Categorical fuzziness: a comparison between crisp and fuzzy class boundary modelling for mapping salt-affected soils using Landsat TM data and a classification based on anion ratios

This paper investigates whether the use of fuzzy, instead of crisp class salinity boundaries, improves the accuracy on the detection of salt types from remote sensing data. To this end, a classification of salt types based on anion ratios, and a supervised maximum likelihood classification technique, where the membership grades of the saline fuzzy classes are incorporated as prior probabilities to classify a Landsat TM image acquired over a salt-affected area of Bolivia are applied. assification system based on anion types has been developed by Russian soil scientists (Plyusnin, 1964). In this approach, salt-affected soils are classified on the basis of salt types, in terms of chloride, sulphate and carbonate anion ratios present in the soil saturation extract. It is of interest to test this approach because not all salts are equally harmful, and do require different reclamation and management measures. Consequently, it is valuable to know the spatial distribution of salt-affected soils and their composition. modelling of the information categories and the incorporation of certainty factors during the classification procedure allowed overcoming low accuracy results. Identification accuracies improved by as much as 44% for chloride-sulphate and sulphate-chloride soils with similar proportions of both anions. Higher accuracies were achieved for soda-sulphate soils, as compared to the sulphate-chloride types. This is attributed to the fact that carbonates and sulphates used in the ratio for deriving soda-sulphate soils, have absorption features in the infrared and thermal ranges of the spectrum.