CTDS-bootdht return differs from dht2

[Amira Salom]

Hi everyone!

Im using camera trap distance sampling (CTDS) to estimate the density of two large mammals (red deer and goats). I have several doubts in some issues regarding my outputs.

With one species (goats) everything seems to be working fine, but when estimating density with a geographical stratification I obtain different total density estimates from dht2 and bootdht (which I run to obtain the 95% CI).

dht2:

Density estimates:

 ambiente Estimate    se    cv    LCI     UCI     df

   bosque   0.6794  0.385 0.567 0.2285  2.0201 24.055

   junco   0.5793   0.756 1.305 0.0552  6.0763  7.117

 pastizal   4.1889   2.121 0.506 1.1733 14.9546  4.477

  peatbog   1.0838 0.565 0.521 0.3922  2.9944 22.238

    Total   1.0551 0.335 0.317 0.5645  1.9723 40.654

bootdht:

median mean   se  lcl   ucl   cv

bosque     0.69 0.78 0.47 0.13  1.91 0.67

junco      0.57 0.62 0.65 0.00  2.15 1.16

pastizal   4.24 4.50 3.34 0.29 11.86 0.79

peatbog    1.05 1.11 0.64 0.11  2.55 0.61

Total      6.75 7.01 3.62 1.55 14.94 0.54

As you can see habitat estimates are similar, just total density is very different. Does dht2 and bootdht have different ways of estimating this?

Another (probably silly) question is which estimates to report, the dht2 density estimates and the bootdht 95% CI and CV? Or only the 95% CI from the bootdht?
  
My second issue arise with red deer estimates. While going through the selection process of a detection function I inspect out of curiosity the density estimates i would have obtain for each detection function tested using dht2 (for both species). For red der i found estimates could vary a lot, here some examples:

with uniform with 1 ajustment term: D=4.5, 95% CI=3.09-6.56

halfnormal with 1 adjustment term: D=17.52, 95% CI=12.02-25.55

hazard rate with 1 adjustment term: D=24.16, 95% CI=16.55-35.25

I checked the duiker results from the CTDS example an estimates doesnt differ so much with the detection function selected and I dont find the same issue with my other species. In my study site red deer is reactive to the cameras (mainly attraction), and therefore distance observations of this behaviour were discarded it and now im going for the method recommended in Delisle et al. 2023 and directly ignoring HR key functions. Nonetheless, there is still the issue on just how different my estimates are between the reamining key functions. It makes me doubt just how much i can trust my estimates. Have some had this isue before? Any hint of what could be hapening and how relevant this could be?

I hope I have made myself clear.
Thanks in advance,

Amira Salom

[Eric Rexstad]

Amira

Welcome to the list; there is a lot to unpack in your questions.

Starting with your question regarding your estimates for goats. From the output you provided, it appears dht2​ has provided a density estimate for the entire study area that is a weighted average, with weighting (presumably) by stratum size. This is what is supposed to happen when strata are geographically defined.

On the other hand bootdht​ has merely summed the stratum-specific density estimates and more absurdly, summed the standard errors, etc. Clearly the study area wide estimates coming from bootdht​ are not legitimate.

bootdht​ is naive about types of stratification; it simply knows how to resample camera stations within strata. It does not do anything sensible with regard to totals. Those results from bootdht​ should be disregarded.  We have a note of these challenges in our list of issues

https://github.com/DistanceDevelopment/Distance/issues/155

`bootdht` and `dht2` incompatibility · Issue #155 · DistanceDevelopment/Distance For analyses requiring dht2, e.g. non-geographic stratification or multipliers (not using the activity package, bootstrapping cannot be performed. This is because bootdht only accepts as input dsmo... github.com

Regarding what to report (stratum-specific estimates): use the point estimates from dht2​; use precision measures produced by bootdht​ (totals disregarded).

Red deer

I can't make definitive statements about results for this species without more details about the study design and detection distance distributions. Rather than make an assessment of the merits of a detection function from the density estimates they produce, far better to examine the shape of the fitted model and the detection distances. I don't have experience with censoring of detections based upon perceived responsiveness to the camera.

---

From: distance...@googlegroups.com <distance...@googlegroups.com> on behalf of Amira Salom <amira...@gmail.com>
Sent: 14 June 2024 14:58
To: distance-sampling <distance...@googlegroups.com>
Subject: [distance-sampling] CTDS-bootdht return differs from dht2

 

--
You received this message because you are subscribed to the Google Groups "distance-sampling" group.
To unsubscribe from this group and stop receiving emails from it, send an email to distance-sampl...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/distance-sampling/05f6a457-0f5f-4836-9dfb-6dd515ede3b3n%40googlegroups.com.

[Amira Salom]

Hi,
After showing Eric the detection distance distribution, he pointed out that my truncation distance was too big, and that stronger truncation should enhance the robustness of modelling  the detection function and therefore estimates from the various models should be closer together as a result.  As you can see, I was truncating at 20m, though detection probability was >0.1 beyond 12.5m.

I refitted the models with truncation at 12.5m and indeed estimates from the various detection functions became closer together.

Amira Salom / Becaria doctoral CONICET 
amira...@gmail.com

Centro Austral de Investigaciones Cientí­ficas (CADIC-CONICET) 
Bernardo Houssay 200Ushuaia - Tierra del Fuego - Argentina 
www.cadic-conicet.gob.ar/