Surface Routing: Finding Real Gravel and Avoiding Unrideable Roads
20+ years testing gear in Colorado backcountry
How to Find Gravel Roads for Bikepacking That Are Actually Rideable
To find gravel roads for bikepacking, filter a map for unpaved surface tags to build a candidate list, then verify every unpaved segment against satellite imagery and one independent second source before it goes into your route file. The filtering takes ten minutes. The verifying is the part that decides whether the trip works.
Here is why the second step exists. Your routing app does not know what a road is made of. It knows what a person typed into a database field, possibly six years ago, possibly from a car window, possibly never. Surface data is crowd-contributed metadata, not observation.
So "unpaved" on your screen is a claim of unknown provenance. The skill is testing that claim cheaply — about ninety seconds per segment — instead of riding into it. Everything else in route planning and navigation for bikepacking assumes the line under your cursor is real.
What the Map Is Actually Telling You
Whichever planner you use, the surface shading on your screen comes from a field somebody filled in by hand. In the open-data world that field usually lives in OpenStreetMap, whose documentation is public — so that is where to learn to read it properly, whatever app renders it.
OSM does not have one surface field. It has three, and they ask three different questions.
surface asks what it is made of. The wiki defines the key as describing the physical surface of roads, footpaths and other surface features, particularly regarding material composition and structure.
tracktype asks how firm the roadbed is. The tracktype page calls it a simple classification of tracks and other unpaved ways, considering surface firmness, how much the road has been improved relative to the surrounding ground, and how well maintained it is. It runs grade1 (solid, usually paved) to grade5 (soft, an unimproved track lacking hard materials, uncompacted, surfaced with soil, sand or grass).
smoothness asks how rideable it is. The smoothness page describes a classification scheme for the physical usability of a way for wheeled vehicles, particularly regarding surface regularity and flatness. Its documented values run excellent, good, intermediate, bad, very_bad, horrible, very_horrible, impassable.
Those three are independent. A road can be gravel, sit on a firm bed, and still be bad because it is washboarded into your fillings. Most apps then flatten all three into a few coarse buckets — paved, unpaved, path — so the detail is gone before it reaches your screen.
The surface values that matter on a loaded bike
| OSM value | What the wiki says it is | What that means loaded |
|---|---|---|
compacted | Larger and smaller material compacted together, more stable than loose gravel | The good stuff. Rail-trail-grade rolling |
fine_gravel | Fine loose gravel, described as unsuitable for road bicycles | Fast on 40mm plus, tiring on anything narrower |
gravel | Anything from track ballast down to a compacted surface | The widest range on the list. Always verify this one |
dirt | Exposed soil, also called earth | Excellent dry, a different road wet |
ground | No special surface; the ground itself shows marks of human or animal usage | Not a road. Someone drove across a field |
sand | Grains smaller than two millimetres | Walking, usually |
unpaved | The wiki says the value gives only a rough description | Tells you nothing except that it is not sealed |
The gap between compacted and gravel will change your day plan more than any other pair there, and it is the pair mappers most often get wrong.
The trap: unknown gets drawn as ordinary
Here is the single most useful thing to internalise. A road with no surface tag is not a paved road. It is a road nobody has described. Most renderers style that absence exactly like an ordinary minor road — a thin line in a neutral colour, indistinguishable from the county blacktop two miles east.
The tag tells you something. The missing tag tells you nothing, and it looks exactly like something.
An untagged road is not a paved road. It is a road nobody has looked at, drawn in the same colour as one somebody has.
Why the Line Lies Even When It Is Tagged
A tag is a snapshot from whenever a mapper last cared. Every OSM object carries the date it was last edited, but an edit is not an inspection — somebody can fix a road's name from an armchair without ever standing on its surface. Four failure modes account for most of the trouble.
The seasonal gate. A forest road that is genuinely a road, genuinely gravel, and genuinely locked behind a gate from October to June for mud season or wildlife closure. The tag is correct. The road is unavailable. Maps have no seasons.
The two ruts through a field. This is tracktype=grade5 doing its job honestly, and a router using it anyway because it is classified as a road. In August it is a pleasant farm crossing. Two days after rain on clay it is a bike-carry — which is why weather rewrites a surface plan faster than any other variable.
The ford where a bridge used to be. The road was continuous when it was mapped. The bridge went out in a spring flood and nobody edited the map, because the person who would have is not out there.
The decommissioned through road. It shows as connecting two valleys. On the ground it was closed, ripped and replanted a decade ago and the last mile is alder. Nothing removed the line.
The Four-Source Stack
Four sources, because each catches a different category of failure and none catches all four. Imagery shows you the physical world but not the law. A heatmap shows you human behaviour but not remote country. The agency map shows you the official record but not the condition. A trip report shows you the condition, but only on the day somebody rode it.
1. Satellite imagery at maximum zoom
Zoom until the tool stops zooming, then follow the segment end to end. You are looking for specific tells, not a general impression.
Canopy closing over the line. Continuous tree cover where the road should be means either leaf-on imagery hiding a fine road or a road being reclaimed. Both are worth knowing.
No tyre tracks. A maintained gravel road shows two faint wheel tracks and a lighter graded surface. A road with no visible tread has probably not been driven since the imagery was captured — though canopy or image resolution can hide tracks that are there.
A turnaround at the end of what should be a through road. A widened circle or hammerhead where the line continues on the map is the classic signature of a road that dead-ends in reality.
If your viewer keeps a historical archive, step back a few captures: a road clearly graded in an older image and faint in the current one is more likely telling you about maintenance than about image quality. What imagery cannot tell you is whether you may be there, or what the surface does wet.
2. Heatmaps, and how to read a dark segment
A heatmap answers the one question no tag can: do people actually ride here. Strava's Global Heatmap, per Strava's own description, shows where users have been active over the last year as heat on a map, is built from aggregated de-identified activity data, is updated monthly, and does not show heat in an area until multiple users have uploaded activities traversing it in the last year.
That last detail is the reading skill. A bright line means rideable and popular. A faint line may mean remote rather than bad. No line at all on a road within reach of a populated area is the real warning sign — people live nearby, so somebody would have ridden it if it went through.
Name the bias out loud, because it changes how you weight the evidence: a heatmap shows where app users ride. It under-represents remote country and over-represents commuter corridors. So a dark segment inside an otherwise bright route deserves far more attention than a dark segment in the desert. The first is a gap where people stop; the second is nobody being there.
Which tools expose surface data and heatmap layers at all varies a lot; we compare them in the best bikepacking navigation apps.
3. Official agency sources
Land managers publish the authoritative answer on whether a road exists in the system. On US national forest land that is the Motor Vehicle Use Map. The Forest Service's MVUM information page explains that an MVUM displays National Forest System routes and areas designated as open to motorized travel, with allowed vehicle classes and seasonal allowances, and states plainly that routes not shown on the MVUM are not open to public motor vehicle travel.
Read that scope carefully, because it is easy to over-read. An MVUM is a motor vehicle designation document. A road appearing on one is strong evidence it physically exists and is a recognised, administered route — which is what you wanted to know. It is not a statement about bicycle access and not a condition or surface report; the same page notes the maps do not currently show mechanized, non-motorized and over-snow use. A road can be on the MVUM and still be a washboarded mess, and can be missing from it for reasons unrelated to bikes.
County road departments and state land agencies publish comparable inventories of varying quality. Use them the same way, and never as a substitute for asking the land manager what the current rules are.
4. Trip reports, and the phone call nobody makes
The fourth source is people. Route repositories, bikepacking write-ups, forum threads and local club posts often describe the exact segment worrying you. Search the road number with the year rather than the region — "FR 2620" gets you further than "gravel riding in the national forest."
Then the underused one. Call the ranger district office or the county road department. Ask three things: is the gate open right now, is the bridge in, when was it last graded. One four-minute call beats an hour of squinting at pixels, and the person answering has usually driven the road.
Rail Trails and the Almost-Finished Problem
Rail trails are the best surface in bikepacking when complete and one of the sharpest planning traps when not. A trail can be named, signed and mapped and still have unbuilt gaps in the middle — and a gap does not strand you, it quietly routes you onto the parallel highway, which is worse than it sounds when that highway is the valley's only road.
The scale is not hypothetical. The Rails-to-Trails Conservancy's 2026 route assessment for the Great American Rail-Trail puts the preferred route at more than 3,700 miles, of which approximately 2,085 miles are existing trail and approximately 1,670 miles are identified gaps still to be developed. That is the country's most visible rail-trail project, publicly tracked and funded, and still a long way from continuous.
So verify continuity, not existence. Follow the trail end to end on imagery across the section you plan to ride, and check the managing agency's page for closed or unbuilt segments. If you want a low-stress trip rather than a research project, an established beginner route has had this work done for you — a legitimate thing to buy with a route choice.
When You Cannot Verify It
Sometimes the stack comes back ambiguous: imagery is leaf-on, the heatmap is blank because the country is empty, the agency map is silent, nobody has written about it since 2019. That is a decision point, not a research problem, and it has three honest answers.
Route around it. Cheapest by a distance, and the one we take most often. A detour costs ten minutes at the kitchen table and a few miles on the bike — a better trade than it looks, and how to build a GPX route covers reshaping a line without rebuilding the day.
Keep it, but buy insurance. Find the bail-out before you leave: the last junction where you can cheaply turn around, and the parallel road you would fall back to. Then ride the segment early in the day, with daylight in hand and legs that can absorb a surprise. A bad surface costs you hours rather than miles, which is the whole argument in how many miles per day to plan and why an unverified segment should never sit at mile 55. Save the map layers offline too, because the segment you most want to re-check is the one with no signal.
Accept it as the point. On a trip where finding out is the objective, the unverified segment is the good part. Just be honest that you chose it: expect hike-a-bike and go in on tyres you trust, since chunky unknown surfaces are where sealant earns its keep and where tubeless problems surface. What this is not is a first trip — a first bikepacking trip has enough unknowns without a road that may not exist.
Verify the Segment, Not the Map
The shift that makes surface routing work is small and slightly deflating: stop asking the map what a road is, and start asking which roads to go check. The tag is a lead. Imagery, a heatmap, the land manager and a human being are the confirmation, and you only need all four on the segments that fail the ninety-second look.
That costs about half an hour per trip. Skipping it costs the afternoon you spend pushing a loaded bike back up a road that ended in a turnaround circle — and the rest of that day's plan with it.
For the rest of the workflow this sits inside, from drawing the line to loading it on a device to navigating it when the signal goes, see our complete bikepacking route planning guide.
