H3 Analysis

Assessing the suitability of Uber's H3 grid for crime concentration and hotspot analysis.

Pros and Cons of H3

✅ Good for consistent and extremely efficient spatial indexing
✅ Homogeneous neighbour relationship (edge-edge), all neighbours (almost) equidistant
✅ Extremely fast conversion of points to cells (non-spatial operation)
❌ Fixed resolutions with a sevenfold size factor
❌ Not truly hierarchical
❌ Affected by areal distortion

Investigations

We focus on two of the cons in this notebook:

H3 cells vary in size non-significantly, due to icosahedral distortion. We assess the impact of this on enumerating crime hotspots.
How likely are points to be non-hierarchical? A point in a child cell may be outside the parent cell, or conversely a point in a parent might not be in any of it's 7 child cells. This will affect aggregation - we cannot assume summing the counts in child cells will get a correct parent cell count.

References

https://www.uber.com/gb/en/blog/h3/

This notebook uses data from the safer-streets project's Azure blob storage. It requires a connection string, to which AZURE_STORAGE_CONNSTR should be set. Use of a .env file is strongly encouraged.