Low inherent detection probability despite many detections

[gausec21]

Hi all,

I am an M.S. student, and I used a Royle-Nichols model in PRESENCE to estimate mean site abundance of a species (to account for heterogeneity in detection probability). I surveyed 58 points 3 times in one season (98 positive detections and 76 surveys with 0 detections). The majority of my sites had at least 1/3 surveys with a positive detection.

Summary of detection histories:
000, 001, 010, 011, 100, 101, 110, 111
Frequency 12, 3, 2, 5, 9, 4, 3, 20

Proportion of sites with at least one detection: 0.79
Frequencies of sites with detections:
sampled detected
Season-1 58 46


My top model has three site covariates and detection probability is constant.


The first thing I noticed is that the output for my model is missing site-specific detection probabilities (r). I believe this was included in previous versions of the software, but I could be wrong. I do have three estimates (I am assuming 1 per survey) of c(1) (all are ~0.12) which I think is detection probability.

Output from PRESENCE:
============================================================

Individual Site estimates of <c(1)>
Site , estimate , Std.err , 95% conf. interval
c(1) 1 1 , : 0.1168 , 0.1143 , 0.0149 - 0.5369
c(2) 1 1 , : 0.1168 , 0.1143 , 0.0149 - 0.5369
c(3) 1 1 , : 0.1168 , 0.1143 , 0.0149 - 0.5369

============================================================
Am I interpreting this correctly in understanding that the inherent detection probability for my species is 12%? I find this value surprisingly low. In a prior study within the same region using PRESENCE and the same model for the same species, the researchers obtained much higher estimates of site-specific inherent detection probabilities despite having fewer survey points and far fewer positive detections.

Out of curiosity, I also ran the model in R using the unmarked package and obtained very similar results. Is it possible that I made an error in my analysis?

Thank you!
Carol

[jhines]

The "c" parameter is not the species detection probability. It's the probability of detecting an individual of the species. So, if there are 10 individuals available for detection at a site, the species detection probability would be 1 - (1-c)^10 (ie., 1 - prob all individuals were missed). If c=.14, that would be 1 - (1-.14)^10 = 1 - .86^10 = 0.78. If you model the detection parameter (c) as survey-specific with a covariate, you should get site/survey-specific estimates in the output. The lambda parameter is the estimate of the number of individuals at each site.

The "r" parameter is the same sort of detection probability for the Royal point-count model. That models uses the counts at each site/survey, instead of just presence/absence as in the Royal-Nichols heterogeneity model. The models are similar, but have different assumptions.

[gausec21]

Thank you, and I apologize if I am not understanding correctly:

Individual/ inherent detection probability (r) is ~12% for all sites?

The study I am hoping to compare to (that also used PRESENCE) reported estimates of site-specific individual/ inherent detection probability (r) from the Royle/Nichols Abundance Induced Heterogeneity Model and site-specific species detection probability (p) from a a simple single-season occupancy model. I would like to report the same estimates from my models, but I am struggling to understand how to estimate site-specific r.

They reported two estimates of r, one for each of their survey areas (each comprising approximately 20 survey points that were surveyed 3x each). Specifically, Location #1 had an individual detect. prob of 0.42, while Location #2 r= 0.62. Both of these estimates are much higher than 0.12. I assume they calculated site estimates by averaging the 20 site-specific individual detection probabilities, or could I solve for r using 1 - (1-r)^Ni where Ni is an individual site estimate of λ?

I think I might be wrong, and site abundance (Ni) can't be interpreted as mean site abundance (λ).

Thank you again for your help.