RCITI @ UNSW SYDNEY × DEPARTMENT OF TRANSPORT AND PLANNING, VICTORIA
Safe Local Roads and
Streets Program
Risk data products for Victoria's road network — Infrastructure Risk Rating,
Collective Risk and Personal Risk
Capability briefing
Prof. Taha Rashidi · Research Centre for Integrated Transport Innovation (rCITI), UNSW Sydney · 30 July 2026
Who you'd be working with
A research centre that ships decision-ready data products
rCITI at UNSW Sydney works end-to-end: from raw state datasets to validated, documented, decision-ready platforms. Road-safety risk analytics is a core, active capability for the centre — not a one-off study. The examples in this briefing are live systems built on Victorian data.
Victorian data fluencyVicmap TR_ROAD, DTP open data, crash records, SCATS and telematics — handled daily in live projects.
Methods that match the manualsAusRAP / iRAP protocols, Austroads guidance and HSM-aligned evaluation — applied, not just cited.
Delivery-grade engineeringReproducible pipelines, technical documentation, metadata and handover packages as standard outputs.
University-based independenceTransparent methods with quality flags and caveats the Department can stand behind publicly.
What the Program needs
Three risk products, one curated network
ROAD-ATTRIBUTE DATA · PROACTIVE
Infrastructure Risk Rating
Whole-of-network rating from curated input attributes agreed with the Department — score, band, supporting fields, plus confidence and imputation flags where data is sparse.
CRASH DATA · REACTIVE
Collective Risk
FSI and FSI-equivalent crash density — score, bands and rankings, with corridor / midblock and intersection views and documented calculation methods.
CRASH ÷ EXPOSURE
Personal Risk
Exposure-adjusted crash rate per distance travelled — with AADT inputs carrying explicit source and confidence fields per segment.
Supporting products in scope: curated VicMap TR_ROAD network layer · validated five-year crash summaries · AADT attribution with source & confidence · technical documentation, data dictionaries and a reproducibility & handover package
This briefing is structured around the mandatory data products in the scope.
The foundation
We know these five measures inside-out — and teach them
Two data lanes, one goal.Reactive: crash history drives Collective and Personal Risk. Proactive: coded road attributes drive Star Rating and IRR. Keeping the lanes straight is what makes the outputs defensible.
We wrote the explainer.Our team maintains a public, referenced guide to exactly these five measures — with a live Q&A assistant — used by practitioners to un-confuse them.
Why it matters for SLRSP.Outputs only change decisions when DTP, TAC and councils can read them with confidence. We bring the literacy layer with the data.
From our published explainer: the road-risk family — crash-based and attribute-based measures, and the systemic method spanning both.
Collective & Personal Risk
Crash-risk analytics we've already built for Victoria
Corridor FSI analytics, the DTP way.Five-year validated crash windows, FSI-equivalent severity weighting, percentile banding, corridor / midblock vs intersection views.
Austroads / HSM-aligned evaluation, live.A working before-after platform on the full Victorian network: Empirical Bayes with comparison-group and naive baselines, severity and road-user filters, per-AADT normalisation.
Built to be interrogated.Every estimate ships with its window, method, confidence interval and significance — nothing hides in a spreadsheet.
Live evaluation on a 60 km Victorian arterial — EB effect estimate with confidence intervals, severity split, and method / normalisation controls.
The exposure problem
Personal Risk lives or dies on AADT — we close that gap
Local roads rarely have traffic counts — yet Personal Risk needs exposure on every segment. Our approach attributes AADT across the entire network with an explicit source and confidence tier on each segment, so low-confidence values are flagged, never hidden.
1
MEASUREDPermanent counters and tube surveys — the ground-truth spine.
2
EXPANDEDSCATS signal detectors calibrated against measured sites.
3
PROBEConnected-vehicle telematics, penetration-corrected.
4
MODELLEDNetwork and land-use context inference — envelope-checked against statewide travel totals.
~1.27M
VicMap TR_ROAD segments statewideAttribution designed for full-network coverage — every segment carries its AADT, its source, and its confidence.
The question you asked
IRR on local roads: we can source or generate every input
| IRR input attribute | How we deliver it for local roads |
|---|
| Road stereotype & seal | VicMap TR_ROAD attributes — class, divided / undivided, lanes, seal, urban–rural |
| Alignment & curvature | Derived from network geometry, computed statewide |
| Carriageway width | Imagery- and LiDAR-assisted measurement; modelled where sparse, flagged as such |
| Roadside hazards (left / right) | Tree-canopy and roadside feature datasets — in active use today for hazardous objects near the carriageway |
| Land use & access density | Planning-zone layers, cadastre and access inference |
| Intersection density | Network topology, computed statewide |
| Traffic volume | Tiered AADT attribution with per-segment source & confidence |
| Operating speed (desirable scope) | Probe-speed data and speed-limit / operating-speed comparison — already built |
Curated inputs, agreed with the Department — every field lands with quality, confidence and imputation flags. And we're open to any dataset you can bring: asset registers, surveys, SCATS, LGA data.
Beyond the minimum
Network-wide risk modelling, with QA you can audit
Statewide screening layers.Machine-learning crash-risk models over the full Victorian open crash record — complementing IRR and crash-risk mapping with pattern-level screening.
Validation as culture.Holdout testing, envelope reconciliation, documented caveats — data-quality and confidence fields are first-class outputs, exactly as the scope requires.
Reproducible to the end.Versioned pipelines, data dictionaries and metadata — the reproducibility and handover package is how we work by default.
Statewide risk screening and hotspot analytics built on the Victorian open crash record (2012–2025).
How we'd work with you
A delivery partnership, not a black box
01
Confirm inputs togetherDataset confirmation with the Department at commencement. We identify gaps, limitations and inconsistencies up-front — in writing, before they become surprises.
02
Deliver in stagesCurated network layer → exposure → Collective & Personal Risk → IRR. Each stage documented, spot-checkable and usable on its own.
03
Leave you self-sufficientReproducibility and handover package: your team can re-run, audit and extend every output — with data dictionaries and metadata included.
Prof. Taha Rashidi · Research Centre for Integrated Transport Innovation (rCITI), UNSW Sydney