Efford et al (2015) - tiger data?

questions concerning anlysis/theory using program DENSITY and R package secr. Focus on spatially-explicit analysis.

Efford et al (2015) - tiger data?

Postby leopard1 » Sat Jun 04, 2016 2:09 am

Dear Murray

I hope you are well and thanks for your continuous support in running this fantastic forum! :D

I recently read your tiger, ovenbird and possum paper and found it an absolutely intriguing reference and model for home-range related camera-trap problems. I'm currently trying to figure out how to apply this to leopards and am re-running your simulations to get a grip on the analysis.

It seems I can get the ovenbird and possum data to work perfectly, however I am struggling to get the tiger data to run! Is the data actually included in the supplementary section of your Ecography paper? The lines of code below don't seem to allow me access to re-run the code:

## Load package ‘secr’ and data (previously prepared multi-session capthist
## object CHc28 and list of masks masks28)
library(secr)
load (’dataobjects.RData’)
## Fully reserve-specific parameters
## -- output is a list of fits, rather than an ‘secr’ object
fit.0c28 <- mapply(secr.fit, CHc28, masks28, SIMPLIFY = FALSE, trace =
FALSE, MoreArgs = list(model = list(D~1, lambda0~1, sigma~1),
detectfn = ’HHN’, verify = FALSE, start = c( -7.7, -4.2, 8.1),
binomN = 1))
## extract estimates from reserve-specific model fits
lapply(fit.0c28, predict)
## Fully reserve-specific, with k parameterization
fit.0c28k <- mapply(secr.fit, CHc28, masks28, SIMPLIFY = FALSE, trace =
FALSE, MoreArgs = list(model = list(D~1, lambda0~1, sigmak~1), detectfn =
’HHN’, verify = FALSE, start = c( -7.7, -4.2, 4.2), binomN = 1))
## extract estimates from reserve-specific model fits
lapply(fit.0c28k, predict)
## Multi-reserve model with constant sigmak
fit.0csl28 <- secr.fit( CHc28, mask = masks28, model = list(D~session,
lambda0~session, sigmak~1), detectfn = ’HHN’, verify = FALSE, binomN = 1)
predict(fit.0csl28)

Best

Alex
leopard1
 
Posts: 4
Joined: Sun Aug 25, 2013 10:04 am

Re: Efford et al (2015) - tiger data?

Postby murray.efford » Sat Jun 04, 2016 2:18 am

The raw tiger data (dataobjects.RData) were not provided with the paper - only the R code. I'm sorry to disappoint, and you're not the first to notice this. If you have a research interest in tigers that requires those data then Drs Jhala and Qureshi may be able to help.
Murray
murray.efford
 
Posts: 686
Joined: Mon Sep 29, 2008 7:11 pm
Location: Dunedin, New Zealand

Re: Efford et al (2015) - tiger data?

Postby oksana.grente » Fri Apr 21, 2017 12:45 pm

Dear Murray Efford,

As Alex, I read your paper about "Density-dependent home-range size revealed by spatially explicit capture–recapture" and I found it very interesting. I am trying to apply this model to data on a jaguar population in Belize, with 13 annual surveys. However, I wonder about three things:

1. What is the difference between running your model with the hazard half-normal detection function (HHN) or with the half-normal detection function (HN) ? You said "we specify the model in terms of lambda(d) rather than g(d) because lambda(d) is expected to have a linear relationship to the probability density of the activity distribution" . Does the use of HHN only simplify the computation of the model? Or does the model with sigmak only work with HHN? I ran the models with my data and with the two detection functions HHN and HN, and for HN I had the following warning message:
In new.param(details, model, CL) : Using parameterization details$param= 4

2. I obtain k=54.46 for constant sigmak model (with HN detection function). Because density is given in ha in secr pacakge, does it means that k is in fact 0.5446, to relate density in km^-2 with sigma in km? Which would be pretty close of the k you obtain for the tiger dataset.

3. For the following models, iteration limit exceeded:

fitA <- secr.fit(mydata, mask = mymask, model =list(D ~ session, sigma ~ session, lambda0 ~ 1), detectfn = ’HHN’)

fitB <- secr.fit(mydata, mask = mymask, model =list(D ~ session, sigmak ~ session, g0 ~ 1), detectfn = ’HN’)

I had not this problem for the following model:

fitC <- secr.fit(mydata, mask = mymask, model =list(D ~ session, sigmak ~ 1, g0 ~ 1), detectfn = ’HN’)

Does this mean that fitA and fitB results are not informative because the models are too complex to compute with our dataset? However, I found weird that the number of reached iterations is not exactly the same.

Thank you in advance.

Best wishes,

Oksana Grente
oksana.grente
 
Posts: 1
Joined: Thu Apr 20, 2017 1:34 pm

Re: Efford et al (2015) - tiger data?

Postby murray.efford » Tue Apr 25, 2017 5:31 am

Hello Oksana

Thanks for your interest in our Ecography paper. I'm sure there is lots more to learn about how these parameters co-vary in different populations.

Trying to address your specific questions:

1. My motivation for using detection functions expressed in terms of hazard (e.g. HHN) rather than probability (e.g., HN) concerns interpretation rather than computation per se. It seems to me that the hazard of detection should scale with the probability density of the activity distribution. It seems just a little clumsy to model the probability of detection (a non-additive quantity) as a function of the probability density of activity. For most purposes the distinction is aesthetic (doing what seems most elegant).

2. Yes, the estimate value for k (i.e. sigmak) needs to be divided by 100 because of the different units used for density and sigma in 'secr'. The possible match between jaguar and tiger is interesting!

3. I can't say anything much about your model fitting problems. I think models with sigmak ~ session and sigma ~ session are equivalent. I don't often fit models with so many parameters, and I don't remember reaching the iteration limit. If you really want to pursue that full model I suggest first fitting each session separately with lapply(mydata, secr.fit, ...) and then using the session-specific estimates of D and sigma or sigmak as starting values. That's just a suggestion, and it will be fiddly to construct the vector of start values.

Regards
Murray
murray.efford
 
Posts: 686
Joined: Mon Sep 29, 2008 7:11 pm
Location: Dunedin, New Zealand


Return to analysis help

Who is online

Users browsing this forum: No registered users and 3 guests