View article

Artificial Neural Network (ANN) is a type of empirical data-driven model that can be developed and run for flood forecasting purposes very quickly compared to physically based or conceptual models. Since the Northeast region of Bangladesh is vulnerable to flash floods in the pre-monsoon season, this study has investigated the potential of using ANN to forecast the river stages at different lead times of Jadukata River at Laurergarh Saktiarkhola gauging station of Bangladesh Water Development Board (BWDB) in Sunamganj. The ANN model was calibrated and validated with 3-hourly accumulated rainfall data from the Tropical Rainfall Measuring Mission (TRMM) and 3-hourly observed river stage data from BWDB. The ANN model where both the variables of past rainfall and river stage are included as input data can forecast the future river stages better than ANN models where only one variable, either the past rainfall or river stage, is used instead. Also, the performance of an ANN model for forecasting future water levels does not deteriorate significantly if the lead time is increased from 6 hours (R2= 0.94) to 48 hours (R2= 0.89). Overall, this study demonstrates that ANNs hold promising capability in real-time flash flood forecasting.