First of all thanks for your time and input!
Yes. Thanks for pointing that out. I do have Lat Long coordinates for my problem. Then Ill just need to manipulate the ExPQuad kernel method to also do some trigonometry. Either way, my conceptual problem right now is how to include the linear sub-model inside of the GP. Would this be considered a mean function m(x',x') ? Somthing like GP(0,k(x,x) ) + X' \beta where \beta are parameters to be learned from the data and X' is the design matrix. Williams and Rassmussen 2006 on p.46 equation 240 talk of adding a linear model as a mean function, however, they do so in the context of modeling the residuals. Do you think there are significant time improvements by reparameterizing inside the SparseApprox method?