# How to predict the Geo-data such as elevation, given a set of data for training, what would be the best approach?

Lets say that I would like to predict or forecast the elevation of a certain coordinate on a map. What I have is the data of random points in, lets say, 50km radius from the point I would like to predict (or forecast) the elevation for. Lets say it is all located in a CSV file, and headers for variables are:

Longitude(decimal), Latitude(decimal), Elevation(meters)

Which algorithm or a chain of algorithms would give the best results finding the unknown elevation when Longitude and a Latitude are known?

EDIT: The one I am aware of is http://cs.stanford.edu/people/karpathy/convnetjs/ I believe there could be a better approach.

## 1 Answer

If the only information you have is the coordinates and elevation, I think you're better off just interpolating the elevation of the unknown point from the nearby known points. What kind of spatial interpolation would be best, is probably a better question for https://gis.stackexchange.com/.

Another thing you could try is calculating some other features based on the coordinates. Assuming you can get some additional (spatial) data you could calculate things like distance to the nearest coastline, distance to nearest fault-line and some dummies (Country/Continent/Hemisphere). These features will probably have some predictive value for elevation. If you base the prediction on these features and not the coordinates themselves, normal predictive methods should work unless there are some spatial autocorrelation issues I'm not thinking off.

• Thanks, that is exactly what I wanted to do... to add as much data as I can, but the basic idea is to have an algorithm that would work the best using the longitude and latitude only, and then build on top of that. So, linear interpolation would do it then. That would make my app much quicker, thanks! – Damir Olejar Sep 6 '16 at 16:19