Showing papers in "Journal of hydrology in 2022"

Research on Particle Swarm Optimization in LSTM Neural Networks for Rainfall-Runoff Simulation

Abstract: Flood forecasting is an essential non-engineering measure for flood prevention and disaster reduction. Many models have been developed to study the complex and highly random rainfall-runoff process. In recent years, artificial intelligence methods, such as the artificial neural network (ANN), have attempted to construct rainfall-runoff models. The more advanced deep learning methods of long short-term memory (LSTM) network have been proved to better predict hydrological time series. However, the selection of LSTM hyperparameters in the past mostly relied on the experience of the staff, which often led to failure to achieve the best performance. The aim of this study is to develop a method to improve flood forecast accuracy and lead time. A deep learning neural network model based on LSTM networks and particle swarm optimization (PSO) is proposed in this paper. The PSO algorithm was used to optimize the LSTM hyperparameter to improve the ability to learn data sequence features. The model focuses on the Jingle Watershed in the Fenhe River and the Lushi Watershed in the Luohe River and was used to predict flood processes using rainfall and runoff observation data from stations in the watersheds. We evaluated the performance of the model with the Nash Sutcliffe efficiency coefficient, root mean square error, and bias. The results show that the PSO-LSTM model outperforms the M-EIES, ANN, PSO-ANN, and LSTM at all stations in the watersheds. The PSO-LSTM model improves the flood forecasting accuracy at different lead times, especially for those exceeding 6 h, and has higher prediction accuracy and stability. The PSO-LSTM model could be used to improve accuracy in short-term flood forecast applications.

Do ERA5 and ERA5-land precipitation estimates outperform satellite-based precipitation products? A comprehensive comparison between state-of-the-art model-based and satellite-based precipitation products over mainland China

Abstract: • Hourly ERA5-Land precipitation products are firstly evaluated over mainland China. • ERA5 and ERA5-Land share similar performance and have their own advantages. • Model-based products outperform satellite-based estimates in some circumstances. • Product recommendations are proposed for different application scenarios. Accurate precipitation retrievals with fine spatio-temporal resolutions are critical in global and regional analyses. As an important alternative of satellite-based precipitation products, model-based precipitation estimates have undergone rapid development over the past few decades. With the recent public release of the fifth generation of atmospheric reanalysis by the European Centre for Medium Range Weather Forecasts (ERA5) and ERA5-Land, it is necessary to verify whether these two latest model-based precipitation products outperform satellite-based precipitation products. This study comprehensively evaluates the performances of state-of-the-art satellite-based and model-based precipitation products, including the Integrated Multi-satellitE Retrievals for Global Precipitation Measurement (IMERG), Global Satellite Mapping of Precipitation (GSMaP), ERA5, and ERA5-Land, over mainland China from 2016 to 2019. The main findings are as follows: (1) Satellite-based products generally outperform model-based products, but the latter significantly perform better than the former over high-latitude regions and in winter. (2) ERA5 and ERA5-Land share similar spatio-temporal patterns and have their own advantages in terms of different types of metrics. (3) Satellite-based products perform best over subregions of subtropical and tropical monsoon climate (ST), whereas model-based products show highest performance over subregions of temperate monsoon climate (TM) and temperate continental climate (TC); both types of products show the poorest performance over subregions of plateau mountain climate (PM). (4) IMERG-Final performs best in terms of precipitation events, while GSMaP-Gauge tends to overestimate the duration of precipitation events, and model-based products tend to underestimate the mean precipitation rate of the events. These findings provide valuable insights into the error characteristics of state-of-the-art model-based and satellite-based precipitation products in the recent years, and will also serve as useful reference for the potential improvement of precipitation retrieval algorithms in the next generation.

Improving streamflow prediction in the WRF-Hydro model with LSTM networks

Abstract: Researchers have attempted to use machine learning algorithms to replace physically based models for streamflow prediction. Although existing studies have contributed to improving machine learning methods, they still have weaknesses, such as large dataset requirements and overfitting. Therefore, we propose an approach that combines the Weather Research and Forecasting hydrological modeling system (WRF-Hydro) and the Long Short-Term Memory (LSTM) network, i.e., WRF-Hydro-LSTM, to improve streamflow simulations. In this approach, LSTM was employed to predict the residual errors of WRF-Hydro; in contrast, the conventional approach with LSTM predicts streamflow directly. Here, we performed numerical experiments to predict the inflow of Soyangho Lake in South Korea using WRF-Hydro-LSTM, WRF-Hydro-only, and LSTM-only. WRF-Hydro-LSTM and LSTM-only showed better results (NSE = 0.95 and R greater than 0.96) compared to WRF-Hydro-only (NSE = 0.72 and R = 0.88); however, in terms of the percent bias, WRF-Hydro-LSTM had a better value (1.75) than LSTM-only (17.36). While the LSTM-only follows objective functions and not physical principles, WRF-Hydro-LSTM simulates residual errors and efficiently decreases uncertainties that are inherent with conventional methods. Furthermore, a sensitivity test on the training dataset indicated that the correlation coefficient and NSE value were not overly sensitive, but the PBIAS value differed substantially depending on the training set. This study demonstrates that WRF-Hydro-LSTM is particularly useful for representing real-world physical constraints and thus can potentially improve streamflow prediction compared to using either of the two approaches exclusively.

Runoff changes in the major river basins of China and their responses to potential driving forces

Abstract: Identifying the effect of climate variability and human activities on runoff changes is scientifically essential for understanding hydrological processes and sustainable water resources management. This study selected 64 catchments located in the mainland of China to quantify the effects of different driving forces on runoff changes. Results showed that annual runoff in the Haihe river basin, Liaohe river basin, and Yellow river basin exhibited significantly decreasing trends from 1965 to 2018 (P < 0.05), whereas the Northwest river basin had positive trends in the annual runoff. Meanwhile, the Pettitt test method was applied to detect abrupt changes in annual runoff. Compared to the rivers in Southern China, the northern rivers had significant abrupt changes in annual runoff and mostly occurred in the 1990 s. The Choudhury-Yang equation based on the Budyko hypothesis was used to assess the sensitivity of runoff to precipitation (P), potential evapotranspiration (ET0), and the land surface (n) changes. The results showed that runoff was more sensitive to P and n, compared to ET0. Attribution analysis revealed that P was the dominant factor in the Northwest river basin, Southwest river basin, Yangtze river basin, Southeast river basin, and Pearl river basin, whereas the changes in n were responsible for runoff changes in the Liaohe river basin, Haihe river basin, Yellow river Basin, Songhuajiang river basin, and Huaihe river basin. The land surface changes (n) were resulted from vegetation restoration, urbanization expansion, construction of reservoirs/check dams, and surface water withdrawals, leading to significant changes in river runoff in recent years. The findings can provide good insight for water resources management across China.

Daily flow duration curve model for ungauged intermittent subbasins of gauged rivers

Abstract: In this study, we develop a daily flow duration curve model for ungauged intermittent subbasins of gauged rivers. The long-term mean streamflow of the river basin, one of the central parts in the model, is calculated with a regression model of annual precipitation and physical characteristics of the river basin; e.g., drainage area, basin relief, topographical slope, drainage density. The input data of the model are normally accesible, and this makes the model applicable to ungauged points within the river basin. Another central part of the model is the cease-to-flow point which makes the model significant for intermittent rivers. The daily streamflow discharge recorded in gauging stations over the river basin is nondimensionalized by dividing each with their own long-term mean, and their collection is transformed to fit the normal probability distribution; i.e., the normalized nondimensional daily streamflow data are used in the model. The streamflow data are inverted back to the original distribution for any given exceedance percentage by incorporating the cease-to-flow point, and are dimensionalized finally by using the empirically-derived long-term mean streamflow discharge. The model is applied on hundreds of thousands station-day daily streamflow data from three river basins in different geographical regions in Turkey. Results of the case studies are found promising to propose the model as a good foundation for the daily flow duration curve at an ungauged intermittent subbasin of gauged rivers: However, it is noticeable that the model might have low performance in some particular gauging stations where the hydrological behavior deviates from the general characteristics of the river basin. This can be overcome by developing empirical models better approaching the observed long-term mean streamflow, which is a key issue of the model.

Prediction of estuarine water quality using interpretable machine learning approach

Abstract: Estuaries are principal sources of pollution in coastal areas. Estuarine water quality prediction models can provide early warnings to prevent major disasters in coastal ecosystems. In this study, several machine learning models—multiple linear regression, artificial neural networks, random forest, and extreme gradient boosting (XGBoost)—were developed to predict NH4+-N in the Xiaoqing River estuary, China. The results show that there is a strong nonlinear relationship between estuarine NH4+-N and NH4+-N of the upper reaches. The shapely additive explanations method was used to interpret the XGBoost model and discover the influence of the upper reaches of the river on the estuary. These explanations showed that two stations monitoring water quality in the upper reaches (Shicun and Sanchakou) had a critical impact on estuarine water quality. If NH4+-N concentration of the upper reaches is below 2 mg/L, estuarine NH4+-N would not be negatively influenced by the upper reaches. These results can support pollution warnings for improving estuarine water quality and the integrated environmental management of the river and costal area.

A Critical Review of Real-Time Modelling of Flood Forecasting in Urban Drainage Systems

Abstract: • Last recent works on real-time flood forecasting in urban drainage systems are reviewed. • A bibliometric and in-depth critical review is conducted. • All points are classified in data collection and preparation, model development and performance assessment. • Real-time data requirements and developed real-time urban flood forecasting models are discussed. There has been a strong tendency in recent decades to develop real-time urban flood prediction models for early warning to the public due to a large number of worldwide urban flood occurrences and their disastrous consequences. While a significant breakthrough has been made so far, there are still some potential knowledge gaps that need further investigation. This paper presents a comprehensive review of the current state-of-the-art and future trends of real-time modelling of flood forecasting in urban drainage systems. Findings showed that the combination of various real-time sources of rainfall measurement and the inclusion of other real-time data such as soil moisture, wind flow patterns, evaporation, fluvial flow and infiltration should be more investigated in real-time flood forecasting models. Additionally, artificial intelligence is also present in most of the new RTFF models in UDS and consequently further developments of this technique are expected to appear in future works.

Application of a modern multi-level ensemble approach for the estimation of critical shear stress in cohesive sediment mixture

Abstract: Exploration of incipient motion study is significantly important for the river hydraulics community. The present study, along with experimental investigation, considered a new multi-level ensemble machine learning (ML) to determine critical shear stress (CSS) of gravel particles in a cohesive mixture of clay-silt-gravel, clay-silt-sand-gravel, and clay-sand-gravel. The multi-level ensemble ML included a voting-based ensemble meta-estimator integrated with three modern standalone ensemble techniques, namely extreme gradient boosting (XGBoost), Adaptive boosting (Adaboost), and Random Forest (RF), and performance is compared with three standalone ensemble models for prediction of CSS values. Besides, the optimum input combinations were explored using the forward stepwise selection method, as a correlation-based feature selection, and mutual information theory. The outcomes of simulation indicated that the multi-level ensemble machine learning (voting) model in terms of correlation coefficient (R = 0.9641), and root mean square error (RMSE = 0.2022) was superior to the standalone ensemble techniques, i.e., XGBoost (R = 0.9482, and RMSE = 0.2375), Adaboost (R = 0.9496, and RMSE = 0.2387), and RF (R = 0.9392, and RMSE = 0.2739) for accurate estimation of CSS.

An Integrated Inversion Framework for Heterogeneous Aquifer Structure Identification with Single-Sample Generative Adversarial Network

Abstract: Generating reasonable heterogeneous aquifer structures is essential for understanding the physicochemical processes controlling groundwater flow and solute transport better. The inversion process of aquifer structure identification is usually time-consuming. This study develops an integrated inversion framework, which combines the geological single-sample generative adversarial network (GeoSinGAN), the deep octave convolution dense residual network (DOCRN), and the iterative local updating ensemble smoother (ILUES), named GeoSinGAN-DOCRN-ILUES, for more efficiently generating heterogeneous aquifer structures. The performance of the integrated framework is illustrated by two synthetic contaminant experiments. We show that GeoSinGAN can generate heterogeneous aquifer structures with geostatistical characteristics similar to those of the training sample, while its training time is at least 10 times faster than that of typical approaches (e.g., multi-sample-based GAN). The octave convolution layer and multi-residual connection enable the DOCRN to map the heterogeneity structures to the state variable fields (e.g., hydraulic head, concentration distributions) while reducing the computational cost. The results show that the integrated inversion framework of GeoSinGAN and DOCRN can effectively and reasonably generate the heterogeneous aquifer structures.

Seepage Characteristics in Loess Strata Subjected to Single Point Water Supply

Abstract: • A Conical-like cavity was formed near the water source, and the preferential passage of water flow appeared below the cavity. • The wetting body shape gradually transformed from oblate ellipsoid to long ellipsoid after the formation of the passage. • Seepage process can be defined as four periods: rapid infiltration, stable infiltration, slow infiltration, and preferential infiltration. • A two-dimensional seepage diffusion calculation model was established and preliminarily verified by example calculations.

Flash-flood hazard using Deep Learning based on H2O R package and fuzzy-multicriteria decision-making analysis

Abstract: The present study was done in order to simulate the flash-flood susceptibility across the Suha river basin in Romania using a number of 3 hybrid models and fuzzy-AHP multicriteria decision-making analysis. It should be noted that flash-flood events are triggered by heavy rainfall in small river catchments. To achieve the proposed results, a total of 8 flash-flood predictors (slope angle, plan curvature, hydrological soil groups, land use, convergence index, profile curvature, topographic position index, aspect) along with a sample of 111 torrential phenomena points were used as input datasets in the next four algorithms: Fuzzy-Analytical Hierarchy Process (FAHP), Deep Learning Neural Network -Analytical Hierarchy Process (DLNN-AHP), Multilayer Perceptron - Analytical Hierarchy Process (MLP-AHP) and Naïve Bayes - Analytical Hierarchy Process (NB-AHP). The Analytical Hierarchy Process was used to calculate the coefficients for each class/category of flash-flood predictors. The torrential points sample was split into training (70%) and validating samples (30%). The modelling was done in Excel, SPSS and R software (H2O package), while the result mapping was performed in ArcGIS 10.5 software. The analysis revealed that the high and very high susceptibility degrees are spread over a maximum of 35.01% of the study area. The best performances, demonstrated by an AUC-ROC of 0.984, are associated with the Deep Learning Neural Network – Analytical Hierarchy Process model, followed by Naïve Bayes – Analytical Hierarchy Process model (AUC = 0.976), Multilayer Perceptron - Analytical Hierarchy Process model (AUC = 0.882) and Fuzzy-Analytical Hierarchy Process (AUC = 0.807). These results indicates that Deep Learning Neural Network is a promising machine learning model which can provide outcomes with very high precision. Also, according to the present research results the deep learning neural network, having many hidden layers, is able outperform the multilayer perceptron that contains a single hidden layer. The main novelty of the present research is the application of the three ensemble models (DLNN-AHP, MLP-AHP and NB-AHP) and also the use of H2O package for the first time in literature, to evaluate the flash-flood susceptibility in small river catchments.

A robust deep learning workflow to predict multiphase flow behavior during geological <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" altimg="si15.svg"><mml:mrow><mml:mi mathvariant="bold">C</mml:mi><mml:msub><mml:mi mathvariant="bold">O</mml:mi><mml:mn>2</mml:mn></mml:msub></mml:mrow></mml:math> sequestration injection and Post-Injection periods

Abstract: • A robust deep learning (DL) workflow has been developed to predict the long term process of geologic C O 2 sequestration. • Different sets of features fed to DL have been evaluated to obtain the most accurate prediction for pressure and saturation. • The DL workflow shows great efficiency with a speedup of 250 times compared to full reservoir simulation. Simulation of multiphase flow in porous media is essential to manage the geologic C O 2 sequestration (GCS) process, and physics-based simulation approaches usually take prohibitively high computational cost due to the nonlinearity of the coupled physics. This paper contributes to the development and evaluation of a deep learning workflow that accurately and efficiently predicts the temporal-spatial evolution of pressure and C O 2 plumes during injection and post-injection periods of GCS operations. Based on a Fourier Neural Operator, the deep learning workflow takes input variables or features including rock properties, well operational controls and time steps, and predicts the state variables of pressure and C O 2 saturation. To further improve the predictive fidelity, separate deep learning models are trained for C O 2 injection and post-injection periods due to the difference in primary driving force of fluid flow and transport during these two phases. We also explore different combinations of features to predict the state variables. We use a realistic example of C O 2 injection and storage in a 3D heterogeneous saline aquifer, and apply the deep learning workflow that is trained from physics-based simulation data and emulate the physics process. Through this numerical experiment, we demonstrate that using two separate deep learning models to distinguish post-injection from injection period generates the most accurate prediction of pressure, and a single deep learning model of the whole GCS process including the cumulative injection volume of C O 2 as a deep learning feature, leads to the most accurate prediction of C O 2 saturation. For the post-injection period, it is key to use cumulative C O 2 injection volume to inform the deep learning models about the total carbon storage when predicting either pressure or saturation. The deep learning workflow not only provides high predictive fidelity across temporal and spatial scales, but also offers a speedup of 250 times compared to full physics reservoir simulation, and thus will be a significant predictive tool for engineers to manage the long-term process of GCS.

Integrated variable weight model and improved DRASTIC model for groundwater vulnerability assessment in a shallow porous aquifer

Abstract: Accurate and effective assessment of groundwater vulnerability is very important for ensuring a healthy groundwater ecosystem. We evaluate the vulnerability of shallow porous aquifers by modifying the DRASTIC model and verify the effectiveness of the variable weight model (VWM). Firstly, topography, aquifer media, and the impact of the vadose zone are replaced by land-use type, aquifer thickness, and the hydraulic resistance of the vadose zone. Second, the weighting is optimized using the analytical hierarchy process (AHP) and variable weight theory (VWT). Thirdly, the original and the improved DRASTIC methods were used to evaluate groundwater vulnerability. Finally, three sets of samples hydrochemical parameters (NO3-, Mg2+,COD) were used to verify the improved frameworks using the Receiver Operating Curve (ROC) method. The improved model increases the dispersion degree of groundwater vulnerability. Compared with the original DRASTIC method, the correlation of the VWM is significantly improved in terms of the area under curve (AUC) for NO3- (0.786), COD (0.753), and Mg2+ (0.831). In short, it is necessary to optimize the parameters and weights of the model in order to realize reliable estimations of groundwater vulnerability. In particular, the use of VWM brings the results in line with the actual situation by changing the weights.

Hydrogeochemical characteristics and processes of groundwater in an over 2260 year irrigation district: A comparison between irrigated and nonirrigated areas

Abstract: A comprehensive understanding of groundwater chemistry and its evolution in irrigation districts is essential for irrigation management. In this study, with emphasis on a comparison between irrigated and nonirrigated areas, the groundwater chemistry and hydrogeochemical processes of a typical irrigation district with an irrigation history of over 2260 years were studied to clarify the effects of long-term irrigation on groundwater quality. Based on 107 water samples collected from across the study area, a comprehensive analysis was conducted using multivariate statistics, stable isotope analysis, and hydrogeochemical modeling. The results showed that the variation range and average concentrations of almost all the ions in the irrigation district are much greater than those in the nonirrigated areas. The groundwater in the nonirrigated areas could be characterized by low TDS and HCO3 or HCO3·SO4 types, whereas the groundwater in the irrigation district could be characterized by complex water types and high TDS. Stable isotopes of hydrogen and oxygen indicated that the groundwater in the irrigation district experienced strong evaporation. The calculated groundwater residence time showed that it takes approximately 2180 years for the groundwater to flow through the irrigation district. Some old irrigation water was present, and it influenced the current groundwater chemistry in the study area. The processes forming the current groundwater chemistry in the irrigation district can be formulated as: mixing → evaporation → water–rock interaction. For mixing, the proportions of rainfall, irrigation water from the Jing River, and lateral recharge from the nonirrigated areas were found to be approximately 30%, 49%, and 21%, respectively. For evaporation, the ratio based on the TDS was 3.83 when comparing the groundwater in the irrigation district with the mixed water after evaporation. Hydrogeochemical modeling showed that the irrigation area is a potential carbon source. The dissolution of halite and gypsum, the precipitation of calcite and dolomite, CO2 degassing and Na-Ca exchange are the main chemical reactions in the geochemical evolution of groundwater. The lesson learned from this irrigation district is that long-term irrigation will lead to salinization and complexity of the groundwater. Hence, in the water resource management of irrigation districts, attention should not only be paid to the balance of water quantity, but also to the balance of salts and the hydrochemical evolution of groundwater.

Stacked machine learning algorithms and bidirectional long short-term memory networks for multi-step ahead streamflow forecasting: A comparative study

Abstract: Prediction of river flow rates is an essential task for both flood protection and optimal water resource management. The high uncertainty associated with basin characteristics, hydrological processes, and climatic factors affecting river flows make streamflow prediction a very challenging problem. These reasons, together with the increasingly wide availability of data relating to flow rates and rainfall, frequently lead to a preference for data-driven models over physically based or conceptual forecasting models. This study shows the results of an in-depth comparison between two different daily streamflow prediction models: a novel simpler model based on the stacking of the Random Forest and Multilayer Perceptron algorithms, using the Elastic Net algorithm as meta-learner, and a more complex model based on bi-directional Long Short-Term Memory (LSTM) networks. Bayesian optimization was employed in selecting hyperparameters. The two prediction models were compared through the analysis of four different case studies: the Bacchiglione River, the Raccoon River, the Wilson River, and the Trent River. The two models showed comparable forecasting capabilities. In the most favourable cases, both models demonstrated very high accuracy (R2 greater than 0.93, Mean absolute percentage error approximately equal to 10%). The stacked model outperformed the bi-directional LSTM network model in several cases in predicting peak flow rates but was less accurate in forecasting low flow rates. In addition, its computation times are significantly shorter. The prediction accuracy of both models decreased as the forecast horizon increased. The length of the time series plays an essential role in developing models with satisfactory forecasting capabilities. In this study, the effectiveness of forecasting models was not influenced by the river regime. However, a high variance of the input dataset and a large number of outliers in the time series can reduce the accuracy of prediction models.

Global rainfall erosivity projections for 2050 and 2070

Abstract: • Global Rainfall erosivity increase on average 26.2–28.8% (2050) and 27–34.3% (2070). • Use of 19 climate models and three RCP scenarios (2.6, 4.5, 8.5) for 2050 and 2070. • 80–85% of global land surface will have an increasing trend in rainfall erosivity. • Large projected changes in rainfall erosivity are expected in the Northern Hemisphere. • Climate change and the increase in rainfall erosivity will drive high erosion rates. The erosive force of rainfall (rainfall erosivity) is a major driver of soil, nutrient losses worldwide and an important input for soil erosion assessments models. Here, we present a comprehensive set of future erosivity projections at a 30 arc-second (∼1 km 2 ) spatial scale using 19 downscaled General Circulation Models (GCMs) simulating three Representative Concentration Pathways (RCPs) for the periods 2041–2060 and 2061–2080. The future rainfall erosivity projections were obtained based on a Gaussian Process Regression (GPR) approach relating rainfall depth to rainfall erosivity through a series of (bio)climatic covariates. Compared to the 2010 Global Rainfall erosivity baseline, we estimate a potential average increase in global rainfall erosivity between 26.2 and 28.8% for 2050 and 27–34.3% for 2070. Therefore, climate change and the consequential increase in rainfall erosivity is the main driver of the projected + 30–66% increase in soil erosion rates by 2070. Our results were successfully compared with 20 regional studies addressing the rainfall erosivity projections. We release the whole dataset of future rainfall erosivity projections composed of 102 simulation scenarios, with the aim to support further research activities on soil erosion, soil conservation and climate change communities. We expect these datasets to address the needs of both the Earth system modeling community and policy makers. In addition, we introduce a modeling approach to estimate future erosivity and make further assessments at global and continental scales.

Directed Graph Deep Neural Network for Multi-step Daily Streamflow Forecasting

Abstract: Reliable and accurate multi-step streamflow forecasting is of vital importance for the utilization of water resources and hydropower energy system. In this paper, a spatial deep learning model, directed graph deep neural network, is proposed for multi-step streamflow forecasting. The proposed model uses spatial information capture process and feature aggregation process to exploit multi-site hydrological and meteorological information. The spatial information capture process consists of multiple convolutional layers to extract the precipitation information of meteorological stations. And the feature aggregation process uses the multi-layer perceptron to aggregate the precipitation information and the streamflow information. The proposed model is applied in a real-world case study in the upstream of Yangtze River basin. Experimental results demonstrate that the proposed model significantly outperforms artificial neural network, Long Short-Term Memory Network, Gated recurrent unit and Convolutional Neural Network in terms of forecasting accuracy. In addition to the forecast accuracy, the hidden Markov regression is employed to quantify the forecasting uncertainty given by the directed graph deep neural network. The uncertainty estimation result demonstrates that the hidden Markov regression is able to handle the heteroscedastic and non-normal forecasting uncertainty given by directed graph deep neural network.

Emerging organic contaminants in karst groundwater: A global level assessment

Abstract: A quarter of the world's population uses groundwater from karst aquifers. A range of emerging organic contaminants (EOCs) are considered a potential threat to water resources and dependant ecosystems, and karst aquifers are the most vulnerable groundwater systems to anthropogenic pollution. This paper provides the first global compilation (based on 50 studies) of EOCs in karst aquifers and explores EOC occurrence and the use of EOCs to understand karst systems. Of the 144 compounds detected in the reviewed studies, the vast majority in karst groundwater are pharmaceuticals and pesticides. Maximum concentrations of compounds varied over five orders of magnitude, and nearly half of the detected compounds exceed 100 ng/L. Karst groundwater is shown to have lower frequency of detection and lower concentrations compared to surface waters and local shallow intergranular aquifers, but overall higher concentrations compared to other major aquifer types. A growing number of studies have demonstrated the utility of EOCs and some legacy compounds for groundwater quality assessment and as tracers for characterising karst systems. They can improve understanding of vulnerability, storage, attenuation mechanisms, and in some cases have been used to assist with catchment delineation. This is a growing research area for karst hydrogeology, and more research is needed to understand EOC contamination of karst aquifers, and to develop EOCs as tracers within karst to improve our understanding of this critical water resource.

Future projection of seasonal drought characteristics using CMIP6 in the Lancang-Mekong River Basin

Abstract: As the most widespread natural hazard, drought has significant impacts on the livelihood and ecosystems in the Lancang-Mekong River Basin (LMRB). However, few studies focus on the seasonal characteristics of drought in the LMRB, especially under future climate projections by the Sixth Phase of Coupled Model Inter-comparison Project (CMIP6). This study filled the knowledge gap using SPI and SPEI, based on eight GCMs of CMIP6 under three scenarios (i.e., SSP1-2.6, SSP2-4.5, SSP5-8.5). Our results show that the LMRB tends to experience wetter wet season and drier dry season with the rising temperature considered (based on SPEI), while the temporal trend of dry-season SPI is not significant. The future trends of SPEI are -0.006 per year and -0.011 per year under SSP2-4.5 and SSP5-8.5, respectively. The trend magnitudes demonstrate spatial heterogeneities. Our evaluation based on SPEI shows that the most notable increases of dry-season drought (in terms of duration and intensity) are distributed in the middle reaches of LMRB. The upper Lancang and middle Mekong basin will likely experience more wet-season droughts. The dry-season drought accounts for 60% of total drought events in the near future (i.e., 2021–2055) and more than 80% in the far future (i.e., 2061–2095) under SSP2-4.5. Effective strategies are needed to enhance food and drinking water security in the LMRB, especially for the dry seasons under a changing climate.

Multiscale Petrophysical Characterization and Flow Unit Classification of the Minnelusa Eolian Sandstones

Abstract: Integration of petrophysical and geological information is critical to simulation of subsurface carbon storage (GCS). In this sense, two depositional facies were identified from the core description and well-log interpretation, namely massive (MS) and cross-bedded (CB) facies groups. Additionally, pore-scale characteristics were studied by a combination of techniques, e.g. Nuclear Magnetic Resonance (NMR) and mercury intrusion capillary pressure (MICP). Scanning electron microscope (SEM) and petrographic analyses show that the pore structure is dominantly controlled by the depositional environment and dolomite cementation. NMR-T2 distributions of MS and CB facies show triple and quadruple modes, respectively. In addition, MICP of high- and low-permeability MS facies samples, and their CB facies group mixtures were collected. The MS sample pore-throat size distribution is uni-modal, while the triple-modal characteristic of the mixtures indicates heterogeneous pore structures at the sub-core scale for CB facies. The reliably estimates of porosity and permeability for both facies groups via NMR techniques and the MLR (Multiple Linear Regression) approach demonstrate the applicability of these techniques to eolian sandstone. Moreover, irreducible water saturation via the T2-cutoff method correlates strongly with T2LM instead of porosity. Finally, the rock quality index and flow zone indicator were calculated based on Combinable Magnetic Resonance (CMR) log interpretations. This provides direct connection to properties measured in the well. Four flow units were classified for both facies groups. Results show that better reservoir quality with significant heterogeneities is observed in the CB facies. This study highlights the importance integrating multiscale petrophysical properties including facies, pore architecture and diagenesis analysis with core- to log-scale property characterization. The results herein validate our reservoir characterization and flow unit classification in eolian reservoirs.

Hydraulic modelling of inland urban flooding: recent advances

Abstract: • Research in urban flood modelling has received a major attention in last 4 years. • The topics cover flow process understanding and innovation in numerical modelling. • Many new experimental set-ups and numerical models were developed. • Innovative calculations aim at improving prediction quality or efficiency. • Challenges remain for an efficient and precise urban flood calculations. This review provides a synthesis of advances in our understanding of urban flood processes and their modelling over the last four years (2018–2021). Four aspects are covered: knowledge of urban flood flow and transport processes, stability of humans and objects within flooded streets, reliability of computational modelling and approaches for speeding-up computations of urban flood event. New laboratory setups have shed light on previously unexplored processes such as flow intrusion into buildings or contaminant exchanges between surface and underground drainage. The stability of a single pedestrians or object (e.g., vehicles, waste containers) under urban flooding was analysed, but not group effects such as clogging. Improvements in computations were achieved by new strategies for merging and processing various sources of high quality topographic and forcing data (e.g., precipitation), the incorporation of more and more details on the drainage systems (e.g., effect of gullies), and 3D instead of 2D simulations. Computational efficiency was enhanced based on massive parallelization, adaptive mesh, porosity models, surrogate models as well as machine learning. Finally crowd-sourced data are shown to offer an avenue for next generation model validation methods. Remaining knowledge gaps and guidance for future research are proposed and predict that additional research work will be performed in following years.

Development of a fine-resolution snow depth product based on the snow cover probability for the Tibetan Plateau: Validation and spatial–temporal analyses

Abstract: Accurate remotely sensed snow depth (SD) data are essential for monitoring and modeling hydrological processes in cold regions. While the available passive microwave SD data have been widely used by the community, the coarse spatial resolution (typically at 0.25°) of these data impedes the explicit representation of the hydrological processes in snow-dominated regions, especially in mountainous regions with complex terrain. To improve the spatial resolution and quality of passive microwave SD data for the Tibetan Plateau (TP), we develop a spatial–temporal downscaling method to produce a 19-year, daily 0.05° SD product by combining the existing high temporal resolution daily SD data and the high spatial resolution 8-day cloud-free Moderate Resolution Imaging Spectroradiometer (MODIS)-based snow cover probability (SCP) data, the latter of which were produced using an new advanced temporal filter algorithm. Validations against the observed SD data from 92 meteorological stations suggest that the newly-developed 0.05° SD product greatly improves upon the original 0.25° version. Based on this 0.05° SD product, we found that higher SD values are mainly distributed on the southeastern and eastern TP as well as the Himalaya and Karakoram, while much lower SD values occur on the inner TP. During 2000–2018, the TP-averaged annual SD showed a slight (p > 0.05) increasing trend because there were little changes in SD for most grids across the TP. Regarding different basins within TP, the annual SD during 2000–2018 slightly increased over most basins except for the Amu Dayra, Ganges, Brahmaputra, and Inner TP, where the basin-scale SD showed insignificant decreasing tendencies. In general, the spatial–temporal variations in the SD across the TP were very heterogeneous because SD was affected by multiple climatic factors. The newly-developed 0.05° SD product could facilitate our understanding of the hydrological processes on the TP through a more explicit representation of the gridded-based snow water information.

A Short-term Flood Prediction based on Spatial Deep Learning Network: A Case Study for Xi county, China

Abstract: • A Convolutional Long Short Term Memory Network is used to predict the flood events based on deep learning techniques. • The spatial and time characteristics of floods in China are well modeled to overcome the shortcomings generated by merely relying on time-series analysis. • Different from traditional methods, the hydrological area is gridded into different watersheds for future processing using image processing methods. Floods cause substantial damage across the world every year. Accurate and timely prediction of floods can significantly minimize the loss of life and property. Recently, numerous machine learning models have been used for flood prediction, showing that their performance is preferable to traditional statistical models. However, the existing models neglect the spatial features of floods, which drive flood generation and concentration. In this paper, the area of interest is divided into grids based on longitude and latitude, and the rainfall and discharge collected by stations are combined into tensors according to station coordinates. Different from one-dimensional time series, our input feature is a two-dimensional time series with spatial information. Hence, combining a Convolutional Neural Network (CNN) with a Long Short Term Memory Network (LSTM), we propose the convolution LSTM (ConvLSTM) to extract spatiotemporal features of hydrological information. The methodology is demonstrated using the hydrological data collected at the Xi County stations, located on the Huai River in Henan Province, China. Numerical results indicate that the relative error of arrival time is within 30%, and the relative error of peak discharge is within 20%, satisfying the 2005 Chinese Water Resource Standard on flood prediction permit error. The experiments also show that the ConvLSTM outperforms the recent models in terms of flood arrival time and peak discharge, thereby proving a promising alternative.

Experimental study on simultaneous heat-water-salt migration of bare soil subjected to evaporation

Abstract: Evaporation in shallow soil has a great influence on the properties of surface soil. Therefore, under evaporation conditions, research on the rule and mechanism of heat-water-salt migration in surface soil has great significance for engineering geology and environmental science. In this study, with full consideration of the initial water content, salt content, atmospheric temperature, and groundwater supply, a series of indoor evaporation tests were performed on shallow saline soil to observe the heat-water-salt migration process. In addition, before and after the evaporation test, the soil microstructure at different depths was observed and analysed to explore the influence of heat-water-salt migration on the soil. The results show that during the evaporation process, salt carried by water migrates in the soil column, and the water-vapour phase transition causes salt accumulation, which also leads to the electrical conductivity (EC) profile showing a V-shape, and the salt content in the middle of the soil column is lower than that in the end. In addition, high atmospheric temperature significantly reduces the surface water content but alleviates salt accumulation. Under the condition of groundwater supplementation, the salt accumulation in the soil column surface is rapid. By observing the microstructure, it can be seen that a flocculent structure exists in the soil before evaporation. After evaporation, the surface microstructure changes significantly, and pores and salt crystals can be observed. The results provide a reference for understanding heat-water-salt transport in arid areas.

Short-term flood probability density forecasting using a conceptual hydrological model with machine learning techniques

Abstract: 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.

An upscaling approach to predict mine water inflow from roof sandstone aquifers

Abstract: Underground coal mining suffers from groundwater intrusion from the aquifers overlying coal seams. Therefore, developing methods for the accurate prediction of roof water inflow is urgently needed to design a safe drainage system. In this study, we developed a novel upscaling framework to predict roof water inflow by integrating the multiscale hydrogeological properties of roof aquifers. In this framework, we imaged rock samples via scanning electron microscopy and performed pore-scale analysis based on fractal theory. A fractal model of permeability was introduced to calculate the seepage capacity of the pore structure in the samples. The effect of fractures was further evaluated via core-scale pneumatic experiments. Subsequently, we derived an upscaling formula of hydraulic conductivity used for predicting roof water inflow at the field scale. The proposed upscaling approach was demonstrated using data from a coal mine in Northern China. The results indicate that the actual water inflow (21 m3/h) is within the predicted range of our upscaling framework (9.32–92.78 m3/h), and the initial line fracture rate dx is distributed between 0.02 % and 0.03 %. Therefore, these findings can guide the development of methods for considering micropores and fractures simultaneously and scaling them up to the field scale for effective prediction of water inflow from roof aquifers.

Short-term runoff prediction using deep learning multi-dimensional ensemble method

Abstract: Recently, deep learning models have been widely used in water conservancy engineering forecasting problems, due to their excellent ability to deal with the complex interactions between various hydrological factors. However, the existing research mainly focus on model structure adjustment and input feature selection, ignoring the influence of model ensemble on prediction. In this paper, a novel multi-model ensemble method, namely deep learning multi-dimensional ensemble method, is proposed to deal with multi-step runoff prediction problem. The method, which can significantly improve the runoff prediction performance, couples two ensemble techniques with different functional dimensions, namely snapshot ensemble and attention ensemble. Snapshot ensemble technique is used to enhance model generalization capabilities from single-model dimension. While, attention ensemble technique is employed to increase model prediction accuracy from multi-model dimension. Furthermore, a novel data-driven model, called deep learning multi-dimensional ensemble model, is proposed by combining three different deep learning neural networks with deep learning multi-dimensional ensemble method. The proposed model is applied in a real-world case study in the upstream of Yangtze River basin. Three evaluation indicators and ten comparative models are used to test the model performance. The test results not only show the superiority of proposed model over other comparison models, but also prove the effectiveness of the deep learning multi-dimensional ensemble method. The study highlights the power of the ensemble of deep learning model and the promising prospect of our deep learning multi-dimensional ensemble method in hydrological predictions.

Boosted artificial intelligence model using improved alpha-guided grey wolf optimizer for groundwater level prediction: Comparative study and insight for federated learning technology

Abstract: Modeling groundwater level (GWL) is a challenging task particularly in intensive groundwater-based irrigated regions due to its dependency on multiple natural and anthropogenic factors. The main motivation of the current investigation is to develop a new advanced artificial intelligence (AI) model for GWL simulation. An Adaptive Neuro-Fuzzy Inference System (ANFIS) optimized by Improved Alpha-Guided Grey Wolf optimization (IA-GWO) algorithm is proposed in this study for reliable prediction of GWL in an intensively irrigated region of Northwest Bangladesh. Natural and anthropogenic factors including rainfall, evapotranspiration, groundwater abstraction, and irrigation return flow were considered as input variables for the development of the models. The efficacy of the proposed model was compared with standalone ANFIS and ANN models and their hybrid versions using particle swarm optimization (ANFIS-PSO) models. Both standard statistical metrics and visual inspection of scatter plots, violin plots, and Taylor diagrams were employed for performance evaluation. Thirty-one years (1981–2011) monthly groundwater level data were used for the calibration and validation of the models. The results revealed the better performance of ANFIS-IA-GWO with normalized root mean square error (NRMSE) of 0.06–0.11 and Kling-Gupta efficiency (KGE) of 0.96–0.98 compared to ANFIS-PSO (NRMSE ∼ 0.38–0.55 and KGE ∼ 0.70–0.86) and ANN-IA-GWO (NRMSE ∼ 0.42–0.57 and KGE ∼ 0.75–0.91) and ANN-PSO (NRMSE ∼ 0.50–0.63 and KGE ∼ 0.63–0.83). The visual comparison of results showed that ANFIS-IA-GWO model was able to replicate the mean, distribution, interquartile range, and standard deviation of observed GWL more appropriately compared to other models

Vegetation restoration improves soil hydrological properties by regulating soil physicochemical properties in the Loess Plateau, China

Abstract: Soil hydrological properties play a key role in soil hydrological processes. However, the effect of long-term vegetation restoration on soil hydrological properties and the corresponding influencing mechanisms remains poorly understood. Here, three soil hydrological properties including saturated water-holding capacity (SWHC), field capacity (FC) and saturated hydraulic conductivity (Ks), as well as several basic soil properties in the Zhifanggou watershed of the Loess plateau were investigated. The variations in SWHC, FC and Ks under different vegetation restoration types and their dominant influencing factors were analyzed. Moreover, we collected available Ks data from peer-reviewed publications to determine the land use with the largest Ks across the entire Loess Plateau. The results showed that SWHC FC and Ks were increased after 20 years of vegetation restoration. The higher Ks was found in shrubland and forest in the whole Loess Plateau. Compared with cropland, Ks in shrubland was increased by 87.10% at 0–20 cm, 48.89% at 20–40 cm, and 18.37% at 40–100 cm, respectively, indicating that the impact of revegetation on Ks were most obvious in the upper soil layer. Bulk density (BD), total porosity (TP), capillary porosity (CP), noncapillary porosity (NCP) and soil organic matter (SOM) had a significant effect on SWHC, FC and Ks for different land-use types (P < 0.01). Soil porosity (i.e., TP, CP and NCP) and BD, soil chemical properties (i.e., SOM and pH), and soil particle composition explained 93.8%, 59.2%, and 13.4% of the total variance in soil hydrological properties (i.e., SWHC, FC and Ks), respectively. This indicates that soil porosity and BD are the dominant factors affecting soil hydrological properties. Moreover, soil particle composition played an important role in regulating Ks, with the contribution of 38.6%. The established pedotransfer function (PTF) of Ks using BD, clay and silt content had a better performance than two existing PTFs. This research provides a more systematic and comprehensive understanding of the soil hydrological effect of vegetation restoration in the Loess Plateau.

Short-term Rainfall Forecasting using Machine Learning-Based Approaches of PSO-SVR, LSTM and CNN

Abstract: Short-term rainfall forecasting plays an important role in hydrologic modeling and water resource management problems such as flood warning and real time control of urban drainage systems. This paper compares the performances of three machine and deep learning-based rainfall forecasting approaches including a hybrid optimized-by-PSO support vector regression (PSO-SVR), long-short term memory (LSTM), and convolutional neural network (CNN). The approaches are used to develop both 5-minute and 15–minute ahead forecast models of rainfall depth based on datasets of Niavaran station, Tehran, Iran. Results of applying the models to all data points indicated that PSO-SVR and LSTM approaches performed almost the same and better than CNN. Subsequently, rainfall events were divided into four classes depending on their severity and duration using K-nearest neighbor method, and a separate forecast model was built for each of the classes. Classification of the events improved the forecast models accuracy where PSO-SVR and LSTM were the best approaches for the 15-minute and 5-minute ahead rainfall forecast models, respectively. Investigating the impact of more predictors on the forecast quality, adding differences of rainfall depths to model predictors improved the accuracy of PSO-SVR approach for the 5-minute ahead forecast model up to 13%. Furthermore, depending on the rainfall event, additional input variables considering rainfall depth fluctuations over shorter time periods than the forecast lead time increased the performances of the PSO-SVR and LSTM approaches between 3–15% and 2–10%, respectively.