Spatiotemporal estimation of groundwater and surface water conditions by integrating deep learning and physics-based watershed models.

The impacts of climate change on hydrology underscore the urgency of understanding watershed hydrological patterns for sustainable water resource management. The conventional physics-based fully distributed hydrological models are limited due to computational demands, particularly in the case of large-scale watersheds. Deep learning (DL) offers a promising solution for handling large datasets and extracting intricate data relationships. Here, we propose a DL modeling framework, incorporating convolutional neural networks (CNNs) to efficiently replicate physics-based model outputs at high spatial resolution. The goal was to estimate groundwater head and surface water depth in the Sabgyo Stream Watershed, South Korea. The model datasets consisted of input variables, including elevation, land cover, soil type, evapotranspiration, rainfall, and initial hydrological conditions. The initial conditions and target data were obtained from the fully distributed hydrological model HydroGeoSphere (HGS), whereas the other inputs were actual measurements in the field. By optimizing the training sample size, input design, CNN structure, and hyperparameters, we found that CNNs with residual architectures (ResNets) yielded superior performance. The optimal DL model reduces computation time by 45 times compared to the HGS model for monthly hydrological estimations over five years (RMSE 2.35 and 0.29 m for groundwater and surface water, respectively). In addition, our DL framework explored the predictive capabilities of hydrological responses to future climate scenarios. Although the proposed model is cost-effective for hydrological simulations, further enhancements are needed to improve the accuracy of long-term predictions. Ultimately, the proposed DL framework has the potential to facilitate decision-making, particularly in large-scale and complex watersheds.

Assessment of hydraulic and thermal properties of the Antarctic active layer: Insights from laboratory column experiments and inverse modeling.

To better understand the changes in the hydrologic cycle caused by global warming in Antarctica, it is crucial to improve our understanding of the groundwater flow system, which has received less attention despite its significance. Both hydraulic and thermal properties of the active layer, through which groundwater can flow during thawing seasons, are essential to quantify the groundwater flow system. However, there has been insufficient information on the Antarctic active layer. The goal of this study was to estimate the hydraulic and thermal properties of Antarctic soils through laboratory column experiments and inverse modeling. The column experiments were conducted with sediments collected from two lakes in the Barton Peninsula, Antarctica. A sand column was also operated for comparison. Inverse modeling using HydroGeoSphere (HGS) combined with Parameter ESTimation (PEST) was performed with data collected from the column experiments, including permeameter tests, saturation-drain tests, and freeze-thaw tests. Hydraulic parameters (i.e., Ks, θs, Swr, α, ß, and Ss) and thermal diffusivity (D) of the soils were derived from water retention curves and temperature curves with depth, respectively. The hydraulic properties of the Antarctic soil samples, estimated through inverse modeling, were 1.6 × 10-5-3.4 × 10-4 cm s-1 for Ks, 0.37-0.42 for θs, 6.62 × 10-3-1.05 × 10-2 for Swr, 0.53-0.58 cm-1 for α, 5.75-7.96 for ß, and 5.11 × 10-5-9.02 × 10-5 cm-1 for Ss. The thermal diffusivities for the soils were estimated to be 0.65-4.64 cm2 min-1. The soil hydraulic and thermal properties reflected the physical and ecological characteristics of their lake environments. The results of this study can provide a basis for groundwater-surface water interaction in polar regions, which is governed by variably-saturated flow and freeze-thaw processes.

Effects of soil heterogeneity and preferential flow on the water flow and isotope transport in an experimental hillslope.

Soil water movement plays vital roles in hillslope runoff generation and groundwater and surface water interaction. However, there are still knowledge gaps about the impacts of soil heterogeneity and preferential flow on the internal water flow and transport process. In this study, the vertical soil heterogeneity focused on the variations in soil retention capacity, and the consideration of lateral preferential flow emphasized the higher hydraulic conductivity. We combined isotopic tracing and numerical modeling in an artificial hillslope, focusing on monitored processes of the artificial rainfall and isotopic tracing experiment. The results showed that the soil moisture quickly accumulated at the bottom of the hillslope during rainfall events, while the 2H enrichment occurred in the topsoil derived from enriched isotope injection in the second artificial rainfall. The evaporation process slowed down the mixing of new water in the topsoil and old water in the lower layer. We found that the vertical soil heterogeneity had significant influences on the internal water and isotope transport paths within the hillslope. The lateral preferential flow played an important role in the water flux and transport time to the seepage face. The coupling of isotopic tracing, which reflects the water transport and mixing with the hillslope, effectively improved the model simulation and mechanism analysis of hillslope water flow. Our findings provide new insights into the mechanisms governing soil water flow and transport dynamics in hillslopes, taking into account vertical soil heterogeneity and lateral preferential flow.

Evaluating backward probability model under various hydrogeologic and hydrologic conditions.

Contaminant source identification improves the understanding of contaminant source characteristics including location and release time, which can lead to more effective remediation and water resources management plans. The backward probability model can provide probabilities of source locations and release times under various contaminant properties and hydrogeologic conditions. The backward probability model has been applied to numerous synthetic and real contamination sites for locating possible contaminant sources, but it is also important to evaluate the reliability of the backward probability model through rigorous verification analyses. Here, we present a model verification framework for the backward probability model using a stepwise approach from simple to complex model settings: comparison with previous studies, transient saturated flow under various hydrogeologic conditions, and transient variably-saturated flow conditions. As a simple condition, one-dimensional homogeneous problems under steady-state and transient flow conditions were verified by comparing with previous studies. Model verifications with complex conditions were conducted by comparing forward and backward probability simulation results. The verification results demonstrate that the backward probability model performs well for homogeneous problems. For heterogeneous problems, the backward probability model results in slightly different backward travel times due to differences in solute decay and boundary conditions assigned for both forward and backward probability simulations, but the backward travel time at the maximum probability can be reproduced well.

Estimating cumulative wastewater treatment plant discharge influences on acesulfame and Escherichia coli in a highly impacted watershed with a fully-integrated modelling approach.

Wastewater treatment plant (WWTP) discharge is often considered a principal source of surface water contamination. In this study, a three-dimensional fully-integrated groundwater-surface water model was used to simulate the transport characteristics and cumulative loading of an artificial sweetener (acesulfame) and fecal indicator bacteria (Escherichia coli) from WWTPs within a 6800 km2 mixed-use, highly impacted watershed in Ontario, Canada. The model, which employed 3.5 × 106 computational nodes and 15 layers, facilitated a comprehensive assessment of groundwater-surface water interactions under high and low flow conditions; processes typically not accounted for in WWTP cumulative effects models. Simulations demonstrate that the model had significant capacity in reproducing the average and transient multi-year groundwater and surface water flow conditions in the watershed. As a proxy human-specific conservative tracer, acesulfame was useful for model validation and to help inform the representation of watershed-scale transport processes. Using a uniform WWTP acesulfame loading rate of 7.14 mg person-1 day-1, the general spatial trends and magnitudes of the acesulfame concentration profile along the main river reach within the watershed were reproduced; however, model performance was improved by tuning individual WWTP loading rates. Although instream dilution and groundwater-surface water interactions were strongly dependent on flow conditions, the main reach primarily consisted of groundwater discharge zones. For this reason, hydrodynamic dispersion in the hyporheic zone is shown as the predominant mechanism driving acesulfame into near-stream shallow groundwater, while under high flow conditions, the simulations demonstrate the potential for advective flushing of the shallow groundwater. Regarding the cumulative impact of the WWTPs on E. coli concentrations in the surface flow system, simulated transient E. coli levels downstream of WWTPs in the watershed were significantly lower than observed values, thus highlighting the potential importance of other sources of E. coli in the watershed.

Natural Stimuli Calibration with Fining Direction Regularization in an Integrated Hydrologic Model.

The interaction between surface water and groundwater during flood events is a complex process that has traditionally been described using simplified analytical solutions, or abstracted numerical models. To make the problem tractable, it is common to idealize the flood event, simplify river channel geometry, and ignore bank soil heterogeneity, often resulting in a model that only loosely represents the site, thus limiting its applicability to any specific river cross-section. In this study, we calibrate a site-specific fully-integrated surface and subsurface HydroGeoSphere model using flood events for a cross-section along the South River near Waynesboro, VA. The calibration approach presented in this study demonstrates the incorporation of fining direction regularization with a highly parameterized inversion driven by natural stimuli, to develop several realistic realizations of hydraulic conductivity fields that reflect the depositional history of the system. Specifically, we calibrate a model with 365 unique material zones to multiple flood events recorded in a dense well network while incorporating possible fining sequences consistent with the depositional history of the riverbank. Over 25,000 individual simulations were completed using calibration software and a cloud platform specifically designed for highly parallelized computing environments. The results of this study demonstrate the use of fining direction regularization during model calibration to generate multiple calibrated model realizations that account for the depositional environment of the system.

Determination of rate constants and branching ratios for TCE degradation by zero-valent iron using a chain decay multispecies model.

The applicability of a newly-developed chain-decay multispecies model (CMM) was validated by obtaining kinetic rate constants and branching ratios along the reaction pathways of trichloroethene (TCE) reduction by zero-valent iron (ZVI) from column experiments. Changes in rate constants and branching ratios for individual reactions for degradation products over time for two columns under different geochemical conditions were examined to provide ranges of those parameters expected over the long-term. As compared to the column receiving deionized water, the column receiving dissolved CaCO3 showed higher mean degradation rates for TCE and all of its degradation products. However, the column experienced faster reactivity loss toward TCE degradation due to precipitation of secondary carbonate minerals, as indicated by a higher value for the ratio of maximum to minimum TCE degradation rate observed over time. From the calculated branching ratios, it was found that TCE and cis-dichloroethene (cis-DCE) were dominantly dechlorinated to chloroacetylene and acetylene, respectively, through reductive elimination for both columns. The CMM model, validated by the column test data in this study, provides a convenient tool to determine simultaneously the critical design parameters for permeable reactive barriers and natural attenuation such as rate constants and branching ratios.

A semi-analytical solution for simulating contaminant transport subject to chain-decay reactions.

We present a set of new, semi-analytical solutions to simulate three-dimensional contaminant transport subject to first-order chain-decay reactions. The aquifer is assumed to be areally semi-infinite, but finite in thickness. The analytical solution can treat the transformation of contaminants into daughter products, leading to decay chains consisting of multiple contaminant species and various reaction pathways. The solution in its current form is capable of accounting for up to seven species and four decay levels. The complex pathways are represented by means of first-order decay and production terms, while branching ratios account for decay stoichiometry. Besides advection, dispersion, bio-chemical or radioactive decay and daughter product formation, the model also accounts for sorption of contaminants on the aquifer solid phase with each species having a different retardation factor. First-type contaminant boundary conditions are utilized at the source (x=0 m) and can be either constant-in-time for each species, or the concentration can be allowed to undergo first-order decay. The solutions are obtained by exponential Fourier, Fourier cosine and Laplace transforms. Limiting forms of the solutions can be obtained in closed form, but we evaluate the general solutions by numerically inverting the analytical solutions in exponential Fourier and Laplace transform spaces. Various cases are generated and the solutions are verified against the HydroGeoSphere numerical model.