GSD Calculator and Drone Mapping Accuracy Guide

The formula

Ground sample distance is the ground distance covered by a single pixel. It is the number that determines what you can resolve, and it is the first thing to fix when planning a flight.

GSD (cm/px) = (sensor width mm x flight height m x 100) / (focal length mm x image width px)

Everything on the right is either a camera specification or a decision you make. Only the flight height is yours to change.

Worked examples, real hardware

DJI Phantom 4 RTK — 1-inch sensor, 13.2 mm wide, 8.8 mm focal length, 5472 px wide.

At 100 m: (13.2 x 100 x 100) / (8.8 x 5472) = 132,000 / 48,153.6 = 2.74 cm/px

DJI Mavic 3 Enterprise — 4/3 sensor, 17.3 mm wide, 12.29 mm focal length, 5280 px wide.

At 100 m: (17.3 x 100 x 100) / (12.29 x 5280) = 173,000 / 64,891 = 2.67 cm/px

Both land near 2.7 cm at 100 m, which is why the two aircraft are broadly interchangeable for survey work at that altitude. The difference between them is endurance and wind tolerance, not resolution.

GSD by altitude, Phantom 4 RTK

Note what happens to coverage. Halving the height halves the GSD and roughly quadruples the image count for the same area — because you cover half the width and half the length per frame. A 1.1 cm GSD survey of a 50-hectare site is not a slightly bigger job than a 2.74 cm one; it is around six times the data.

From GSD to accuracy

GSD sets the ceiling. It does not deliver accuracy on its own.

With RTK or PPK georeferencing and properly deployed ground control, the working expectation is:

• Horizontal accuracy: 1 to 3 x GSD
• Vertical accuracy: 2 to 3 x GSD

At 2.74 cm GSD that is roughly 3–8 cm horizontally and 5–8 cm vertically. Aeroyantra achieves 2–3 cm horizontal accuracy on typical RTK survey datasets with good control.

The multiplier is where the craft lives. What moves you from 3x to 1x:

• Ground control quality. Well-distributed, precisely surveyed GCPs, including points at the edges and corners of the block, not clustered in the middle.
• Overlap. 70–80% forward and 60–70% side for survey work. Thin overlap produces a weak bundle adjustment regardless of GSD.
• Camera calibration. A rolling shutter on a moving aircraft introduces distortion that no amount of resolution corrects.
• Cross-hatch or oblique passes. A single-direction grid over flat, low-texture ground is where vertical error creeps in. Adding a perpendicular pass costs one battery and often halves vertical residuals.

Working backwards from the deliverable

This is the calculation that actually matters on a job, and most people do it in the wrong direction.

Do not start from "how low can I fly". Start from the contour interval or tolerance the client specified.

1. Contour interval → vertical accuracy. Vertical accuracy should be roughly one third to one half of the contour interval. 25 cm contours → 8–12 cm vertical accuracy.
2. Vertical accuracy → GSD. Divide by the 2–3x multiplier, then keep margin. 8–12 cm vertical → roughly 2–3 cm GSD.
3. GSD → flight height. Rearrange the formula. At 2.74 cm/px, a Phantom 4 RTK flies at 100 m.
4. Flight height → flight time and image count. Now you know whether the job is one battery or nine, and you can price it.

Quick reference

These are planning figures, not guarantees. Verify against independent check points on every project — see below.

Why "reported accuracy" is not accuracy

Photogrammetry software reports an RMS error after the bundle adjustment. That number is the software grading its own work: it tells you how well the solution fits the control points it was given, which is not the same as how well it fits reality.

If you used every marker as a control point, you have no independent measure at all. A suspiciously low reported RMS on a dataset with poor control usually means the adjustment has absorbed the error rather than resolved it.

Hold back check points. Survey more markers than you need, feed some to the adjustment as control and withhold the rest. The residuals at the withheld points are your real accuracy. A common split is two thirds control, one third check, with the check points spread across the block rather than bunched.

Aeroyantra reports check-point residuals alongside the adjustment result, so the independent figure is in the deliverable rather than something you have to compute separately.

Common mistakes

Flying lower to fix an accuracy problem. If the constraint is control or overlap, a finer GSD adds processing time and achieves nothing. Diagnose before you re-fly.

Using the 35 mm equivalent focal length. The formula needs the actual focal length. A Phantom 4 RTK is 8.8 mm, not the 24 mm equivalent. Using 24 mm understates GSD by a factor of nearly three.

Assuming RTK removes the need for ground control. RTK gives precise camera positions, which fixes the horizontal datum well. Vertical accuracy still benefits substantially from at least a few checked points, and you need them anyway to verify the result.

Mixing sensor width with sensor diagonal. A "1-inch sensor" is 13.2 mm wide and 16 mm on the diagonal. Use the width with the image width in pixels.

Ignoring terrain relief. Flight height is height above the ground, not above the take-off point. On a site with 60 m of relief, a nominal 100 m flight gives 2.2 cm GSD on the hilltops and 4.4 cm in the valley. Terrain-following flight planning fixes this; a flat grid over varied terrain does not.

Frequently Asked Questions

How do you calculate GSD for a drone?

Ground sample distance in cm per pixel equals the sensor width in mm multiplied by the flight height in metres multiplied by 100, divided by the focal length in mm multiplied by the image width in pixels. For a DJI Phantom 4 RTK at 100 m, that is (13.2 x 100 x 100) / (8.8 x 5472), which gives 2.74 cm per pixel. Halving the flight height halves the GSD and roughly quadruples the number of images required to cover the same area.

What accuracy can I expect from a drone survey?

With RTK or PPK georeferencing and properly deployed ground control, expect horizontal accuracy of roughly 1 to 3 times the GSD and vertical accuracy of roughly 2 to 3 times the GSD. At a 2.74 cm GSD that means about 3 to 8 cm horizontally and 5 to 8 cm vertically. Aeroyantra achieves 2-3 cm horizontal accuracy on typical survey datasets, and reports independent check-point residuals rather than only the software's own reported error.

What GSD do I need for a topographic survey?

Work backwards from the contour interval. A widely used rule is that the GSD should be no coarser than one fifth of the required vertical accuracy, and the vertical accuracy should be around one third to one half of the contour interval. For 25 cm contours you want roughly 8-12 cm vertical accuracy, which points to a GSD of about 2-3 cm per pixel. For 50 cm contours a 4-5 cm GSD is usually sufficient, which lets you fly higher and cover more ground per battery.

Does a lower GSD always mean a more accurate survey?

No. Below a certain point, GSD stops being the limiting factor and ground control, camera calibration and image overlap take over. Flying lower to halve the GSD roughly quadruples the image count, lengthens processing, and increases the chance of a weak bundle adjustment from motion blur or poor overlap. A 2 cm GSD with bad control is less accurate than a 4 cm GSD with good control.

What is the difference between GSD and accuracy?

GSD is the ground distance covered by one pixel, so it sets the finest detail that can be resolved. Accuracy is how closely the measured position of a feature matches its true position. GSD bounds what accuracy is achievable but does not deliver it: georeferencing, ground control, overlap and camera calibration determine how much of that potential you actually realise.

Related

• GCP placement for survey-grade accuracy
• Process RTK drone images without GCPs
• Aeroyantra pricing — $0 to start, $2.50 per credit