We need to develop a procedure to sample representative stations to use in fitting and validation. In many areas (including Oregon), the station locations are highly spatially clustered and so selecting random subsets for model fitting is likely to lead to spatially (and therefore environmentally) biased data. We need to develop a procedure to use existing environmental data (climate and elevation) to select a stratified sample for model fitting. The environmental data used here will not be used in the interpolation procedure, only to ensure that we are getting a sample that is as representative as possible.