A fatal & serious injury (FSI) risk-screening product for signalised intersections — fusing SCATS per-leg traffic volumes with connected vehicle near-miss data to rank every covered site by exposure-controlled residual risk, refreshed monthly. Live for Melbourne metro today; built to scale statewide.
Serious casualty crashes at any single intersection are statistically rare events. Blackspot programmes must therefore wait years for crash history to accumulate before a site qualifies for treatment — by which time the harm has already occurred. Volume and crash history alone predict essentially none of the FSI variation at fine scale. Victoria already collects, every day, two data assets that together change this.
SCATS loop detectors sit on every leg of every signalised intersection in Victoria, logging 15-minute counts published as daily volume files. That yields per-approach AADT — an exposure denominator at intersection-leg resolution that routine crash-history screening rarely has access to.
Connected vehicle telematics captures harsh-event near-misses across the network — a leading indicator observed thousands of times more often than serious crashes. Raw density is largely a volume echo; the signal lives in the residual above exposure expectation.
Each month: every covered signalised intersection scored by its near-miss residual — what remains after SCATS exposure expectation is subtracted, never raw density — ranked and mapped as a prioritisation layer for signal-operation tweaks, low-cost treatments and blackspot funding.
The core question — does near-miss data add FSI signal beyond traffic volume? — has been tested and published in the Near-Miss × Crash Divergence study (Melbourne metro, 2023–25, exposure-hardened with real AADT).
In a Poisson model with a real AADT exposure offset, exposure alone explains essentially none of the FSI variation. Adding near-miss features lifts it to 0.054 — a modest, grid-level (1 km) effect, but it is signal where volume alone predicts nothing. Translating it to per-intersection decisions is the pilot's validation task.
The severity link strengthens under exposure control (from ρ = 0.181 raw) — harder braking intensity correlates with a higher share of severe outcomes, independent of volume.
Honest finding: raw near-miss density adds little once traffic volume is controlled. This is exactly why SignalWatch is built on the residual — and why SCATS per-leg exposure is the non-negotiable backbone of the design.
Full method, interactive maps and all findings: nearmiss.crashstats.com.au
Validation pathway: operational decisions follow prospective validation in Phase 1 — lift measured against a volume + crash-history baseline under spatially-blocked cross-validation.
Build the monthly pipeline for covered metro signalised intersections. Deliverables: monthly ranked screening lists, a prospective validation report against held-out future FSI, and a fleet-penetration bias study.
Per-leg exposure refinement, turn-movement attribution via micro-simulation, coverage expansion, integration with treatment catalogues for signal-operation and low-cost countermeasures.
Extend to all Victorian signalised intersections as telematics coverage allows; embed in blackspot and safe-system investment cycles with measured prevented-FSI reporting.
TAC funds the infrastructure and pays the claims. Reactive lists spend the treatment dollar only after casualties accumulate. SignalWatch ranks the covered signalised network by exposure-controlled residual risk every month, so investment can be screened toward the sites carrying elevated surrogate risk — with every operational use conditional on prospective validation first.