Analysis of environmental thresholds for runoff and erosion through qualitative reasoning methods

Abstract

Qualitative reasoning methods have proven useful to extract qualitative information from quantitative data in a wide variety of applications. In this work, we analysed a 30 months record of runoff and sediment yield from an experimental station in Benidorm, SE Spain, comprising a total of 104 rainfall events. We adopted a decision-tree approach to identlfy the environmental variables, basically rainfall characteristics, whlch were the most informative to predict the production of runoff or sediment and, for these variables, to determine thresholds for chfferent levels of production. A set of 11 ramfall parameters, most of them highly correlated, were simultaneously presented to the C4.5 software tool. Using the variable AI30, which combines the information of height and intensity, 98% of the cases were correctly classified as productive or non-productive for runoff, with an estimate of 5% for the expected error. For the production of sediments, the kinetic energy of the rainfall and the storm duration were incorporated into the prechction vector, giving a 1% of rnisclassifications and an expected error of 6%. For a subset of cases, the length of the dry period and the antecedent soil moisture were also considered, but no improvement was detected. We also considered the level of production, which was automatically classified as low, medium or hgh, and obtained decision trees of lugher order with higher error rates. The usefulness and drawbacks of the methods for the problems in erosion research considered in hs work are discussed.