Rainfall forecasting using an artificial neural network model to prevent flash floods

Flash floods are a dangerous natural disaster as they have killed more people than any other natural disaster and caused millions of ringgit in property damage. This paper presents a new approach for modeling rainfall forecasting using the artificial neural network technique (ANN). Daily actual data from the years 2007 to 2010, collected from 3 main stations in Selangor, were used to develop the ANN model. Intended to provide near real time forecasting, different network types with different hidden neurons were tested. Preliminary tests show that a Feed-Forward Back propagation ANN model using a hyperbolic tangent sigmoid transfer function achieved the most accurate rainfall forecasting. Sets of data consisting of meteorological parameters (wet bulb temperature, relative humidity, wind speed, cloudiness and air pressure) were used as inputs and rainfall collected by rain gauges at selected stations were used as the target. The results for rainfall forecast were fed into a SCADA-based system as a solution for preventing flash floods.