buffer choice for SECR

[Aditya Ghoshal]

Hi everyone, 

I am conducting secr analysis for leopard populations in different protected areas in southern India. Since these are highly mobile animals the suggest.buffer function gives me a value of ~9700m. I used the shapefile of the protected area (which is about ~1000 sq km in area) and used the "polybuffer" argument to make a mask around it of 10km. width (we had a lot of camera trap stations on the edge so the buffer area extends beyond the border of the park; using "trapbuffer" truncated it within the park boundary which I felt was not appropriate). The park is not fully fenced and there are other forested areas and agricultural lands around it thus leopards can move in and out of the park (Leopards are known to use human-use landscape for foraging on cattle and stray dogs). We detected 87 leopards in our survey and about 1/4th of them had only a single capture. So I think we have a decent dataset. But I am a bit unsure about the mask I am making. I have another option of delineating habitat non-habitat in a 10km radius around the park but I want to know whether that is required, given my primary objective is to estimate abundance and density of the leopards within the park. It was also suggested to me that it is better to estimate density without any habitat mask and to use it only for abundance estimation. Any thoughts or suggestions are welcome. 

P.S. I am using a hybrid model with sex covariate for analysis (using either hn or ex detection functions). 

Thank you, 

Thank you, 

[Murray Efford]

If the leopards you sample with cameras potentially have activity centres outside the park then you need to allow for that in defining the mask; otherwise density estimates will be biased upwards. The mask must be at least as large as the area from which they can come, but may be larger without biasing the estimates. A spatial model always has a mask or 'state space' (horrible term) although that may be constructed automatically in secr.fit by specifying the buffer argument.

If you ultimately want to estimate population size in a subset of the masked area then you can specify the subset area later.

[Aditya Ghoshal]

Thank you Murray. Actually the park itself is a part of a larger contiguous habitat, and includes three more parks that houses leopards and tigers and we do know animals are moving between these parks. Currently I am putting a polybuffer of 10km around my park to get a robust estimate of density and then  to getting an abundance estimate within the park boundaries. Is that a reasonable approach? 

2 further questions:

1. since my park boundary is of interest to me (actually the forest dept.) I am using the shapefile and creating a mask of type traprect (because that way all the habitat area within my park gets covered and nothing outside it) and using this to estimate the abundance (even though the model used a 10k polybuffer). 

2. The Estimated N within this smaller area is 50 individuals even though we detected 87 in out CT arrays. I am guessing this is because when I make a mask of 10k polybuffer, I am assigning 87 animals to this larger area and thus the abundance estimate within a smaller area is lower than the number of animals I detected. I am planning to explain this to the forest dept. as having about 50 "resident" leopards within the park boundaries but other animals still using the park area transiently. 

Is my approach and my reasonings correct? 

Thank you. 

[Murray Efford]

That sounds broadly correct to me. For a publication I would want to sharpen some of your terminology. In (1) we don't care how you achieve a map (mask) of habitat within the park, so long as it is correct (plot it and overlay the park boundary to be sure). In (2) the general idea is right, but the terms 'resident' and 'transient' have particular biological meanings that don't really fit here. Also, you cannot tell which animals are centred inside. Perhaps you are better to bite the bullet and talk about estimated number of animals with activity centres inside and outside the park, even though that notion may be new to the Forest Dept. The problem is common - I vaguely remember a case with tigers in south India where adjacent areas both 'claimed' all the animals that appeared on their cameras, even though the lists overlapped. Perhaps one of Karanth's studies.

[Aditya Ghoshal]

Thank you so much! I have been struggling with these doubts and how to explain things to the forest dept. for a while now. 

Regarding the overlap between parks, we are also working in the same state as Karanth did. There are multiple protected areas that are connected by narrow corridors. I'll steer away from using the terms resident to avoid any misinterpretations.

Thank you again for the help. 

--
You received this message because you are subscribed to a topic in the Google Groups "secr" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/secrgroup/Ep_nFUObdRM/unsubscribe.
To unsubscribe from this group and all its topics, send an email to secrgroup+...@googlegroups.com.
To view this discussion on the web, visit https://groups.google.com/d/msgid/secrgroup/49774fb3-d43b-4485-8875-0be4aa296925n%40googlegroups.com.

[Aditya Ghoshal]

Sorry for bugging you again but I have one small question. I read in one of the posts in phidot.org that for a half-normal function 95% of the movement is limited within 2.95 sigma. What is the function in secr package to get the values for each detection function. I have been searching and I couldn't find the post again (forgot to save it!)  

[Murray Efford]

I think you're looking for the secr function circular.r. The magic number is 2.45 or maybe 2.24, not 2.95. It all depends how you think about detection functions, and don't ask me to explain.

[Aditya Ghoshal]

Thanks! And I won't ask you to explain :-D . I decided to stick to half-normal functions anyway since that one is still easily explainable.