Week 7: Fencing, signage and wildlife crossings

Why the interventions with the best headline results are also the ones hardest to fund, and what that means for recommending one.

5 April 2027

Open the slides

Exclusion fencing paired with underpasses cuts recorded strikes further than almost anything else in the literature — and costs more per kilometre than a local road authority will usually approve without an unusually strong case. This lecture works through what an engineering measure actually changes about a roadside, what the evidence says about each candidate, and why “the best measure” and “the fundable measure” are frequently different documents.

Four ways to change what happens at the roadside#

Every engineering measure in this space does one of four things, and it is worth naming them separately before comparing options, because a measure that looks weak against one goal can be the right one for another:

1. exclude — physically stop an animal reaching the traffic lane at all (fencing);
2. redirect to a safer crossing — give the animal a route across that isn’t the road surface (underpasses, overpasses, culverts sized for wildlife);
3. change driver behaviour — get the driver to slow down, look, or brake earlier (static signage, dynamic driver-activated warnings);
4. change road-use conditions — alter the road environment itself rather than either party’s behaviour (verge clearance improving the sightlines Week 5 discussed, lighting, speed-limit changes).

Most of the measures you’ll compare this week are combinations of the first two, on the theory that exclusion alone just moves the problem.

Fencing works until it doesn’t#

Exclusion fencing that runs out — at a property boundary, a side road, or simply where the funded section ends — doesn’t remove animals from the area, it relocates where they cross. An animal that would have crossed somewhere along a long fenced stretch instead reaches the fence, walks along it, and crosses at the end. This “end of fence” effect is a genuine displacement problem, not a hypothetical one, and it means a fencing recommendation has to specify where the fence ends and what happens there, not just how long it is.

It’s also a source of measurement noise in the literature you’re about to read: a study that places its wildlife-crossing counters near the fence itself, rather than along the whole corridor, can record a large apparent reduction that is partly a real effect and partly animals being pushed to an unmonitored stretch just past the fence line. Keep that in mind when a fencing result looks unusually strong.

Usage is not the same measurement as collisions#

A camera trap at an underpass that shows animals using it regularly is evidence the structure works as a crossing — it is not, on its own, evidence that road collisions on that stretch have gone down. Those are two different measurements: one counts animal movements through a structure, the other counts vehicle-animal collisions on the road. A structure can be well-used and still coincide with no measurable change in collision counts, if animals were already avoiding that exact spot, or if the same population simply crosses more often overall once a safe route exists. Recommending a crossing structure on usage data alone is recommending on the wrong measurement.

Signage: weak evidence, still specified#

Static warning signage is one of the cheapest, fastest-to-install measures available, and the evidence for it changing driver behaviour enough to reduce collisions is thin. It still gets specified on almost every road authority’s standard hazard-mitigation list, for reasons that aren’t really about effectiveness: it’s inexpensive, it requires no ongoing maintenance budget beyond replacement, and — as Week 8 will make uncomfortably explicit — installing a sign is also a road authority creating a public record that it was aware of a hazard. Weak evidence and low cost keep it on almost every list regardless.

What the evidence actually says#

Rytwinski, T., Soanes, K., Jaeger, J. A. G., Fahrig, L., Findlay, C. S., Houlahan, J., van der Ree, R., & van der Grift, E. A. (2016), a meta-analysis of road-mitigation studies published in PLoS ONE, is the single most useful source for this week because it covers both ends of the evidence spectrum in one place:

Read those three rows the way the authors themselves want them read, not the way a press release would summarise them. The authors flag substantial heterogeneity between the studies they pooled, a dataset that is 82% North-American and 90% focused on a narrow set of mammal taxa (“strongly limits the conclusions” the review draws for other taxa or regions — kangaroos included), a signal consistent with publication bias favouring reported successes, and the fence-end monitoring problem described above inflating some of the fencing estimates. The 83% figure is the strongest single result in this literature and it still comes wrapped in every one of those caveats. None of that makes the finding useless — it makes it a finding about fencing-plus-crossing structures in mostly North American, mostly large mammal contexts, that a kangaroo-strike proposal has to argue is likely to generalise, not assume it does.

Comparing candidate measures#

The evidence-strength column and the approval-complexity column run in roughly opposite directions. That tension is the whole reason this week exists.

Illustrative planning estimates — fictional teaching figures, not a real quote#

These numbers are invented for this exercise, deliberately rounded, and do not resemble any actual engineering quote. Use them to compare shape — which measure is cheap to try and which is a genuine capital commitment — not to cost a real project.

Why “install everything” isn’t a recommendation#

Fencing, a crossing structure, signage, dynamic warnings and reflectors together aren’t a stronger recommendation than any one of them alone — a list of every available measure is not a scoped proposal, it’s a refusal to choose. Every measure on that list also competes for the same limited capital-works budget and the same finite approval bandwidth at a road authority; recommending five things is often a reliable way to get none of them funded. A defensible recommendation names the road, states which mechanism (exclude / redirect / change behaviour / change conditions) the site actually needs, and argues for one primary measure — exactly what Week 12’s final proposal will ask you to do at the scale of an entire intervention.

Three sites, three different answers#

This week’s session works through three real entries from this unit’s synthetic road dataset — Bunjalung Ridge Road, Coorabin Road and Merricourt Highway — each with a different traffic volume, reporting profile and approving authority. The same evidence table above supports a different recommendation at each one, because the constraint that matters most changes from site to site: budget at one, approval-pipeline length at another, ecological connectivity at the third.

Where this goes next#

Whatever measure gets recommended for a site, someone still has to approve paying for it — and if a hazard was known and nothing was done, someone has to answer for that too. Week 8 picks up exactly where a sign, a fence, or the absence of either, turns into a question of who is liable.

Reading#

Rytwinski, T., Soanes, K., Jaeger, J. A. G., Fahrig, L., Findlay, C. S., Houlahan, J., van der Ree, R., & van der Grift, E. A. (2016). “How effective is road mitigation at reducing road-kill? A meta-analysis”, PLoS ONE, 11(11), e0166941. The source for every figure in this lecture, including the authors’ own caveats about geographic and taxonomic bias — read the limitations section, not just the headline numbers.