#cap=… — or ask for a new one. Part 1, the five captured farms, needs no link:
open it here.
Part 2 of 2 · Live
Point at any farm on Earth
We read the imagery directly: Sentinel-2's ten 10 m and 20 m bands and Landsat's thermal measurement, with radar predicted from them. Vegetation indices are off by default; tick “Add customer indices” if you want one, shown as a labelled derived view.
- Pick a point — type coordinates, click the map, or choose one of the five farms from part 1.
- We fetch the newest satellite image of that spot — Sentinel-2, free and public, a new pass every few days.
- About a minute later you get four views of the imagery itself: true colour, a sharpened preview, and near-infrared and short-wave infrared false colour.
- Every view says what it is — measured by the satellite, derived from its bands, or predicted by a model.
- Then the honest part — what a satellite alone can and cannot tell you, and why.
← Part 1: five real farms, already processed
Why the labels matter
False colour is still a measurement: the satellite's own bands, shown in colours your eye does not see. A vegetation index, if you switch one on, is arithmetic over those bands rather than an observation, and it is labelled as a derived view. A sharpened preview is a model's reconstruction, not a finer measurement. The list at the bottom shows, without softening it, everything the product cannot yet tell you.
Click the map or type coordinates. Every request covers the same fixed square around the point; its real width is printed with the result.
Why no size control
There is no area-width control, because there is no choice to offer: the engine's combined call reads one fixed window per acquisition, so every request returns the same width whatever is asked for. The width and pixel count you actually get are printed with the result, read from the response — never from what was requested. It is an acquisition request, not a saved field boundary.
Click the map to drop the point, or drag the marker. The map picture is Esri World Imagery, used only to choose a point — every result below comes from Sentinel-2. Without a pointer, the latitude and longitude boxes above do the same job.
No point selected yet.
Fetching this acquisition
Not started.
The highlighted stage is a local timer, not progress reported by the server: one request is in flight and it answers once. The elapsed seconds are real, and the measured timings replace this estimate as soon as the response lands.
One acquisition
This spot, as the satellite saw it
What this spot shows
Scene details
Measured pixels beside reconstructed pixels
Drag across the image to compare. Keyboard: focus it and use the arrow keys.
Four views of the same ground
- True colour — what the satellite measured, at 10 m per pixel.
- Sharpened preview — a model's reconstruction at 2.5 m. Finer grid, not a finer measurement.
- Near-infrared false colour — measured 10 m bands; growing plants show red.
- Short-wave infrared false colour — measured 20 m bands; wet and dry ground, plants and soil show apart.
Development build — what that means here
These panels are this development build's output, not accepted product. The three product capabilities whose ground they cover — a field scene preview, a sharpened preview and an index layer library with its band composites — appear below among the things this product cannot yet do, each with the prerequisite acceptance is waiting on. Both statements hold at once: what is on screen is development-grade, and the product catalog has accepted none of it.
Customer-requested indices (opt-in, derived)
- Shown because you asked — you ticked “Add customer indices”; they are off by default.
- Derived, not measured — arithmetic over the measured bands. Proxies, not crop measurements.
- Never an input to anything else here — nothing else on this page reads these two.
Readings across the whole window, not across a field
- Averaged over the whole square — roads, farmyards, neighbouring fields and water included.
- Shown as a range (10th–90th percentile), so a mixed square does not read as one verdict.
Why this is not a field statistic
No boundary exists here, so neither reading is a field statistic. Each one is the mean over every pixel of the fixed window the response delivered — roads, farmyards, neighbouring fields and water included — with its 10th–90th percentile spread, so you can see how wide the mixture is rather than read one number as a verdict. The capability that would supply a traced boundary to average inside is listed below as not demonstrable from a browser, which is precisely why this is a window statistic. Both readings are derived indices — ratios computed from Sentinel-2 bands, one of which is area-averaged onto the other's grid for NDMI — a proxy rather than a measurement, with no cloud or shadow mask applied to either.
The iPhone app's current NDVI rule: a baseline we measure against, not our method
- Shown because you asked — you ticked the outline rule; it is off by default.
- A learned boundary model is in development (boundary track); this rule is what it is measured against.
Everything else the models see here
Each view is labelled with how far to trust it. Most are research-grade predictions.
Scene quality
The part most demos leave out
What a satellite can't tell you (yet)
The full product capability list
Loading the capability catalog…