Incorrect parameter estimates when estimating "richness

[bayne]

Hi All.

I have been using Presence to estimate species richness while accounting for detectability. I have 120 sites with 6 visits each with 60 species detected and 10 more that I think should exist in species pool but have not been detected in my sample. Many of the species are rare and have low detectability. I have tried running this with and without species with no detections (i.e. species that should exist in species pool but have not been detected in my actual data).

Running Presence using a model with species entered via dummy variable coding I have been able to get a solution that has a stable solution (i.e. Numerical convergence was reached - although variance/ covariance matrix was not computed successfully). However, many of the rare species (single detection quite often) have very high positive parameter estimates resulting in a predicted occupancy rate of 1 which is not possible.

Any suggestions on how to get realistic estimates of occupancy for these species? Talking the inverse for rare species gets you a value that is more realistic for rare species (i.e. occupancy rate much closer to zero) and provides a richness estimate that is believable. Ultimately I want to model the effects of habitat covariates and compare model fit using AIC but want to make sure that the fit derived simply for a "species matrix" is a valid starting point for comparison.

Thanks in advance for any advice

Erin Bayne

[darryl]

Hi Erin,
So are trying to fit a psi(Species)p(Species) model? If so, then it's possibly a sparse data issue (very few detections for individual species) so you might want to try grouping some species together and form covariates representing species groups. These groups could be just rare species, or species that have similar detectabilities when they are present at a location.

You do have to think carefully about exactly what you mean when say 'rare'. Something could be rare on the landscape, but locally abundant when it is there (low psi, high p); or common on the landscape, but low density (high psi, low p); or both (low psi, low p).

Cheers
Darryl

[bayne]

Hi Darryl, thanks for the input. Yes I am trying to create a Psi(spp) p(spp) model. Admittedly the species for which the sign seems "reversed" are those with low psi & low p so that it is likely a sparse data issue as the other types of rare you describe generally make sense.

One more question however. In the case of richness I could see creating p(spp_groupings) making sense but would you still have a complete species matrix for psi?

Theoretically I guess if you had ecologically equivalent species and collapsed them together the increased value of psi that would presumably count as "more species" in the richness estimator simply because you "increased the value of psi overall for those two species combined? However would this be a better estimate than leaving as individual species?

Thx

Erin