Runoff forecasting is important for water resource management. Although deep learning models have substantially improved the accuracy of runoff prediction, the temporal and feature dependencies between rainfall–runoff time series elements have not been effectively exploited. In this work, we propose a new hybrid deep learning model to predict hourly streamflow: SA-CNN-LSTM (self-attention, convolutional neural network, and long short-term memory network). The advantages of CNN and LSTM in terms of data extraction from time series data are combined with the self-attention mechanism. By considering interdependences of the rainfall–runoff sequence between timesteps and between features, the prediction performance of the model is enhanced. We explored the performance of the model in the Mazhou Basin, China; we compared its performance with the performances of LSTM, CNN, ANN (artificial neural network), RF (random forest), SA-LSTM, and SA-CNN. Our analysis demonstrated that SA-CNN-LSTM demonstrated robust prediction with different flood magnitudes and different lead times; it was particularly effective within lead times of 1–5 h. Additionally, the performance of the self-attention mechanism with LSTM and CNN alone, respectively, was improved at some lead times; however, the overall performance was unstable. In contrast, the hybrid model integrating CNN, LSTM, and the self-attention mechanism exhibited better model performance and robustness. Overall, this study considers the importance of temporal and feature dependencies in hourly runoff prediction, then proposes a hybrid deep learning model to improve the performances of conventional models in runoff prediction.
19 May 2023 in Journal of Environmental Management
Urbanization has significant impacts on watershed hydrology, but previous studies have been confirmatory and not comprehensive; in particular, few studies have addressed the impact of urbanization on flooding in highly urbanized watersheds. In this study, this effect is studied in Chebei Creek, a highly urbanized watershed in the Pearl River Delta, southern China. Landsat satellite images acquired in 2015 were used to estimate land use and cover changes using the Decision Tree (DT) C4.5 classification algorithm, while the Liuxihe model, a physically based distributed hydrological model (PBDHM), is employed to simulate watershed flooding and hydrological processes. For areas with high degrees of urbanization, the duration of the flood peak is only 1 h, and the flood water level shows steep rises and falls. These characteristics increase the difficulty of flood modeling and forecasting in urbanized areas. At present, hydrological research in urbanized watersheds generally focuses on the quantitative simulation of runoff from urban areas to the watershed, flood flows, peak flood flow, and runoff depth. Few studies have involved real-time flood forecasting in urbanized watersheds. To achieve real-time flood forecasting in urbanized watersheds, PBDHMs and refined underlying surface data based on remote sensing technology are necessary. The Liuxihe model is a PBDHM that can meet the accuracy requirements of inflow flood forecasting for reservoir flood control operations. The accuracies of the two flood forecasting methods used in this study were 83.95% and 97.06%, showing the excellent performance of the Liuxihe model for the real-time flood forecasting of urbanized rivers such as the Chebei Creek watershed.
The uneven distribution of meteorological stations in small and medium-sized watersheds in China and the lack of measured hydrological data have led to difficulty in flood simulation and low accuracy in flood forecasting. Traditional hydrological models no longer achieve the forecasting accuracy needed for flood prevention. To improve the simulation accuracy of floods and maximize the use of hydrological information from small and medium-sized watersheds, high-precision hydrological models are needed as a support mechanism. This paper explores the applicability of the Liuxihe model for flood simulation in the Caojiang river basin and we compare flood simulation results of the Liuxihe model with a traditional hydrological model (Xinanjiang model). The results show that the Liuxihe model provides excellent simulation of field floods in Caojiang river basin. The average Nash–Sutcliffe coefficient is 0.73, the average correlation coefficient is 0.9, the average flood peak present error is 0.33, and the average peak simulation accuracy is 93.9%. Compared with the traditional flood hydrological model, the Liuxihe model simulates floods better with less measured hydrological information. In addition, we found that the particle swarm optimization (PSO) algorithm can improve the simulation of the model, and its practical application only needs one representative flood for parameter optimization, which is suitable for areas with little hydrological information. The study can support flood forecasting in the Caojiang river basin and provide a reference for the preparation of flood forecasting schemes in other small and medium-sized watersheds.
20 November 2024 in International Journal of Disaster Risk Reduction
Precipitation fluctuations in the Pearl River Basin (PRB) have a significant impact on river runoff, causing huge economic losses and casualties. However, future precipitation variations in the PRB remain unclear. Therefore, we explored the projected changes in precipitation in the PRB based on the coupled model intercomparison project phase 6 (CMIP6) model via three shared socio-economic pathways scenarios (SSP1-2.6, SSP2-4.5, and SSP5-8.5). In our study, the optimal ensemble of global climate models in the PRB was identified using the comprehensive rating index (CRI), which is based on climatology, spatial variation, and interannual variability, and it was used to analyze potential precipitation changes in the basin in the period 2025–2100. The results showed that the CMIP6 models underestimated precipitation in the PRB; the consistency between the observations and the multi-model ensemble mean of the four best models was higher than those of any other ensembles, and the CRI value was highest (0.92). The annual precipitation in the PRB shows a significant increasing trend under three scenarios from 2025 to 2100 (p < 0.01), with the highest rate of precipitation increase being seen under the high-emission scenario. By the end of the 21st century, the regional mean precipitation in the PRB will increase by 13%, 9.4%, and 20.1% under SSP1-2.6, SSP2-4.5, and SSP5-8.5 scenarios, respectively. Spatially, the entire basin is projected to become wetter, except for a slight decrease of less than 6% in the central part of the basin and the Pearl River Delta in the near term in the 21st century, and the highest increases are projected to occur in the Xijiang River basin.