GSD — ground sample distance — is the ground size of one pixel in your photograph. It is the single number that decides whether a drone survey can resolve a fence line or only a field boundary, and it is fixed before the flight, not after. The good news is that it depends on four things you already know about your camera, and the arithmetic is one multiplication. The bad news is that three separate mistakes, all of them common, quietly turn a correct formula into a wrong answer.
Every length must be in the same unit before you divide, and the answer comes out in that unit per pixel. In practice you want millimetres per pixel or centimetres per pixel, so the usual working form keeps height in metres, sensor width in millimetres, focal length in millimetres, and converts at the end:
The same expression gives the footprint — how much ground one frame covers — if you drop the image width from the divisor:
That is the whole model. GSD is also footprint across divided by image width, which is the identity
worth remembering: GSD = footprint ÷ pixel count. If two routes through the arithmetic
disagree, one of your four inputs is wrong.
Take the camera in a DJI Mavic 3 Enterprise: a 4/3 sensor 13.2 mm wide and 8.8 mm high, a 12.29 mm focal length, and 5280 × 3956 pixels. Fly at 120 m above the ground:
footprint across = 120 × 13.2 ÷ 12.29 = 128.9 m,
footprint along = 120 × 8.8 ÷ 12.29 = 85.9 m, so one frame covers 128.9 × 85.9 m,
about 1.11 hectares.
GSD = 128.9 m ÷ 5280 px = 0.0244 m/px, that is 2.44 cm/px. Check it the other way:
120 × 1000 × 13.2 ÷ (12.29 × 5280) = 1 584 000 ÷ 64 891 = 24.41 mm/px. Same
answer, which is the point of the identity.
Now halve the height to 60 m and everything halves: footprint 64.5 × 43.0 m, GSD 1.22 cm/px. GSD is directly proportional to height above ground and to nothing else on that list. Double the height, double the GSD, and you need four times as many photos to cover the same field — which is exactly why asking for a finer GSD costs flight time quadratically, not linearly.
| Height AGL | GSD | Footprint | Ground per 100 px | Typical use |
|---|---|---|---|---|
| 30 m | 0.61 cm/px | 32.2 × 21.5 m | 0.61 m | Inspection, crack and defect mapping |
| 60 m | 1.22 cm/px | 64.4 × 43.0 m | 1.22 m | Detail survey, cadastral boundaries |
| 90 m | 1.83 cm/px | 96.7 × 64.4 m | 1.83 m | Topographic survey at 1:500 |
| 120 m | 2.44 cm/px | 128.9 × 85.9 m | 2.44 m | Standard mapping, 1:1000 detail |
| 180 m | 3.66 cm/px | 193.3 × 128.9 m | 3.66 m | Large-area orthophoto, 1:2000 |
| 300 m | 6.10 cm/px | 322.2 × 214.8 m | 6.10 m | Reconnaissance, vegetation and volume work |
If a specification asks for 1.5 cm/px from the same camera:
(15 mm × 12.29 × 5280) ÷ 13.2 = 973 560 ÷ 13.2 = 73 755 mm, so 73.7 m above
ground. Note the units: the GSD and the focal length and the sensor width were all in millimetres,
and the answer came out in millimetres before the final division by 1000. Mixing millimetres and
metres here is the single most common arithmetic error in the whole calculation.
Spec sheets quote two focal lengths. One is the real optical focal length in millimetres; the other is the 35 mm equivalent, which describes the field of view as if the lens were on a full-frame camera. The formula needs the real focal length, paired with the real sensor width. The Mavic 3 has a real focal length of 12.29 mm and a 35 mm equivalent of 24 mm; using 24 in the formula gives a GSD of 1.25 cm/px instead of 2.44 — a factor of two, in the optimistic direction, which is the worst direction for a survey you have already flown.
A phone camera makes the trap sharper still: the 35 mm equivalent is often quoted while the real focal length and the sensor width are never published. In that case the sensor width in the formula must be the full-frame 36 mm, because the equivalent figure already assumes it. Use the two as a matched pair or not at all.
Every formula on this page takes height above the ground. Drone telemetry usually reports height above the take-off point, and take-off points are frequently on the highest part of a site. Fly a quarry or a hillside from the rim and the drone reads 120 m while the ground beneath it is 150 m below: the real GSD at the bottom of the pit is 1.9 times the planned figure. Terrain following exists precisely to hold the distance constant, and it is worth enabling whenever the site relief is a significant fraction of the flight height.
Vertical datum confusion compounds this. A height reported from a GNSS receiver may be ellipsoidal, while the map you are matching is orthometric. Over most of the world the two differ by tens of metres — see geoid vs ellipsoid — which is far larger than the error you are trying to control. Convert to a common vertical datum before comparing anything.
The model is a pinhole: no distortion, no rolling shutter, no lens profile, perfectly perpendicular camera. Real results differ by percent, not by factor, and that is acceptable — but only if you know which direction. Lens distortion is worst at the frame edges, so the effective GSD at the corners is coarser than the centre value the formula gives. Rolling shutter smears pixels during motion, and at a fast ground speed the along-track sampling gets worse than the arithmetic predicts. Auto-exposure and focus hunting at low light soften the image without changing a single number.
Then there is processing. A photogrammetric pipeline resamples the orthophoto to whatever pixel size you ask for, and asking for less than the capture GSD invents detail that was never captured. Output resolution should never be finer than the flight GSD; equal is the honest setting, and half the value is a common and defensible compromise.
Sanity-check the GSD against something you can measure. Put a levelling staff, a surveyor's tape or a chequerboard target in the overlap area before the flight. Measure it in pixels in the delivered orthophoto and the true GSD falls out directly: staff length divided by its pixel length. One target costs a minute and validates the whole chain — formula, datum, distortion and processing.
GSD tells you the scale; it does not tell you how many photographs. That needs overlap: the forward overlap between successive shots along a line, and the side overlap between parallel lines. At 70% forward overlap the ground advances by 30% of the footprint along-track between shots; at 70% side overlap the lines are 30% of the footprint across apart. Everything else — line count, photos, flight distance, flight time — follows from those two numbers and the site dimensions, and is worked through in how to plan a mapping flight.
A rough budget helps before you commit: area divided by (footprint × 0.3 × 0.3) gives the order of magnitude of the photo count at 70/70 overlap. A 1 km² site at 120 m with the camera above works out at 1 000 000 ÷ (128.9 × 85.9 × 0.09) ≈ 1000 photographs, which is a 25-minute flight and a two-hour processing run on a laptop. Knowing that before you leave the office is the difference between a survey brief and a guess.
CoordCalc publishes free surveying and coordinate calculators. Results are for reference — confirm against the governing standard before construction or procurement. About · Privacy & cookies · Terms of use · Developers