Blog

How many ground control points does a stereo satellite pass need?

Aerial view straight down over a pine forest split by a single road.

Every GIS lead who's priced out a VHR stereo tasking order asks this before signing off: how much ground control do we need to trust the elevation model that comes back? Fewer points than you'd guess, placed smarter than you'd guess. It depends less on the satellite and more on what you're asking the canopy surface model to do.

What ground control fixes

A stereo pair doesn't arrive raw. The vendor's rational polynomial coefficients (RPCs) already encode the satellite's orbit, attitude, and timing well enough to place most pixels within a few meters of their true position, no survey crew required. Ground control doesn't build that geolocation from scratch. It corrects the systematic bias left in that model: a consistent tilt or offset across the scene, not the pixel-level noise in any single height return.

A bias error reads straight through as a biomass error. If the model sits a meter high on the east side of a stand and a meter low on the west side, that tilt rides into the height-to-volume regression and shows up as skewed standing volume across the property, not just as a cosmetic elevation offset.

How many points, and where they earn their keep

For a single stereo scene over a compact, low-relief block, three to five well-distributed ground control points are usually enough to pin the bundle adjustment: one near each corner and one somewhere in the middle, ideally spanning the full elevation range on the property rather than clustering on the flattest, most accessible part of it. That last part trips up a lot of first-time flights. A crew sets GCPs where the truck can park, which tends to be the low, flat pad by the gate, and then wonders why the model drifts on the ridgeline half a mile away.

Terrain complexity and project extent change the math. A holding with 300 feet of relief, or a carbon project area that spans several watersheds, needs control distributed across each elevation band and each sub-area the model has to report on separately. More points, placed to bracket the terrain the model has to cover, beats a dense cluster near the access road.

You don't always need a fresh survey crew, either. If there's a known hard point nearby, a bridge deck, a section corner, a parking lot pad with a surveyed elevation, that can serve as control without standing up a new GPS occupation for every pass.

Why canopy work needs more than bare-ground control

A standard orthorectification job stops at bare-ground control. A GCP set on exposed ground, a road, a clearing, a log deck, tells you the bias in the digital surface model relative to true elevation. It says nothing about whether the canopy height model built on top of that surface, the part that drives your standing-volume estimate, is reading tree height correctly.

To check that, you need independent height checks that sit under canopy: cruise plot data, known tree heights from a recent timber cruise, or LiDAR-derived check points if an older flight exists for part of the holding. Ground control gets the surface model's geometry right. A canopy height check tells you whether the height-to-volume regression built on it holds up for your species mix and stand structure.

None of this calls for a full LiDAR campaign to validate. It calls for knowing the difference between correcting a satellite's geometry and validating a biomass estimate, and budgeting ground truth for the second question even after the first one's handled.

If you're scoping a stereo tasking order and want the ground control plan matched to a defined parcel or carbon project boundary rather than guessed at, see how Canopy Height Mapping builds a standing-volume layer from VHR stereo for a holding or project area. Reach out if you want that control plan sized to your specific acreage and relief before the next tasking window.

Start a pilot

← Back to the blog