Survey-grade data. Cinema-grade footage.

The automated pass compares each frame against what a clean surface should look like at that resolution — an edge that shouldn't be there, a texture break, a contrast shift where the material changes. It's good at finding candidates. It is not good at knowing what it found.

A shadow from a parapet reads the same as a hairline crack to an edge detector. A water stain reads like corrosion. A bird on a panel reads like damage until someone looks at the next three frames and watches it fly away. The flagging pass exists to make sure nothing gets missed — not to make the call.

So every flagged frame gets a certified pilot's eyes before it reaches a report, and roughly a third of flags get dismissed on review. That ratio is the point: a system that flagged less would be missing things, and one we trusted without review would be guessing. The report you get has already been argued with.

Ground sample distance is the size of one pixel on the ground. At 120 m with a 20 MP sensor that lands near 2.4 cm — every pixel covers a 2.4 cm square of dirt.

It sets the ceiling on everything downstream. You cannot measure a crack narrower than a pixel, and you cannot label an asset you cannot resolve. If the deliverable needs 1 cm detail we fly lower and accept more flight lines, more batteries, and a longer day.

The trade is unforgiving: halving GSD roughly quadruples the image count. So the useful question at scoping isn't how sharp we can get — it's the smallest thing you need to see. We fly to that.

The tell of an amateur aerial is speed. A drone moving at the speed it can move reads as a drone; one moving at the speed a crane or helicopter would reads as cinema.

Three things fix most of it: a wider radius than feels necessary, a slower stick than feels productive, and a gimbal that leads rather than chases. Radius does the heavy lifting — at 15 m the parallax swings hard and the subject wobbles in frame; at 40 m the same orbit becomes a slow reveal.

We usually shoot the orbit twice, once at the speed we want and once slower, because a 60 fps pass conformed to 24 hands the editor a speed ramp that a real-time pass can't.

Part 107 covers most commercial work in the US with no special paperwork: up to 400 ft above ground level, within visual line of sight, under 100 mph, and clear of people who aren't part of the operation.

Two things people assume need waivers but generally don't. Night flight is permitted with anti-collision lighting visible for three statute miles and current training. Controlled airspace is usually a LAANC authorization that comes back in minutes rather than months.

What still needs real lead time: beyond visual line of sight, flight over open-air assemblies, and some operations from a moving vehicle. If your site sits under a Class B shelf or beside a hospital helipad, tell us early — that's the constraint most likely to move your date. Rules change, and we confirm current requirements per mission rather than from memory.

Three ways to make a map that holds its position. Ground control points are surveyed targets on the ground that processing ties the model to. RTK corrects the aircraft's position live from a base station or network. PPK records raw satellite data and corrects it afterward.

They aren't interchangeable. RTK is fastest and fails quietly when the radio link drops. PPK survives a lost link because the correction happens later, which is why it's our default over water and in terrain. Ground control is slower and still the most defensible — if the deliverable is going into a legal record, we want targets on the ground regardless of what the aircraft was doing.

Most jobs get a hybrid: PPK for the flight, with four or five checkpoints to prove the result rather than produce it.

Leaving US airspace changes the paperwork more than the flying. Every island is its own civil aviation authority, most want an operator permit rather than a nod to Part 107, and lead times run from a few days to several weeks.

Gear is the other half. Batteries over 100 Wh draw questions at check-in, spares need terminal covers, and some jurisdictions want a carnet for professional equipment entering temporarily. That goes into the schedule rather than getting discovered at customs.

Then there's salt. Airframes flown near surf get a freshwater wipe and a motor inspection the same day, not at the end of the trip. Corrosion doesn't announce itself until a bearing is already rough.

A photograph taken from above still has perspective in it. Buildings lean away from the centre, terrain compresses toward the edges, and any measurement you take is wrong in a way that changes across the frame.

An orthomosaic removes that. Every pixel is corrected to a true top-down view using the elevation model built from the same flight, then hundreds of corrected images are blended into one. The result is a picture with a map's geometry — a distance measured in one corner means the same thing as a distance measured in the other.

That's the whole difference. It's why an orthomosaic drops into CAD with a coordinate system attached, and why a beautiful aerial photo — which we're glad to take as well — can't do that job.

Log profiles look wrong out of camera on purpose: flat, grey, low contrast. What they're doing is spending the sensor's range on information instead of on a look.

The payoff is the sky. A bright overcast against a dark roofline is the shot most likely to clip, and a baked-in Rec.709 profile throws away that highlight detail before anyone sees the file. Shot in log, the sky comes back in the grade.

The cost is discipline — log needs correct exposure and ND filtration, and it needs someone to actually grade it. We deliver both: a graded master for immediate use, and the log files if your colourist would rather start from source.

NDVI compares how strongly vegetation reflects near-infrared against red light. Healthy leaves bounce back a lot of NIR and absorb red; stressed ones don't. The index turns that ratio into a picture, and stress shows up in it before it's visible to the eye.

What it doesn't tell you is why. Drought, compaction, nitrogen deficiency and root disease can all read as the same dull patch. NDVI is very good at pointing at a place and nearly useless at naming a cause — it narrows down where to walk, and someone still has to walk it.

It also saturates. Once a canopy closes the index flattens and stops separating good from very good, which is why late-season maps often disappoint growers who loved the early ones.