Issue with SMR models in secr

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

Issue with SMR models in secr

Postby pstrampelli » Wed Jan 29, 2020 6:13 am

Dear Murray,

I hope this finds you well.

My name is Paolo, and I am a PhD student attempting to estimate African lion density with camera trap data from a number of purpose-built grids in Tanzania. I posted here once in the past, but with a different account. I’m writing as I’ve run into an issue with the mark-resight models.

I initially employed classic secr models, with no issue. However, due to the relatively high number of individuals which could not be IDd, I decided to fit mark-resight models, as per Rich et al. 2014 & 2019. Based on the mark-resight guide, and other threads on here, I treated it as a sighting-only dataset, with unknown pre-marking (knownmarks=FALSE), and without specifying pID (so model 3A in Table 1 in the guide). I treated all unmarked as marked but unknown (nonID, so Tm), with no unmarked (Tu) file specified.

I ran the models both with and without the ‘knownmarks’ parameter (as I forgot to include this the first time). However, both of these versions seem to have an issue. The density estimate from the model without specifying that knownmarks=FALSE is realistic (based on the results from the secr model), but the SE is 0. The results from the run with knownmarks=FALSE are even more strange, with parameter estimates being NA.

I re-ran the analysis with a bigger buffer, in case that could be a cause of the issue, but this did not make a difference. I am running secr 3.2.1.

Any idea why this could be happening? Even though I’ve played around with it for a while I’m a bit stumped – so would be extremely grateful for any help you could provide on this! The data is 'normal', with a five '1s' and two '2s' in the Tm file, so nothing that should be too strange. I’ve copied the outputs from both runs below, and would of course be happy to send over the input files if that might be of help.

Thanks so much and all the best,
Paolo


-----------------------------------------------------------------------------------------------------------------------
secr.0 (without knownmarks=FALSE)
-----------------------------------------------------------------------------------------------------------------------
secr.fit(capthist = Rungwa.lion.SMR.v1.capthist, model = list(g0 ~
1, sigma ~ 1, D ~ 1), buffer = buffer, detectfn = "HN",
binomN = 1, hcov = "V5", details = list(minprob = 1e-200),
trace = FALSE)
secr 3.2.1, 12:40:37 28 Jan 2020

Detector type proximity (101)
Detector number 40
Average spacing 3457.8 m
x-range 630379 660060 m
y-range 9228623 9258932 m

Usage range by occasion
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41
min 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1
max 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79
min 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
max 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101
min 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 0 0
max 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1

Marking occasions
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

N animals : 24
N detections : 70
N nonID sghting : 19 (c-hat 1)
N occasions : 101
Mask area : 480519.8 ha

Model : D~1 g0~1 sigma~1 pID~1 pmix~h2
Mixture (hcov) : V5
Fixed (real) : none
Detection fn : halfnormal
Distribution : poisson
N parameters : 5
Log likelihood : -553.658
AIC : 1117.316
AICc : 1120.649

Beta parameters (coefficients)
beta SE.beta lcl ucl
D -8.299959 0.0000000 -8.299959 -8.299959
g0 -5.463857 0.2398960 -5.934044 -4.993669
sigma 9.540185 0.1359532 9.273721 9.806648
pID 6.740249 0.2585893 6.233424 7.247075
pmix.h2M 1.500000 0.0000000 1.500000 1.500000

Variance-covariance matrix of beta parameters
D g0 sigma pID pmix.h2M
D 0 0.000000e+00 0.000000e+00 0.000000e+00 0
g0 0 5.755007e-02 -1.925953e-02 -3.806325e-05 0
sigma 0 -1.925953e-02 1.848327e-02 1.070616e-05 0
pID 0 -3.806325e-05 1.070616e-05 6.686842e-02 0
pmix.h2M 0 0.000000e+00 0.000000e+00 0.000000e+00 0

Fitted (real) parameters evaluated at base levels of covariates

session = 1, h2 = F
link estimate SE.estimate lcl ucl
D log 2.485270e-04 0.000000e+00 2.485270e-04 2.485270e-04
g0 logit 4.219305e-03 1.007923e-03 2.640760e-03 6.735070e-03
sigma log 1.390752e+04 1.899542e+03 1.065433e+04 1.815404e+04
pID logit 9.988190e-01 3.050220e-04 9.980411e-01 9.992883e-01
pmix logit 1.824255e-01 0.000000e+00 1.824255e-01 1.824255e-01

session = 1, h2 = M
link estimate SE.estimate lcl ucl
D log 2.485270e-04 0.000000e+00 2.485270e-04 2.485270e-04
g0 logit 4.219305e-03 1.007923e-03 2.640760e-03 6.735070e-03
sigma log 1.390752e+04 1.899542e+03 1.065433e+04 1.815404e+04
pID logit 9.988190e-01 3.050220e-04 9.980411e-01 9.992883e-01
pmix logit 8.175745e-01 0.000000e+00 8.175745e-01 8.175745e-01

-------------------------------------------------------------------
secr.0 (with knownmarks=FALSE)
-------------------------------------------------------------------

secr.fit(capthist = Rungwa.lion.SMR.v1.capthist, model = list(g0 ~
1, sigma ~ 1, D ~ 1), buffer = buffer, detectfn = "HN",
binomN = 1, hcov = "V5", details = list(minprob = 1e-200,
knownmarks = FALSE), trace = FALSE)
secr 3.2.1, 13:20:07 28 Jan 2020

Detector type proximity (101)
Detector number 40
Average spacing 3457.8 m
x-range 630379 660060 m
y-range 9228623 9258932 m

Usage range by occasion
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41
min 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1
max 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79
min 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
max 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101
min 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 0 0
max 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1

Marking occasions
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

N animals : 24
N detections : 70
N nonID sghting : 19 (c-hat 1)
N occasions : 101
Mask area : 480519.8 ha

Model : D~1 g0~1 sigma~1 pID~1 pmix~h2
Mixture (hcov) : V5
Fixed (real) : none
Detection fn : halfnormal
Distribution : poisson
N parameters : 5
Log likelihood : -1e+10
AIC : 2e+10
AICc : 2e+10

Beta parameters (coefficients)
beta SE.beta lcl ucl
D NA NA NA NA
g0 NA NA NA NA
sigma NA NA NA NA
pID NA NA NA NA
pmix.h2M NA NA NA NA

Variance-covariance matrix of beta parameters
NULL

Fitted (real) parameters evaluated at base levels of covariates

session = 1, h2 = F
link estimate SE.estimate lcl ucl
D log NA NA NA NA
g0 logit NA NA NA NA
sigma log NA NA NA NA
pID logit NA NA NA NA
pmix logit NA NA NA NA

session = 1, h2 = M
link estimate SE.estimate lcl ucl
D log NA NA NA NA
g0 logit NA NA NA NA
sigma log NA NA NA NA
pID logit NA NA NA NA
pmix logit NA NA NA NA
pstrampelli
 
Posts: 3
Joined: Tue Jan 28, 2020 9:58 am

Re: Issue with SMR models in secr

Postby murray.efford » Wed Jan 29, 2020 7:33 am

To me it seems unlikely you'll get anything sensible from a mark-resight analysis with no correctly identified unmarked animals.
murray.efford
 
Posts: 686
Joined: Mon Sep 29, 2008 7:11 pm
Location: Dunedin, New Zealand

Re: Issue with SMR models in secr

Postby pstrampelli » Wed Jan 29, 2020 11:54 am

Hi Murray,

Thanks for your reply.

I am sorry, I don't understand exactly what you mean. I have identified 24 animals - as I would for a normal secr analysis - and, in addition, I am supplementing this by including information on marked but unknown animals (through the Tm file). This seems to me what Rich et al. 2014 and 2019 did (treating all unidentifiable animals as marked but unknown, and not having any as unmarked) - and which is what I am trying to implement here

This is also what I understood you had recommended to another poster here, who was running an analyses on badgers, where you said: "If the natural marks on some are inadequate for them ever to be identified then you have a pre-marked mark-resight study with unknown fraction marked (model 3 or 3a in that scheme)" - which, again, is what I am going for. The majority of pictures are of identified animals, but I am trying to include the ones of the non-identifiable ones.

Apologies for the confusion and thanks again for your help.

All the best,
Paolo
pstrampelli
 
Posts: 3
Joined: Tue Jan 28, 2020 9:58 am

Re: Issue with SMR models in secr

Postby murray.efford » Wed Jan 29, 2020 4:18 pm

In Rich et al 2014 I read "Counts of unmarked individuals are used to inform
detection parameters and the proportion of marked but not identifiable photos per known individual is used as a correction factor for the encounter rates of marked individuals".
murray.efford
 
Posts: 686
Joined: Mon Sep 29, 2008 7:11 pm
Location: Dunedin, New Zealand

Re: Issue with SMR models in secr

Postby pstrampelli » Thu Jan 30, 2020 8:02 am

Hi Murray,

Thanks for your reply.

I must be confusing myself. I think I was thrown off from by your message above, where you said that "it seems unlikely you'll get anything sensible from a mark-resight analysis with no correctly identified unmarked animals."

I don't understand what you mean by identifying unmarked animals - isn't the whole point of these particularly models to incorporate data from the animals in the study which cannot be identified at an individual level?

My intention was to do as Rich et al. (2019) did: "To implement SMR models for these species, we had to classify unresolved detections as (1) marked but unidentifiable or (2) unmarked. Making this distinction proved to be extremely difficult, so we adopted an approach aimed at minimizing positive bias (i.e., the chance of misidentifying a marked individual as an unmarked individual). For lions, we classified all unresolved detections as marked but unidentifiable."

So my understanding of the above is that they classified all unmarked individuals as marked but unidentifiable (Tm) and none as unmarked (Tu). This is what I tried to do (which led to the issues detailed above), but your first message seems to imply that I need to identify unmarked animals? I am not sure I fully understand this.

I apologise for the further confusion.

All the best,
Paolo
pstrampelli
 
Posts: 3
Joined: Tue Jan 28, 2020 9:58 am

Re: Issue with SMR models in secr

Postby murray.efford » Thu Jan 30, 2020 4:46 pm

By 'identify' I really meant 'classify'. Observations will comprise counts of known marked animals (some identified to individual, some not) and known unmarked animals.
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 10 guests