Journal of Hydrologic Engineering

The Nash–Sutcliffe efficiency (NSE) is now the most widely used criterion for measuring the goodness of fit between the hydrological model simulation and corresponding observation. Because there is substantial sampling uncertainty regarding hydrological simulation and observation, the NSE is a random variable. A probability density function (PDF) of NSE variable was derived based on the assumption of the simple linear regression model between the observation and simulation from the hydrological model. To avoid a subjective interpretation of the hydrological model performance, the confidence interval of the NSE variable was determined by its PDF. Because the difference in NSE variables ($\mathrm{\Phi }$) between two time periods or two hydrological models is often used for comparing their performances and can also be taken as a random variable, hypothesis testing should be implemented to determine whether the difference is adequate or whether the difference is a chance variation. Because the PDF of the difference can be derived based on the joint PDF of the every NSE random variables, a procedure of the hypothesis testing for the difference between two NSE variables is then proposed for comparing the performances between the hydrological models or between different time periods for a hydrological model. The proposed hypothesis testing has been applied in the abcd and dynamic water balance model (DWBM) hydrological models as case studies to illustrate the procedure of assessing the performances of a hydrological model and justifying its superiority to another model according to both the estimated NSE value and its confidence level due to the sampling uncertainty. Therefore, the proposed hypothesis testing can provide hydrological model end-users with rational model assessment.

Abstract

Several reservoirs across the United States are filling with sediment, which jeopardizes their functionality and increases maintenance costs. USACE developed the Reservoir Sedimentation Information (RSI) system to assess reservoir aggradation and track dam operation suitability for water resource management and dam safety. The RSI data set contains historical elevation-capacity data for approximately 400 dams (excluding navigation structures), which correspond to less than 1% of dams across the United States. Thus, there is a critical need to develop methods for estimating reservoir sedimentation for unmonitored sites. The goal of this project was to create a generalized method for estimating reservoir sedimentation rates using reservoir design information and watershed data. To meet this objective, geospatial tools were used to build a refined composite data set to complement the RSI system’s data with precipitation and watershed characteristics. Nine deep learning models were then used on the benchmark data set to determine its accuracy at predicting capacity loss for the RSI reservoirs: four supervised machine learning models, four deep neural network (DNN) models, and a multilinear power regression model. A DNN model, containing a progressively increasing node and layer construction, was deemed the most accurate, with $R^2$ values from its calibration and validation data sets being 0.83 and 0.70, respectively. The best model was recalibrated over the entire data set, which showed greater accuracy on the prediction of the RSI reservoir’s capacity loss, with an $R^2$ of 0.81. This predictive model could be used to evaluate the capacity loss of unmonitored reservoirs, forecast sedimentation rates under future climate conditions, and identify reservoirs with the highest risk of losing functionality.

Practical Applications

Many communities depend on reservoirs for a variety of socioeconomic benefits, such as providing reliable water sources and flood mitigation. Rivers entering reservoirs are a constant source of silt, sand, and gravel particles (i.e., sediments) that deposit slowly, filling the reservoirs over time, thus reducing their volume capacity and effectiveness. Surveys to measure reservoir capacities are labor intensive and expensive and many sites go unmonitored. This study provides prediction tools to estimate capacity loss over time using data derived from publicly available data sources (e.g., digital elevation models, monthly precipitation data, and the National Inventory of Dams). A key role of this tool is to forecast reservoir capacity loss over time under varying climate change conditions and relate the associated sedimentation processes to local and regional conditions. The tool can also be applied broadly to hindcast capacity loss for reservoirs with or without prior surveys for identifying high-risk sites that should be investigated further. USACE plans to use these prediction tools with the RSI database to conduct a national assessment of reservoir impacts, which will inform distributions of federal resources to address water security concerns related to reservoir sedimentation.

Variational modal decomposition (VMD) has proven to be an effective technique for improving the accuracy of runoff prediction and has been widely used in runoff analysis and prediction. However, the presence of noise and outliers can complicate the runoff decomposition, so obtaining purely stochastic and deterministic components is difficult. Therefore, exploring the chaotic properties of the components obtained from decomposition can provide a new perspective to reveal intrinsic variability patterns and enhance our understanding of the unity of determinism and stochasticity. In this study, a two-stage VMD framework is proposed to decompose the monthly runoff, with the chaotic characteristics analyzed by employing an unthreshold recurrence plot and the largest Lyapunov exponent. Additionally, a chaotic prediction model is developed using phase space reconstruction, Volterra filter, and wavelet neural network methodologies. The model is validated using monthly runoff data from four stations in the Yellow River Basin, China. The results demonstrate that, although the VMD method effectively isolates trend components in monthly runoff, it still exhibits challenges in separating periodic and random elements, leading to the identification of chaotic components characterized by the amalgamation of periodicity and randomness. Notably, the two-stage VMD-phase space reconstruction-Volterra-wavelet neural networks model outperforms the VMD-phase space reconstruction-Volterra-wavelet neural networks model, with a substantial increase in Nash–Sutcliffe efficiency during validation, rising from an average of 0.9271–0.9508 and reaching a maximum of 0.96 across the four stations. Overall, this study demonstrates the potential of VMD for improving the monthly runoff prediction accuracy and elucidates the interplay between determinism and stochasticity in runoff analysis, offering valuable insights for further research in this area.

Accurate and reliable runoff prediction is essential for the efficient operation of hydropower systems. This paper presented a runoff probability prediction model that utilizes an enhanced long short-term memory (LSTM) network. The model incorporates a combination of a long and short-term memory network, a quantile regression module and an interval correction module. The proposed model utilizes the LSTM network to effectively capture the time-series characteristics of the runoff data. By incorporating the quantile regression module, the model allows for probability predictions without the need for prior assumptions. Furthermore, the inclusion of an interval correction module helps refine the prediction results, leading to improved accuracy and a narrower prediction interval. The integration of these three modules greatly enhances the precision of the predictions and brings the probability estimates closer to the true distribution. By incorporating the quantile regression module, the model allows for probability predictions without the need for prior assumptions. Jinsha River and Lancang River were selected to evaluate the performance of the model because of the availability of long-term reliable data, geographical representation, and socioeconomic importance. The prediction results demonstrate superior performance compared with other existing models. Moreover, the model enables obtaining probabilistic predictions with appropriate prediction intervals and high reliability.

The Neches River Tidal floodplain has been historically prone to flooding, resulting in critical shutdowns of refineries and port transit. This study implements a one-dimensional (1D) and a coupled 1D/two-dimensional (2D) Hydrological Engineering Center’s River Analysis System (HEC-RAS) model including the bathymetry data of the main channel to develop flood inundation maps under different hydrological and hydraulic conditions. Comparing the 1D model with the FEMA Base Level Engineering (BLE) model, the results show that flood inundation maps are highly sensitive to the bathymetry data, which is essential to simulate accurate flood inundation maps for a river with dredged waterways. The inundation maps simulated by the 1D/2D model have better agreement with the high-water marks than the 1D model during Hurricane Harvey. In conclusion, coupled 1D/2D modeling provides a more realistic depiction of flood threats with the limited available data and is superior to the FEMA BLE model for producing accurate flood inundation maps of the Neches River Tidal floodplain.

The rise of groundwater level and the acceleration of flow velocity in the discharge section of a geothermal system is difficult to explain by gravity-driven groundwater system theory alone. The temperature of groundwater increases after being heated in the geothermal systems, resulting in a decrease in groundwater density, an increase in salinity, and a decrease in viscosity. These changes create additional pressure heads, which are collectively recorded as geothermal driving forces. Based on the hydrochemistry and isotopes of hot springs in the Heyuan fault zone, South China, the geothermal driving force was evaluated as a case study. The geothermal driving force, which is mainly caused by the increase of groundwater temperature, increases with the increase of groundwater temperature and decreases with the increase of groundwater salinity. At the deepest position of groundwater circulation in the Heyuan geothermal field, the groundwater temperature reaches a maximum of 113.42°C, and the geothermal driving force generated here also reaches a maximum of $+255.88\mathrm{m}$. The geothermal driving force gradually decreases from deep to shallow in the Heyuan fault zone, but it makes the average vertical flow velocity of groundwater increase from $3.3\times {10}^{-3}\mathrm{m}/\mathrm{d}$ in the recharge section to $16.8\times {10}^{-3}\mathrm{m}/\mathrm{d}$ in the discharge section. Therefore, the geothermal driving force exists in the discharge section of the geothermal system, and its additional effect accelerates the flow of geothermal water in the discharge section.