Smoothened CNN-LSTM Hybrid Model with Spatio Temporal Attention Mechanism for River Water Level Prediction

Flooding in river basins has become increasingly frequent in the state of Kerala, particularly during the monsoon period. Several factors like prolonged rainfall, meteorological factors, topography of the surrounding region, stream flow rate etc. influence the river water levels. Predictive deep learning models can be leveraged to develop accurate and efficient flood early warning systems. Timeseries forecasting technique using deep learning models are nowadays widely used in flood forecasting. This work is an attempt to forecast the water level of Manimala river using a multivariate time series prediction approach. A Smoothened CNN-LSTM hybrid model with spatio temporal attention mechanism is developed in this work. Performance of the model is evaluated using the metrics-Root Mean Square Error(RMSE), Mean Absolute Error(MAE), R2 error, total training time and memory used for training