I am using the Optimized Hotspot Analysis tool in ArcGIS Pro to identify clustering of high and low values in my incident point data. After running the tool iteratively for multiple input point datasets, I have noticed that the cell sizes and distance bands used vary per iteration. I can understand that this is because the number of points and their geographic distribution vary per dataset. This influences the cell sizes and distance bands because the Optimized Hotspot Analysis tool employs multiple methods to derive an optimal value for each of these two parameters. For e.g., the distance band would vary per dataset because the Z-score distance peaks identified by the Incremental Spatial Autocorrelation tool are different. If no peaks are found the tool will calculate the average distance to 30 nearest neighbours and use that as the distance band.
I now wish to standardize the cell size and distance bands used across each iteration. To achieve that, I am trying to first understand how the tool performs when I vary each of these parameters while keeping the other constant. However, I have realized that my frame of thought is incorrect because the two parameters (cell size and distance band) are not independent variables. Instead, the distance band is influenced by the cell size chosen.
I have a few questions regarding how the tool operates.
1. A hexagon with a cell size of 3000 m means that the centroids of each hexagon are 3000 m apart. Does it make sense to apply a distance band of 4000 m when the cell size is 3000 m? Or should the distance bands be multiples of the cell size (i.e. 6000 m, 9000 m, 12000 m etc.)? These questions are premised by the assumption that the distance band is measured from the centroid of the hexagon.
2. I am trying to find guidance on methods by which I can derive an optimal cell size and distance band for multiple incident point data. Can anyone point me in the right direction regarding this, because I am strongly against the idea of arbitrarily standardizing the cell size and distance band when they are so influential on the results, but at the same time , I wish to have the results comparable and find it challenging when the outputs are hexagons of different cell sizes. Should I only specify a common cell size but keep the distance band to be calculated automatically by the software?
Thank you, and please correct me if I have mentioned something that is untrue, especially if it is a flaw in my understanding of the tool. All in all, I feel like it seems that I have to switch over and use the Hot Spot Analysis tool instead, since that one would allow for more customization. Hoping to hear from the community soon.
Hi there, thank you for replying. I am not too sure I understand what your question is sorry, could you elaborate further what you mean?
The way I understand the aggregation process is as follows; after a cell size is calculated (or specified), hexagon grids will be generated across the map, where each hexagon feature will have a value for the number of points within its hexagon boundary. This means that similar to a spatial join function, the hexagons will range from 0 on to the highest value for point count. Fortunately, in the Optimized Hotspot Analysis tool, it only preserves hexagons that have non-zero values, before running the calculations to determine the distance band and subsequent Getis-Ord Gi* calculation.
So, in my case, I have hexagons that range from 1 (minimum), all the way to tens of thousands. If you are asking about the minimum point requirement of 30 points, my count far exceeds that.
As a resolution of this thread, question number 1 does not have any major bearing, and for question number 2, there are many resources online and in publications, where researchers investigate the effect of cell size and bandwidth when employing local indicators of spatial autocorrelation, such as the Getis-Ord Gi* used in the Optimized Hotspot Analysis tool in ArcGIS.