Level crossings and emergency access: first results from portfolio-scale screening

A research note on measuring emergency-access exposure around level crossings. It pairs a reproducible national exposure layer with a worked isochrone example at Betton, and argues that this exposure is becoming computable at portfolio scale.

Level crossing closures are often discussed as traffic delay. For emergency response, the same closure can have a sharper meaning. A fire engine, ambulance, or rescue unit may have to wait at the barrier, take a longer route around the railway, or start from a station located on the wrong side of the line. In those cases, the question is not only whether the crossing is safe in the railway sense. It is whether the surrounding road network remains permeable when the crossing is unavailable.

That question has usually been answered one site at a time, after an incident or during a local consultation. A companion write-up argued that it deserves to be measured across the whole network, ahead of incidents. This note turns that argument into a first measurement exercise.

The measurement works in two layers. The first is a national exposure layer, a reproducible count of how many people live close enough to an active level crossing for closures to matter. The second is a worked example at one crossing, where access isochrones in the open and closed states show the change in reachable territory. Together they follow a single line of reasoning, from a closed barrier, to a longer emergency route, to the population that route serves, to the access-time gap at the asset, to a screen that ranks assets before any field review.

The two layers carry different weight, and the difference matters. The exposure figures are computed and reproducible. The Betton example is exploratory and illustrative. It shows how the method behaves on real road and rail geometry, not what is happening at Betton. The method assumptions stay open, and the sections below keep that line visible.

1. Exposure first: who lives close enough for closures to matter?

A screening method starts with exposure. Before any closure is modelled, it helps to know how many residents live near enough to an active crossing that a closure could touch their access to emergency care. That count sets the size of the problem, and it decides whether a screening method is worth building at all.

The crossings come from the SNCF Réseau open inventory. Keeping only the genuinely active ones, meaning public road crossings whose nearest track on the same line is still operated, narrows 17,387 recorded crossings to about 11,233 active level crossings. The figure sits close to the commonly quoted "12,000", and it comes straight from the data.

The population comes from the GHSL GHS-POP grid, a 100 m equal-area raster of resident counts published for epochs from 2000 to 2030. For each radius around the active crossings, the catchment circles are unioned so each resident counts once, and the raster population inside is read off. Distances run through Lambert-93, so a metre on the map is a metre on the ground. Raw Web-Mercator at French latitudes would stretch them by about 1.44.

For the 2025 epoch, the count rises with radius as it must.

So roughly 8.6 million people live within one kilometre of an active level crossing, about 13% of metropolitan France. At 5 km the figure reaches about 60% of the country, which simply reflects how rail corridors thread through populated areas.

Exposure is not impact. Proximity to a crossing does not mean an emergency trip is delayed. Only a fraction of calls need to cross the railway, and only a fraction of those meet a closure. These figures do not say that all residents are affected by closures. They define the population for which closure-related access effects are worth testing. Exposure is the denominator a screen narrows down, and the worked example below begins that narrowing.

2. From exposure to access-time gap: the Betton worked example

The Betton case is used here as a worked example. It shows the unit of computation. One crossing, one assumed response origin, one open state, one closed state, and the change in reachable territory.

The crossing sits on the Rennes – Saint-Malo line at Betton (Ille-et-Vilaine, 35). The assumed response origin is the local fire and rescue station (SDIS 35). Over the OpenStreetMap road network, access isochrones map the territory reachable within 2.5, 5, and 10 minutes, in two states. The open state keeps the crossing available. The closed state forces traffic to reroute. The access-time gap, Δt = h₁ − h₀, the closed state minus the open state, is the object of interest. It is the territory that moves further away in time once the barrier is down.

Read left to right, the matrix keeps three things distinct. The first column is reachability when the route is clear. The second is reachability once the crossing is closed and the route takes another way over the railway. The third column is the gap, where the closure lengthens access, concentrated on the side of the line the detour swings around. Read top to bottom, the same territory appears at tightening time budgets, so the gap can be read against the time responders have.

The Betton result describes one crossing, one assumed origin, and one pair of states. It is a worked example of the method, not a verdict on Betton. What it establishes is narrower and firmer. The access-time gap can be computed for a single crossing on real geometry. That matters because the crossing is the asset-level unit on which a portfolio method can act.

3. What this makes possible

The value of the method is not to replace field diagnosis. It is to reduce the search space. Most level crossings will not be emergency-access priorities. A portfolio screen helps identify the smaller subset where local review, emergency-service consultation, or engineering assessment may be justified.

Several extensions follow from the same unit of computation. The Betton calculation can run across the whole active portfolio, ranking crossings and corridors by expected access-time gap. Each crossing can be tested against more than one response origin, since a single station rarely tells the whole story. Closure duration and frequency can enter later, turning a static gap into an expected delay. Population, road permeability of the detour, and railway closure assumptions can then be read together, so geometry and the people it serves inform the same ranking.

The exposure layer can also be sharpened. A finer population base than the 100 m grid resolves the closest rings, and an age-aware view adds weight where it belongs, since time-critical medical risk concentrates in older residents and the GEOSTAT age bands open that view at 1 km and above.

None of this produces a final diagnosis on its own. The point is to identify where diagnosis is worth doing.

4. Why this matters for infrastructure managers

For an infrastructure manager, a level crossing is an asset, with a location, a line, a category, and a renewal history. Emergency access cuts across that view. It is territorial rather than railway-internal, since it depends on the road network around the crossing and on where response units sit. A closure, a suppression, or a renewal that changes how the crossing operates can shift the routes emergency vehicles depend on.

A portfolio screen brings the two views together. It does not decide which crossings to close, protect, or grade-separate, and it does not replace local engineering or emergency-service knowledge. It provides a reproducible layer that flags where access effects deserve review before design or prioritisation decisions are taken. SAMRoute supplies that layer, not a verdict.

5. Conclusion

This is still a screening layer, not a diagnosis. But it changes the starting point. Emergency-access exposure no longer has to remain a qualitative concern attached to a few known sites. It can be measured consistently across a portfolio, tested on real road geometry, and used to identify where more detailed review is warranted. Identification is now computable.

6. Sources and method notes

• SNCF Réseau. Level-crossing open dataset (Liste des passages à niveau). Active set scoped to public road crossings on operated track. URL
• European Commission, JRC. GHSL GHS-POP R2023A, a multitemporal population grid (1975–2030), 100 m, World Mollweide (ESRI:54009). DOI
• Eurostat / GISCO. GEOSTAT 2021 census population grid, 1 km² (ETRS89-LAEA, EPSG:3035), with age bands for the later age-split view. URL
• OpenStreetMap contributors. Road and rail network for the Betton isochrone computation. © OpenStreetMap, ODbL. URL
• Method. Population catchment computed as the dissolved union of per-crossing buffers in Lambert-93 (EPSG:2154), summed inside the raster engine. Estimates are bracketed by lower (fully-inside) and upper (touching) cell bounds. Isochrones computed from the response origin over the OpenStreetMap network. The companion write-up develops the clinical time-sensitivity and societal-value chain (out-of-hospital cardiac arrest, Quinet value of statistical life).

Fabrice Colas, founder of Oriskami SAS / SAMRoute.