Problem with selected area in a multi-criteria design

[Angélica]

Hello,

What is the best method to choose conservation areas for species and vegetation types in a multi-criteria analysis?

I ran several multi-criteria solutions to find areas of connectivity between protected areas by assigning different targets to species distribution and leaving areas with primary vegetation as permanently included areas.

However, solutions include only part of the vegetation (and not the whole area initially assigned as permanently included).

In addition to these biological variables, I have other human variables in which I assigned different weights when running solutions. Initially, I tried to assign restrictions and / or costs to the integration of vegetation and the rest of the variables, but consNet was interrupted or took many days without offering a solution, apparently without much progress.

My question then is what is the best way to include information of vegetation and species distributions (i.e. in order to have different targets of species and vegetation types), if the main objective is to maximize the representation of biodiversity in a minimal well-connected area, while including wihth restrictions /or benefits other human variables.

Thanks in advance for guidance!

[Michael Ciarleglio]

Hi Angelica,

(For any of these issues, I can give you more help if you provide the .quickload file, or the original input files).

What is the best method to choose conservation areas for species and vegetation types in a multi-criteria analysis?

For a quick overview of multi-criteria analysis in ConsNet, start with worksheets 2, 3, 4, and 5.  I have attached them here for convenience (the website is down right now).

worksheet 2:  setting up the objective

worksheet 3:  setting up a multiple criteria minimum area problem

worksheet 4:  setting up a multi-criteria maximum representation problem

worksheet 5:  performing the analysis

I ran several multi-criteria solutions to find areas of connectivity between protected areas by assigning different targets to species distribution and leaving areas with primary vegetation as permanently included areas.

However, solutions include only part of the vegetation (and not the whole area initially assigned as permanently included).

ConsNet should not be removing any of the permanently included cells.  Try the following.  On the INIT SOLNS tab (initial solutions), build the solution with 'NO cells selected'.  When this builds a solution, it will not select any cells (except for those that are permanently included).  Does that solution look like you expected?  If not, then there might be a problem with the input file.  Otherwise, I may need to look at your .quickload file, and try to reproduce the issue.  

In addition to these biological variables, I have other human variables in which I assigned different weights when running solutions. Initially, I tried to assign restrictions and / or costs to the integration of vegetation and the rest of the variables, but consNet was interrupted or took many days without offering a solution, apparently without much progress.

Here are some follow-up questions:

• How large is your dataset (number of species, number of cells)
• Is the problem that the search is running slowly?  Usually, iterations should progress quite rapidly . . . does the search progress appear to hang up on a single iteration?
• Or is the problem that the search is running quickly but not finding improvements to the solution?

Here are some troubleshooting options:

• Read Chapter 8 in the manual (all about Performance).  It is a little bit more technical, but it covers most of the known performance problems.  The most likely reason that ConsNet crawls: Java runs out of memory.  How large is your data set?  And how much RAM is available on your computer?
  
• Searches converge faster if you start from a decent heuristic solution (see worksheet 5)
• I've observed some cases where ConsNet doesn't do a good job at significantly improving the solution.  This is more related to the objective function formulation.  It is a little harder to diagnose.

My question then is what is the best way to include information of vegetation and species distributions (i.e. in order to have different targets of species and vegetation types), if the main objective is to maximize the representation of biodiversity in a minimal well-connected area, while including wihth restrictions /or benefits other human variables.

ConsNet has two types of objective functions.  I believe that type 1 is better tested and a little more predictable:

1. If you need to guarantee a specific representation target, while optimizing for shape and other costs, then you should use the 'multiple criteria minimum area objective' (worksheet 3)
   
2. If you have an upper bound on the cost or number of cells, and you need to maximize shape, representation, and other costs, then you should use the 'multiple criteria maximum representation objective' (worksheet 4)

Let me know if you have more questions!

mc

[Angélica]

Hi Michael,

Thanks a lot for your kind help and the disposition to check the .quickload file! In fact it is not such a big area (just 103 714). Actually, the problem is that the search is running quickly but not finding improvements to the solution.

I already read the slected chapters you suggest. However, I'll look over more carefully the worksheets, because my problem can be as a result of have used many constraints. Also I'll check the input file as you suggest.

Thanks again, hope to solve this issue and i'll tell you if I need more help. 

Angélica

[Michael Ciarleglio]

Hi Angelica,

Here is how you can tell if the constraints are causing the problem.  On the search tab there is a graph which shows the objective score.  If the solution does not meet all of the constraints (infeasible), then it is shown with a gray circle.  If all of your solutions have gray circles, then the search is not finding solutions which meet the constraints.

Also, there are definitely times when I have found that the objective scaling and weights makes a big difference in the search.  Since ConsNet considers adding and removing cells one at a time, it can show local affinity for your top criteria (while missing the global optimum).  Sometimes, if you revisit your preference weights, you may find that you have assigned too much importance to one of the criteria.  But sometimes, the problem is in the search itself.  I can usually figure it out and come up with workarounds.

I have the day off tomorrow.  If you want me to take a look, zip up BOTH the quickload file and the objective folder.  That will enable me to see your objective and run a search.

mc

[Angélica]

Hi Michael!

I checked my input files and one of them had missing data, so that should be the reason for the difference between total areas reported. I guess I lost some data at some point during the processing of information in the GIS.

ConsNet did not reported gray circles, so that should not be the problem. I modified the weights (because in fact I was assigning too much importance to one of the criteria) and it is running very well! I'll see the results.

Thank you very much again for your kind attention Michael!

[Michael Ciarleglio]

That good news!  Let me know if you run into any more difficulty.

mc