estimating Relative Species Richness

[gbferreira]

Hi all. I have some questions on the use of Presence to estimate relative species richness (RSR). I’ve found some topics on this issue, but none of them addressed my doubts.

I want to use the occupancy approach to understand the effect of local site covariates on large mammals’ RSR in a protected area. More specifically, I want to see if the type of land use (3 different classes of land use) before the establishment of the protected area affect species richness. I’m fairly familiar with single season, single species occupancy analysis on Unmarked, but as far as I understood it is not possible to use site characteristics (in my case, land use in the past) to model species richness in this package (although you can use species features as detection covariates, such as trophic guild or mass).

I’m trying to run the analysis using Presence, but my main difficult is on how to incorporate land use as an occupancy (or RSR) covariate. As my matrix is transposed for the RSR analysis, I have the regional species pool (28 species) as rows and each camera trap site (n=50) as columns. So, in my Presence data form, species features (mass and trophic guild) are in the site covariate table, whereas land use in each camera trap site is in the sample covariate table.

I would like to run the following models:
RSR(.),p(.)
RSR(.),p(mass)
RSR(.),p(guild)
RSR(land use),p(.)
RSR(land use),p(mass)
RSR(land use),p(guild)

I was able to the models with no covariate for RSR, but couldn’t run the other models - I imagine that’s because land use are in the sample covariate table and, thus, I cannot model it as RSR covariate.

So my questions is: How do I assemble the “Design Matrix” window to be able to incorporate land use as a RSR covariate? Or am I doing something wrong in the initial steps and should change my approach?

Thanks in advance for the attention.
Gee.

[darryl]

Hi Gee
The way you've set up your data, you are essentially estimating richness for the entire set of camera sites, ie 'richness' is the number of species present somewhere across the 50 camera-trap sites. This doesn't allow you to look at 'local richness' at an individual site. You could do that though in PRESENCE if you enter your camera trap data at some sort of finer-scale rather than that detected or not across the whole period cameras were out, e.g., use daily or weekly detection/nondetection for your cameras (just like people often do for camera-trap data).

The way to set your data up would be to create a single-species detection matix for each species (ie 50 rows and K columns, where K is the number of periods you divide your camera trap data into), then enter all of the data into a single data file (ie 28*50 rows and K columns). You can then use covariates to keep track of which rows of data belong to which species as a 'site' covariate, then also the local site covariates which will obviously be repeated 28 times. You should be able to use the same concept in unmarked as doing it within R would likely be much easier as in PRESENCE you will need to set up an indicator variable for each species to allow for species-level effects in detection etc., which can be a bit of a pain to do (but is possible).

Doing it within OpenBUGS or JAGS would even be better as you have a finite number of species of interest.

Cheers
Darryl

[gbferreira]

Hi Darryl. Thanks a lot for your answer. It was really useful, but with new knowledge comes new doubt… I did as you suggested and assembled a matrix of 28*50 rows and 9 one-week sampling occasions. I now have two questions:

1) Should I do this analysis considering the regional species pool (28spp) or only the species actually recorded (18spp)?
I did both ways and I found the results for the 28spp matrix a bit awkward when I use the model with land use covariates. The psi for each matrix and each land use type are as follows:

Regional pool (28spp) matrix: type A 0.26; type B 0.32; type C 0.43
Recorded species (18spp) matrix: type A 0.45; type B 0.30; type C 0.42

There is a large decrease in psi for land use type A from one matrix to another, while psi for the other types is roughly the same. The results from the 18spp matrix resemble more closely the observed data, where the highest richness was observed in land use type A.

I wonder if this decrease was not artificially caused by adding to the matrix 10 species from the regional pool that was not recorded, resulting in 500 (10spp*50sites) rows of no records (only 0 or NA). Because each type of land use had a different number of camera traps (type A: 25 sites; B: 15 sites; C: 10 sites), these rows of no records are not equally divided between land uses, and land use A “gets” half of them. Which could explain the reason why in the 28spp matrix this land use has the smallest psi and in the 18spp matrix it has the largest psi.
If this is the case, I believe I should use the 18spp matrix (only recorded species). Right?

2) How should I interpret psi in this analysis?
For the RSR analysis where the matrix is transposed (i.e. species are sites and sites are occasions), psi is the proportion of the regional species pool that occur in the study area. But in this new matrix I`m not sure what psi means. Is the psi value the proportion of sites occupied by any of the species? Or the proportion of the regional species pool in a given site? Or something else?

Sorry for the long message, I tried to keep it as short as possible. Thanks, Gee.

[darryl]

Hi Gee,
1) It's up to you, how do you want to define the species pool? If you use the 18 species then (theoretically), you could have species-specific occupancy and detection probabilities (as well as habitat effects). If you use the 28, then you have to make some assumptions about occupancy and detection for those species never seen anywhere. The psi is lower with 28 species isn't worrying to me. It's basically suggesting that those additional species where probably absent. Note that 0.26*28 is close to 0.45*18, i.e., the number of species is similar.

2) psi is interpreted as normal, probability of a species begin present at a site. You can get site-level richness by adding up the psi-hat values for each species at a particular site. Using the conditional psi values at the end of the PRESENCE output (Pr(occupancy|detection history)) would be even better. Note taht you can include species-level covariates with this approach as well.

3) note that a slightly better way of doing this within PRESENCE could be to use the multi-scale occupancy model (called multi-method in PRESENCE), where psi is Pr(species in pool), theta is Pr(species present at site| in pool) and p is detection in survey. Data format would be 1 row per species then K columns for site 1, K columns for site 2, etc. So data matrix would be 28 rows, and K*50 columns. You'll have to think carefully about how you set up the covariates and the design matrices, but and advantage of this approach is a bit more reality in that if a species is not in the pool, it has to be absent from all sites by definition.

Cheers
Darryl