Flood Hazard Mapping by Using Spatio-temporal Prediction Model

The flood hazard mapping has been applied in flood early warning systems. The flood hazard mapping requires the water discharge data and the spatial data of the flood risk area. The water discharge data of the river can be collected from the telemetry stations. However, the water discharge data is scarce due to the few number of the telemetry stations in the remote areas. This paper proposed to apply the spatio-temporal prediction model to improve the flood hazard mapping process. The proposed spatio-temporal prediction model is composed of two modules which are 1) the temporal prediction module applied to find the predicted water discharge by using the Long Short-Term Memory (LSTM) technique and 2) the spatial prediction module applied to interpolate the discharge value around the watershed area using the Inverse Distance Weighted interpolation (IDW) technique. In this research, the water discharge data is collected from two telemetry stations in Phayao Province, Thailand. The hazard map covers 270 square kilometers. The evaluation results show that the proposed model can predict the water discharge value with a mean of MAPE 5.449%. Then, the flood hazard map is generated by the spatial data and water discharge data that has an accuracy of 96.7% prediction compared to the actual flood situation