Assessing temporal and spatial changes of salinity using fuzzy logic, remote sensing and GIS. Foundations of an expert system

Abstract

This paper presents the foundations of an expert system to map landscape features related to salinity, based in a South American case study. Salinity distribution is mapped using an approach that integrates multi-temporal classification of remotely sensed data, physical and chemical soil properties and landform attributes. Change detection maps are derived in a post-classification change detection. These maps represent the changes in the distribution of salinity type and its severity, and are used as inputs to model the nature, magnitude and reliability of salinity-related changes that occurred in the area. Three rule-based expert systems using fuzzy sets and fuzzy linguistic rules to formalise the expert knowledge about the actual possibility of changes to occur are designed and implemented within a geographical information system (GIS). The systems use a fuzzy semantic import approach that enables the integration of multi-disciplinary knowledge in basic sets of fuzzy rules. The outputs of the fuzzy knowledge-based system are three maps representing ‘likelihood of changes’, ‘nature of changes’ and ‘magnitude of changes’. These maps are then combined with landscape information and analysis of the spatial association among these variables, represented in different GIS layers, is undertaken to derive an exploratory hazard prediction model. The coefficient of areal correspondence is used to assess the degree of association among the variables and landscape positions. The sum of the coefficients of areal association between the variables considered represents evidence of salinity hazard within a specific landscape unit. Higher coefficient values indicate higher hazard to salinity processes. Because the classification model differentiates among saline and alkaline areas, it is possible to evaluate the nature of the salinity changes, i.e. whether an area may become more saline, alkaline, or saline–alkaline. This information is important for decision-makers and land planners, because different reclamation measures can be adopted according to the salinity type. The approach provides a fast way of assessing the likely extent of salinity at regional level, enabling the integration of a variety of data sources and knowledge. Moreover, this monitoring model can help to evaluate the effectiveness of salinity control and management action plans.