Search

Nonstationary calculation of sediment load is an important basis for reservoir operation and management. The Jinsha River Basin (JRB) is the most important sediment-producing section of the Yangtze River, and the main source of sediment into the Three Gorges Reservoir (TGR). In this study, three sediment transport nonstationary analysis models were constructed through the Generalized Additive Models for Location, Scale and Shape (GAMLSS) model, namely Mode0, Mode1 and Mode2. The study found that the correlation coefficients between sediment load with runoff and rainfall are 0.78 and 0.73, respectively. Runoff and precipitation can be used as covariates for the nonstationary analysis of sediment transport. The optimal edge distribution of Mode0, Mode1, and Mode2 is Gumbel, Weibull, and Logistic, respectively. Mode2 can accurately describe the nonstationary change of sediment load, and Mode1 can basically reflect the nonstationary change of sediment load. While Mode0 is a consistent model, which cannot reflect the nonstationary change of sediment load over time. Mode2 has a good ability to simulate the quantile changes of sediment load during the training period and the test period, while Mode1 has a weak generalization ability. Mode0 cannot reflect inconsistent changes, and the simulation result is a fixed value. This study provides an important reference for the nonstationary analysis of sediment load under inconsistency conditions in the JRB. [ABSTRACT FROM AUTHOR]

Runoff prediction plays an important guiding role in water resources planning and management, flood and drought prevention. As the Hanjiang River Basin (HRB) is the water source of the middle line of the South-to-North Water Diversion Project, it has higher requirements for water resources accurate prediction. In order to analyze the prediction capabilities of different prediction methods for the HRB runoff, this study constructed 12 prediction models to simulate and predict the runoff of four hydrological stations in the HRB. Furthermore, the Markov Chain Monte Carlo (MCMC) method was used to analyze the transition probability of runoff from low-to-high (high-to-low). The results showed that the runoff of four hydrological stations in the HRB all showed a downward trend, and most of the sudden changes occurred in the 1980 s. The smoother the runoff changes, the easier it is to make accurate prediction. Among the 12 models, the quadric spline Markov forecasting model (QS-MFM), moving average Markov forecasting model (MA-MFM), Markov forecasting model (MFM), deep neural networks (DNN), and cubic exponential smoothing (CES) methods have stronger generalization ability and can more accurately predict the runoff of the HRB. The average relative error during the validation period is 0.27, 0.28, 0.33, 0.34 and 0.39, respectively. The logistic model can accurately simulate the change of runoff status in the HRB. The wet threshold of Baihe (BH), Huanglongtan (HLT), Huangjiagang (HJG), and Huangzhuang (HZ) is 819.9 m3/s, 207.4 m3/s, 1313.9 m3/s and 1681.7 m3/s, and the dry threshold is 480.4 m3/s, 130.6. m3/s, 817.8 m3/s and 1083.4 m3/s, respectively. The MCMC method can accurately estimate the parameters of the logistic model, and the low-high (high-low) runoff transition probability model constructed in the HRB can accurately calculate the low to high (high to low) runoff transition probability. [ABSTRACT FROM AUTHOR]