0

Transition Potentials all have probability of 0

Katie Fensome

December 09, 2012 21:46

None

Follow

I have been using Land Change Modeler (LCM) in IDRISI for a few weeks, experimenting with projecting deforestation in the Ecuadorian Amazon. I am now trying to perform the final series of projections, but suddenly every time I run the submodel and create transition potentials, every one of the potentials has a probability of 0, or an extremely low potential for change. I have changed none of the project parameters. I have also created new projects with the same parameters, and have tried using new images in case the ones I had been using had somehow become corrupted. 

However, nothing I try is working, and I wonder if there could be a problem with LCM itself?

I would like to try uninstalling an reinstalling IDRISI, but am worried that I will need my license number and password that I was sent originally or may have problems with it- I can't afford to lose it completely for a while!

Has anyone had a similar experience, or has anyone tried uninstalling and reinstalling IDRISI before?

Cheers,

Katie

• Facebook
• Twitter
• LinkedIn

6 comments

• 0
  
  Clark Labs Tech Support Team December 10, 2012 17:03

  Hi Katie,

  If your output demonstrates a low potential for change, this is an indicator that the driver variables you are using are not explaining the change in the landscape.  I would suggest you re-examine the driver variables you have selected for this model. 

  Once you have selected the driver variables, you might want to look at the sample size and the number of iterations if you are using MLP.  Increasing the number of sample size and iterations may help raise the accuracy levels.

  Best,
  
  Laura Hansen

  Clark Labs Technical Support Staff

• 0
  
  Katie Fensome December 20, 2012 15:19

  Hi, thank you for your response but unfortunately I don't think the problem can be rectified in that way because I have used the exact same project parameters and variables several times before and got much higher transition potentials than I am now getting. The variables I am using all have significant explanatory power, but even so the transition potentials all come out with a value of 0 or one single, low value across the region. As I haven't changed anything about the project I am finding it very difficult to work out what the problem could be!

  Any ideas would be hugely appreciated, thanks.

• 0
  
  Katie Fensome December 28, 2012 17:35

  I recently noticed that when I run the submodel in the transition potential box, the MLP stops improving after only 2 or so iterations. This is probably why the potential either ends up as 0 or 1 across the whole area: if it only repeats twice it either gets a 'yes' or 'no' rather than a range of possibilities. 

  I can't work out how to improve this-  I have tried changing the start and end learning rates but that hasn't helped. 

  Again- any help would be hugely appreciated!

  Katie

• 0
  
  Tomaso Ceccarelli January 05, 2013 17:42

  Hi, I am experiencing a similar problem. I have tested several submodels and driver variables and in some instances I am getting accuracy rates of about 82%. However I always get transition potential maps with values of 0. Maybe the underlying problem is indeed the week explanatory power of the drivers, but, as for Katie, the tests through Cramer's V were encouraging.

  And in any case, I don’t understand why would I get these levels of accuracy rates.

   

  Thanks a lot on my side as well for any suggestion you may have on  this !

  Tomaso

   

• 0
  
  Gabriela M Luque May 06, 2013 18:39

  Have yo tried creating transition potential maps for one transition at a time instead of multiple transitions? It worked for me even though i barely varied the explanatory variables.

   

  Good luck with that!

• 0
  
  Derek Vollmer February 19, 2014 09:30

  Yes, I had the same luck as Gabriela when I created transition potential maps one transition at a time. It defeats the purpose of having a submodel, but obviously in the end you want reasonable maps.