Multi-time-scale input approaches for hourly-scale rainfall–runoff modeling based on recurrent neural networks

This study proposes two straightforward yet effective approaches to reduce the required computational time of training process for time-series modeling through a recurrent neural network (RNN) using multi-time-scale data as input. One approach provides coarse and fine temporal resolutions input RNN in parallel. The other concatenates over before considering them RNN. In both approaches, first, finer resolution are utilized learn scale behavior target data. Next, coarser expected capture long-duration dependencies between variables. proposed were implemented hourly rainfall-runoff at snow-dominated watershed by employing long short-term memory (LSTM) network, which is newer type Subsequently, daily meteorological input, flow discharge was considered results confirm that can significantly (up 32.4 times). Furthermore, one improves estimation accuracy.