Flood water level prediction modeling using NNARX structure for Sg Pahang basin

There were a total of 58 events of natural disaster in Malaysia for the period between years 1980 to 2010 that claiming a total of 1,239 lives of the 640,000 people affected. These data were based on statistics provided by United Nation Officer for Disaster Risk Reduction (UNISDR). From all different categories of natural disasters considered, flood accounted for over half the registered events. Floods contribute to 8 out of 10 disaster events with the highest human exposure and affect over 85 % of all the disaster-stricken people. Floods are thus the primary hazard which affecting Malaysia, in particular the west coast of Peninsular. Therefore, an accurate and reliable flood prediction model is very much needed to provide early warning for residents nearby flood locations for evacuation purposes. However, current trends in flood prediction only involve flood modeling because no prediction time was mentioned and discussed. Furthermore, in Malaysia there is none of flood model or flood prediction model developed yet. An existing system in the Department of Irrigation and Drainage Malaysia is only the alarming system which alarms the users only when the water level exceeds the danger limit. Based on these scenarios, the research objective is to obtain a flood water level prediction model for Pahang flood prone area using Neural Network Autoregressive Model with Exogenous Input (NNARX) structure. The samples used for model training, model validation and model testing were carefully selected. In order to obtain good flood water level prediction model, all samples must be the data when flood events happened. All samples were real-time data that were obtained from the Department of Irrigation and Drainage Malaysia upon special request. From carefully selected samples, several optimal flood prediction times were suggested for flood location in Pahang. Model validation and model testing were conducted to observe the prediction performances. The optimal prediction time was selected based on the results of prediction performances. Results show NNARX model successfully predicted flood water level ahead of time.