Flood Detection Using Backpropagation Neural Network Method

Abstract

Lack of river and watershed management will cause problems and disasters. One of it is the flood that can cause physical, social and economic loss. So countermeasures or flood anticipation are needed by using the Early Warning System (EWS) to provide early information if a flood is going to occur. This study uses five input indicators: temperature, humidity, water discharge, water surface altitude and rainfall data that will produce output in the form of notifications and alarms for the Early Warning System (EWS). Then the input and output data configuration will be processed using a Backpropagation Neural Network. Data used is data recorded in real-time on the research object for two weeks with the composition of training and testing data with a percentage of 80% and 20%. The best backpropagation neural network model used has the input of 5 neurons layer architecture, 15 neurons as the hidden layer and three neurons as the output layer. The flood prediction result uses the Backpropagation Neural Network method, has an RMSE score performance of 2.16e-21 and a percentage success testing system of 91.33%. It shows that the model has an excellent accuracy level.