JOURNAL OF ENVIRONMENTAL MANAGEMENT, vol.359, pp.1-13, 2024 (SCI-Expanded)
A deep learning architecture, denoted as CNN s LSTM, is proposed for hourly rainfall-runoff modeling in this study. The architecture involves a serial coupling of the one-dimensional convolutional neural network (1DCNN) and the long short-term memory (LSTM) network. In the proposed framework, multiple layers of the CNN component process long-term hourly meteorological time series data, while the LSTM component handles shortterm meteorological time series data and utilizes the extracted features from the 1D -CNN. In order to demonstrate the effectiveness of the proposed approach, it was implemented for hourly rainfall-runoff modeling in the Ishikari River watershed, Japan. A meteorological dataset, including precipitation, air temperature, evapotranspiration, longwave radiation, and shortwave radiation, was utilized as input. The results of the proposed approach (CNN s LSTM) were compared with those of previously proposed deep learning approaches used in hydrologic modeling, such as 1D -CNN, LSTM with only hourly inputs (LSTM w Hour), a parallel architecture of 1D -CNN and LSTM (CNN p LSTM), and the LSTM architecture, which uses both daily and hourly input data (LSTM w D p H). Meteorological and runoff datasets were separated into training, validation, and test periods to train the deep learning model without overfitting, and evaluate the model with an independent dataset. The proposed approach clearly improved estimation accuracy compared to previously utilized deep learning approaches in rainfall = runoff modeling. In comparison with the observed flows, the median values of the Nash-Sutcliffe efficiency for the test period were 0.455-0.469 for 1D -CNN, 0.639-0.656 for CNN p LSTM, 0.745 for LSTM w Hour, 0.831 for LSTM w D p H, and 0.865-0.873 for the proposed CNNsLSTM. Furthermore, the proposed CNN s LSTM reduced the median root mean square error (RMSE) of 1D -CNN by 50.2%-51.4%, CNN p LSTM by 37.4%-40.8%, LSTM w Hour by 27.3%-29.5%, and LSTM w D p H by 10.6%-13.4%. Particularly, the proposed CNN s LSTM improved the estimations for high flows ( >= 75th percentile) and peak flows ( >= 95th percentile). The computational speed of LSTM w D p H is the fastest among the five architectures. Although the computation speed of CNN s LSTM is slower than LSTM w DpH's, it is still 6.9-7.9 times faster than that of LSTM w Hour. Therefore, the proposed CNN s LSTM would be an effective approach for flood management and hydraulic structure design, mainly under climate change conditions that require estimating hourly river flows using meteorological datasets.