Geographic Information Systems Stack Exchange is a question and answer site for cartographers, geographers and GIS professionals. It's 100% free, no registration required.

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

I currently use a "classic" raster based approach of dividing a study area into cells. All input layers are converted to rasters at the same cell resolution and given a suitability rating.

alt text

A final suitability rating for each cell is calculated by combining each layers rating, with weightings to reflect the factors importance.

A final mask is applied to exclude any areas, such as waterbodies, that are unsuitable for the proposed land use.

Problems with this approach include:

  • choosing a cell resolution that is either to large to provide meaningful results, or a high resolution giving a false sense of accuracy.
  • finding weightings for each of the input parameters.

Are there any other problems or alternatives to producing land suitability maps?

share|improve this question
up vote 7 down vote accepted

An alternative that is well known in some circles but seems not to be known at all within GIS is Multi-attribute Value Theory. This is a theoretically well-grounded way to establish accurate scoring methods involving two or more characteristics (attributes). It proceeds by systematically considering trade-offs among attributes. With suitability problems, for example, you would consider what change in elevation would be needed to compensate for a given change in slope in order to retain the same suitability, with similar considerations for all possible pairs of attributes.

Insights provided by the theory include:

  1. It is possible for the weights of one subset of attributes to vary with the levels of another subset of attributes. When this happens, a simple weighting system is not possible--more complex formulas are needed.

  2. When such dependencies do not hold (or are not strong), it is often possible to find how to re-express attributes (such as taking their logarithms or square roots or reciprocals) in such a way that a simple weighted scoring system correctly represents the value of every combination of attributes. (The simple test for this is called the "corresponding trade-offs condition".)

I don't think I have ever seen a report of a GIS scoring application (which includes all suitability studies) that recognizes the need to check for independence of attributes (1) or bothers to assess the correct way to express the attributes (2). Unless this work is done, no scoring system has a legitimate claim to general accuracy or usefulness in making decisions.

This issue is far more important than resolution or the MAUP in terms of producing a product that is genuinely useful for making siting decisions.

share|improve this answer

One term used to describe this issue is the 'Modifiable Areal Unit Problem' and a paper I read on this topic is Excess commuting and the modifiable areal unit problem. The authors' approach is to examine the analysis at a few different spatial scales to look at what point convergence occurs.

This is a satisfactory solution for examining one parameter, but when there are many, it becomes more complicated. In this case, perhaps you could use ModelBuilder or Python for your analysis and run it several times varying the cell size to examine whether you have noticeably different results. Depending on your time availability (and computing power) you could look for convergence mathematically (stop when the difference is less than a certain percentage) or judge it more qualitatively.

share|improve this answer
sorry - put the wrong paper/link in there first time. It's fixed now! – djq Nov 10 '10 at 22:08

Aspect, height, and slope all come from the same raster source originally, so the nice thing about continuing to use rasters is you can keep the same resolution for these inputs without losing information due to re-sampling. (This paragraph mostly null & void if you're using other sources of data at a lot of other resolutions. :) )

A useful extension beyond coming up with weights by hand is to use known occurrences of the thing you're modelling suitability for & passing it along into a statistics program, as in:

That way you train your suitability using known sites, rather than WAGs. Of course it's rather more involved...

share|improve this answer
+1 for introducing the idea that weights can actually be derived from data rather than merely being reasonable guesses! – whuber Nov 21 '10 at 21:55

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.