Testing location data with edge cases, not just happy paths
An address in a well mapped city center tells you almost nothing about how your system handles a rural route, a disputed border, or a query near the poles. Test the hard cases deliberately.
How elevation accuracy changes with terrain type
Two elevation readings with the identical stated resolution can carry very different real-world accuracy, depending entirely on the terrain underneath them. This is one of the more counterintuitive facts about elevation data, and it matters whenever you are relying on a number to be close to the truth rather than just plausible.
On flat, open terrain, such as agricultural land, deserts, or a calm coastline, elevation is straightforward to capture accurately. There is little variation within any reasonably sized grid cell, so the sampled value tends to sit close to the true height across the whole cell, and the stated resolution comes close to describing the actual achievable accuracy.
Steep terrain behaves very differently. On a mountainside or in a canyon, elevation can change by tens of meters within the footprint of a single grid cell, which means the sampled value is necessarily an approximation of a range rather than a precise reading of one spot. The same resolution figure that was nearly exact on flat ground becomes a much rougher estimate here, simply because there is more real variation for the sampling process to average or miss.
Vegetation and built structures add a separate complication. Depending on how the underlying elevation data was captured, dense forest canopy or tall buildings can raise the recorded surface height above the actual ground level beneath them. This is not an error in measurement so much as a question of what surface was actually measured, the top of the canopy or roofline versus bare earth. It is most noticeable in dense urban centers and heavily forested regions, and far less relevant in open terrain where there is nothing tall enough to matter.
The practical implication is to treat elevation accuracy as terrain-dependent rather than as a single fixed number that applies everywhere. If your use case is sensitive to this, such as engineering, agriculture, or hazard modeling, sample multiple nearby points and look at consistency between them as a sanity check, since a smooth run of nearby values in flat terrain is a good sign, while sudden jumps in steep or heavily built terrain may reflect genuine variation, not necessarily bad data.
For most applications, such as displaying rough elevation for a location or flagging general terrain characteristics, this level of care is unnecessary and the stated resolution is a fine guide on its own. Our elevation lookup accepts a list of points in one request, which makes it easy to sample a small cluster around a location of interest and check for this kind of consistency when the stakes are high enough to warrant it.