Research on Runoff Simulations Using Deep-Learning Methods

Runoff simulations are of great significance to the planning management water resources. Here, we discussed influence model component, parameters and input on runoff modeling, taking Hanjiang River Basin as research area. Convolution kernel attention mechanism were introduced into an LSTM network, a new data-driven Conv-TALSTM was developed. The analyzed based Conv-TALSTM, results suggested that optimal greatly affected by correlation between data output data. We compared performance variant models (TALSTM, Conv-LSTM, LSTM), found can reproduce high flow more accurately. Moreover, comparable when trained with meteorological or hydrological variables, whereas peak values closer observations. When two datasets combined, better. Additionally, also ANN (artificial neural network) Wetspa (a distributed for Water Energy Transfer Soil, Plants Atmosphere), which verified advantages in simulations. This study provides direction improving accuracy, simplifying structure shortening calculation time