Computer model emulation in agriculture
Computer models are seeing expanded use in agricultural applications from crop yield to soil loss to nutrient runoff. Statistical emulators of these computer models can provide practitioners with enhanced capabilities to solve their calibration, optimization, and design needs. We will discuss some of our recent work in developing Gaussian process (GPs) emulators for agricultural applications including dealing with functional inputs and large number of computer model runs. For functional inputs, we introduce the asymmetric Laplace functional weight for a parsimonious distance function between two functional inputs and show that it is comparable to automatic relevant determination despite using a small number of parameters. When many computer model runs are available, we introduce a one-at-a-time approach to knot selection in a knot-based approximation to a full GP and show it is comparable to the full GP at a fraction of the computational expense.