Applying distance/area dependent procedures in geographic coordinates over large areas?

Question

I'm currently doing some niche modelling with data in geographic coordinates spanning all of Latin America. I am using R for my analysis. Some of the tasks I have to perform are area/distance dependent and yield incorrect results when applied in geographic coordinates. For example, spatial sampling (sp::spsample) leads to a bias towards more points in southern South America and buffering points and polygons (rgeos::gBuffer) leads to stretching/skewing of the buffer away from the equator.

What's the best way to address this issue? Should I project my data, apply the necessary procedure, then project back? If so, what projection would work for such a large area? Ideally, I'm looking for a general approach that could be applied to any region, but I fear this doesn't exists.

Answer 1

The only general and accurate approach, for large regions, would be to keep all spatial calculations in geodetic space, i.e., on the ellipsoid. However, you probably don't have this option.

The "projection" approach can work, but it depends entirely on what spatial calculations you wish to use. A different projection might be required for each case: A conformal projection will be good for directions (azimuths) but will require extensive use of scale factors to correct distances. And as whuber already said, an equal-area projection will work for areal calculations (such as sampling).