Control Surveys

A control survey establishes a small set of points whose positions you trust more than anything else on the project — then everything else (topo shots, stakeout, drone ground control, scan registrations) is measured relative to them. Get the skeleton right and a season of daily work hangs on it cleanly. Get it wrong and every deliverable inherits the wound.

• Horizontal control. Positions (northing/easting or lat/long) of stable monuments. Established by static GNSS, repeated RTK, or traverse from existing control.
• Vertical control. Elevations carried by differential leveling (still the gold standard) or GNSS + geoid model where the tolerance allows. Benchmarks are vertical control with a published pedigree.
• The hierarchy. Control comes in orders: national reference stations (CORS) at the top, then primary project control, then secondary points densified from it, then everyday measurements. Each tier is measured tighter than the tier it supports — error flows downhill and must never flow up.

“Accurate” isn't a number. Standards bodies define control quality two complementary ways, and a professional deliverable states which one it's claiming.

• Relative (proportional) accuracy. The classic order/class system: closure expressed as a ratio, e.g. 1:10,000 — one part of error per 10,000 parts of distance measured. Natural for traverse and leveling loops, where error accumulates with distance.
• Positional (FGDC / network) accuracy. The modern GNSS-era statement: a point’s position is within X cm of the datum at 95% confidence. Natural for GNSS networks, where a point 50 km from the CORS can be as good as one 5 km away.
• Local vs network accuracy. Two GCPs can each be 4 cm from truth (network) yet 1 cm relative to each other (local). Drone mapping and construction live on LOCAL accuracy — consistency between your control points — which is why one bad point in an otherwise consistent set stands out so loudly.
• NGS datasheets. Published control comes with a datasheet: coordinates, datum/epoch, accuracy class, monument description, recovery history. Reading one before you trust a monument is the difference between control and folklore — the monument may have been disturbed since publication.

The design principle behind every good network is redundancy: more measurements than the minimum, so errors have a way to reveal themselves. A point measured once is an assertion. Measured twice, it's a check. Measured from two independent setups, it's control.

• Anchor to something above you. OPUS, CORS, or published monuments — so your project lives in a frame the rest of the world can reproduce.
• Set pairs, not singles. A lone control point can’t orient a total station. Pairs with clear intervisibility, 150 m+ apart where the site allows, give you orientation plus an instant distance check.
• Build closed figures. Loops and braced figures force errors to show up as misclosures. An open-ended spur hides its own blunders.
• Independent occupations. For GNSS control: occupy each point twice, hours apart or on different days, with re-initialization between. Same answer twice from different satellite geometry is the cheapest strong evidence there is.
• Monument for the project’s life. Rebar + cap in undisturbed ground, out of the construction footprint, described well enough that a stranger can recover it. The best network in the world dies when the dozer eats it.

Here's the whole loop compressed into one site visit — setting the points, observing them, and checking the numbers before anything else gets measured.

Drone mapping didn't invent new geodesy — GCPs and checkpoints are a control survey wearing high-vis paint. Everything above applies directly:

• GCPs carry the frame. They put your reconstruction in the project datum. 5–10 well-placed GCPs beat 30 careless ones; corners + center beats a row down the middle.
• Checkpoints prove it. Points measured like GCPs but withheld from processing. The reconstruction’s fit to its own GCPs is self-flattering; checkpoint RMSE is the number a client can trust — report it.
• Same redundancy rule. Two GNSS occupations per GCP. One bad GCP quietly bends the whole block; two occupations make the bad one confess early.
• Same metadata rule. A GCP file without datum, projection, geoid, and units is a future emergency. The deep dive on flight-side accuracy lives in the Drone Surveying guide at /free/drone-course.