ABSTRACT
This chapter provides further reproducible examples on how to select optimal spatial predictive methods, including optimizing relevant parameters, selecting predictive variables for each method based on predictive accuracy and stabilizing the accuracy using the parameters optimized and predictive variables selected; and also systematically compares the performance of these spatial predictive methods to further demonstrate the procedures, rules and pitfalls in spatial predictive modelling. The following three response variables are used: 1) zinc (i.e., continuous data) in the meuse data set in the sp package; 2) species.richness (i.e., count data ) in the sponge2 data set; and 3) gravel (i.e., percentage data) in the petrel data set. The R code for parameter optimization, variable selection and assessment of predictive accuracy are provided.
