From the FAQ:
I found a cache with a faulty elevation
Sadly there are a lot of geocaches with these issues. The elevation data does not come from Groundspeak since they don't have that data. We have therefore build our own SRTM-service for the purpose, it serves both SRTM1 and SRTM3. As a fallback for the areas these doesn't cover, we fallback to several over services, for example: Geonames, Google and MapQuest.
If you find that a value in Project-GC isn't correct. Please read below how our elevation data works, and also check other services like for example Geonames SRTM3. If you find out that other services has about the same fault tolerance, then there is nothing to do.
We do not do manual changes in the data. But if you find an entry that seems to be off by a lot, feel free to report it so that we can investigate.
How is the elevation data calculated?
This is actually quite advanced, but we will try to simplify it a bit.
First off, we use different data sets for different areas on the earth. For most of the planet we are using data from the SRTM1 and SRTM3 databases created by Nasa. Nasa has created this data by measuring the height using satellites. SRTM1 has a higher resolution, but that data only exists around USA. SRTM3 exists for other parts of the world, except closer to the poles.
The SRTM1 data has ONE measure point per 30x30 meters, and the SRTM3 data has ONE measure point per 90x90 meters. What this means, is that there is no measurement for every coordinate, and therefore not for every geocache location. So what we do is that we interpolate between the 4 closests values to get a weighted average for the geocache location. In an area which is very hilly, like mountains, this will give a quite big fail factor and almost always a too low value. There is however no better way to do this, since the measurements just doesn't exist, well, except manually adding data for all the geocaches, which we do not do.
Other services like for example Geonames might have slightly different approaches to how they calculate the data, therefore we will have smaller differences. But they do have similar solutions.
Before using SRTM as our primary source we used AsterGDEM which is created by radar measurements. This data has measure points for every 30x30 meters in a larger area of the earth, which in one way makes it better. But, AsterGDEM is also known for being affected by radar shadows, which makes the data quite useless for some areas. We have found out that switching to SRTM has giving us more precise data.
For those areas where there is no SRTM1 or SRTM3 data we are falling back to other services which relies on other data.