Skip to content

Layered Risk Heatmap

Tool — instantiates Vulnerability Hotspot Mapping and Hardening

Overlays exposure and susceptibility layers on one shared unit grid so the cells where several risks pile up light up as hotspots the average hides.

A Layered Risk Heatmap is the visual substrate of hotspot reasoning: it registers every risk layer to one common grid of units and stacks them, so a cell that carries high exposure and high susceptibility and thin protection glows while its neighbours stay cool. Its defining move is co-location made visible — where a single-number average reports "risk is moderate," the overlay shows the two or three cells where disadvantages coincide. It is a display, not a test: it renders where layers meet and leaves the questions of is this cluster real and why to other mechanisms. That restraint is what makes it distinct from a spatial cluster detector (which runs a statistical test) or a stratification table (which cross-tabs categories rather than mapping space).

Example

A city wants to place cooling centers before summer. Averaged citywide, heat risk looks unremarkable. The heatmap registers, to a shared grid of census blocks, four layers: land-surface temperature (exposure), tree-canopy cover (a protective layer, inverted), the share of residents over 65 with a chronic condition (susceptibility), and home air-conditioning access. Each is normalized to the same scale and stacked. Three blocks light up hot on every layer at once — old, treeless, low-AC, and running several degrees warmer than the citywide mean. None of them was the single hottest or the single oldest block, so no one-variable ranking had flagged them. The map moves two planned cooling centers onto those blocks. The heatmap didn't decide anything; it made the pile-up legible enough that the decision became obvious.

How it works

What distinguishes this tool is the registration and overlay step, not any single layer:

  • Common unit grid. Every layer is snapped to the same set of units so cells are directly comparable and stackable — the grid is the shared coordinate system all reasoning hangs on.
  • Normalize, then overlay. Each layer is put on a comparable scale and rendered as a translucent stack; the composite darkens exactly where multiple layers coincide.
  • Read the intersections, not the marginals. The output of interest is the co-location — cells hot on several layers — not the single worst cell on any one layer.

It surfaces candidates. It does not certify them.

Tuning parameters

  • Grid resolution — block vs. tract vs. grid-square. Finer units resolve micro-hotspots but add noise and can fragment a real cluster; coarser units smooth it away.
  • Layer selection and weighting — which layers enter, and equal vs. weighted stacking. Weights quietly encode a theory of what matters; equal weighting is a choice too.
  • Normalization and scale — percentile vs. absolute. Percentile scaling always paints a "worst 5%," even in a uniformly safe system.
  • Color binning / thresholds — where the breaks sit. Move a threshold and a warm cell becomes hot; binning choices can manufacture or hide a hotspot.
  • Overlay order and opacity — which layers sit on top, changing what the eye reads first.

When it helps, and when it misleads

Its strength is legibility: it converts an average that hides concentration into a picture a mixed room of stakeholders can read together, and it forces the question of where layers coincide rather than where any one layer peaks. As a shared artifact it aligns people fast.

Its central failure mode is being mistaken for evidence. A bright cell can be noise — the map shows concentration but never tests it, so a hotspot must still clear statistical validation before anyone spends on it. Because scale, weights, and binning all shift what looks hot, the tool is easily run backwards: choose the normalization that reddens the block you already wanted funded. And a cell describes the unit, not each person in it — reading individual risk off an aggregate cell is the ecological fallacy.[1] The discipline is to fix grid, weights, and scale before looking, and to route bright cells to validation and ground-truth rather than acting on the picture alone.

How it implements the components

Layered Risk Heatmap realizes the mapping-and-overlay side of the archetype — the components that make co-location visible, and no others:

  • protected_unit_grid_or_set — defines the common grid every layer registers to; the shared canvas.
  • exposure_layer_overlay — renders each exposure and susceptibility layer as a normalized, stackable overlay.
  • co_location_intersection_map — the composite is the map of where layers intersect; the hot cells are the intersections.

It does NOT define the harm metric or the susceptibility inventory it displays (those come from Intersectional Stratification Table and Vulnerability Index Construction), nor test that a bright cell is real — that guard belongs to Multiple-Testing Holdout Check.

Notes

The grid you choose is upstream of every hotspot the map can show: redraw blocks into tracts and clusters merge or vanish. This sensitivity to how units are cut and scaled is the modifiable areal unit problem[2] — fix the unit deliberately before mapping, not after seeing which cells you like.

References

[1] Ecological fallacy — inferring an individual's risk from a group-level rate. A hot cell characterizes the unit, not every resident in it; targeting an intervention still needs within-cell reasoning. ↩

[2] Modifiable areal unit problem — a well-known geography caution that statistical results depend on how areal units are drawn and at what scale, so hotspot patterns can be an artifact of the zoning. ↩