Geostatistical modelling of rainfall in Fars Province of Iran using non-Gaussian spatial process

Prediction of response values is a primary goal in many applications. The standard approach to this problem is kriging which is essentially a linear prediction using optimal least squares interpolation of the random field. However, the optimal predictor is not necessarily a linear one unless the geostatistical data support the Gaussian model. As data often exhibit non-normality, some of the most effective spatial processes are reviewed in the current study. The usefulness of the presented models is demonstrated based on the prediction of rainfall levels in Fars Province, Iran. The measurements were taken from 100 stations. To assess the predictive performance of the evaluated models, 15 stations were randomly withheld. Subsequently, the predicted values at these locations were evaluated against the measured ones. The results of the study indicated that, comparing to some well-known models, the skew Gaussian model introduced in this article demonstrated a better performance in the prediction.

Automated summary

Prediction of response values is a primary goal in many applications. The standard approach to this problem is kriging which is essentially a linear prediction using optimal least squares interpolation of the random field. However, the optimal predictor is not necessarily a linear one unless the geostatistical data support the Gaussian model. As data often exhibit non-normality, some of the most effective spatial processes are reviewed in the current study. The usefulness of the presented models is demonstrated based on the prediction of rainfall levels in Fars Province, Iran. The measurements were taken from 100 stations. To assess the predictive performance of the evaluated models, 15 stations were randomly withheld. Subsequently, the predicted values at these locations were evaluated against the measured ones. The results of the study indicated that, comparing to some well-known models, the skew Gaussian model introduced in this article demonstrated a better performance in the prediction..