Search

Accurate real-time flood predictions play a vital role in flood early warning systems, which further helps in mitigating the damage and saving lives. Error correction using machine learning (ML) in physics-based models (alternatively known as physicallybased models) has been widely considered and recommended in the literature to improve forecast accuracy. This study mainly focuses on evaluating the ability of novel tree-based ML methods and Bidirectional LSTM (BLSTM) at different lead times and high flow conditions. Also, the performance of these methods is compared with the traditionally used autoregression (AR), Multilayer perceptron (MLP), and naïve models. So overall, we evaluated six data-driven models and one naïve model on Storå river to correct the errors in the physics-based model: Two tree-boosting ML models (XGBoost, Gradient boosting), two deep learning-based models (MLP, BLSTM), and then simple models like autoregression (AR) & persistence (or naïve). Then, a stacked model combining XGBoost, and AR is developed and tested. Hyperparameter tuning is performed using Bayesian optimization. Results on the independent test set show that all the methods can improve the discharge simulations from a physics-based model. However, the Bidirectional LSTM and stacked model are consistently performed slightly better than other models in all lead times. At shorter lead times, tree-boosting approaches marginally underperformed. While gradient boosting performed better at longer lead times and produced results comparable to BLSTM and stacked models, XGBoost continues to underperform but gave better results than AR and PERS & MLP. The BLSTM and stacked models performed well under high flow conditions as well. Even though the difference is minor, they consistently outperformed all the other models. Furthermore, while tree-based methods (XGBoost & gradient boosting) fared somewhat worse than BLSTM & stacked model, they outperformed basic methods (AR/Pers) and MLP at high f

Linking biogeochemical processes to water flow paths and solute travel times is important for understanding internal catchment functioning and control of water quality. Base cation weathering is a process closely linked to key factors affecting catchment functioning, including water pathways, soil contact time, and catchment characteristics, particularly in silicate-dominated areas. However, common process-based weathering models are often calibrated and applied for individual soil profiles, which can cause problems when trying to extrapolate results to catchment scale and assess consequences for stream water and groundwater quality. Therefore, in this work, base cation export was instead modelled using a fully calibrated 3D hydrological model (Mike SHE) of a boreal catchment, which was expanded by adding a relatively simple but still reasonably flexible and versatile weathering module including the base cations Na, K, Mg, and Ca. The results were evaluated using a comprehensive dataset of water chemistry from groundwater and stream water in 14 nested sub-catchments, representing different catchment sizes and catchment characteristics. The strongest correlations with annual and seasonal observations were found for Ca (r = 0.89-0.93, p < 0.05), Mg (r = 0.90-0.95, p < 0.05), and Na (r = 0.80-0.89, p < 0.05). These strong correlations suggest that catchment hydrology and landscape properties primarily control weathering rates and stream dynamics of these solutes. Furthermore, catchment export of Mg, Ca, and K was strongly connected to travel times of discharging stream water (r = 0.78-0.83). Conversely, increasing Na export was linked to a reduced areal proportion of mires (r = -0.79). The results suggest that a significant part (~45%) of the catchment stream export came from deep-soil weathering sources (>2.5 m). These results have implications for terrestrial and aquatic water quality assessments. If deep soils are present, focusing mainly on the shall

The impact of precise river flood forecasting and warnings in preventing potential victims along with promoting awareness and easing evacuation is realized in the reduction of flood damage and avoidance of loss of life. Machine learning models have been used widely in flood forecasting through discharge. However the usage of discharge can be inconvenient in terms of issuing a warning since discharge is not the direct measure for the early warning system. This paper focuses on water level prediction on the Storå River, Denmark utilizing several machine learning models. The study revealed that the transformation of features to follow a Gaussian-like distribution did not improve the prediction accuracy further. Additional data through different feature sets resulted in increased prediction performance of the machine learning models. Using a hybrid method for the feature selection improved the prediction performance as well. The Feed-Forward Neural Network gave the lowest mean absolute error and highest coefficient of determination value. The results indicated the difference in prediction performance in terms of mean absolute error term between the Feed-Forward Neural Network and the Multiple Linear Regression model was 0.003 cm. It was concluded that the Multiple Linear Regression model would be a good alternative when time, resources, or expert knowledge is limited.

A groundwater plume containing high concentrations of pharmaceutical compounds, mainly sulfonamides, barbiturates and ethyl urethane, in addition to chlorinated ethenes and benzene was investigated. The contamination originating from a former pharmaceutical industry discharges into a multilayered aquifer system and a downgradient stream. In this study, geological and hydrogeological data were integrated into a numerical flow model to examine identified trends using statistical approaches, including principle component analysis and hierarchal cluster analysis. A joint interpretation of the groundwater flow paths and contaminant concentrations in the different compartments (i.e. groundwater and hyporheic zone) provided insight on the transport processes of the different contaminant plumes to the stream. The analysis of historical groundwater concentrations of pharmaceutical compounds at the site suggested these compounds are slowly degrading The pharmaceutical compounds migrate in both a deep semi-confined aquifer, as well as in the shallow unconfined aquifer, and enter the stream along a 2 km stretch. This contrasted with the chlorinated ethenes, which mainly discharge to the stream as a focused plume from the unconfined aquifer. The integrated approach developed here, combining groundwater flow modelling and statistical analyses of the contaminant concentration data collected in groundwater and the hyporheic zone, lead to an improved understanding of the observed distribution of contaminants in the unconfined and semi-confined aquifers, and thus to their discharge to the stream. This approach is particularly relevant for large and long-lasting contaminant sources and plumes, such as abandoned landfills and industrial production sites, where field investigations may be very expensive.

Contaminant mass discharge is a commonly applied tool to evaluate the environmental impact of contaminated sites on water resources. At large contaminated sites with heterogeneous sources, such as landfills, the number of wells available is often not sufficient, leading to a high uncertainty of mass discharge estimates. In this study, we tackle the uncertainty of the contaminant mass discharge due to low sampling densities by interpolating limited water‐sample data with the support of surface Direct Current resistivity and Induced Polarization geophysical data. The method relies on finding a conceptual link between the bulk conductivity imaged from geophysics and the contaminant concentrations. We investigate the link between: 1) imaged bulk and electrical water conductivity, 2) water conductivity and conservative ionic species, 3) water conductivity and redox‐sensitive species, 4) water conductivity and semi‐persistent organic species, 5) water conductivity and biodegradable organic compounds. The method successfully identify similarities between the distribution of the bulk conductivity and chloride and pharmaceutical compounds in a landfill leachate plume and between the bulk conductivity data and benzene and chlorinated ethenes for a contaminant plume from a former pharmaceutical factory. Contaminant concentrations were interpolated through regression kriging, using geophysical data as the dependent variable. The distribution of concentration determined with the novel method showed a lower mean relative estimation error than the traditional method of kriging only contaminant concentration data. At large sites, the method can improve contaminant mass discharge estimates, especially if surface geophysical measurements are integrated in the site investigation at an early stage.

Det danske samfund bruger hvert år mange midler på at undersøge, risikovurdere og oprense forurenede grunde. Det er derfor vigtigt, at vi konstant forbedrer de undersøgelser, der ligger til grund for, hvorledes vi risikovurderer og prioriterer de mange forurenede grunde til oprensning.

At contaminated sites, knowledge about geology and hydraulic properties of the subsurface and extent of the contamination is needed for assessing the risk and for designing potential site remediation. In this study, we have developed a new approach for characterizing contaminated sites through time-domain spectral induced polarization. The new approach is based on: (1) spectral inversion of the induced polarization data through a reparametrization of the Cole–Cole model, which disentangles the electrolytic bulk conductivity from the surface conductivity for delineating the contamination plume; (2) estimation of hydraulic permeability directly from the inverted parameters using a laboratory-derived empirical equation without any calibration; (3) the use of the geophysical imaging results for supporting the geological modelling and planning of drilling campaigns. The new approach was tested on a data set from the Grindsted stream (Denmark), where contaminated groundwater from a factory site discharges to the stream. Two overlapping areas were covered with seven parallel 2-D profiles each, one large area of 410 m × 90 m (5 m electrode spacing) and one detailed area of 126 m × 42 m (2 m electrode spacing). The geophysical results were complemented and validated by an extensive set of hydrologic and geologic information, including 94 estimates of hydraulic permeability obtained from slug tests and grain size analyses, 89 measurements of water electrical conductivity in groundwater, and four geological logs. On average the IP-derived and measured permeability values agreed within one order of magnitude, except for those close to boundaries between lithological layers (e.g. between sand and clay), where mismatches occurred due to the lack of vertical resolution in the geophysical imaging. An average formation factor was estimated from the correlation between the imaged bulk conductivity values and the water conductivity values measured in groundwater, in order to convert t

In this study we present the prediction of permeability from time‐domain spectral induced polarization (IP) data, measured in boreholes on undisturbed formations using the El‐log logging‐while‐drilling technique. We collected El‐log data and hydraulic properties on unconsolidated Quaternary and Miocene deposits in boreholes at three locations at a field site in Denmark, characterized by different electrical water conductivity and chemistry. The high vertical resolution of the El‐log technique matches the lithological variability at the site, minimizing ambiguity in the interpretation originating from resolution issues. The permeability values were computed from IP data using a laboratory‐derived empirical relationship presented in a recent study for saturated unconsolidated sediments, without any further calibration. A very good correlation, within one order of magnitude, was found between the IP‐derived permeability estimates and those derived using grain size analyses and slug‐tests, with similar depth‐trends and permeability contrasts. Furthermore, the effect of water conductivity on the IP‐derived permeability estimations was found negligible in comparison to the permeability uncertainties estimated from the inversion and the laboratory‐derived empirical relationship.

Leaching of organic and inorganic contamination from landfills is a serious environmental problem as surface water and aquifers are affected. In order to assess these risks and investigate the migration of leachate from the landfill, 2D and large scale 3D electrical resistivity tomography were used at a heavily contaminated landfill in Grindsted, Denmark. The inverted 2D profiles describe both the variations along the groundwater flow as well as the plume extension across the flow directions. The 3D inversion model shows the variability in the low resistivity anomaly pattern corresponding to differences in the ionic strength of the landfill leachate. Chemical data from boreholes agree well with the observations indicating a leachate plume which gradually sinks and increases in size while migrating from the landfill in the groundwater flow direction. Overall results show that the resistivity method has been very successful in delineating the landfill leachate plume and that good correlation exists between the resistivity model and leachate ionic strength.

Contaminants such as chlorinated solvents and pesticides, as well as new classes of compounds or emerging micropollutants are extensively produced, utilized and then discarded in society and subsequently released to streams from multiple point and diffuse sources. Sustainable management of water resources requires assessment of multiple contamination sources within a watershed in order to assess their direct impact on water quality. Determination of flow paths and groundwater fluxes are essential for evaluating the transport, fate and potential impact of contaminant plumes discharging to streams. This implies that investigators have the tools to evaluate the governing parameters, including an appreciation of the scale of variability, as well as conceptual and numerical models that incorporate the various mechanisms affecting flow and transport. A major multidisciplinary field scale investigation of the Grindsted stream area including geology, hydrogeology, geophysics, environmental chemistry, ecology and environmental engineering was carried out in 2012-2017, to develop the scientific basis for conducting risk assessments for contaminated sites impacting surface waters. The Grindsted stream area is a well-studied site, affected by many polluting sources including the plume from a former pharmaceutical factory. Our overall aim of the field investigations was to (i) test the applicability of different methods for mapping groundwater pollution as it enters streams at a complex site, and (ii) perform a source identification and risk assessment of the stream’s chemical and ecological status. The study included development of a geological and hydrogeological model, numerical modeling of the flow and transport, mapping of the contaminant plume, and detailed field investigations at the main entry point of the plume. We quantified the contaminant mass discharge and attenuation of the plume at the groundwater-surface interface by different approaches (contro

Geophysical methods are increasingly being used in contaminant hydrogeology to map lithology, hydraulic properties, and contaminant plumes with a high ionic strength. Advances in the Direct Current resistivity and Induced Polarization (DCIP) method allow the collection of high resolution three dimensional (3D) data sets. The DC resistivity can describe both soil properties and the water electrical conductivity, while the IP can describe the lithology and give information on hydrogeological properties. The aim of the study was to investigate a large contaminant plume discharging to a stream from an old factory site by combining traditional geological, hydrological, and contaminant concentration data with DCIP surveys. The plume consisted of xenobiotic organic compounds and inorganics. The study assesses benefits and limitations of DCIP geophysics for contaminated site investigations. A 3D geological model was developed from borehole logs and DCIP data as framework for the complex transport pathways near the meandering stream. IP data were useful in indicating the continuity and the changes in thickness of local clay layers between the borehole logs. The geological model was employed to develop a groundwater flow model describing groundwater flows to the stream. The hydraulic conductivity distribution was based on IP data, slug tests and grain size analysis. The distribution of contaminant concentrations revealed two chemically distinct plumes, separated by a clay layer, with different transport paths to the stream. The DC resistivity was useful in mapping ionic compounds, but also organic compounds whose spatial distribution coincided with the ionic compounds. A conceptual model describing the contaminant plume was developed, and it matched well with contaminant concentrations in stream water and below the streambed. Surface DCIP surveys supported the characterization of the spatial variability in geology, hydraulic conductivity and contaminant concentr

Domain transformation methods are useful techniques for solving problems on non-stationary domains. In this work, we consider the evolution of the water table in an unconfined aquifer. This nonlinear, time-dependent problem is greatly simplified by using a mapping from the physical domain to a reference domain and is then further reduced to a single, (nonlinear) partial differential equation. We show well-posedness of the approach and propose a stable and convergent discretization scheme. Numerical results are presented supporting the theory.

Contaminants in leachate plumes from landfills and other contaminated sites are a threat to the environment. Efficient site characterization methods are needed. The perspectives of the IP method are investigated in combination with geological sampling and chemical analyses of water samples. Along a leachate plume from a landfill hosting both household and chemical waste, borehole IP data, geological samples, grain size, and contaminant concentrations in water samples are examined for correlations related to geology and concentrations of contaminants. Results relating the Cole-Cole parameters with sediment types and pore water resistivity representing the concentrations of the contaminants show that the formation resistivity primarily is controlled by the contaminant concentrations while the IP parameters primarily are related to the clay content and grain size distribution of sandy sediments at the site.

Several mathematical methods for modelling free surface aquifers are available. Aquifer-stream interaction is an important application of these models, and are challenging to simulate because stream interaction is described by a highly variable head boundary, which can cause numerical instabilities and errors. In addition, when streams are sinuous, groundwater flow is truly 3-dimensional, with strong vertical flows and sharp changes in horizontal direction. Here 3 different approaches to simulating free surface aquifers are compared for simulating groundwater-stream interaction. The aim of the models was to investigate the effect of meander bends on the spatial and temporal variability of aquifer-stream interaction, and to develop a new 3D conceptual model of groundwater-stream interaction. Three mathematical methods were tested, representing the three main methods available for modeling 3D unconfined aquifers: a saturated-unsaturated flow model, moving mesh, and a new coordinate transformation. The saturated/unsaturated model couples the saturated groundwater flow equation with a solution of Richards equation. The moving mesh solves the saturated groundwater equation with a free surface and deformable numerical finite element mesh. Finally, the new coordinate transform method employs a coordinate transform so that the saturated groundwater flow equation is solved on a fixed finite element mesh with a stationary free surface. This paper describes in detail the new coordinate transform method. It employs a transformation of the vertical coordinate, so that the top surface remains stationary. The transformation introduces non-linearities into the saturated groundwater flow equation, with the hydraulic conductivity becoming a function of the head at the top boundary. Mathematical analysis is then applied to show well posedness, and provide stability and linear convergence results. Numerical results confirm the mathematical analysis. The three methods we

It is important to understand the fate and transport of contaminants in limestone aquifers because they are a major drinking water resource. This is challenging because they are highly heterogeneous; with micro-porous grains, flint inclusions, and being heavily fractured. Several modeling approaches have been developed to describe contaminant transport in fractured media, such as the discrete fracture (with various fracture geometries), equivalent porous media (with and without anisotropy), and dual porosity models. However, these modeling concepts are not well tested for limestone geologies. Given available field data and model purpose, this paper therefore aims to develop, examine and compare modeling approaches for transport of contaminants in fractured limestone aquifers. The model comparison was conducted for a contaminated site in Denmark, where a plume of a dissolved contaminant (PCE) has migrated through a fractured limestone aquifer. Multilevel monitoring wells have been installed at the site and available data includes information on spill history, extent of contamination, geology and hydrogeology. To describe the geology and fracture network, data from borehole logs was combined with an analysis of heterogeneities and fractures from a nearby excavation (analog site). Methods for translating the geological information and fracture mapping into each of the model concepts were examined. Each model was compared with available field data, considering both model fit and measures of model suitability. An analysis of model parameter identifiability and sensitivity is presented. Results show that there is considerable difference between modeling approaches, and that it is important to identify the right one for the actual scale and model purpose. A challenge in the use of field data is the determination of relevant hydraulic properties and interpretation of aqueous and solid phase contaminant concentration sampling data. Traditional water sampl