Transport Ireland Infrastructure (TII) commissioned a feasibility study to develop Accident Predictive Models (APMs) for its trunk road network. The study, carried out in Phase 1 by TRL Limited, aimed to determine whether existing road and traffic data could support regression‑based models that predict collision frequency from physical road characteristics. The ultimate goal is to provide engineers with tools to assess safety performance, identify sections that require investigation, and estimate crash modification factors for proposed design changes, thereby informing economic appraisal of new schemes.
The technical work focused on assembling a comprehensive, georeferenced dataset. The core of the dataset is the TII GIS base layer, which contains the national road network, junctions, and road categories. Additional layers were sourced from Ordnance Survey Ireland (OSI) PRIME2, providing detailed road and structure information; from the Road Surface Profiler (RSP) and Laser Crack Measurement System (LCMS) surveys, offering geometric parameters such as crossfall, gradient, and SCRIM coefficients; and from the PMS lane‑width database, delivering paved width for 100‑metre sections. PMS asset inventory data supplied information on hard shoulders and other carriageway features. All layers were linked within a single GIS framework to create a central modelling dataset.
Collision data for the period up to 2021 were matched to the GIS network, producing 41,456 recorded incidents, of which 6,758 involved injury. The distribution of collisions by road type revealed that 40 % of all collisions occurred on non‑legacy single carriageway roads, and nearly half of injury collisions were on the same road class. Mainline motorways accounted for 14.6 % of all collisions and 13.1 % of injury collisions, with a collision density of 6.0 per kilometre and 11.6 per 10^8 vehicle‑kilometres per year. Dual carriageways had 17.1 collisions per kilometre and 29.0 per 10^8 vehicle‑kilometres, while legacy roads recorded 4.3 collisions per kilometre and 19.1 per 10^8 vehicle‑kilometres. Link roads exhibited the highest collision density at 22.0 per kilometre and 97.4 per 10^8 vehicle‑kilometres, with 7.6 % of all collisions and 10.3 % of injury collisions occurring on them.
Segment length was a key modelling consideration. When the network was divided into 100‑metre sections, 67 % contained no collisions; this proportion fell to 27 % for 500‑metre sections and 10 % for 2‑kilometre sections. The study therefore recommended adopting a minimum segment length of 500 metres to mitigate zero‑inflation issues while retaining sufficient spatial resolution for practical engineering use.
The collaboration structure involved TII as the client and primary funder, TRL Limited as the project lead, and data providers such as OSI, SCRIM, and the PMS system. The project was conducted over a period that encompassed the collection and processing of 2021 data, with a systematic literature review covering APM developments from the 1980s to the early 2000s to benchmark methodology. The outcome of Phase 1 is a validated dataset and a set of preliminary statistical relationships that will inform the development of full APMs in subsequent phases, enabling TII engineers to predict collision outcomes for proposed road design changes and to manage safety across the trunk network more effectively.