Santa Barbara Wildlife Crossings

Why did the mule deer cross the road?

Because the culvert was too long and too dark.

Community scientists have logged hundreds of animals killed on the county's roads. Combined with Caltrans drainage inventories, traffic, terrain and statewide connectivity maps, those records point to a short list of highway stretches where crossings would do the most good, and to the kind of crossing each one needs.

Garrett CraigAnalysis code and data: github.com/garrettgcraig/CA_crossings

Context

A new law asks Caltrans to plan for wildlife

On September 30, 2026, Governor Newsom signed SB 1250 (Cortese). It directs Caltrans to fold wildlife connectivity into routine highway planning: wildlife crossings, culverts, habitat corridors and fencing now belong in the department's asset management plan, and Caltrans must keep an inventory of connectivity needs jointly with the California Department of Fish and Wildlife.

That inventory needs evidence about where animals actually try to cross. Most of the existing structures under Santa Barbara County's highways are drainage culverts built to move water, and most are far too small or too enclosed for anything larger than a skunk. Replacing them on the usual maintenance cycle is the cheapest moment to make them work for wildlife, if the right ones are identified in time.

This analysis scores every half-kilometer of state highway in the county, ranks the stretches where a crossing would help most, and matches each to a practical solution: retrofitting a bridge or culvert that is already there, upsizing a culvert at replacement, or building a new underpass or overpass where the terrain allows.

Results

US 101 and SR 154 lead, and Gaviota Pass checks out

Half of the county's 96 highest-need segments are on US 101, mostly in two runs: the Gaviota Coast and Gaviota Pass (postmiles 29 to 52), and the stretch from north of Buellton past Los Alamos (postmiles 62 to 79), where 23,000 to 31,500 vehicles a day pass through oak woodland and rangeland. Two-thirds of all top segments already have a culvert or bridge that could be retrofitted or upsized.

SR 154 between San Marcos Pass and Lake Cachuma (postmiles 10 to 23) holds another 23 top segments. Oak woodland and chaparral meet the road on both sides, and CDFW lists the route as a movement barrier (W165). US 101 at Gaviota Pass checks out as well: 8 of its 21 segments land in the top decile (median: 88th percentile), carried by habitat and connectivity because only four roadkill records fall there. That matches the site Caltrans and CDFW chose independently.

SR 1 adds 20 top segments, most of them north of Lompoc (postmiles 24 to 34), including the county's top overpass site, a deep cut on a CDFW-listed barrier (W048). Click a row to see it on the map.

Top candidate zones

Adjacent top-decile segments merged; ranked by need

#LocationkmRoadkillStrongest option

What is being killed

Cleaned roadkill records by group, both sources (gray groups are left out of crossing analyses)

Recommended solution, top-decile segments

Based on existing structures, embankment height and fencing

Roadkill rises with the speed limit

Roadkill per km of state highway, by posted speed limit

Wildlife-vehicle collisions are concentrated after dark. Across more than a million deer-vehicle collisions in the United States, collisions were 14 times more frequent in the two hours after sunset than in the two hours before, and the autumn switch to standard time, which pushes evening commutes into darkness, raised collisions by 16% (Cunningham et al. 2022, Current Biology). Faster traffic leaves less time to stop for what headlights reveal, since stopping distance grows with the square of speed (Fambro et al. 2000), and speed is among the strongest predictors of collisions with large animals (Seiler 2005).

Fence ends

How stable is the ranking?

The choices no study can fix barely move the ranking: habitat strips of 100 m or 500 m instead of 300 m keep 85–88% of top-decile segments, and swapping speed for a night sight-distance measure keeps 99%. Weights matter more: unequal weightings keep 50–79%. A separate ranking that adds roadside barriers for the surveyed routes agrees closely with the main one (rank correlation 0.91).

Methods

What went into the ranking

Every state highway in the county was cut into 953 segments of about 500 m. Each segment carries everything known about that stretch of road and its surroundings, and the ranking combines those measures into a single need score. The full analysis is a set of reproducible R and Quarto notebooks in the project repository.

  1. Clean the roadkill

    Merge the survey's two form versions with iNaturalist, assign species to groups by the size of crossing they'd use, and keep dead animals within 10 m of a road.

  2. Inventory the structures

    Join Caltrans' 19,691 culvert records to their end treatments, keep the 632 county culverts open and ungrated at both ends under the highway, and size them for wildlife by height and openness (area ÷ length).

  3. Describe each segment

    Traffic and lanes from FHWA's 2024 HPMS, speed limits from HPMS and OpenStreetMap, embankment and side slopes from 10 m USGS elevation, and habitat and protected land on each side.

  4. Find roadkill hotspots

    Count records within 500 m of each segment and compare with 999 random scatterings of the same records along the network.

  5. Score need

    Average the percentile ranks of seven components, each described below.

  6. Match a solution

    Choose retrofit, upsize, underpass or overpass from what already crosses the road and whether it sits on fill or in a cut, then merge adjacent top segments into zones.

What each rule rests on

Every threshold and weight, with the peer-reviewed study behind it

RuleBasis
Roadkill kept within 10 m of a roadPhone GPS error had a median of 5–8.5 m outdoors (Zandbergen & Barbeau 2011)
Birds left out of crossing analysesEvidence that fencing with crossings cuts roadkill comes from mammals (Rytwinski et al. 2016)
Hotspots: 999 random placements, 95% bound2D HotSpot method (Coelho et al. 2008)
Hotspot radius 500 mScale where hotspots of different groups coincide most (Teixeira et al. 2013)
Equal component weightsUnit weights are robust when weights can't be estimated (Dawes 1979)
Habitat on the weaker sideCrossing use depends on habitat at the entrances (Clevenger et al. 2001a; Cain et al. 2003; Ng et al. 2004)
Protected landCrossings pay off most where they join conservation areas (Sijtsma et al. 2020)
Traffic and speedJaeger et al. 2005; Seiler 2005; Fambro et al. 2000
Flat approaches, drainages favoredFewer collisions where roads cross steep slopes, more in drainages (Gunson et al. 2011)
Mesocarnivore culvert ≥ 1.0 m tallCarnivores crossed through 1 m and 1.5 m culverts (Mateus et al. 2011)
Deer culvert ≥ 3 m tall, openness ≥ 0.6Reed et al. 1975; Sawyer et al. 2012
Underpass where the road is ≥ 3 m above its surroundingsRoom for a 3 m-high deer underpass (Sawyer et al. 2012)
Fencing with every structureFences plus crossings cut large-mammal roadkill 83% (Rytwinski et al. 2016)
Barriers ≥ 8 ft block deer2.4 m fences stopped every deer jump in trials (VerCauteren et al. 2010)
Fence-end zone 1 kmCollisions were greatest within 1 km of fence ends (Clevenger et al. 2001b); checked at 300 m and 500 m

Conventions with no study behind the exact value: 500 m segments; 300 m habitat strips (checked at 100 m and 500 m); the 30–150 m slope strip; the top-decile cutoff and 0.5 mi zone merge (reporting choices); the 3 m cut test for overpasses; the 50% fenced-length test; and the rules for joining mapped barrier pieces.

Road curvature and a fixed 35° slope cutoff were dropped: studies disagree on curves (deer collisions were more frequent on straighter roads in Bashore et al. 1985), and no study sets a slope cutoff.

Caveats

  • Survey effort is treated as equal across roads, though volunteers drive some routes far more than others. Only the survey's 2025 records are used; iNaturalist records from every year are included.
  • Roadside barriers were mapped from Street View on parts of US 101, SR 154, SR 246 and SR 1 only; they shape recommendations and a separate ranking, not the main score.
  • Speed limits come from sampled HPMS sections and OpenStreetMap; where neither has a value, a class-based estimate fills in.
  • The Caltrans culvert tables say nothing about culvert floors (substrate, sediment or standing water), which affect use; each candidate needs a field check.
  • Each candidate zone needs a site visit before design: what the data shows as a culvert or bridge may differ on the ground.

What goes into the need score

Each component is converted to a percentile rank across the county's 953 highway segments, so 80 means a segment scores higher than 80% of segments on that measure. The seven percentiles are averaged with equal weights: no study supplies weights for these criteria, and equal weights are a robust default when weights can't be estimated from data (Dawes 1979). Select any highway segment on the map, or open an example, to see its breakdown.

Roadkill hotspots

1/7
Measures
Dead animals within 500 m of the segment, counting groups that use crossings (mammals, reptiles and amphibians; birds and unidentified animals left out).
How it's scored
The same records are scattered at random along the network 999 times. A segment whose count beats the 95th percentile of those scatterings is a hotspot, and its score is how far it exceeds that line (Coelho et al. 2008). The 500 m radius is where hotspots of different animal groups overlap most (Teixeira et al. 2013).
Source
Room to Roam survey (2025 records) and iNaturalist (all years).
Why it matters
The most direct evidence of where animals try to cross and fail. Its weakness is effort: places people drive and report more look worse.

Connectivity

2/7
Measures
CDFW's connectivity rank (1–5) for the landscape around the road.
How it's scored
Percentile of the rank. CDFW's list of priority movement barriers is used to check the results (Gaviota Pass) rather than scored.
Source
CDFW Areas of Conservation Emphasis, terrestrial connectivity (ds2734).
Why it matters
A statewide view that doesn't depend on anyone's survey effort, so it catches important corridors where few people report roadkill.

Habitat on both sides

3/7
Measures
Share of natural land cover in a 300 m strip on each side of the road, taking the weaker side.
How it's scored
Tree, shrub, grassland, bare ground and wetland count as natural; cropland, buildings and open water do not. The strip width is a convention; 100 m and 500 m strips keep 85–88% of the top decile.
Source
ESA WorldCover 2021 (10 m).
Why it matters
Studies tie crossing use to habitat and cover at the entrances (Clevenger et al. 2001a; Cain et al. 2003; Ng et al. 2004). A crossing has an entrance on each side, so the weaker side sets the score.

Protected land

4/7
Measures
Share of protected land in the same 300 m strips, weaker side.
How it's scored
Percentile, like habitat. About 90% of segments have no protected land on at least one side, and tied segments share a rank, so a value of zero still scores about the 45th percentile.
Source
California Protected Areas Database (CPAD 2024a).
Why it matters
Crossings are most cost-effective where they join conservation areas (Sijtsma et al. 2020), and protected land keeps the habitat at both ends from being developed.

Traffic volume

5/7
Measures
Average annual daily traffic (AADT).
How it's scored
Percentile of AADT, so only the order of segments matters.
Source
FHWA Highway Performance Monitoring System 2024.
Why it matters
Traffic volume drives how strongly roads repel and divide animal populations (Jaeger et al. 2005). Collisions rise with traffic at low and moderate volumes but can level off on the busiest roads as animals stop trying to cross (Seiler 2005; Grilo et al. 2015); either way, a busier road is a bigger obstacle.

Speed

6/7
Measures
Posted speed limit.
How it's scored
Percentile of the speed limit.
Source
HPMS sample sections and OpenStreetMap (via Overture Maps), with a class-based estimate where both are missing.
Why it matters
Speed is among the strongest predictors of collisions with large animals, and lower speeds are a recommended fix (Seiler 2005). Stopping distance grows with the square of speed (Fambro et al. 2000), which matters most after dark, when deer collisions are 14 times more frequent than just before sunset (Cunningham et al. 2022). Locally, roadkill per km rises from 0.27 at 45 mph or less to 0.51 at 65 mph or more.

Flat approach

7/7
Measures
Mean slope of the steeper roadside strip (30–150 m from the road).
How it's scored
Percentile of flatness (the negative of mean slope). Segments where a waterway bridge or a culvert at least 1 m tall carries a drainage under the road get the flattest value.
Source
USGS 3DEP 1/3 arc-second elevation (about 10 m).
Why it matters
Collisions are less likely where a road crosses steep slopes and more likely in level terrain and where it crosses drainages (Gunson et al. 2011); cougars in southern California travel along canyon bottoms (Dickson & Beier 2007).

Do the components overlap? Only modestly. Across all segments, the strongest rank correlations are between connectivity and habitat (0.41), since CDFW's connectivity ranks draw partly on natural land cover, and between traffic and speed (0.34). Protected land looks independent overall (0.16 with connectivity), but 90% of segments have none on at least one side; among the 98 segments with protected land on both sides, it tracks connectivity at 0.42. No pair comes near the 0.7 or so where two measures would duplicate each other.

LayerSourceUsed for
RoadkillForestWatch Room to Roam survey (2025 records); iNaturalist (Dead Animals of Santa Barbara County, California State Roadkill, GLOBAL Roadkill)Hotspots, roadkill component
Culverts & end treatmentsCaltrans District 5 drainage inventory (provided by Caltrans; not published online)Existing passages, size classes
Highways, postmilesCaltrans GIS Data Library: State Highway Network, postmilesSegments, locations
BridgesCaltrans: State Highway BridgesExisting structures
Traffic countsCaltrans: Annual Average Daily Traffic 2022Traffic component (with HPMS)
Lanes, AADT, speed (sampled)FHWA HPMS 2024Traffic and collision components
Speed limitsOpenStreetMap contributors, via Overture MapsCollision component
ElevationUSGS 3DEP 1/3 arc-secondGrade, embankment, steep approaches
Land coverESA WorldCover 2021Habitat on each side
Connectivity & barriersCDFW Areas of Conservation Emphasis; Wildlife Movement Barriers (ds2867); Big-Game Priority Areas (ds2889)Connectivity component
Protected landCalifornia Protected Areas Database 2024aProtected component
Roadside barriersForestWatch barrier survey (not published online)Fence ends, recommendations

3D map

Eight kinds of solutions, eight places

Each pin is a real stretch of highway that illustrates one type of fix. Select one to fly there, then use the buttons at the top right to rotate and tilt the view (or right-drag with a mouse, Shift + arrow keys, or two fingers on a touch screen). Click any road, culvert, bridge or animal for details.

Loading imagery and terrain…
Imagery: USDA NAIP · Terrain ×1.4