Underfoot

What's physically on the ground at a GPS point, or all along a path. Road, trail, forest, lawn, building, water, wetland, rail, crops, bare ground or snow: Underfoot fuses ten free open-data sources, today's weather and the newest satellite pass among them, and returns a probability for each, with its working shown.

Open the app · Download underfoot.html · How it works · Roadmap · Changelog

A 900 m line across Yosemite Valley: from the Ansel Adams Gallery over two roads, through Cook's Meadow, across the Merced River and into the forest to the Valley Loop Trail. Each point along it is a sounding.

No API keys, no account, no server. It runs in your browser, on a phone, or from a single HTML file on disk.

One line, every crossing

Draw a line, paste coordinates, or drop in a GPX, CSV or GeoJSON track. Stations are spaced along it, and every mapped road, trail, rail line and stream the line crosses gets a station of its own, named from OpenStreetMap. Where the line follows a mapped trail or road, its stations move onto it and read as what it is, and a GPS track's weave across the trail it's on doesn't count as crossings. The transect underneath shows the posterior along the whole line over its elevation profile, and a route card above it sums the line up: how much of it is forest, grass, path or road, what it crosses, how much it climbs, and which stations it's unsure about. A station worth checking on the ground, because the call is close or the sources disagree, gets an amber halo on the map and a mark on the transect's doubt band, and the answer says why: grass 49% or forest 48% · the map says grass, land cover says forest. The five most worth a look, 40 m or more apart, make a numbered check list for your next walk, which exports as GPX waypoints.

Right or wrong. Under any answer, say whether it's right, wrong or you're not sure, what's really there and how you know. A mark keeps what each source said at that moment and stays in your browser; the Marks menu lists, edits and exports them. With ten or more, Refit fits the source weights to your marks, shows how many each set gets right on marks it never saw, and lets you choose; Reset puts the defaults back exactly.

On the demo line, the gallery comes out building 97%, Northside and Southside Drives paved 98%, Cook's Meadow grass at up to 98%, the Merced River water 98%, and the Valley Loop Trail, under 55% tree canopy in an evergreen-forest pixel, path 91%. Smoothing along the line cleans up one-station flicker inside a run of forest or meadow but never blurs a road, building or crossing.

An area, in acres

Outline a lot and Underfoot reads all of it. Field maps of 2 m cells tile across the outline, each cell taking the map, the imagery and the land cover, canopy and satellite pixels under it. The Area card gives acres per class, the map shows each cell's call as a mosaic, and GeoJSON (the outline and each class's cells) and CSV take it away. A 5-acre lot on Cook's Meadow reads in about 5 s: forest 2.5 acres, grass 1.9, the trail across it 0.16. The classes always sum to the outline's own area. One outline takes up to 64 tiles, about 180 acres.

Getting across

Every station is also rated for crossing it on foot, on an ATV and in a truck: a speed and a band, go, slow or no-go. The rating reads what the ground probably is, its grade and steepness, its roughness, today's soil water on soil that drains poorly (USDA's soil survey in the US), the trees overhead, the type of mapped path, and what the line crosses. The transect gets a band per preset, the route card sums the line (Foot 2.6 km/h, blocked at 654 m: the river, no bridge), and each station's panel shows every factor. It's a model of the ground, not of access or the vehicle. How it works, and what it doesn't know.

Over the years

Under the answer, the time machine reads every distinct picture Esri's archive of past imagery holds of the spot (World Imagery Wayback, releases since 2014 with captures back to about 2010). Most of its 197 releases reuse the tile underneath, so asking each release where its tile came from walks back through only the dozen or so that changed it. Each picture is scored by the imagery classifier, the captures are smoothed across the years, and a change is flagged between the two captures either side of it: at Giga Texas, changed between Nov 2020 and Jan 2022: bare ground → building. Tap a picture to see that capture on the map. The past is the photo alone, with no historical map, and the panel says so. On six places that changed and five that didn't, it flags 5 of the 10 changes and none of the unchanged places, with two false flags at a gravel lot (every place).

This is the beginning of the README. Read the rest at the original. Read the full README ↗