Spatial Prediction of Rainfall Using Skew-Normal Processes
Modeling spatial data with Gaussian processes is the common thread of all geostatistical analyses due to their close relationship with kriging (Cressie, 1993). In some instances, the assumption that the distribution is Gaussian is clearly violated. One basic objective of this chapter is to develop spatial models to handle data showing non-Gaussian characteristics, such as skewness. For example, modeling rainfall data from a statistical perspective presents the challenge of dealing with a skewed distribution. Several sophisticated models have been developed to capture the underlying physical dynamics that govern rainfall. Smith and Robinson (1997) presented a likelihood-based approach and also used a Bayesian method to analyze it. A diﬀerent approach is to use a truncated normal distribution (Stidd, 1973; Sanso´ and Guenni, 2000) to model the rainfall distribution. De Oliveira, Kedem, and Short (1997) used a Bayesian transformed Gaussian (BTG) model to analyze a rainfall data set.