Detection species interaction factor

questions concerning analysis/theory using program PRESENCE

Detection species interaction factor

Postby Karla_Garcia » Thu Aug 31, 2017 6:03 pm

Hello!
I'm using two-species multi season occupancy models to study species interaction with the psiBa/rBa parameterization. I have 33 sampling points, that have been surveyed 3 times within each season during year. I have built models testing different scenarios, evaluating within-season effects and an individual detection covariate.

I initially limited the analyses to patterns of detection because the site spacing was not distant enough to ensure spatial independence between survey units for species A. For this reason, parameters for occupancy probabilities (i.e. ΨA, ΨBA and ΨBa) were modelled independently. However, I'm not sure if it's right to build models testing different scenarios for psiBA and psiBa, given that for species B was ensure spatial independence between sampling points.

I built models that assumed that the detection probability of the subordinate species was influenced by the presence (pB≠rBa; pB≠rBA) or detection (rBa≠rBA) of the dominant species or was independent of the dominant species (pB = rBa = rBA). I also evaluate a set of models where, give than both species are present, the detection of the dominant species was influenced by the detection of the subordinate species (pA≠ rA) or independent (pA= rA). Preliminary results showed that the detection of the both species was independent of the detection of the other species (i.e., pA = rA, pB=rBa=rBA). However, I found no consistent interaction in the detection process at sites occupied by both species (δ = 0.29). I calculated the detection species interaction factor (δ) adapted from the formula for phi from Richmond et al (2010):

δ=(rA*rBA)/(rA*((rA*rBA)+((1-rA )*rBa))

It’s right? /Is there any other way of obtaining detection species interaction factor?

Thank you very much in advance for any comment!
Karla_Garcia
 
Posts: 4
Joined: Wed Jan 22, 2014 5:28 am

Re: Detection species interaction factor

Postby jhines » Tue Sep 12, 2017 3:37 pm

Hi Karla,

The formula for delta looks correct.

What do you mean by "no consistent interaction in the detection process" with delta=.29? I assume that value came from a model where rBA not = rBa, although you said that the results indicated that pB=rBa=rBA. A delta value of 1.0 meaans no interaction. Values < 1.0 mean the detections of both species are fewer than expected than if detections were independent.

You can have PRESENCE estimate the species interaction factor (phi for occupancy, delta for detection) by using the "phi/delta" parameterization (top option button in the run dialog window).

Jim
jhines
 
Posts: 599
Joined: Fri May 16, 2003 9:24 am
Location: Laurel, MD, USA

Re: Detection species interaction factor

Postby Karla_Garcia » Mon Sep 18, 2017 10:32 pm

Thanks Jim!
In fact, the results of the best model indicated independence between the probabilities of detection (pA = rA, pB=rBa=rBA), however, when estimating delta showed a value <1.0, suggesting that the detections of both species are less than expected than if detections were independent (i.e. avoidance). I don't know why this incongruity. I'll try the phi/delta parameterization option.
Karla_Garcia
 
Posts: 4
Joined: Wed Jan 22, 2014 5:28 am

Re: Detection species interaction factor

Postby Karla_Garcia » Tue Oct 03, 2017 10:44 pm

Hi again,
I'm using the phi/delta parameterization option, since my interest is to determine the detection species interaction factor (delta) and its variation according to the seasonality. However, and after building different models opcion, the best model (according to AIC) suggest the following estimates for the "p" of one of the species:
pA [1] 1 site1: 0.6733 0.1420 0.3677 - 0.8796
pA [2] 1 site1: 0.6733 0.1420 0.3677 - 0.8796
pA [3] 1 site1: 0.6733 0.1420 0.3677 - 0.8796
pA [4] 1 site1: 0.0000 0.0000 0.0000 - 1.0000
pA [5] 1 site1: 0.0000 0.0000 0.0000 - 1.0000
pA [6] 1 site1: 0.0000 0.0000 0.0000 - 1.0000
pA [7] 1 site1: 0.0563 0.0855 0.0025 - 0.5828
pA [8] 1 site1: 0.0563 0.0855 0.0025 - 0.5828
pA [9] 1 site1: 0.0563 0.0855 0.0025 - 0.5828
pA [10] 1 site1: 0.0000 -1. # IND 1. # QNB - 1. # QNB
pA [11] 1 site1: 0.0000 -1. # IND 1. # QNB - 1. # QNB
pA [12] 1 site1: 0.0000 -1. # IND 1. # QNB - 1. # QNB

In this case, the model only correctly estimates the "pA" for two of the seasons (winter and summer), while for the other two (spring and autumn) it shows these values.

What should I do in these cases? ignore this model?
Thank you in advance for the help
Karla_Garcia
 
Posts: 4
Joined: Wed Jan 22, 2014 5:28 am

Re: Detection species interaction factor

Postby jhines » Wed Oct 04, 2017 7:42 am

Hi Karla,

In response to your previous post... you gave a value of delta of 0.29, but the best model indicated independence of detection probabilities. Did the confidence interval for delta include 1.0? Even though the estimate was < 1.0, if the confidence interval includes 1.0, then that model is not giving strong evidence that the detection probabilities are not independent.

A value of zero for pA is usually due to either lack of data (no sites where only species A occurs), or no detections of species A when only species A occurs. In the first case, you have no information about pA for those surveys. This could happen if species B occurs everywhere species A occurs. In that case, I'd suggest reporting "NA" for detection probability (pA) in those surveys.

In the 2nd case, the estimate of zero is correct, based on the data. With an estimate of zero, the standard error is undefined.

I see no problem using this model, based on the output listed if you can verify the cause of the zeros. If you'd like to send me your data off-list (jhines@usgs.gov) I'd be happy to double-check it.
jhines
 
Posts: 599
Joined: Fri May 16, 2003 9:24 am
Location: Laurel, MD, USA

Re: Detection species interaction factor

Postby Karla_Garcia » Wed Oct 04, 2017 5:13 pm

Hi Jim,

Thank you very much for your prompt response and help.

Indeed, I have reviewed the data, and for spA no sites were recorded where only A occurs, at least for three season. There is only one site during winter where only A occurs. Therefore, an estimate of pA is only expected for winter (the model also showed a pA for summer, although with a very low value pA = 0.06). In this case, as you suggest, I could report "NA" for detection probability (pA) in those surveys.

I have also sent you the data so that you can double-check it. Thank you again.
Karla_Garcia
 
Posts: 4
Joined: Wed Jan 22, 2014 5:28 am


Return to analysis help

Who is online

Users browsing this forum: No registered users and 15 guests