DATA MINING APPLIED FOR DETERMINING STREAM FLOW PERMANENCE
Keywords: Flow Permanence, Mini-basins, Data Mining, Decision Tree, Morphometric Attributes
Abstract. Streamflow permanence is an aspect of great legal importance in Brazil, because streams, depending on their flow regime, are protected or not by law. Various methods, from field methods to computational methods, are used to determine the flow regime of streams, however some are too time spending and computational methods usually need gaging information as input. Furthermore, computational methods used to extract drainage networks do not identify the flow regimes of streams, and modelled drainage networks always need to be refined manually, as some authors indicate that up to 55% of modelled drainage length is ephemeral in some cases. This work proposes a semi-automatic computational method to determine the flow regime of first order streams, which uses 11 morphometric attributes of the mini drainage basins of these streams to develop a classification model using decision tree algorithms. WEKA package was used to perform the data mining process, which resulted on the development of a compact 8 node decision tree. Ten-fold stratified cross-validation was used to validate the model, which obtained an accuracy rate of 70%. The drainage network of the study area extracted with the classical approach was refined after the result of the classification was obtained. Quantitative analysis of channel length by Strahler order shows an overall reduction of 25% in channel length after refinement was undertaken, and for 1st order streams, as much as 31% were classified as ephemeral. Modelling the drainage areas of headwater streams represents a new approach to determining stream flow permanence, and inclusion of new attributes in the model may yield better results in future research.