Search

• Four hydrological models calibrated with four objective functions are compared. • The Granger Ramanathan average variant C (GRC) method performs the best. • Using more than nine ensemble members does not further improve performance. • Combinations of models and objective functions are better than the single model and objective. • Averaging outperforms the ensemble members except in low-flow simulations. The multi-model combination is a technique to improve the performances of hydrological streamflow simulations. An area that has not been investigated much is the performance dependence of combination techniques on the hydrological model calibration strategy and ensemble size. This study aims at investigating the joint effect of the hydrological models, calibration strategies and ensemble sizes on combination abilities for selecting the most appropriate multi-model combination method. The ensemble members were constructed by applying four hydrological models and four objective functions over 383 catchments in China. The ensemble members were combined by using nine commonly used methods, which are Equal Weights (EWA), Akaike Information Criterion (AICA), Bayes Information Criterion (BICA), Bates and Granger (BGA), Granger Ramanathan A, B, and C (GRA, GRB, and GRC), Bayesian Model Averaging (BMA) and Multi-model Super Ensemble (MMSE). The GRC is found as the best multi-model combination method for hydrological simulations. Adding ensemble members by either multiple hydrological models or calibration strategies could help to improve the simulation abilities. Specifically, the increase of ensemble members can obviously enhance the performance of multi-model combinations when the ensemble size is less than six, while only limited improvement is achieved when the ensemble size is more than nine. The combination of ensemble members with various calibration strategies is hard to compensate for the weakness of hydrological model structures. As well, the application of a single calibration strategy in ensemble members only emphasizes single discharge periods and neglects other important discharge periods. This study found that various models with different objective functions are more robust and efficient. The combination performs better than any individual model in terms of Nash–Sutcliffe efficiency (NSE) for approximately 70% catchments, but the multi-model combination is less efficient in terms of low-flow simulations. [ABSTRACT FROM AUTHOR]

• Intelligent adaptation parameter method improves the hydrological model performance. • LightGBM builds flood characteristics-parameter intelligent adaptation relationship. • The XAJ-IAP model can provide accurate and reliable flood simulation. • LightGBM analysis reveals importance of flood characteristics and parameters. Traditional hydrological modeling methods use a set of parameters to simulate flood processes with complex causes and variable intensity, which can easily lead to parameter instability. To address the problem of parameter instability, this study proposes an approach integrating the hydrological model with Intelligent Adaptation Parameters (IAP), whose intelligent adaptation relationship is established by the light gradient-boosting machine (LightGBM) based on individual calibration parameters by each flood event and flood characteristics including flood-caused rainstorm information and initial soil moisture. A widely used hydrological model, Xin 'anjiang (XAJ) model, is chosen to be integrated with IAP (XAJ-IAP) in this study, which has a relatively complex structure and a total of 15 model parameters. The obtained findings demonstrate that: (1) recalibrating the sensitive runoff concentration and separation parameters with a single flood leads to a notable enhancement in simulation accuracy, while simultaneously considering the model's physical significance; (2) the XAJ overestimates large floods and underestimates small floods. Compared with the XAJ, the XAJ-IAP has a better rain-flood response relationship and simulation accuracy for floods of different magnitudes, solving the problem of parameter instability that exists in XAJ; and (3) evaluated in terms of information gain, sensitive parameters contribute the most to the establishment of the intelligent adaptation relationship in the LightGBM compared to flood-caused rainstorm information and initial soil moisture, indicating that sensitive parameters are the most important input features of the LightGBM. It can be concluded that the intelligent adaptation system can not only solve the problem of parameter instability that exists when traditional hydrological models simulate complex and changeable floods, but also further reveal the relationship between the model and floods. [ABSTRACT FROM AUTHOR]

Hydrological impact assessments are increasingly performed at fine spatial and temporal resolutions in order to resolve local-scale changes under a future climate. Apart from the uncertainty represented by different climate models, emission scenarios and post-processing methods, the local-scale internal variability of the climate can be a major source of uncertainty for hydrological projections. To assess the latter at the catchment scale, this paper presents a methodology which is particularly suitable for spatially distributed hydrological models. An ensemble of daily precipitation and daily mean temperature realizations on a high-resolution grid is simulated from stochastic weather generators (WGs) trained on historical data and equipped with climate change information obtained from a regional climate model. Based on the resulting simulated daily runoff data, the significance of changes in the runoff regime is assessed using a statistical hypothesis test, and the variability contributed by the two input variables is quantified using the analysis of variance (ANOVA). As a proof of concept, simulations on a 1-km grid over a period of 19 years are carried out for nine catchments in central Norway. Significant changes in runoff regimes are found, indicating that the trends introduced in the WGs are not overwhelmed by the local-scale internal variability. Variability in the runoff simulations varies substantially throughout the year; it is highest in periods with potential snowmelt and lowest during winter. Temperature is the dominant source of variability in the colder months (November–March) due to its influence on rainfall and snowmelt. High variability in May–June is contributed comparably by both temperature and precipitation. In summer and early autumn the runoff variability is precipitation dominated. The results are in line with findings in the literature where the runoff variability is driven by the large-scale internal climate variability. This indicates that ignoring the local-scale internal variability may yield an underestimation of the overall variability in runoff projections and projected changes. • Gridded climate input data simulated using stochastic weather generators. • Significance of changes in runoff regime assessed using a statistical test. • Runoff variability due to input variables quantified using ANOVA. • Results in line with findings based on large-scale internal climate variability. • Runoff projections need to consider local-scale internal variability of climate. [ABSTRACT FROM AUTHOR]

• A longer calibration period is required for catchments with nonstationary rainfall-runoff relationships to achieve stable simulations. • With the increase in the length of the interval between the calibration and transfer periods, the transferability of the model decreases gradually. • When forecasting runoff under nonstationary rainfall-runoff relationships, the closeness of the total rainfall amount between calibration and transfer periods is more important than the similarity in rainfall processes. Changing climatic conditions have changed the stationary rainfall-runoff relationships in many basins. In this context, the value of the model parameters will depend more on the selection of the calibration period, which directly affects the accuracy of runoff forecasting. However, systematic exploration and testing of the impact of calibration conditions on the transferability of hydrological models under stationary and nonstationary climatic conditions require more effort. The present study investigates the impact of four calibration conditions on model transferability, including the length of the calibration period, the length of the interval between the calibration and transfer periods, the difference in climate conditions as measured by the total rainfall amount between the calibration and transfer periods, and the difference in the similarity of rainfall processes between the calibration and transfer periods. Two catchments with stationary and nonstationary climatic conditions, and five models, including XAJ, HBV, IHACRES, SIMHYD and GR4J, are used in this study. The results show that (1) a longer calibration period is required for catchments with nonstationary rainfall-runoff relationships to achieve stable simulations; (2) with the increase in the length of the interval between the calibration and transfer periods, the transferability of the model decreases gradually, and the degree of reduction is greater for catchments with nonstationary climatic conditions and rainfall-runoff relationships; and (3) when forecasting future runoff under nonstationary rainfall-runoff relationships, the closeness of the total rainfall amount between the calibration and transfer periods is more important than the similarity in rainfall series between the calibration and transfer periods. This study provides insight into the impact of calibration conditions on the transferability of hydrological models in the context of climate change. [ABSTRACT FROM AUTHOR]

• A novel Encoder-Decoder with an Exogenous input (EDE) structure is proposed. • Four models are evaluated and compared from different perspectives. • The EDE structure is more suitable for long lead-time flood forecasting. • The LSTM-EDE model improves the multi-step-ahead flood forecasting accuracy. Accurate and reliable multi-step-ahead flood forecasting is beneficial for reservoir operation and water resources management. The Encoder-Decoder (ED) that can tackle sequence-to-sequence problems is suitable for multi-step-ahead flood forecasting. This study proposes a novel ED with an exogenous input (EDE) structure for multi-step-ahead flood forecasting. The exogenous input can be the outputs of process-based hydrological models. This study constructs four multi-step-ahead flood forecasting approaches, including the Xinanjiang (XAJ) hydrological model, the single-output long short-term memory (LSTM) neural network with recursive strategies, the recursive ED combined with the LSTM neural network (LSTM-RED), and the LSTM-EDE models. The performance of these four models is evaluated and compared by the long-term 3 h hydrologic data series of the Lushui and Jianxi basins in China. The results show that the LSTM-RED model that integrates recursive strategies into the training process of neural networks is more advantageous than the LSTM model. The proposed LSTM-EDE model can overcome the exposure bias problem, simplify its model structure, increase the computational efficiency in the validation process, and improve the multi-step-ahead flood forecasting accuracy, as compared to the LSTM-RED model. [ABSTRACT FROM AUTHOR]

• Machine learning assists hybrid model to promote flood forecasting and early warning. • Hybridizing MCQRNN with XAJ model for flood probability density forecasting. • XAJ-MCQRNN conquers overfitting and biased-prediction bottlenecks. • XAJ-MCQRNN improves accuracy and reliability of flood probability density forecasts. Making accurate and reliable probability density forecasts of flood processes is fundamentally challenging for machine learning techniques, especially when prediction targets are outside the range of training data. Conceptual hydrological models can reduce rainfall-runoff modelling errors with efficient quasi-physical mechanisms. The Monotone Composite Quantile Regression Neural Network (MCQRNN) is used for the first time to make probability density forecasts of flood processes and serves as a benchmark model, whereas it confronts the drawbacks of overfitting and biased-prediction. Here we propose an integrated model (i.e. XAJ-MCQRNN) that incorporates Xinanjiang conceptual model (XAJ) and MCQRNN to overcome the phenomena of error propagation and accumulation encountered in multi-step-ahead flood probability density forecasts. We consider flood forecasts as a function of rainfall factors and runoff data. The models are evaluated by long-term (2009–2015) 3-hour streamflow series of the Jianxi River catchment in China and rainfall products of the European Centre for Medium-Range Weather Forecasts. Results demonstrated that the proposed XAJ-MCQRNN model can not only outperform the MCQRNN model but also prominently enhance the accuracy and reliability of multi-step-ahead probability density forecasts of flood process. Regarding short-term forecasts in testing stages at four horizons, the XAJ-MCQRNN model achieved higher Nash-Sutcliffe Efficiency but lower Root Mean Square Error values, while improving Coverage Ratio and Relative Bandwidth values in comparison to the MCQRNN model. Consequently, the improvement can benefit the mitigation of the impacts associated with uncertainties of extreme flood and rainfall events as well as promote the accuracy and reliability of flood forecasting and early warning. [ABSTRACT FROM AUTHOR]