cluster connectivity

[Nicolas Urbina-Cardona]

Hi Michael
I'm running the ConsNet for my conservation problem which have 287,987
cells (134,758 deforested-excluded cells and 5217 NPA-included cells)
and 538 species. I have begin with ILV4 adjacency (the best solution
for min cells and shape objective) and I'm refining the CAN with
400,000 iterations, "escape with spatial neighborhood" and "adaptative
tabu reactor 1". To date the program has run 340,000 iterations but
the CAN still having a lot of independent small spots all along the
planning region. I'm thinking to re-refine the solution by using
"large neighborhood only" and another 300,000 iterations. Could you
suggest me another way to improve connectivity or cluster grouping?
Thank's "echele cuñao"!!!! (Sanchez-Cordero
com. pers. 2008)
Best
Nicolas

[Michael Ciarleglio]

Hi Nicolas,

By 340,000 iterations, the search should have greatly reduced the
number of separate conservation areas. There may be a few reasons why
the search is not finding better spatial solutions:

(1) In the objective box, make sure that you are using the ITS
objective (for your specific target set). If you have multiple
objectives, it is sometimes easy to overlook.

(2) Depending on how they are distributed, the excluded cells (and
there are a lot in this dataset) could be preventing the search from
creating a spatially compact solution. Look at the map of excluded
cells (by building the initial solution called "ALL cells selected").
If the excluded cells are highly scattered\, then it may be difficult
to build a compact solution around these excluded cells.

(3) Another way to improve the spatial configuration is to use a
multi-criteria objective for the minimum area problem. You will want
to:
-minimize the number of cells
-maximize the total representation
-minimize the shape
-minimize the number of clusters.
In this analysis, you should place a fairly strong emphasis on shape
and the number of clusters.

If you are still having problems, I can suggest other advanced
techniques to improve the shape of the solution.

Cheers
Michael

[Sahotra]

I think that, in general, for this type of problem, a full-fled multi-
criteria analysis (Michael's third option) will prove to be best. I
would put the strongest emphasis on the number of clusters (to be
minimized, perhaps with a weight of 9 on the ratio scale compared to
minimizing the area). Play around with maximizing the total
representation.

Also, right after loading the data, visually inspect the map to see
how the deforested excluded areas are distributed. If they are widely
and uniformly dispersed, you won't be able to come up with a solution
of high compactness. You may then wish not to exclude them but leave
them in with very low expectations for surrogates.

-SS.

On Jun 6, 5:18 pm, Michael Ciarleglio <michael.ciarleg...@gmail.com>
wrote:

[Nicolas Urbina-Cardona]

Thank you very mucho Sahotra and Michael!!
I realized that my planning region has 47% of deforested area, and
anthropogenic vegetation types (pastures, induced forest (palms and
pinus plantations), human infrastructure) are distributed all along
the region. So it's impossible to better compact my CAN than my
current solutions. I have runed some Multi-criteria based on
Sahotras'sugggestion and my CAN didn't changed significantly (just 12
cells). I'm going to better explain this situation in the paper but
keep the CAN of the second run (with basics using large neighborhood
only).
Thank you very much
Nicolas