Methods to specify species covariance about mobsim HOT 3 OPEN

from mobsim.

Hey @jmchase sounds good! Mike Palmer mentioned to me that he really liked the mobr framework as well but he also saw this as a potentially big gap. So I think folks are going to be asking us about this as time goes on. I personally see it as a mob v2.0 priority. During initial brainstorms I recall that we had thought about this in two ways:

1. a new metric in the framework something that explicitly measure species co-variance but is expressed as a function of scale (e.g., Wagner's work on the spatial decomposition of the variance ratio pdf link. I've worked on this a bit with Allen Hurlbert trying to improve the approach - this is still unpublished by I did setup a git repo with some of the core code: https://github.com/mcglinnlab/vario

2. an ordination kind of analysis that explicitly looked at how composition was shifting due to the treatment and scale. A multi-scale ordination kind of approach which is closely tied to the approach laid out in option 1 may be the most appropriate here given our concerns about scale dependence.

from mobsim.

Hi @dmcglinn and @jmchase , funny and interesting that several people come up with the same idea at the same time. I guess this indicates relevance? I also discussed the issue with Petr and offered to think about an implementation of INTERspecific aggregation/segregation in mobsim. Petr said it is not urgent from his side and the same is true for me.

Here are three ideas for the implementation:

1. Species pair-wise aggregation

We define a species x species matrix, which includes zeros and ones. If there is a one, the species share the same mother points in the Thomas-process simulations, so they are aggregated. If there is a zero they are independent. Maybe the proportion of ones could be a community-level parameter?
Obviously this cannot be used to model species segregation.

2. Higher level spatial-process for aggregation/segregation

We could add another spatial process, which simulates the distribution of mother points (which are random at the moment). When we use aggregation of mother points (across species) e.g. by a Thomas-process, there is interspecific aggregation, when we simulate a regular pattern of mother points (e.g. by Strauss-Hardcore process), there is segregation.

This offers more options, but is also more difficult to simulate and parameterize than 1.

3. Habitat types

We could define the distribution of some discrete habitat types (or of a continuous habitat variable) and the associations of species to the habitat.

All this is possible and should not be too tricky to extent mobsim towards these directions. It is not a priority for me at the moment, but let me know when this becomes an important next step.

from mobsim.