From transition probability to absolute connectivity

[KRNA01]

As site selection in ADRIA currently runs now, we're using the larval transition probabilities provided by @BarbaraRobson.

These are probabilities of transport for larvae, not projections of the absolute number of larvae in transport.

Converting from probability (as a proportion) to a likely number of larvae moving from site A to B, would improve ADRIA's site-selection skill.

We can do this by simply multiplying (cell-wise) the transition probability matrix by relative coral cover (proportion) and site (or reef) area.

Here's an illustration of this, comparing centralities (outdegrees) for the raw connectivity matrix (left) with those for the likelihoods of absolute transport (area * coral cover).

Keen to discuss if we can do this easily.

@ConnectedSystems @Rosejoycrocker, for the purpose of helping IPMF tomorrow, can we just pull in a temporary converted (absolute) matrix into the site-selection functions directly for now, i.e. not touching the rest of the architecture of ADRIA?

Or do you think the effect is only marginal, and we should do later?

PS: Note that I have used k values instead of covers for this illustration, because I didn't have covers handy. Using covers will like shift this.

[Rosejoycrocker]

Hi @KRNA01, We could use an absolute connectivity matrix but this would have to be converted to centrality using matlabs network functions first. Does it make sense to calculate centrality from 'absolute transition probabilities', or should it be calculated as centrality first, then used in centralityareak?

[KRNA01]

Hi @Rosejoycrocker, yes, you're right.

Given how similar the two figures above look, and how busy we are, I'd suggest we push the solution to this till after the Feb deadline.

@ConnectedSystems, do you agree?

[Rosejoycrocker]

also sorry I forgot markdown makes asterix into italic writing, that's supposed to say centrality times area times k

[KRNA01]

Thanks @Rosejoycrocker,

I figured that 👍 .

Note that coral_cover (proportion) is a better metric to use here than k. (cover represents state, k represents scope)

Anyway, given our deadlines, let's kick this till March.

[ConnectedSystems]

Agreed

[KRNA01]

@Rosejoycrocker and @ConnectedSystems ,

Thanks for progressing this.

Quick questions:

1. have you limited this to the DMCDA process?
2. In other words, this does not affect connectivity as used to predict recruitment?

The latter needs to be updated too, but not now. A job for @KRNA01 next.

[Rosejoycrocker]

Hi @KRNA01, this change is purely within the DMCDA function for now. It just multiplies the old centr criteria by area and sumcover (and divided by the max (area times centr times sumcover))

[KRNA01]

@Rosejoycrocker, great - thank you , perfect.

On that note, I remember putting (<1) constraints on centrality measures (C1) in siteConnectivity. Suggest we remove these (if you havent already) as you normalise in the DMCDA function anyway. Do you agree?

[Rosejoycrocker]

@KRNA01 good point, thanks Ken. I'll remove that constraint now

[Rosejoycrocker]

Update: looks like it was taken out at some point, so no problem :)

[ConnectedSystems]

Hi @Rosejoycrocker can we close this issue in light of your recent commits, or were those a step towards resolving this (and we keep this open)?

[Rosejoycrocker]

Hi @ConnectedSystems , feel free to close :)