Where every number comes from

Parallel Maps puts other people's maps side by side and adds a read of its own. This page says what each thing on the map is based on, how it is calculated, and what it cannot tell you. Back to the map →

Trails and tags

The trail lines, their names and their tags come from OpenStreetMap, fetched live for the area on screen through the Overpass API. OSM is licensed under the ODbL; the map corner carries the attribution.

The tags that do work here: surface, smoothness, sac_scale (hiking grade), mtb:scale, trail_visibility, width, steps, wheelchair, and the access tags access, foot, bicycle, mtb, horse. A survey of six US trail regions in August 2026 found difficulty tags on about 1% of trails and surface tags on 20 to 71%, which is why the map colours by surface and why the verdicts lean on surface and access rather than on rated difficulty. The trail card lists what OSM does not record for that trail.

Elevation and grade

Elevation comes from the USGS 3D Elevation Program (3DEP), lidar-derived and 1 m where it has been flown, sampled along a trail or route at 200 points. One-metre lidar is noisy point to point, so elevations are smoothed over about 25 m before any grade is read. The headline grade is the steepest sustained pitch over a rolling 100 m window, which predicts whether you can get up something better than the instantaneous maximum does. The profile's colour bands are the local grade after that smoothing.

Terrain character

This is Parallel Maps' own layer: what moving through the ground is like, computed for every 50 m of trail in the areas where it has been run. It uses only public-domain inputs, so nothing here is traced from a commercial map or satellite view. The satellite basemap the site can display is never read.

USGS 3DEP bare-earth elevation at 1 m, exported from the national mosaic. Slope, grade, cross-slope, roughness, relief, curvature.
USDA NAIP aerial imagery at 0.6 m with a near-infrared band, read from Microsoft's Planetary Computer mirror. Vegetation cover, canopy shadow, whether the tread is visible from the air and how wide it reads.
USGS National Hydrography Dataset. Streams, lakes and wetlands: distances and crossings.
OpenStreetMap gives the line the segments are cut from. The metrics themselves are computed from the three sources above, and the trail line's own error, typically a few metres, is the biggest known source of uncertainty.

How a segment is measured

Each way is cut into pieces of about 50 m and sampled every 2 m. Because a mapped line often sits a few metres off the tread, each sample point first moves to the flattest strip within 5 m across the line, the bench a trail cuts into a hillside. Grade is the slope of the elevation profile along the line, smoothed over about 18 m, never the hillside's own gradient; cross-slope is the hillside gradient across the line. Roughness at 2, 10 and 50 m is how much the ground departs from a smoothed copy of itself at that scale. Where the profile climbs and drops far more than the two ends of the segment differ, on a steep side-hill, the segment is flagged as uncertain and the card says so.

The metrics

Every metric shown on the map, with its unit and the scale the colours run over. This table is read from the same file the map uses, so it is always the current set.

Metric	Unit	Scale	What it is	Source
Loading the registry…

The derived indices

Technicality, runnability, enclosure, exposure and water interaction are labelled v0 because they are plain rules on fixed physical scales, not fitted models. They exist so the layer has an experiential reading before enough hand ratings exist to fit one; the Rate segments tool collects those ratings. The rules, where ramp(x, a, b) runs from 0 at a to 1 at b:

technicality = 0.45·ramp(roughness 2 m, 0.015–0.07 m) + 0.30·ramp(footing variability, 1.5–7°) + 0.25·ramp(steepest grade, 8–30%)

runnability = (1 − technicality) × (1 − 0.8·ramp(mean grade, 3–22%))

enclosure = 0.7·vegetation cover 25 m + 0.3·ramp(canopy shadow, 0–0.4)

exposure = 0.5·ramp(steep ground nearby, 0.05–0.5) + 0.3·ramp(drop within 20 m, 3–12 m) + 0.2·ramp(cross-slope, 10–40%)

water = 0.6·min(crossings, 1) + 0.4·(1 − ramp(stream distance, 0–150 m))

The scales came from the first area computed, Lake Padden: asphalt sits at 0.009 m of 2 m roughness, smooth dirt at 0.016 to 0.021, the rooty ridge trails at 0.06. They will be revised as more areas and more ratings come in.

What a trail is good for

One verdict per activity, on the trail card and, one activity at a time, as a map colouring. They are suitability calls, not difficulty ratings: a rooty, steep trail is a poor walk and a good mountain-bike ride at the same time. The levels are good, OK (with caveats), poor, not allowed and unknown, and each row says what it is based on and whether that came from OSM tags, from terrain character, or both.

Access first. If OSM says an activity is not allowed (foot=no, bicycle=no, access=private…), that is the verdict, in red.
Made surfaces. On a paved, compacted or fine-gravel trail, or any footway or cycleway, the tread is smooth by construction, so the lidar roughness beside it (a rock cut, a wall, a ditch) does not count against running, walking or a stroller; grade still does. The Interurban under Chuckanut Drive read "very technical" before this rule.
Hiking is good on any trail open to foot traffic; only tagged exposure (sac_scale T3 and up), a faint tread or exposed ground nearby lower it.
Trail running turns OK above technicality 0.4 or 14% mean grade and poor above 0.7 or 22%, from terrain where computed; from tags alone, steps and T3+ ground are poor, T2 and rocky MTB grades are OK.
Walking wants a gentle, untechnical trail: good under 8% mean grade and technicality 0.35, OK to 15% and 0.6, poor beyond.
Stroller wants a made surface (paved, compacted, fine gravel) on a gentle grade with no steps; loose gravel or dirt is OK at best; wheelchair=yes counts as good, wheelchair=no as poor.
Mountain bike is good on natural-surface trails and tracks open to bikes, technical or not, and says how technical; paved paths are only OK; alpine ground is a hike-a-bike.
Gravel bike wants a track, a cycleway or a firm surface. The smoothness and surface tags give the tyre-width reading; natural-surface singletrack is OK at best; a purpose-built mountain-bike trail (designated or one-way for bikes), tight corners in the lidar, over 4.5 cm of 2 m roughness or technicality 0.55 is poor.
Road bike is good on paved surfaces, OK on compacted or fine gravel with a caveat, poor on anything looser, and poor on any purpose-built mountain-bike trail.
Horse is judged from tags alone: bridleways and horse=yes/designated are good, horse=no not allowed, anything else unknown.

Roads

Roads are fetched as connectors so a route can link two trails by a lane, and the map can say something about them. From OpenStreetMap: the class (residential, tertiary, secondary…), the posted speed, lane count, sidewalks, bike lanes, lighting, surface and the rare shoulder tag. Where terrain character has been run, the same imagery read that measures a trail's tread measures a road's paved width, and subtracting the lanes (tagged, otherwise two, or one for a one-way or service road, at 3.4 m each) gives an estimate of what is left per side. Under about half a metre the card says little or no shoulder; under 1.2 m, a narrow shoulder. On residential and service streets, where that space is parking and kerb rather than shoulder, it says room beyond the lanes instead. Grade, curvature (as a sight-line proxy) and tree cover come from the same segments.

The road verdicts follow from those facts: a sidewalk or a quiet street is good on foot; a busy class or a posted speed of 40 mph and up with no room beyond the lanes is poor on foot and on any bike; a bike lane, a tagged shoulder or a measured one of 1.2 m and up counts as room; an unpaved road is a ride in itself for a mountain or gravel bike and poor for road tyres. None of this is a safety assessment. It is what the tags and one date of imagery show, stated so you can check it on the ground; shoulders, sight lines and traffic change.

Routing and time

Routes are solved in your browser over the trails on screen, snapping each click to the nearest trail or road and running A* over the network. With Prefer trails on (the default), a road costs four times its length in the search, a service or living street three, a sidewalk or crossing two and a half, so a route takes a lane only where nothing else links two trails; the distance shown is always the true length, and the route card says what share is on roads. Access rules follow the Travelling by choice: trails closed to that way of travelling are left out and marked with red crosses. Where the mapped network has no connection, the leg is drawn as a dotted straight line and the card says so. Time is a Naismith-style estimate: 5 km/h on foot, 11 by mountain bike, 15 by gravel or road bike, plus an hour per 600, 500 or 450 m of ascent. Treat it as a sanity check against daylight, not a prediction.

Search, weather, daylight, ratings

Search uses Photon, re-ranked by distance from the current view.
Weather is the National Weather Service point forecast for the route's start; daylight comes from sunrise-sunset.org, using civil twilight.
Ratings you give in Rate segments stay in your browser until you export them. A shared store, with per-segment means shown on the card, is built but not yet switched on; when it is, ratings are keyed to an anonymous browser id, never a person.
Other maps. The links open Strava, Trailforks, CalTopo, Gaia, AllTrails and OpenStreetMap on the same spot, in your own session on their site. Nothing from them is copied or stored here.

What it cannot tell you

Where the trail actually is. OSM lines are typically a few metres off the tread. Terrain character corrects for this on side-hills by finding the bench, and flags segments where it could not; the honest fix is an independent centreline, which is future work.
Roots and small rocks. A 1 m elevation model separates paved from unpaved and finds incised or rocky trails, but a root is smaller than a pixel. Point-cloud lidar, which is the next input, sees more.
Anything under canopy, from the air. Most Pacific Northwest trails are invisible in aerial imagery. Width and visibility are only reported where the tread shows, and grass beside a path can read as tread.
Conditions. Nothing here knows about mud, snow, blowdown or closures. Surface tags describe a trail in general, not today.
Coverage. Terrain character exists for a few areas; everywhere else the verdicts run on OSM tags alone and say so.
Roads. The paved width is read from imagery on one date; parked cars, shadow and tree cover hide it, and a painted shoulder is not distinguished from an unpainted one. Lane counts are assumed where OSM does not tag them.

Licences

Trail data © OpenStreetMap contributors, ODbL. Elevation (USGS 3DEP), imagery (USDA NAIP) and hydrography (USGS NHD) are works of the United States government in the public domain. Terrain character and the verdicts are computed by Parallel Maps from those sources. Everything on the site is free, and there is no paid tier.