Spatial interpolation of stochastic weather model parameters

A methodology for the spatial representation of the parameters of stochastic weather models is presented. This method provides an efficient method for the space-time modelling of weather data as inputs to other agricultural, ecological and hydrological models used in weather impact assessments. The seasonal variation of the parameters of a point rainfall model, the Rectangular Pulses Poisson model, is represented by a quadratic polynomial spline. The coefficients of this representation are then spatially interpolated as functions of position as well as position and elevation. Thin plate smoothing splines are used in the interpolation procedure, and the methodology is tested with data from 102 stations in the Darling Downs region of South-East Queensland, Australia. Further testing with an independent data set demonstrates the adequacy of the technique in preserving historical statistics from the coefficients of the interpolated surfaces. A useful extension of this methodology to provide stochastic weather data for future climate scenarios is also discussed.