Issues in choosing a method/software

• Is smoothing parameter estimated automatically
• Can other covariates be included in addition to spatial coordinates
  • $Y\sim N(X\beta+g(s_{1},s_{2}),\sigma^{2})$ or $Y\sim N(f_{1}(x_{1})+f_{2}(x_{2})+g(s_{1},s_{2}),\sigma^{2})$
• Can non-normality be handled, e.g., $Y\sim\mbox{Bern}\left(\mbox{logit}^{-1}(X\beta+g(s_{1},s_{2})\right)$)

Figure: 0-1 data

• Is the method designed for anisotropic (directional) data?

Figure: Anisotropy

• Is the method designed for nonstationary (structure varies with location) data?

Figure: Nonstationarity