Search

• Parameter uncertainty varies over time scales due to land use change (LUC). • The calibrated daily runoff model is deficient for the extreme flow changes. • Model structure impacts the extreme runoff direction change for moderate LUC. Hydrological modeling has provided key insights into the mechanisms of land-use change impacts on runoff. However, the uncertainty of this impact is poorly understood. This paper examines the uncertainty originated from hydrological models' parameters and structure in simulating hydrological responses to land use change on daily, monthly and annual time scales. Two hydrological models, SWAT and HSPF, were applied to simulate the runoff response in the Xitiaoxi basin of eastern China. The effects of three land use scenarios (in 1985, 2002 and 2014) were analyzed with respect to the expansion of urban areas and reduction of cropland and forests. The changes in streamflow with multi-timescales between two models were compared at the whole basin and the sub-basin scales. The parameter uncertainty was estimated based on a sequential uncertainty analysis method (SUFI-2). The results indicate that the two models (SWAT and HSPF) could reproduce the observed daily and seasonal flows well, but they cannot accurately reconstruct the extreme flows (such as annual 7-day minimum discharge, R_7dMIN). The effect of parameter uncertainty on streamflow varies over time scales. The simulated annual and monthly runoff change values show large discrepancies but with same trends. The simulated changes for annual maximum discharge (R_1dMAX) and R_7dMIN often show different signs. The average width of the predictive interval in the relative change of different flow characteristics exceeded 100% of the average relative changes for both models. HSPF tends to present larger relative changes for the annual and monthly runoffs compared to SWAT. Moreover, the model choice could affect the direction change in R_1dMAX and R_7dMIN. However, when land use change was significant to a certain degree, both models simulated increased R_1dMAX and decreased R_7dMIN, but with different significance levels. The study suggests that the model structure represents an additional uncertainty, which should be accounted for in the land-use change impact assessment. [ABSTRACT FROM AUTHOR]

• A deep convolutional generative adversarial network (DCGAN) combined with Wasserstein generative adversarial network (WGAN) called DC-WGAN is proposed as a runoff generation model. • A runoff generation method based on Copula function is developed. • The applicability of DC-WGAN in the field of runoff stochastic generation for a reservoir system is verified. • The spatial–temporal correlation of runoff samples of DC-WGAN compared with the Copula function is analyzed. The implicit stochastic optimization (ISO) model is a methodology that has been extensively employed in the mid-long term optimal operation of a reservoir system. The generation of runoff series is the key technology to make up for the poor representativeness and reliability of historical runoff samples and is the premise to guarantee the accuracy of the ISO model. However, the traditional methods are not suitable for the stochastic generation of runoff series, among reservoirs with spatial–temporal correlation. In this work, the generative adversarial network called DC-WGAN is developed as a runoff generation model. DC-WGAN is based on a data-driven model and can capture the correlation of runoff series in time and space dimensions simultaneously. The proposed method is applied to the downstream Jinsha river cascade and Three Gorges cascade reservoirs in China to generate ten-day runoff series and is compared with the stochastic generation method based on Copula function. The focus of this paper is the model's performance in replicating the spatial–temporal correlation of ten-day runoff series. Results are as follows. (1) DC-WGAN has strong learning and generalization ability, which can learn the correct distribution of original runoff samples and generate runoff series similar to that from the validation set. (2) The runoff series among reservoirs with spatial–temporal correlation can be generated through DC-WGAN without assuming the probability distribution of samples in advance. (3) DC-WGAN is better than Copula function in capturing the spatial–temporal correlation, especially the moderate and weak correlation. The proposed methodology has a high practical value and application prospect for the stochastic generation of runoff series with spatial–temporal correlation. [ABSTRACT FROM AUTHOR]

• A new framework based on "feature decomposition-learning reconstruction" was developed for daily runoff forecast. • IPSO algorithm was used to optimize the parameters of DBN model. • PACF was applied to determine the most effective input variables. • VMD-DBN-IPSO model outperforms other comparison models, and it shows consistency in peak flow forecast. Accurate and reliable short-term runoff prediction is of great significance to the management of water resources optimization and reservoir flood operation. In order to improve the accuracy of short-term runoff forecasting, a hybrid model-based "feature decomposition-learning reconstruction" named VMD-DBN-IPSO was proposed. In this paper, variational mode decomposition (VMD) is first used to decompose the original daily runoff series into a set of sub-sequence for improving the frequency resolution. Partial autocorrelation function (PACF) is then applied to determine the input variables of each sub-sequence. The improved particle swarm optimization (IPSO) algorithm is combined with the deep belief network (DBN) model to predict each sub-sequences and finally reconstruct the ensemble forecasting result. Three quantitative evaluation indicators, mean absolute error (MAE), root mean square error (RMSE) and Nash-Sutcliffe efficiency coefficient (NSE), were used to evaluate and compare the established models using the historical daily runoff data (1/1/1988-31/12/2017) at Yangxian and Ankang hydrological station in the Han River Basin of China. Meanwhile, a comparative analysis of the performance of VMD-DBN-IPSO model under different forecast periods (1-, 3-, 5- and 7-day lead time) was performed. In addition, the prediction ability of peak runoff of the VMD-DBN-IPSO model is further verified by analyzing the 10 peak flows during the testing data-series. The results indicate that the VMD-DBN-IPSO model can always achieve the best performance in the training and testing stage, and has good stability and representativeness, the NSE coefficient remains above 0.8, and the prediction error of peak flow is within 20%. It is a preferred data-driven tool for forecasting daily runoff. [ABSTRACT FROM AUTHOR]