Statistical Prediction of the Presence of Salt-Affected Soils by Using Digitalized Hydrogeological Maps

Since spatial databases created from maps in Geographical Information Systems (GIS) do not typically meet the requirements of multivariate tests such as multiple linear regression analyses, we used novel, more robust statistical techniques for predicting the occurrence of certain categories of a soil map. Based on published maps of salt-affected soils (SASs), (hydro) geological conditions (all at the scale of 1:500,000) and regionalization, we showed that the spatial occurrence of SASs in the Great Hungarian Plain can be predicted with an overall precision of 91; 96% for the SASs and 99% for the non-SASs. When we separated the non-SASs into potential SAS and non-SAS categories, the overall precision was 91%, and 87% for the SASs, 92% for the potential SASs and 94% for the non-SASs. The technique of classification trees proved to be better than the classical Multiple Linear Regression, since it does not have limitations regarding the distribution of the data set created from the maps and because it flexibly provided a possibility for incorporating nominal variables through the previous use of homogeneity analysis by means of alternating least squares (HOMALS). Inside the multidimensional space, the first and most important (regarding spatial extent) splits for the separation of non-salt-affected cases from salt-affected ones were realized in the plane of two variables derived with HOMALS from the map of "Textural classes of near-surface geological formations" and the map of "Dominant ion type of groundwater" combined with the map of "Taxonomical division of the regions of Hungary." The plane of these variables was split into five main bands. The clay band has 70% of SASs and the mixed silty layer was divided to a bicarbonatic one with 12% SAS, and a sulfatic one with 65% SAS cover. The remaining sandy band has few SAS, but the dominance of sulfate in the groundwater indicates a higher cover of SASs and these soils constituted the two last bands. Other variables entering the classification were the groundwater level above sea level and the depth to groundwater.

Keywords:

• Classification
• Tree Cross-TABULATION
• Homal
• Hungary
• Saline Regions
• Salinity Risk