Data Dependent versus Data Independent Suitability Models

by Kevin M. Johnston Ph.D. Esri

The concept of data dependency exists within the Reclassify and Rescale by Function tools, two tools commonly used to transform data in suitability models. Data is transformed to model a phenomenon’s preference to criteria. When the values in a study area are input into these tools, the minimum and maximum values are found. The phenomenon’s preference is then “fit” to these values by default.

For example: A phenomenon, in this case wild bobcat, exponentially prefer distances farther away from roads. Within our study area, the minimum distance from roads (0 meters) is assigned the lowest preference of 1 (on a preference scale of 1 to 10) and the maximum distance from roads (about 3500 meters) is assigned the highest preference value of 10. Essentially these minimum and maximum data values are, by default, defined as the lower and upper thresholds of the phenomenon’s preference scale.

Most suitability models map the minimum and maximum values of its criteria to the minimum and maximum of the preference scale – this is the outcome of using the tool defaults within the model. This data dependent form of modeling results in three unintentional consequences. Consider, what would happen if the minimum and maximum values were to change?

Minimum and maximum values changing within the study area

In our bobcat suitability model case study, bobcats prefer areas farthest away from roads. Say a new road is built through the most remote part of our study area, reducing the distance to the farthest location from roads. What would happen to the model?

Because there is a change in the maximum distance from roads in our study area ( reduced from about 3500 meters to 3150 meters) the model re-”fits” the phenomenon preference to the new data values. The highest preference value in our model is still a 10, yet it represents a closer distance to roads. The preference changes as an artifact of the minimum and maximum values changing. Toggle between these layers to see the differences in preference.

Clicking on the map will identify data values and preference values at the point.

Minimum and maximum values changing by altering the study area extent

By accepting the default values in the Reclassify or Rescale by Function tools , this new maximum distance becomes the highest preference for bobcat in the suitability model. Note how the remote areas within the Green Mountains are much less preferable after the extent is changed to include more of Lake Champlain.

Transforming modeled preference to actual preference

If the true phenomenon preference to the criteria is known, regardless within the study area extent or not, then a data independent model can be made (i.e. modeled preference is equal to the actual preference). It is generally accepted that bobcats find areas farther away from roads more suitable, but just how far away exactly?

For demonstration purposes, let us assume bobcats show the highest preference for areas equal to or farther than 4000 meters away from roads. This value can be set as the upper threshold. Preference will increase until this threshold is reached, after which the preference remains constant.

Looking at our study area, all locations are within 4000 meters from the nearest road. Because the threshold is not reached, no areas receive the highest suitability preference.

Reaching the preference threshold

This suitability model was also run on the entire state of Vermont, USA. Certain regions of the southern Green Mountains exceeded the threshold (4000 meters) of maximum preference. Recall that all values equal to or greater than the upper threshold receive the maximum preference value – a 10 on our scale of 1 to 10. The same principle is true for lower thresholds.

Click around on the map to compare distance from roads and preference values.

Comparing suitability models

Distance from roads is just one criteria of many included in our bobcat suitability model. Displayed here is a data dependent model with each criteria’s data values fit to the maximum and minimum preference values. Click the toggles below to view the data independent model which identified thresholds of preferences.

Acknowledgements

We thank Steven Lamonde of Johnson State College and the Vermont Center for Geographic Information for their contributions.