A CASE-BASED REASONING APPROACH FOR RAINFALL-RUNOFF MODELING: USING EXPERIENCE TO PREDICT THE FUTURE

Case-based reasoning (CBR) has been widely used in many real-world applications, such as design, planning, and knowledgebase Internet search engines. In general, CBR systems predict behavior by comparing some given, unknown case to a large library of past cases with the best matching retrieved case(s) serving as an approximate solution to the given problem. Our application here is concerned with predicting the hydrologic response of a stream to precipitation. The cases are developed from historical data on river discharge and meteorological data compiled by USGS and NOAA, respectively. Our demonstration involves a rainfall-runoff analysis for Mill Creek watershed in Ohio. This watershed is located about 10 miles northwest of Columbus and has an area of approximately 180 square miles. In the watershed, glacial drift overlies fractured limestone bedrock. The land use is predominantly agricultural with corn and soybeans as the main crops. We developed a database that contains information on past floods. In the database, these previous floods are characterized using a set of attributes including base flow, time between storms, temperature, and precipitation amount. CBR predicts the peak discharge for a new storm case by comparing these attributes with those of previous cases and adapting the peak discharge for similar case(s). Preliminary results show that this technique can be used to predict peak discharge values for different types of storms, including those following a drought. CBR has promise as a quick alternative to more involved rainfall-runoff modeling.