Quantification of the Salt Content of Soils Under Different Climatic Conditions on a National Scale in South Africa

Abstract

The effect of rainfall, evaporation, and aridity on salt accumulation in the soil, on a national scale, is not straightforward, and other factors such as geology, position in the landscape, and previous climatic conditions should be considered. Good decisions for the management of salt-affected soils on a national scale require good information, derived from raw data. Such data must be generated with specific goals in mind, and it must be stored properly in a format that is easy to access and process. Like any basic resource, the data environment must be managed meticulously. In this study although data verification was previously done on most samples, much effort was devoted to data cleaning. Of the more than 40,000 original data points, only 22,404 data points were used due to the stringent cleaning protocol. A forward selection stepwise regression was used to simplify the various models. In a stepwise regression, variables are added or removed from a regression model one at a time, with the goal of obtaining a model that contains only significant predictors, but does not exclude any useful variables. The accuracy with which EC, ESP, and pH_water were predicted with stepwise multiple linear regression relationships on a national scale is surprising, considering that the various models included all “outlier” values. The R ² statistic indicated that the models as fitted explained the variability in EC and ESP much better for the low rainfall class (<550mm annual rainfall), than for the high rainfall class (>550mm annual rainfall). For EC, <550mm annual rainfall class in the model explains 58% of the variability and for >550mm annual rainfall class 39% of the variability. Values for ESP are 85% for <550mm annual rainfall class and 52% for >550mm annual rainfall class.