Evaluating the Performance of Ordinary Kriging in Mapping Soil Salinity

Abstract

The performance of ordinary kriging (OK) is impacted by different factors that characterize the data sets being interpolated. The following factors were investigated as part of this study to evaluate the performance of OK in mapping soil salinity: (1) sampling density (field scale, subbasin scale, and subbasin scale merged with field scale); (2) spatial point patterns (random, aggregated, and regular); (3) spatial and no spatial autocorrelations; (4) normal and skewed distributions; and (5) homogeneity and heterogeneity. The objective of this study is to evaluate the performance of the OK model against each of the aforementioned factors. To achieve this objective, 36 different data sets were selected from data collected from 1999 to 2008 in a study area in the Lower Arkansas River Valley in Colorado. These data sets were selected to represent the different factors used to evaluate the performance of OK, in which each factor is represented by three different data sets. Assessments of the OK model residuals and the cross-validation techniques were used to evaluate the performance of the model, both analytically and graphically. The results of this study show that the model performance is accurate when using the field-scale data sets and poor when using the subbasin-scale data sets. When the field-scale data sets are merged with the subbasin-scale data sets, the performance of the model improved significantly over the subbasin-scale data sets. The performance of the model is better when using random or aggregated data sets than when using regular data sets. The existence of spatial autocorrelation significantly improves the performance of the model as expected. When there is no spatial autocorrelation, the performance of the model is severely impacted. The model performs better using normally distributed rather than skewed data sets and also using homogeneous rather than heterogeneous data sets.