Terrain follow on uneven sites: what the elevation data does and doesn't know

A waypoint mission normally flies at a fixed altitude relative to your takeoff point. The aircraft climbs to 80 m and stays there, regardless of what the ground underneath is doing.

On flat ground that's fine. On a site with real relief it quietly breaks two things at once.

What goes wrong at fixed altitude

Take a site with 40 m of elevation change, flown at a fixed 80 m above launch. Over the low ground the aircraft is 80 m up. Over the high ground it's 40 m up.

• Resolution changes across the dataset. GSD scales with height above ground, so the high ground is captured at twice the resolution of the low ground. One survey, two standards.
• Overlap changes too. Closer ground means a smaller image footprint, so the fixed spacing between flight lines covers proportionally less of it. Plan 70% side overlap for the valley and you may be getting 40% on the ridge — sometimes below what the software needs to reconstruct at all.

The failure is uneven rather than total, which makes it easy to miss. The dataset looks fine in the areas you happen to check.

And on genuinely steep terrain, fixed altitude is a safety problem before it's a data problem: ground that rises 60 m into an 80 m flight leaves 20 m of clearance, before you count the trees standing on it.

What terrain follow does

Terrain follow adjusts each waypoint's altitude using an elevation model of the site, so that the aircraft holds a roughly constant height above ground instead of a constant height above launch. The flight path becomes a rough copy of the terrain profile, offset upward.

GSD stays consistent. Overlap stays consistent. Clearance stays consistent.

Where the elevation model is wrong

That distinction is the whole reason for caution. Terrain follow is only as good as the elevation data behind it, and that data has known limits.

You may not know which surface it describes. This is the trap. A digital terrain model describes bare earth; a digital surface model describes the top of whatever is standing on it. Global datasets like SRTM, ASTER GDEM, and ALOS AW3D30 are radar- or optically-derived surface models — they partially include canopy and rooftops, inconsistently, depending on how the sensor saw the ground.

Either case is dangerous, in opposite directions. Against bare earth, your clearance vanishes into whatever is growing there: fly 30 m above a terrain model over 25 m trees and you have 5 m. Against a surface model, the aircraft climbs for obstacles that may no longer exist — a felled woodland or a demolished building leaves the profile flying high, which is safe but ruins the consistent height above ground the whole exercise was for.

The resolution is coarse. Those global datasets are posted at roughly 30 m, with each value representing that whole cell. A narrow gully, a levee, a quarry face, or a steep cutting can be smoothed away entirely — the model shows a gentle slope where the ground has a step in it.

It can be out of date. Elevation data has a survey date. Active quarries, construction sites, landfill, and earthworks change faster than the datasets describing them. A site mid-excavation may be tens of metres from its recorded shape.

Vertical datums differ. Elevation data and GNSS altitude are not always referenced to the same surface. SRTM heights, for instance, are orthometric — referenced to the EGM96 geoid — while raw GNSS produces ellipsoidal heights, and the separation between the two varies by location. Mixing them introduces an offset that is usually consistent across a site rather than varying, which makes it easy to overlook and no less dangerous.

Using it safely

Set the takeoff reference point accurately. The whole profile is anchored to the elevation at your launch position. Get that wrong and every waypoint inherits the same error — an offset that persists across the entire mission.

Add margin for what the model can't see. Pick a height above ground that clears the tallest thing standing on the terrain, not the terrain itself. Over woodland that means canopy height plus a real safety buffer.

Read the altitude profile before you fly. Look for steps, spikes, or flat sections where you know the ground isn't flat — those are usually artefacts of coarse sampling rather than real terrain, and they mark where the plan disagrees with reality.

Check the climb rates you've asked for. A profile that tracks steep terrain closely can demand vertical speeds the aircraft can't sustain while holding ground speed. If it can't climb fast enough, it arrives at the next waypoint lower than planned — exactly where the ground is highest.

Verify on site. Confirm visually that clearance matches the plan before committing to an automated run, particularly the first time you fly a given area. On unfamiliar or complex terrain, a conservative test pass at greater height costs one battery and tells you whether the model matches the ground.

DronePoint computes terrain-follow altitudes from elevation data at planning time and shows the resulting profile, so you can inspect the flight path against the terrain before exporting the mission. What the elevation model doesn't contain still needs your eyes on the site.

Follow DronePoint

See more of these notes in your Google Search results.

Add DronePoint as a preferred source