Search

• Basic factors affecting deep learning models for streamflow forecasting. • Hybrid deep learning models for streamflow forecasting. • Input and hyperparameter optimization of deep learning models. • Limitation and prospects in streamflow forecasting. Deep learning has emerged as a powerful tool for streamflow forecasting and its applications have garnered significant interest in the hydrological community. Despite the publication of several review articles on machine learning applications in streamflow forecasting, no review paper has yet focused explicitly on deep learning and its hybrid forms. This paper starts with some characteristics of deep learning models to provide a quick view of deep learning. Next, the configurations and characteristics of hybrid deep learning models, which is a hybridization of modeling techniques with deep learning, are discussed. Another vital role while implementing deep learning modeling is the methods applied for input and hyperparameter optimization. Finally, the limitations encountered in streamflow forecasting using deep learning models and recommendations for further research are outlined. This review covers related studies from 2017 to 2023 to provide the most recent snapshot of deep learning modeling applications in streamflow forecasting. These efforts are expected to contribute to the advancement of streamflow forecasting, potentially enabling more informed decision-making in water resource management. [ABSTRACT FROM AUTHOR]

Summary Finite element generation of complicated fracture networks is the core issue and source of technical difficulty in three-dimensional (3-D) discrete fracture network (DFN) flow models. Due to the randomness and uncertainty in the configuration of a DFN, the intersection lines (traces) are arbitrarily distributed in each face (fracture and other surfaces). Hence, subdivision of the fractures is an issue relating to subdivision of two-dimensional (2-D) domains with arbitrarily-distributed constraints. When the DFN configuration is very complicated, the well-known approaches (e.g. Voronoi Delaunay-based methods and advancing-front techniques) cannot operate properly. This paper proposes an algorithm to implement end-to-end connection between traces to subdivide 2-D domains into closed loops. The compositions of the vertices in the common edges between adjacent loops (which may belong to a single fracture or two connected fractures) are thus ensured to be topologically identical. The paper then proposes an approach for triangulating arbitrary loops which does not add any nodes to ensure consistency of the meshes at the common edges. In addition, several techniques relating to tolerance control and improving code robustness are discussed. Finally, the equivalent permeability of the rock mass is calculated for some very complicated DFNs (the DFN may contain 1272 fractures, 633 connected fractures, and 16,270 closed loops). The results are compared with other approaches to demonstrate the veracity and efficiency of the approach proposed in this paper. [ABSTRACT FROM AUTHOR]

Modelling workflow for BRF-LSTM and BRF-GRU predictive models. [Display omitted] • Deep learning hybrid predictive model (BRF-LSTM) is built for streamflow forecasting. • Climate indices, rainfall and periodicity are factored for accurate predictions. • Adopts feature algorithm (i.e. Boruta) with long short-term memory and gated recurrent unit. • Six gauging sites tested; BRF-LSTM yields > 98% errors in ± 0.015 m, ~1.30% relative error. • The proposed model is useful for hydrological and strategic water resources planning. Long-term forecasting of any hydrologic phenomena is essential for strategic environmental planning, hydrologic and other forms of structural design, agriculture, and water resources management. Climate mode indices, utilising machine learning methods, are frequently considered as predictor variables in order to forecast several different hydrological variables. In this study, a feature selection algorithm based on two different deep learning models, i.e., long short-term memory and a gated recurrent unit, is applied to improve the forecasting capability of streamflow water levels at six gauging stations in the Murray Darling Basin of Australia. This paper therefore aggregates the significant antecedent lag memory of climate mode indices, rainfall, and the monthly factor based on the periodicity as the predictor variables to attain significantly accurate stream water level forecasts. This novel method identifies an improved relationship between the stream water level and climate mode indices through the aggregation of the significant lagged datasets capturing the historical features to predict the future streamflow water level. The boruta feature selection algorithm (BRF) was then applied in a two phase process before and after attaining the significant lagged inputs to screen the optimum predictor variables. The merits of the forecast models were evaluated through different performance evaluation criteria. The results show that the accumulated significant lagged inputs based on climate mode indices, along with the rainfall and periodicity factors are seen to provide improved forecasting of the SWL over the non-BRF deep learning approaches where no prior feature selection was applied. The hybrid LSTM method (i.e., BRF-LSTM model) achieved a unique advantage in terms of SWL forecasting, particularly attaining over 98% of the predictive errors lying within a band of +/-0.015 m with relatively low relative errors (RRMSE ≈1.30% and RMAE ≈ 0.882%), outperforming all of the benchmark models. It is also found that the periodicity factor has a potential influence on the accuracy of the forecast models for the four monitored study stations. This study concludes that the newly developed hybrid deep learning approaches, coupled with the BRF feature selection, provide improved forecasting performance. The hybrid approach developed in this paper can therefore be used to provide a strong provide predictive response algorithm for the hydrological variables that were influenced by the low-frequency variability of the climate model indices in respect to streamflow water level. [ABSTRACT FROM AUTHOR]

• A novel optimization method consists of the PSO-GN algorithm and the RMPSG-method. • The PSO-GN algorithm has a much broader search range than the PSO-Q algorithm. • The RPMSG-method combines partial solutions to get a more rational optimal solution. The discharge flow process for flood control of a gate-controlled reservoir is performed by opening and closing the reservoir floodgates. Determining the optimal operation rules of the reservoir system consisting of fully opened floodgates according to time, taking into account the discharge characteristics of the gate-controlled reservoirs during the flood period, was an important demand raised by many reservoir managers. This paper introduces a novel optimization method based on the PSO algorithm combined with the rearrangement method of the particle solution groups to directly determine the optimal operation rules of the reservoir system which consists of the number of fully opened floodgates according to time. When applying the PSO algorithm, unlike previous research that uses the discharge flow process as decision variables, in this paper, the PSO-algorithm which uses the operating plan of floodgates as the decision variables is employed to determine the initial solution for the minimal flood peak discharge in the downstream. This initial solution is modified as a more reliable and realistic optimal operation rule of the reservoir that simultaneously satisfies the maximum flood peak elimination condition and the minimum operation of reservoir floodgates by the newly developed rearrangement method of partial solution groups. To validate the proposed method in this paper, two reservoirs placed in the upstream of Chongchen River and its tributary in the DPR of Korea are selected as a case study. In addition, the PSO algorithm and progressive optimality algorithm (POA) which use the discharge flow process as the decision variables were conducted to compare the performance with the newly proposed method. The validation results show that the method proposed in the paper is superior to the two methods and sufficiently reflects the demand of reservoir managers, which can be effectively applied to the determination of the optimal operating rule of the gate-controlled reservoirs during the flooding period. [ABSTRACT FROM AUTHOR]

Summary Structural uncertainty is an important source of model predictive errors, but few studies have been conducted on the error-transitivity from model structure to nonpoint source (NPS) prediction. In this study, we focused on the structural uncertainty caused by the algorithms and equations that are used to describe the phosphorus (P) cycle at the watershed scale. The sensitivity of simulated P to each algorithm/equation was quantified using the Soil and Water Assessment Tool (SWAT) in the Three Gorges Reservoir Area, China. The results indicated that the ratios of C:N and P:N for humic materials, as well as the algorithm of fertilization and P leaching contributed the largest output uncertainties. In comparison, the initiation of inorganic P in the soil layer and the transformation algorithm between P pools are less sensitive for the NPS-P predictions. In addition, the coefficient of variation values were quantified as 0.028–0.086, indicating that the structure-induced uncertainty is minor compared to NPS-P prediction uncertainty caused by the model input and parameters. Using the stochastic framework, the cumulative probability of simulated NPS-P data provided a trade-off between expenditure burden and desired risk. In this sense, this paper provides valuable information for the control of model structural uncertainty, and can be extrapolated to other model-based studies. [ABSTRACT FROM AUTHOR]

Abstract: To accurately estimate floods with hydrological models, the model parameters and the initial state variables must be known. Good estimations of parameters and initial state variables are required to enable the models to make accurate estimations. The Xinanjiang rainfall-runoff (RR) model has been widely used in humid and semi-humid regions in China. In this paper, we evaluate the sensitivity of the Xinanjiang RR model parameters and the correlation between the state variables and the output streamflow. In order to reduce the impact of streamflow data error, model structural error and parameter uncertainty, the Xinanjiang RR model is coupled with intelligent optimization algorithms and a data assimilation method to estimate the streamflow in the Luo River in China that was first considered as gauged, and then as ungauged for parameter calibration. Model parameters are estimated in batch using a particle swarm optimization algorithm and three variations of ensemble Kalman Filter data assimilation (the state variable assimilation, parameter assimilation and the dual assimilation at the same time) for the Luo River, and the ungauged basin. In this case, the model parameters are set equal to median values. The results show that when parameter values are determined by an optimization algorithm using 10years of data, the dual ensemble Kalman Filter notably improves the simulated results. When the parameters adopt the median, the state variable assimilation has little effect on the estimation results, but the parameter assimilation positively affects the simulated results. The dual ensemble Kalman Filter also notably improves the simulated results. The time scale of assimilation has little effect on simulation results for the dual assimilation, but it has a large effect on the state variable assimilation and the parameter assimilation. [Copyright &y& Elsevier]

Summary: This paper describes an algorithm for disaggregating daily rainfall into sub-daily rainfall ‘fragments’ (fine-resolution rainfall sequences) under a future, warmer climate. The algorithm uses a combined generalised additive model (GAM) and method of fragments (MoFs) framework to resample sub-daily rainfall fragments from the historical record conditional on daily rainfall amount and a range of atmospheric covariates. The rationale is that as the atmosphere warms, future rainfall patterns will be more reflective of historical rainfall patterns corresponding to warmer days at the same location, or to locations which have an atmospheric profile more reflective of expected future climate. It was found that the daily to sub-daily scaling relationship varied significantly by season and by location, with rainfall patterns on warmer seasons or at warmer locations typically showing more intense rainfall occurring over shorter periods compared with cooler seasons and stations. Importantly, by regressing against atmospheric covariates such as temperature, this effect was substantially reduced, suggesting that the approach may also be valid when extrapolating to a future climate. The GAM–MoF algorithm was then applied to nine stations around Australia, with the results showing that relative to the daily rainfall amount, the maximum intensity of short duration rainfall increased by between 4.1% and 13.4% per degree change in temperature for the maximum six minute burst, and by between 3.1% and 6.8% for the maximum 1h burst. The fraction of each wet day with no rainfall also increased by between 1.5% and 3.5%. This highlights that a significant proportion of the change to the distribution of rainfall is likely to occur at sub-daily timescales, with important implications for many hydrological systems. [Copyright &y& Elsevier]

Summary: The management of water through the impoundment of rivers by dams and reservoirs is necessary to support key human activities such as hydropower production, agriculture and flood risk mitigation. Advances in multi-objective optimization techniques and ever growing computing power make it possible to design reservoir operating policies that represent Pareto-optimal tradeoffs between multiple interests. On the one hand, such optimization methods can enhance performances of commonly targeted objectives (such as hydropower production or water supply), on the other hand they risk strongly penalizing all the interests not directly (i.e. mathematically) included in the optimization algorithm. The alteration of the downstream hydrological regime is a well established cause of ecological degradation and its evaluation and rehabilitation is commonly required by recent legislation (as the Water Framework Directive in Europe). However, it is rarely embedded in reservoir optimization routines and, even when explicitly considered, the criteria adopted for its evaluation are doubted and not commonly trusted, undermining the possibility of real implementation of environmentally friendly policies. The main challenges in defining and assessing hydrological alterations are: how to define a reference state (referencing); how to define criteria upon which to build mathematical indicators of alteration (measuring); and finally how to aggregate the indicators in a single evaluation index (valuing) that can serve as objective function in the optimization problem. This paper aims to address these issues by: (i) discussing the benefits and constrains of different approaches to referencing, measuring and valuing hydrological alteration; (ii) testing two alternative indices of hydrological alteration, one based on the established framework of Indicators of Hydrological Alteration (Richter et al., 1996), and one satisfying the mathematical properties required by widely used optimization methods based on dynamic programming; (iii) demonstrating and discussing these indices by application River Ticino, in Italy; (iv) providing a framework to effectively include hydrological alteration within reservoir operation optimization. [Copyright &y& Elsevier]

Summary: In this paper, we propose an ensemble Kalman filter based on the mutiscale finite element method for the large-scale groundwater problem. The multiscale finite element method is employed to avoid the conventional fine-scale simulation during the forward prediction. The EnKF and the multiscale algorithm are integrated as independent components. They are connected by two modules: one for recovering the fine-scale information and the other for reconstructing the coarse-scale basis functions. The first module interpolates the coarse-scale solution with coarse-scale basis functions. Then the state covariance on the fine-scale mesh is obtained with less computational burden. The second module reconstructs coarse-scale basis functions when the model states have been updated. The proposed method is examined by synthetic cases. The results show that the proposed method with an appropriate coarsening level is able to produce satisfactory estimations for the conductivity field and history matching. The new method offers superior computational efficiency and significant saving in the computer memory. [Copyright &y& Elsevier]

Summary: The Richard’s equation mathematically expressing the infiltration problem is a transport type of equation describing two different phenomena namely advection and diffusion being hyperbolic and parabolic type of differential equation, respectively each having their own physical properties thus a single numerical scheme would not essentially be suitable for both. In this paper a numerical algorithm based on time splitting or fractional step method has been proposed and described. This has allowed designing and applying some different numerical schemes more compatible with the mathematical and physical properties of the corresponding phenomenon. In advection part an implicit characteristic based method has been employed to calculate the cell face fluxes and then the new unknowns have been obtained using finite volume method (FVM) explicitly. Two different speed type quantities namely wave celerity and mass velocity have been distinguished in the advection of infiltrated water to the soil, the magnitude of the first being several times larger than the other. The implicit characteristic based method has been designed specifically to cope with the mentioned high wave celerity. An implicit numerical scheme has been employed in which the space derivative term contributed in definition of cell face flux has been discretized in fourth order accurate manner. The extra cells getting involved due to the higher order discretization have been taken in account in a way not to change the tri-diagonal matrix coefficient shape preserving the efficiency of the algorithm. The performance, accuracy and efficiency of the out coming non-iterative numerical algorithm have been successfully examined by some test cases. The relative importance of the advection and diffusion terms in the concerned transport equation and the ration of their contribution in the overall infiltrated flux also have been discussed in detail. [Copyright &y& Elsevier]

Summary: High resolution, long and reliable rainfall time series are extremely important to assess reliable statistics, e.g. the Depth–Duration–Frequency curves that have been widely used to define design rainfalls and rainfall drainage network dimensioning. The potential consequences of changes in measuring and recording techniques have been somewhat discussed in the literature with respect to a possible corresponding introduction of artificial inhomogeneities in time series. In this paper, we show how to detect another artificiality: most of the rainfall time series have a lower recording frequency than that is assumed, furthermore the effective high-frequency limit often depends on the recording year due to algorithm changes. This question is particularly important for operational hydrology, because we show that an error on the effective recording high frequency introduces biases in the corresponding statistics. We developed a simple automatic procedure to assess this frequency period by period and station by station on a large database. The scaling analysis of these time series also shows the influence of high frequency limitations on the scaling behaviour, leading to possible misinterpretation of the significance of characteristic scales and scale-dependent hydrological quantities. [ABSTRACT FROM AUTHOR]

Summary: Real-time data from on-line sensors offer the possibility to update environmental simulation models in real-time. Information from on-line sensors concerning contaminant concentrations in groundwater allow for the real-time characterization and control of a contaminant plume. In this paper it is proposed to use the CPU-efficient Ensemble Kalman Filter (EnKF) method, a data assimilation algorithm, for jointly updating the flow and transport parameters (hydraulic conductivity and porosity) and state variables (piezometric head and concentration) of a groundwater flow and contaminant transport problem. A synthetic experiment is used to demonstrate the capability of the EnKF to estimate hydraulic conductivity and porosity by assimilating dynamic head and multiple concentration data in a transient flow and transport model. In this work the worth of hydraulic conductivity, porosity, piezometric head, and concentration data is analyzed in the context of aquifer characterization and prediction uncertainty reduction. The results indicate that the characterization of the hydraulic conductivity and porosity fields is continuously improved as more data are assimilated. Also, groundwater flow and mass transport predictions are improved as more and different types of data are assimilated. The beneficial impact of accounting for multiple concentration data is patent. [ABSTRACT FROM AUTHOR]

Summary: Despite all the progress made over the years on developing automatic methods for analysing hydrographs and measuring the performance of rainfall–runoff models, automatic methods cannot yet match the power and flexibility of the human eye and brain. Very simple approaches are therefore being developed that mimic the way hydrologists inspect and interpret hydrographs, including the way that patterns are recognised, links are made by eye, and hydrological responses and errors are studied and remembered. In this paper, a dynamic programming algorithm originally designed for use in data mining is customised for use with hydrographs. It generates sets of “rays” that are analogous to the visual links made by the hydrologist’s eye when linking features or times in one hydrograph to the corresponding features or times in another hydrograph. One outcome from this work is a new family of performance measures called “visual” performance measures. These can measure differences in amplitude and timing, including the timing errors between simulated and observed hydrographs in model calibration. To demonstrate this, two visual performance measures, one based on the Nash–Sutcliffe Efficiency and the other on the mean absolute error, are used in a total of 34 split-sample calibration–validation tests for two rainfall–runoff models applied to the Hodder catchment, northwest England. The customised algorithm, called the Hydrograph Matching Algorithm, is very simple to apply; it is given in a few lines of pseudocode. [ABSTRACT FROM AUTHOR]

Summary: Although neural networks have been widely applied to various hydrological problems, including river flow forecasting, for at least 15years, they have usually been trained by means of gradient-based algorithms. Recently nature inspired Evolutionary Computation algorithms have rapidly developed as optimization methods able to cope not only with non-differentiable functions but also with a great number of local minima. Some of proposed Evolutionary Computation algorithms have been tested for neural networks training, but publications which compare their performance with gradient-based training methods are rare and present contradictory conclusions. The main goal of the present study is to verify the applicability of a number of recently developed Evolutionary Computation optimization methods, mostly from the Differential Evolution family, to multi-layer perceptron neural networks training for daily rainfall–runoff forecasting. In the present paper eight Evolutionary Computation methods, namely the first version of Differential Evolution (DE), Distributed DE with Explorative–Exploitative Population Families, Self-Adaptive DE, DE with Global and Local Neighbors, Grouping DE, JADE, Comprehensive Learning Particle Swarm Optimization and Efficient Population Utilization Strategy Particle Swarm Optimization are tested against the Levenberg–Marquardt algorithm – probably the most efficient in terms of speed and success rate among gradient-based methods. The Annapolis River catchment was selected as the area of this study due to its specific climatic conditions, characterized by significant seasonal changes in runoff, rapid floods, dry summers, severe winters with snowfall, snow melting, frequent freeze and thaw, and presence of river ice – conditions which make flow forecasting more troublesome. The overall performance of the Levenberg–Marquardt algorithm and the DE with Global and Local Neighbors method for neural networks training turns out to be superior to other Evolutionary Computation-based algorithms. The Levenberg–Marquardt optimization must be considered as the most efficient one due to its speed. Its drawback due to possible sticking in poor local optimum can be overcome by applying a multi-start approach. [Copyright &y& Elsevier]

Summary: In this paper, we present a large-scale hydrologic model with a full one-dimensional hydrodynamic module to calculate flow propagation on a complex river network. The model uses the full Saint–Venant equations and a simple floodplain storage model, and therefore is capable of simulating a wide range of fluvial processes such as flood wave delay and attenuation, backwater effects, flood inundation and its effects on flood waves. We present the model basic equations and GIS algorithms to extract model parameters from relatively limited data, which is globally available, such as the SRTM DEM. GIS based algorithms include the estimation of river width and depth using geomorphological relations, river cross section bottom level and floodplain geometry extracted from DEM, etc. We also show a case study on one of the major tributaries of the Amazon, the Purus River basin. A model validation using discharge and water level data shows that the model is capable of reproducing the main hydrological features of the Purus River basin. Also, realistic floodplain inundation maps were derived from the results of the model. Our main conclusion is that it is possible to employ full hydrodynamic models within large-scale hydrological models even using limited data for river geometry and floodplain characterization. [ABSTRACT FROM AUTHOR]

Summary: In the construction of a radial basis function (RBF) network, the key learning strategy is to identify the configuration (i.e. to determine the parameters of radial basis functions), which has significant influence on model performance. To improve the performance of an RBF network, a two-step learning algorithm is proposed in this paper. Two learning algorithms, unsupervised and supervised, are combined to develop the two-step learning algorithm. In the first step, the self-organizing map (SOM) data analysis technique is developed to analyze the input data. Through the SOM data analysis technique, the topological relationships among input data are revealed, and the distribution of the input data in the input space is represented by the synaptic weights. Therefore, all synaptic weights of SOM are regarded as potential centers of hidden neurons of the RBF network. In the second step, each synaptic weight of SOM is examined one by one to judge whether the weight can be adopted as the center of hidden neurons of the RBF network or not. Hence, in a supervised manner, the optimal configuration of RBF network is determined. For evaluating the performance of the RBF-based model developed by the proposed two-step learning algorithm, an actual application is performed to forecast the hourly inflow during typhoons. The results show that the proposed model can successfully yield 1–6-h ahead forecasts of inflow. For comparison, two other neural networks, the well known back-propagation neural (BPN) network and the RBF network with the existing conventional learning algorithm, are also adopted to construct the hourly inflow forecasting models. The cross-validation results demonstrate that the proposed model provides more reliable and accurate inflow forecasts than the two other models. In conclusion, the proposed model is useful to improve the forecasts of inflow to the reservoir. [Copyright &y& Elsevier]

Summary: The paper evaluates the prospect of artificial neural network (ANN) simulation over mathematical modeling in estimating safe pumping rate to maintain groundwater salinity in island aquifers. Feed-forward ANN model with quick propagation (QP) as training algorithm has been used to forecast the salinity under varied pumping rates. The accuracy, generalization ability and reliability of the model are verified by real-time field data. The model is trained with 2 years of real-time field data and prediction on water quality with varying pumping rate is made for a span of 5 years. The same is then compared with both real-time field data and the prediction based on SUTRA (Saturated–Unsaturated Transport) computations. The proposed ANN model has surfaced as a simpler and more accurate alternative to the numerical method techniques. The ANN methodology using minimal lag and number of hidden nodes, along with the optimal number of spatial and temporal variables consistently produced the best performing network based simulation models. The prediction accuracy of the ANN model has been extended for another 5 years to further define the limit of pumping at a permissible level of groundwater salinity. [ABSTRACT FROM AUTHOR]

Summary: The design of an effective groundwater remediation system involves the determination of the source zone characteristics and subsequent source zone removal. The work presented in this paper focuses on the three-dimensional extension and field application of a previously described source zone identification and delineation algorithm. The three-dimensional search algorithm defines how to achieve an acceptable level of accuracy regarding the strength, geographic location and depth of a dense non-aqueous phase liquid (DNAPL) source while using the least possible number of water quality samples. Target locations and depths of potential sources are identified and given initial importance measures or weights using a technique that exploits expert knowledge. The weights reflect the expert’s confidence that the particular source location is the correct one and they are updated as the investigation proceeds. The overall strategy uses stochastic groundwater flow and transport modeling assuming that hydraulic conductivity is known with uncertainty (Monte Carlo approach). Optimal water quality samples are selected according to the degree to which they contribute to the total concentration uncertainty reduction across all model layers and the proximity of the samples to the potential source locations. After a sample is taken, the contaminant concentration plume is updated using a Kalman filter. The set of optimal source strengths is determined using linear programming by minimizing the sum of the absolute differences between modeled and measured concentration values at sampling locations. The Monte Carlo generated suite of plumes emanating from each individual source is calculated and compared with the updated plume. The scores obtained from this comparison serve to update the weights initially assigned by the expert, and the above steps are repeated until the optimal source characteristics are determined. The algorithm’s effectiveness is demonstrated by performing a ‘blind test’ at a field site in the Anniston Army Depot (ANAD) in Alabama. [Copyright &y& Elsevier]

Summary: This paper proposes a method for using major flash flood events occurred at ungauged catchments to reduce the uncertainties in estimating regional flood quantiles. The approach is based on standard regionalization methods assuming that the flood peak distribution rescaled by a site-dependent index flood is uniform within a homogeneous region. A likelihood formulation and a Bayesian Markov Chain Monte Carlo (MCMC) algorithm are used to infer the parameter values of the regional distributions. This statistical inference technique has been selected for its rigorousness – various hypotheses are explicitly formulated in the likelihood function, its flexibility as for the type of data that can be treated, and its ability to compute accurate estimates of the confidence intervals for the adjusted parameters and for the corresponding flood quantiles. The proposed method is applied to two data sets from Slovakia and the South of France that consist of series of annual peak discharges at gauged sites and estimated peak discharges of extreme flash flood events that have occurred at ungauged sites. The results suggest that the confidence intervals of the quantiles can be significantly narrowed down provided that the set of ungauged extremes is the result of a comprehensive sampling over the selected region. This remains valid, even if the uncertainties in the estimated ungauged extreme discharges are considered. The flood quantiles estimated by the proposed method are also consistent with the results of site specific flood frequency studies based on historic and paleoflood information. [ABSTRACT FROM AUTHOR]

Summary: Typically, flash floods are studied ex post facto in response to a major impact event. A complement to field investigations is developing a detailed database of flash flood events, including minor events and null reports (i.e., where heavy rain occurred but there was no flash flooding), based on public survey questions conducted in near-real time. The Severe hazards analysis and verification experiment (SHAVE) has been in operation at the National Severe Storms Laboratory (NSSL) in Norman, OK, USA during the summers since 2006. The experiment employs undergraduate students to analyse real-time products from weather radars, target specific regions within the conterminous US, and poll public residences and businesses regarding the occurrence and severity of hail, wind, tornadoes, and now flash floods. In addition to providing a rich learning experience for students, SHAVE has also been successful in creating high-resolution datasets of severe hazards used for algorithm and model verification. This paper describes the criteria used to initiate the flash flood survey, the specific questions asked and information entered to the database, and then provides an analysis of results for flash flood data collected during the summer of 2008. It is envisioned that specific details provided by the SHAVE flash flood observation database will complement databases collected by operational agencies (i.e., US National Weather Service Storm Data reports) and thus lead to better tools to predict the likelihood of flash floods and ultimately reduce their impacts on society. [ABSTRACT FROM AUTHOR]

Summary: The Standardized Precipitation Index (SPI) has become a popular measure of drought across the globe. Several techniques have been employed in attempts to estimate the time at which any changes in drought patterns occur using categories defined by SPI values. Log-linear models, odds ratios and multiple confidence intervals have all been used while trying to estimate such changepoints and have met with varying degrees of success, in part due to small cell counts and the large number of inferences that must be carried out. The purpose of this paper is to provide an alternative method for the detection of a changepoint in SPI data, specifically in the matrix of transitions from one drought class to another. A resampling algorithm is provided that both detects a potential changepoint and determines its significance under the assumption that at most one changepoint is present. Simulation results are provided to validate the method as well as the suggested metrics for detection. Data from Kenya is used to provide four examples where the technique is implemented on observed rainfall data. [Copyright &y& Elsevier]

Summary: The automatic calibration of hydrological models has traditionally been performed using gauged data. However, inaccessibility to remote areas and lack of financial support cause data to be lacking in large tropical basins, such as the Amazon basin. Advances in the acquisition, processing and availability of spatially distributed remotely sensed data move the evaluation of computational models easier and more practical. This paper presents the pioneering integration of spatial altimetry data into the automatic calibration of a hydrological model. The study area is the Branco River basin, located in the Northern Amazon basin. An empirical stage×discharge relation is obtained for the Negro River and transposed to the Branco River, which enables the correlation of spatial altimetry data with water discharge derived from the MGB-IPH hydrological model. Six scenarios are created combining two sets of objective functions with three different datasets. Two of them are composed of ENVISAT altimetric data, and the third one is derived from daily gauged discharges. The MOCOM-UA multi-criteria global optimization algorithm is used to optimize the model parameters. The calibration process is validated with gauged discharge at three gauge stations located along the Branco River and two tributaries. Results demonstrate that the combination of virtual stations along the river can provide reasonable parameters. Further, the considerably reduced number of observations provided by the satellite is not a restriction to the automatic calibration, deriving performance coefficients similar to those obtained with the process using daily gauged data. [Copyright &y& Elsevier]

Summary: Grey relational method (GRM) has emerged in recent years as a new tool to deal with problems of small samples and insufficient information. Inexact data, short sample and incomplete hydrological data in water research are very commonly encountered. Statistical, fuzzy set and rough set theories have been applied to deal with uncertainty problems in hydrology. In this paper, we propose a new algorithm for grey relational degree and apply it to water environment quality evaluation of the Han Jiang River, one of the major branches of the Yangtze River in China. Our proposed method possesses some essential and desirable properties which ensure that translational properties of relational degree do not exist and it will give a more precise and finer grading of the overall water quality. Our empirical applications of GRM have demonstrated that GRM is a useful tool for analysis of inexact data, short sample and incomplete hydrological data. [Copyright &y& Elsevier]

Summary: The objective of this paper is to present a radar-based quantitative precipitation estimation algorithm and assess its quality over the complex terrain of western Iceland. The proposed scheme deals with the treatment of beam blockage, anomalous propagation, vertical profile of reflectivity and includes a radar adjustment technique compensating for range, orographic effects and variations in the relationship. The quality of the estimated precipitation is remarkably enhanced after post-processing and in reasonably good agreement with what is known about the spatial distribution of precipitation in the studied area from both rain gauge observations and a gridded dataset derived from an orographic precipitation model. The results suggest that this methodology offers a credible solution to obtain an estimate of the distribution of precipitation in mountainous terrain and appears to be of practical value to meteorologists and hydrologists. [Copyright &y& Elsevier]

Summary: With the aim of calibrating a large number of catchments for a semi-distributed, process-based conceptual hydrological model, we introduce a straightforward yet robust automatic calibration procedure. Since identification of a global parameter optimum is not feasible in practical terms, the procedure presents a trade-off between computational time and algorithm complexity to identify, with reasonable effort, a parameter set that is well representative of the catchment’s dynamics. In its standard mode, the calibration combines three efficiency scores which are evaluated both for the entire calibration period and in terms of their annual and monthly variations. These scores are furthermore assessed only in their relevant value range, producing a comprehensive overall acceptability score which is used to guide an iterative parameter search algorithm. An additional set of four flood-sensitive scores is added, thereby improving results in the peak-flow range. Calibration was done for 140 mesoscale (roughly 10–1000km2) catchments in Switzerland, using the hydrological modelling system PREVAH (Precipitation–Runoff-EVApotranspiration-HRU related model) in hourly time steps. For 49 representative catchments with long gauge records, a median Nash–Sutcliffe efficiency (NSE) of 0.75 was achieved for the calibration period in standard mode. The limited loss in efficiency when moving to the validation period (median NSE: 0.72) proves the stability and representativity of the parameter sets identified, while a Monte-Carlo analysis underscores the effectiveness of our procedure. The Nash–Sutcliffe efficiencies for the additional flood calibration are slightly lower, but again almost equally high for the calibration (0.69) and the validation (0.67) period. Despite the concessions made to improve peak-flow results, the simulation’s hydrological plausibility was not compromised. The ultimate goal of our study is flood estimation in ungauged Swiss catchments through continuous simulation using PREVAH. With the extensive calibration task presented in this article, the foundation is laid for regionalisation of the tuneable model parameters, which will be addressed in the companion paper (), along with detailed flood estimation results. [Copyright &y& Elsevier]

Summary: This paper deals with the issue of using scintillometry in conjunction with a simple aggregation scheme to derive area-averaged sensible and latent heat fluxes over a small watershed in Niamey, Niger (Wankama catchment). Data collected in the context of the African Monsoon Multidisciplinary Analysis (AMMA) program has been used to test the proposed approach. For this purpose, a Large Aperture Scintillometer (LAS) was set up over heterogeneous surface transect of about 3.2km spanning three vegetation types. The comparison between scintillometer-based estimates of area-averaged sensible heat fluxes and those measured by a network of the classical eddy covariance (EC) devices showed good agreement, with a relative error of about 20% (R 2 =0.85, RMSD=22Wm−2, and SEE=21.39Wm−2). This is a good result considering the contrast in terms of the footprint scales associated with the two devices, which is amplified by the heterogeneity of the catchment. The results also showed that using the LAS-derived values of area-average sensible heat flux in conjunction with a simple aggregation rule to estimate area-average available energy led to a reasonable prediction of area-averaged latent heat flux (R 2 =0.75, RMSD=64Wm−2 and SEE=50Wm−2), when compared to those measured using the EC network. This is of interest since the LAS can potentially be used to validate estimates of surface fluxes, based on the meso-scale model, as well as those estimated using coarse-scale remote sensing-based algorithm. [Copyright &y& Elsevier]

Summary: The use of artificial neural networks (ANNs) for the modelling of water resources variables has increased rapidly in recent years. This paper addresses one of the important issues associated with artificial neural network model development; input variable selection. In this study, the partial mutual information (PMI) input selection algorithm is modified to increase its computational efficiency, while maintaining its accuracy. As part of the modification, use of average shifted histograms (ASHs) is introduced as an alternative to kernel based methods for the estimation of mutual information (MI). Empirical guidelines are developed to estimate the key ASH parameters as a function of sample size. The stopping criterion used with the original PMI algorithm is replaced with a more computationally efficient outlier detection technique based on the Hampel distance. The performance of the proposed PMI algorithm, in terms of computational efficiency and input selection accuracy, is first investigated by using it to identify significant variables for data series where dependencies of attributes are known a priori. The proposed ASH PMI input variable selection algorithm with the Hampel distance stopping criterion consistently selects the correct inputs, while being computationally efficient. The modified PMI algorithm is then applied to identify suitable inputs to forecast salinity in the River Murray at Murray Bridge, South Australia, with a lead time of 14 days using an ANN approach. The ANN models developed with the inputs selected with the modified PMI algorithm perform very well when compared with results obtained using ANN models with different input sets developed in previous studies. Furthermore, the proposed input variable selection algorithm results in more parsimonious ANN models. [Copyright &y& Elsevier]

Summary: This paper presents an operational approach to integrate a storm classification method into a real-time radar rainfall estimation. A minor modification of a texturing classification algorithm proposed by Steiner et al. [Steiner, M., Houze Jr., R.A., Yuter, S.E., 1995. Climatological characterisation of three-dimensional storm structure from operational radar and rain gauge data. J. Appl. Meteorol. 34, 1978–2007] that can classify each pixel in the radar image as stratiform or convective is used to classify the instantaneous reflectivity field into convective and stratiform components. A method to derive the climatological Z–R relations for convective and stratiform rainfall is proposed. Vertical profiles of reflectivity (VPRs) are used to verify the accuracy of the storm classification. An alternative method for verification of a storm classification scheme based on differences between probability distribution functions of rain gauge rainfall of the two rainfall categories is also presented. A 6-month record of radar and rain gauge rainfall for Sydney, Australia for November 2000–April 2001 is used for training and rainfall events during February 2007–March 2007 are used to evaluate the efficiency and applicability of the proposed methods. The results show that the proposed operational approach for classifying storms in real-time radar rainfall estimates reduces RMSE between radar and rain gauge rainfall from 4.63 to 4.30mmh−1 and from 5.31 to 5.01mmh−1 for the training and verification periods compared to the case where the rainfall is assumed to have a single Z–R relationship. [Copyright &y& Elsevier]

• A short-term reservoir optimal operation considering water propagation was proposed in this paper. • NPOA was applied to the short-term cascade reservoirs optimal operation. • The rationality and feasibility of the model and algorithm were verified. Difficulty remains to represent water propagation impact to construct water balance constraints for the Short-Term Cascade Reservoirs Optimal Operation (SCROO), due to the complex water propagation mechanism existing between cascade reservoirs. The imprecise representation of water propagation impact hinders the implementation of hydropower generation schemes and leads to operational risks. Thus, a modified SCROO model considering water propagation impact was proposed in this study to improve the accuracy of cascade operation. Specifically, water propagation is expressed using the Muskingum method in the modified SCROO model, and characterized by two variables, including propagation time related to the magnitude of the upstream reservoir's outflow and flow attenuation related to the propagation time, which were identified using the successive approximation method under water balance constraints within the modified SCROO model. Subsequently, this research applied the Nested Progressive Optimality Algorithm (NPOA) in the model calculation innovatively, in which the stage benefit function and nest computation could be improved. The modified model and algorithm developed herein were applied and validated on cascade reservoirs along the Yalung River in China by comparing it with traditional methods and algorithms. The results show that: 1) Compared to traditional methods, considering water propagation within water balance constraints as a dynamic parameter can calculate the inflow of downstream reservoir more accurately and mitigate risks in the actual operation of cascade reservoirs. Besides, the reservoir inflow calculated by traditional methods will lead to different magnitudes of benefit loss; and 2) the NPOA can significantly reduce dimensionality problems caused by the water propagation impact and greatly improve the actual hydropower generation than that of a traditional optimal algorithm, Multi-Stage Dynamic Programming (MSDP). Compared to the MSDP, the NOPA is more applicable in a cascade system with a larger number of reservoirs. The modified model and algorithm can significantly improve the effectiveness of cascade reservoir optimal operation schemes and provide more information for decision-makers. [ABSTRACT FROM AUTHOR]

Summary: An accurate forecast of river stage is very significant so that there is ample time for the pertinent authority to issue a forewarning of the impending flood and to implement early evacuation measures as required. Since a variety of existing process-based hydrological models involve exogenous input and different assumptions, artificial neural networks have the potential to be a cost-effective solution. In this paper, a split-step particle swarm optimization (PSO) model is developed and applied to train multi-layer perceptrons for forecasting real-time water levels at Fo Tan in Shing Mun River of Hong Kong with different lead times on the basis of the upstream gauging station (Tin Sum) or at Fo Tan. This paradigm is able to combine the advantages of global search capability of PSO algorithm in the first step and local fast convergence of Levenberg–Marquardt algorithm in the second step. The results demonstrate that it is able to attain a higher accuracy in a much shorter time when compared with the benchmarking backward propagation algorithm as well as the standard PSO algorithm. [Copyright &y& Elsevier]

Summary: An accurate water stage prediction allows the pertinent authority to issue a forewarning of the impending flood and to implement early evacuation measures when required. Existing methods including rainfall-runoff modeling or statistical techniques entail exogenous input together with a number of assumptions. The use of artificial neural networks (ANN) has been shown to be a cost-effective technique. But their training, usually with back-propagation algorithm or other gradient algorithms, is featured with certain drawbacks such as very slow convergence and easy entrapment in a local minimum. In this paper, a particle swarm optimization model is adopted to train perceptrons. The approach is applied to predict water levels in Shing Mun River of Hong Kong with different lead times on the basis of the upstream gauging stations or stage/time history at the specific station. It is shown that the PSO technique can act as an alternative training algorithm for ANNs. [Copyright &y& Elsevier]

Summary: A common problem in coastal aquifers’ management is maximization of groundwater extraction rate without saltwater intrusion. To this end, i.e., to optimize layout of a number of wells and distribution of total flow rate to them, a computational procedure has been developed, combining an optimization and a flow simulation tool. The optimization code is based on genetic algorithms. The respective evaluation function includes numerical simulation of an “equivalent” depth-integrated groundwater flow, by means of a boundary element code. Observance of the main constraint, i.e., zero water inflow through the coast, is checked using the sign of flow rate values at the coastal boundary elements. Validity of this approach is checked with the help of an analytical solution and the balance between accuracy and simplification is discussed. Applications of the proposed method to a coastal aquifer with two zones of different transmissivities, investigating the role of form and magnitude of the penalty function in the optimization process and demonstrating the efficiency of the proposed computational procedure, conclude the paper. [Copyright &y& Elsevier]

Abstract: Distributed watershed models should pass through a careful calibration procedure before they are utilized as a decision making aid in the planning and management of water resources. Although manual approaches are still frequently used for calibration, they are tedious, time consuming, and require experienced personnel. This paper describes an automatic approach for calibrating daily streamflow and daily sediment concentration values estimated by the US Department of Agriculture''s distributed watershed simulation model, Soil and Water Assessment Tool (SWAT). The automatic calibration methodology applies a hierarchy of three techniques, namely screening, parameterization, and parameter sensitivity analysis, at the parameter identification stage of model calibration. The global parameter sensitivity analysis is conducted using a stepwise regression analysis on rank-transformed input–output data pairs. Latin hypercube sampling is used to generate input data from the assigned distributions and ranges, and parameter estimation is performed using genetic algorithm. The Generalized Likelihood Uncertainty Estimation methodology is subsequently implemented to investigate uncertainty of model estimates, accounting for errors due to model structure, input data and model parameters. To demonstrate their effectiveness, the parameter identification, parameter estimation, model verification, and uncertainty analysis techniques are applied to a watershed located in southern Illinois. [Copyright &y& Elsevier]

Haktanir originally introduced the method of self-determined probability weighted moments as an extension of the traditional method of probability weighted moments for parameter estimation. While this method possesses many advantages, his algorithms introduced certain mathematical manipulations for numerical convenience or based upon special knowledge of the behavior of a data sample. Also, some of these algorithms relied upon inputs from numerical tables that are not widely accessible. To improve the usefulness of this method, new algorithms have been developed that directly implement the relevant equations and do not rely upon external results. In this paper, we show that these features extend the applicability of self-determined probability weighted moments without loss of accuracy in the parameter estimates. Examples from flood peak analysis and extreme wind speed estimation are presented. [Copyright &y& Elsevier]

• Studied the effect of catchment size and simulation time steps on parameter sensitivity. • Hydrological model is developed for 10 different watersheds of Narmada River basin. • Two Global Sensitivity Analysis methods used to identify the sensitive parameters. • To avoid over parameterization, most and least sensitive parameters are identified. • Identifying a set of sensitive parameters that covers all the hydrological processes. Hydrological models have many parameters representing various hydrological processes. These parameters are effective at different spatial and temporal resolution. Most of the parameters can be measured such as slope, elevation, area, vegetation type etc., but others cannot be estimated from the available data. Thus, they remain unknown in a hydrological simulation. Local and Global Sensitivity Analyses (LSA and GSA respectively) are therefore used to reduce the number of parameters that need to be equipped with input–output data. GSA also improves the efficiency of the model calibration and validation which in turn increases the reliability of the model. In this paper, we investigate into the effect of size of a catchment (small, medium, and large) time steps of simulation (Daily and Monthly), and the choice of GSA method in performing sensitivity analysis. A Soil and Water Assessment Tool (SWAT) hydrologic model is developed for 10 subbasins of Narmada River basin at daily and monthly temporal simulations from January 1991 to December 2000. The sensitivity of 29 input parameters have been analysed with two GSA methods: the multilinear regression SA used by sequential uncertainty fitting algorithm (SUFI-2) in SWAT – Calibration and Uncertainty procedure (SWAT-CUP), and the first order sensitivity indices driven by Fourier Amplitude Sensitivity Test (FAST). Our results show that a hydrological model set up with only sensitive parameters can also produce similar results as when all the parameters are considered. We concluded that the selection of catchment size and simulation time step strongly influence the parameter sensitivity. Our results show that the sensitivity of some of the parameters like Curve number for moisture condition (CN2.mgt), baseflow alpha factor for bank storage (ALPHA_BNK.rte) and lateral flow travel time (LAT_TTIME.hru) are independent of catchment size and simulation time step. Moreover, baseflow alpha factor for the recession constant (ALPHA_BF.gw), manning's value for overland flow (OV_N.hru) and soil erodibility factor (USLE_K(..).sol) are not sensitive for large and medium range watersheds but those are sensitive for small watershed at a monthly time step. Ground water delay (GW_DELAY.gw), average slope steepness (HRU_SLP.hru) and support practice (p) factor of USLE soil equation (USLE_P.mgt) are not sensitive for large watersheds but sensitive for small and medium range watersheds at a daily time step. Our results strongly recommends that a detailed parameter sensitivity analysis is an important step in setting up any hydrological model to reduce the number of parameters while addressing all relevant hydrological processes. [ABSTRACT FROM AUTHOR]

• Adaptive sine cosine algorithm is proposed for hydropower operation. • Elite mutation strategy is introduced to increase individual diversity. • Simplex dynamic search strategy is used to improve solution accuracy. • Neighborhood search strategy is used to improve the convergence rate. • Simulations demonstrate the superiority of the presented method. Recently, the multiple hydropower reservoirs operation optimization is attracting rising concerns from researchers and engineers since it can not only improve the utilization efficiency of water resource but also increase the generation benefit of hydropower enterprises. Mathematically, the reservoir operation problem is a typical multistage constrained optimization problem coupled with numerous decision variables and physical constraints. Sine cosine algorithm (SCA), a new swarm-based method, has the merits of clear principle and easy implementation but suffers from the premature convergence and falling into local optima. To improve the SCA performance, this paper proposes an adaptive sine cosine algorithm (ASCA) where the elite mutation strategy is used to increase the population diversity, the simplex dynamic search strategy is used to enhance the solution quality, while the neighborhood search strategy is used to improve the convergence rate. The simulations of 25 test functions show that ASCA outperforms several existing methods in both convergence rate and solution quality. The results of a real-world hydropower system in China demonstrate that ASCA betters the SCA method with obvious increase in power generation. Thus, the main contribution of this study is to provide an effective optimizer for multiple hydropower reservoirs operation. [ABSTRACT FROM AUTHOR]