confident.

Cheers

Murray

[baptiste plana]

Hello Murray, Thank you very much for your prompt reply – that’s really very kind of you.

Yes, indeed, I’ve been making the most of ‘Secrbook’, which is a real goldmine for the Secr.

From what I’ve been able to gather in practical terms, to create designs that work well for the Secr, it’s necessary to vary the spacing and extent so that individuals can be recaptured spatially and we don’t miss any movements due to a trapping area that’s too small. So can I take these two factors into account for my simulations without necessarily having to set a fixed number of detectors? Once the ideal extent and spacing have been defined, can I then vary my geometries, if I’ve understood correctly? The part of the simulations that’s unclear to me is which factors need to be fixed before the calculations, which factors will be determined by the others and therefore won’t be fixed, and whether I should group these variables or break them down into separate analyses. I drew particular inspiration from these studies(https://esajournals.onlinelibrary.wiley.com/doi/full/10.1002/ecs2.2217, https://www.researchgate.net/publication/256669243_Estimating_jaguar_densities_with_camera_traps_Problems_with_current_designs_and_recommendations_for_future_studies) in my attempt to establish a configuration logic, but I still cannot find anything robust that would enable me to meet my research objectives regarding the influence of design parameters on accuracy, and subsequently propose an optimal design for tracking the species.

To answer your question, I’ve chosen to use ‘groups’ in the models because the females have a sigma of 33 m whilst the males have a sigma of m. And I feel it’s necessary to take this into account in the simulations so as not to over- or underestimate the detection of a particular sex. Furthermore, in the multi-session models I was able to run previously, it was this one that explained 96 per cent of my data: D~session g0~session + h2 sigma~session + h2 pmix~h2.

I was indeed referring to the size of the network required to effectively cover the population under study.

Thank you for your reply and your help,

Have a great day

Baptiste 

[Murray Efford]

Hello again

1. Yes, you need a broadly suitable spacing and extent, but once you have found one I doubt you need to vary it. The methods are fairly robust, and seeking strict 'optimality' is futile as the exact behaviour of the animals (and lambda0 and sigma) is unknown.

2. Fixing spacing and extent fixes the number of detectors, given a particular geometry (e.g. square grid).

3. By all means draw inspiration from those studies, but understand that what matters is the amount of data (n,r), so long as you don't break into the pathological parts of the design space. Simulation may not even be needed.

4. I stand by my earlier comment on heterogeneity. Probably it is there, but does it matter?

Murray