Risk modelling: land vulnerability expert system

Abstract

The land vulnerability models were constructed partly in accordance with the criteria of the FAO-framework for Land Evaluation. They were developed using the ALES shell (Rossiter, 1990), which is a computer program that allows land evaluators to build expert systems. The expert systems created are based on decision trees, which are hierarchical multiway keys in which the leaves are results such as land quality ratings, and the interior nodes of the tree are decision criteria such as land characteristic values. Decision trees appear very valuable as tools to predict the impact of agricultural activities and global changes on land suitability, vulnerability and sustainability. Further, land evaluation, according to the criteria of the FAO framework, is more accessible and transparent to the user, when decision trees are used rather than the usual matching-tables. In addition to this, some land qualities cannot necessarily be characterised by simulation modelling (Bouma et al., 1993a, b). The decision trees created by this project are formed by using expert knowledge (experience), specialists, land users, and collected literature. It seemed a reasonably accurate way of classifying soil vulnerability (erosion and soil/groundwater contamination) of a particular land unit.