Flood forecasting technology with radar-derived rainfall data using Genetic Programming

Abstract

Implementation of flood forecasting system is crucial for reducing flood disasters urgently and effectively. The authors propose a new method of flood forecasting using genetic programming (GP) and GMDH. Traditional method based on physical model takes time to analyze the hydrologic and hydraulic characteristics of a river, but the new method has potential to make a water level forecasting model from ground-based or radar-derived rainfall automatically by learning the past data of river water level or dam inflow and rainfall, which will be useful in particular for medium-to-small scale rivers. Case studies were conducted for the water-level prediction at the Saba and the Onga Rivers in Japan. The results from both the case studies were encouraging to promote the new method, because the water-level predictions with 6-hour lead time were relatively good. Furthermore, comparative analysis about the incorporation of spatial distribution of rainfall in the upstream brought out the necessity of the combined incorporation of both direct and averaging area for better accuracy.