Expected Values

[Garrick MacDonald]

Hello Michael,

I am interested in using ConsNet for my conservation project. I will need to create an expected representation file and I was wondering if you could provide me with an example.

I have read your manual and I am completing your tutorial with the data provided.

My expected values will be organized into four categories (low, moderate, high, very high). Since your manual says to stay between 0 and 1, would my representation file be 0, 0.25, 0.50, 0.75, 1?

Thank you,

-Garrick

[Michael Ciarleglio]

Hi Garrick,

A representation value of zero means there is no representation, so zero might not be suitable for the low value. I would use zero only to represent complete absence.

Other than that, I think a good place to start is to consider the value as a percentage confidence. Your low values might be 0.1 (10%) and high confidence might be 0.9 or 1.0. In the end, the values will be interpreted on a ratio scale, so 0.75 is 3 times as much representation as 0.25.

This approach is designed to work with the output from a Maxent stastical analysis. I'm not sure whether your approach would be as statistically sound, but it makes sense to me given that we never have perfect information anyway.

[Garrick MacDonald]

Hi Michael,

Thank you very much for your response! Your information was very appreciated and will be implemented in my approach. 

I have another question I was hoping you could answer. When I start the initial solutions using the West Virginia training data, only three different heuristic algorithms appear (all, no, and drop). Do you have any idea why MDS2, RF4 and ILV4 do not become visible? I am using ConsNet version 2.0.

Thank you,

- Garrick

[Michael Ciarleglio]

Hi Garrick,

Some other notes popped into my head regarding expected values.  This approach assumes that your cells are roughly the same size (on a grid).  If you are using irregularly shaped cells, and some are larger than others, you might need to adjust your representation values to reflect the difference in area.  For example, a cell which is one square kilometer should contribute more representation then a cell that is just a few hectares.  We haven't really dealt with data sets like that, so I'm not completely sure about the best practice.

mc

On Thursday, August 20, 2015 at 11:48:31 AM UTC-7, Garrick MacDonald wrote:

[Michael Ciarleglio]

Regarding your question on the why you have no initial solutions.  The goal of these initial solution generators is to build a conservation network which achieves a specific representation target (using the fewest cells possible).  If you don't define any representation targets, then you won't see any 'initial solution' algorithms available.  That's probably what is happening.

mc

On Thursday, August 20, 2015 at 11:48:31 AM UTC-7, Garrick MacDonald wrote:

[Garrick MacDonald]

Ok, that makes sense. I have not defined a representation target, but I will now so I can experiment with the additional initial solutions. 

As always, thank you so much for your assistance. 

- Garrick

[Garrick MacDonald]

Ok, that assumption works for my conservation project. My study area is comprised of a grid with cells of the same size. Thank you for sharing this information.

[Garrick MacDonald]

Hello Michael,

I was hoping you could help me clarify another concern regarding my representation file. Due to my master’s methodology, I am using a preexisting habitat evaluation model as my ConsNet input representation file. I am attempting to use ConsNet to introduce additional spatial and socio-economic costs into the preexisting habitat evaluation model to see how the additional cost effects the overall conservation area network design.

The preexisting habitat evaluation model was created through a combination of the predicted presence of indicator species (Arroyo Southwestern Toad, California Gnat Catcher, etc.) and environmental components (animal corridor analysis, grassland evaluation, etc.). Due to the composite nature of the habitat evaluation model, I am limited to the ordinal classification of very high, high, moderate, and low. With this preexisting habitat evaluation model in mind, would my representation file look something like the example below? Further, is it possible to create a representation file with these characteristics?

Thank you again for your help!

Garrick

Example:

Note: I am using your prior suggestion “Other than that, I think a good place to start is to consider the value as a percentage confidence. Your low values might be 0.1 (10%) and high confidence might be 0.9 or 1.0. In the end, the values will be interpreted on a ratio scale, so 0.75 is 3 times as much representation as 0.25.”

 

Representation File: Regular Distance Grid

4,30,30, very_high, high, moderate, low

1 -77.71887 39.3176 1 0 0 0

2 -77.7272 39.3176 0 0 0 .10 

3 -77.7272 39.32593 0 .75 0 0

4 -77.73553 39.30927 0 0 .30 0

5 -77.73553 39.3176 0 0 0 .10

6 -77.73553 39.32593 0 .75 0 0

7 -77.73553 39.33426 1 0 0 0