Uncovering Trends in Transformations Using Model Comparison with the Suitability Modeler
Uncovering Trends in Transformations Using Model Comparison with the Suitability Modeler
The following is tailored for our Suitability Modeling group. While using Model Comparison, I discovered a pattern that was initially puzzling. However, by leveraging Model Comparison, I was able to pinpoint the underlying cause. I believe you’ll find the results both insightful and intriguing.
This blog demonstrates how trends in transformations between two models can be revealed using the Model Comparison interface in the Suitability Modeler in ArcGIS Pro 3.6. The scenarios presented build on the previous blog, Exploring what-if scenarios using the Model Comparison interface in the Suitability Modeler, and further analyze the siting of early warning emergency sirens.
To maximize the effectiveness of the sirens, the models incorporate two key criteria to assess population distribution.
The first criterion evaluates proximity to existing structures, favoring locations that are closer.
The second criterion targets areas with the highest population density by considering the distance from high-density population centers.
Caption: There are three high-density population locations, with two situated near the intersection of the two rivers.
Ensuring that signals reach the maximum number of people is essential. To assess the impact of the distance from high-density population centers, we used the Model Comparison interface. First, we created a scenario with all criteria weights set to 1. Then, we developed a second scenario increasing the weight for distance from high-density populations to 1.5.
With the Model Comparison interface, we can identify where the two models agree relative to the highest suitability values.
Caption: The percent difference in suitability values between the two models highlights areas of agreement, particularly those with high suitability (shown in lighter green) when the weight for proximity to high-density populations is increased.
The average suitability values between the two models are highest near the high-density areas (light green) and gradually decrease (darker green).
When overlaying the final regions on a map of the hot spot analysis statistic, there is significant agreement in the hotspot map—particularly in the high suitability areas (cold spots), which are shown in green.
Caption: The final locations are shown overlaid on the hotspot analysis map.
The most significant differences between the models appear at the outer edges of the study area, highlighted in red.
When comparing the final locations, the pink markers in the image above represent siren placements from the first scenario, where distance from high-density populations is weighted equally with other criteria. The blue markers indicate siren locations when this weight is increased. Notably, three regions overlap: two in the far east and one in the far west. The sirens northeast and southeast of the river intersection shift closer to high-density population areas, with the blue locations positioned north of the pink markers.
An intriguing ring appears on the hot spot map, where the models converge, shown in green. The reason for this ring becomes clear when examining the transformations applied to the criteria: distance from structures uses a linear transformation, while distance from high-density populations uses an MSSmall transformation (see images below).
Caption: The Linear function is applied to the distance from structures (left image), while the MSSmall function is used for distance from high-density populations (right image). The purple arrow highlights where the MSSmall function rapidly decreases in suitability.
When using the MSSmall function, locations near high-density population areas are assigned a suitability score of 10. These locations also receive a score of 10 in the Linear function for the distance from structures criterion. As we move farther from high-density areas, the MSSmall transformation for distance from high-density populations remains at 10, while the Linear function’s suitability for distance from structures gradually decreases.
This difference in how the scores change as we move away from high-density populations leads to a decrease in model similarity, eventually making the differences between them nonsignificant (highlighted in yellow). At the inflection point in the MSSmall function (see the arrow in the image above), the suitability score drops sharply, whereas the Linear function continues to decrease at a steady rate.
As we move farther away, the MSSmall and Linear functions intersect, assigning identical suitability scores to locations at that distance. This intersection causes the models to become similar again, as shown by the green ring in the image above. Beyond this point, the two functions diverge once more.
We conclude that including high-density population as an additional criterion was somewhat redundant, since distance from water and distance to structures were already input criteria, and high-density population areas are typically located near rivers and existing structures.
When comparing the effects of increasing the weight of the distance to high-density populations criterion, the statistical similarities and overlapping regions indicate that this adjustment does not significantly impact the model. The redundancy of this criterion, as previously noted, is substantial, and assigning it a higher weight has minimal effect.
Through this comparison, we found that the Model Comparison interface not only enhances our understanding of the relationships between the models, but also offers valuable insights into the models themselves.