Monthly Runoff Prediction by Hybrid CNN-LSTM Model: A Case Study

Prediction of runoff plays a vital part in planning and managing water resources. To model hydrological processes, advances in artificial intelligence methods can act as powerful tools. A novel hybrid model named CNN-LSTM is employed by incorporating a long short-term memory and convolutional neural network to improve prediction accuracy. Application of applied hybrid model is exemplified utilising hydrologic data from Barak River Basin in Cachar district of Assam, India, to test model accuracy. The best performing model was chosen on basis of different performance assessment standards, i.e., R-2, IA and RMSE. The outcomes revealed that CNN-LSTM model performs superiorly with R-2 = 0.9875, RMSE = 1.364 and IA = 0.9897 during training phase compared to standalone CNN model. The present study's result suggests that proposed CNNLSTM network is a vital tool for runoff prediction in catchments and assists in providing viable measures for a catchment.