mattyj wrote:As someone who's worked with the National Elevation Dataset recently, I want to explain this a bit more. For most parts of the country, the NED comes in 1/3 arc-second resolution, approx. 30 ft.
Thanks for that explanation, I didn't know that the US NED data were
that accurate.
I've spent a lot of time using Google Earth and Google maps terrain view, both for European and South American mountain ranges. One of my reasons would be to find the location and elevation of a particular summit.
In the beginning, I took the data provided at face value, but after a while, I started noticing discrepancies, both with printed maps and with what I saw with my own eyes, out in the field. I found mountains that showed only one summit, while I knew there were several. And I saw elevations in Google Earth or Terrain view that were much lower than the maps or guidebooks said they were, especially for sharp peaks.
When I started to look for background information, I read about the inaccuracies resulting from sampling grid problems, but until recently, wasn´t quite sure what the grid size was. A friend helped me along the way, and I learned that the elevation data in South America is based on SRTM data, which, according to the SRTM site itself, has a sampling of 1 arc seconds, but outstide the US and Australia, the officially released data is only 3 arc seconds. I´m going out on a limb here and guess that the reason is something military. Anyway, 3 arc seconds implies a lattice of 90 m (270 ft) - at the equator that is, I won´t go into different latitudes. That's roughly one point per football field. Many summits are smaller than that, so it's very likely that the nearest SRTM sampling point is not quite on the summit itself. And the further away, the lower it probably is.
Having sampling points, the next step is to calculate elevations and contour lines. Enter interpolation algorithms. Although I don´t know the algorithm used by Google, looking at Google Earth, it seems to me that the elevation of any one particular point is based on more than only the four closest points.
However, no matter how good the algorithm, I can't see how anything but smoothly curved mountains with big summits could ever be interpolated even reasonably well. For jagged mountains and sharp spikes, there simply isn´t enough data. Only if the raw SRTM data is corrected before Google uses it, better results can be expected. I do not know to what extent such corrections are applied, but from what I saw, especially in South America, there is room for improvement, to put it mildly. A room the size of a football field.
Knowing all this, what does that mean for using Google Earth and Maps? If I don't have a good map, I still use it for summit elevations, but tend to take the value given by Google as a good starting point, a lower bound for the actual value. If I know that the summit is big, I tend to trust the value more. As for the location, if I have photos, I use them to help me pinpoint it. Sometimes, especially for a rocky summit on a glaciated peak, the satellite image is helpful too.
Links:
Shuttle Radar Topography MissionBackground on Google Earth (and SRTM)