Search

Floods affect communities and ecosystems worldwide, emphasizing the importance of identifying their precursors and enhancing resilience to these events. Here, we calculated Antecedent Total Water Storage (ATWS) anomalies from the new 5-day (5D) Gravity Recovery and Climate Experiment (GRACE) and its Follow-On (GRACE-FO) satellite solutions to enhance the detection of pre-flood and active flood conditions and to map post-flood storage anomalies. The GRACE data were compared with ~ 3300 flood events reported by the Dartmouth Flood Observatory (2002–2021), revealing distinct ATWS precursor signals in 5D solutions, in contrast to the monthly solutions. Specifically, floods caused by saturation-excess runoff—triggered by persistent rainfall, monsoonal patterns, snowmelt, or rain-on-snow events—show detectable ATWS increases 15 to 50 days before and during floods, providing a valuable opportunity to improve flood monitoring. These 5D solutions also facilitate a more rapid mapping of post-flood storage changes to assess flood recovery from tropical cyclones and sub-monthly weather extremes. Our findings show the promising potential of 5D GRACE solutions, which are still in the development phase, for future integration into operational frameworks to enhance flood detection and recovery, facilitating the rapid analysis of storage changes relative to monthly solutions. [ABSTRACT FROM AUTHOR]

Physical agricultural drought indices generally use soil moisture to represent root‐zone water availability (RZWA). The uncertainty in root‐zone properties, especially in deep‐rooting regions, may lead to significant uncertainty in the results. This study adopted a conceptual model that requires no specific root‐zone properties to model the RZWA. A RZWA‐based drought index (AgDI) was then developed by standardizing the root‐zone water deficit. A comparison over 20 catchments with different terrain and vegetation demonstrated the effectiveness of AgDI in characterizing agricultural droughts, with the correlations between vegetation indices (normalized difference vegetation index [NDVI] and gross primary productivity [GPP]) and AgDI (mean ~0.54) being significantly higher than those between the same vegetation indices and the Palmer Drought Severity Index (PDSI, ~0.22), Standardized Precipitation‐Evapotranspiration Index (SPEI, ~0.33), and Standardized Soil moisture Index (SSI, ~0.37). Compared to the SSI, the AgDI was not equally advantageous everywhere, but it generally showed better performance in deep‐rooting regions. Plain Language Summary: Agricultural drought can adversely affect agroecosystems; therefore its detection and quantification are essential for hazard mitigation. Limited root‐zone water availability (RZWA) is the direct cause of agricultural drought. Generally, physical agricultural drought indices are based on the soil moisture of several soil layers (e.g., 0–100 cm). However, these methods may result in uncertainties because the soil moisture in such soil profiles may not fully reflect the RZWA, especially in deep‐rooting ecosystems. In addition, estimation of RZWA is challenging because of the uncertainty in root‐zone properties (e.g., roots distribution and hydraulic properties). To overcome these known limitations, we developed a conceptual model that does not require root zone properties to model the RZWA, using which a new agricultural drought index (AgDI) was then developed by standardizing the root zone water deficit. A comparison study over 20 catchments with different terrain and vegetation properties showed that the AgDI is more effective in characterizing agricultural drought than the PDSI, SPEI and SSI. Compared to the soil moisture based SSI, the AgDI is not advantageous everywhere but generally exhibits better performance in deep‐rooting regions. We also found that the meteorological droughts are the main triggers of, but do not always lead to, agricultural droughts. Key Points: RZWA was inferred by using a conceptual model needing no specified root‐zone propertiesOur RZWA‐based drought index (AgDI) is effective in characterizing agricultural droughtsThe AgDI generally performs better in deep‐rooting regions than soil moisture based indices [ABSTRACT FROM AUTHOR]

Carbon capture and storage is being pursued globally as a geoengineering measure for reducing the emission of anthropogenic CO2 into the atmosphere. Comprehensive monitoring, verification, and accounting programs must be established for demonstrating the safe storage of injected CO 2. One of the most commonly deployed monitoring techniques is time‐lapse seismic reservoir monitoring (also known as 4‐D seismic), which involves comparing 3‐D seismic survey data taken at the same study site but over different times. Analyses of 4‐D seismic data volumes can help improve the quality of storage reservoir characterization, track the movement of injected CO 2 plume, and identify potential CO 2 spillover/leakage from the storage reservoirblue. However, the derivation of high‐resolution CO 2 saturation maps from 4‐D seismic data is a highly nonlinear and ill‐posed inverse problem, often requiring significant computational effort. In this research, we apply a physics‐based deep learning method to facilitate the solution of both the forward and inverse problems in seismic inversion while honoring physical constraints. A cycle generative adversarial neural network (CycleGAN) model is trained to learn the bidirectional functional mappings between the reservoir dynamic property changes and seismic attribute changes, such that both forward and inverse solutions can be obtained efficiently from the trained model. We show that our CycleGAN‐based approach not only improves the reliability of 4‐D seismic inversion but also expedites the quantitative interpretation. Our deep learning‐based workflow is generic and can be readily used for reservoir characterization and reservoir model updates involving the use of 4‐D seismic data. Key Points: We develop a cycle generative adversarial neural network (CycleGAN) model for time lapse seismic data inversion in carbon storage reservoirsCycleGAN combines reservoir simulation and rock physics inversion to invert high‐resolution maps of CO 2 saturation from acoustic impedanceWe show that CycleGAN can successfully identify the bidirectional mappings between CO 2 saturation changes and acoustic impedance changes [ABSTRACT FROM AUTHOR]

• Leakage detection is important for the safety of carbon sequestration repositories. • A machine learning method is applied to detecting bottom-hole pressure anomalies. • The method is based on convolutional long short-term memory neural net (ConvLSTM). • ConvLSTM can be used to incorporate both spatial and temporal information. • ConvLSTM successfully identifies pressure anomalies in a real pressure data set. Carbon capture and storage (CCS) has been extensively investigated as a potential engineering measure to reduce anthropogenic carbon emission to the atmosphere. Real-time monitoring of the safety and integrity of carbon storage reservoirs is a critical aspect of any commercial-scale CCS deployment. Pressure-based sensing is cost effective, suitable for real-time monitoring, and scalable to large monitoring networks. However, questions remain on how to best harness intelligent information from the high-frequency pressure monitoring sensors to support real-time decisions. This work presents a deep-learning-based framework for analyzing and detecting anomalies in pressure data streams by using a convolutional long short-term memory (ConvLSTM) neural network model, which allows for the fusion of both static and dynamic reservoir data. In ConvLSTM, the convolutional neural network (CNN) is used for spatial pattern mining and the LSTM is used for temporal pattern recognition. The performance of the ConvLSTM model for real-time anomaly detection is demonstrated using a set of pressure monitoring data collected from Cranfield, Mississippi, an active enhanced-oil-recovery field. The anomaly detection model is trained using bottom-hole pressure data acquired from the base experiment (without leak event) and then tested on pressure data collected during a series of controlled CO 2 release experiments (with artificially created leak events). Results show that the ConvLSTM neural network model successfully detected anomalies in the pressure time series obtained from the controlled release experiments. Inclusion of static information into the model further improves the robustness of ConvLSTM. [ABSTRACT FROM AUTHOR]

Abstract Rapid evolution of Internet-of-Things is driving the increased deployment of smart sensors in environmental applications, contributing to many big data characteristics of environmental monitoring. Most of the current environmental monitoring systems are not designed to handle real-time datastreams, and the best practices for datastream processing and predictive analytics are yet to be established. This work presents a complex event processing (CEP) engine for detecting anomalies in real time, and demonstrates it using a series of real monitoring data from the geological carbon sequestration domain. We show that the service-based CEP engine is instrumental for enabling environmental intelligent monitoring systems to ingest heterogeneous datastreams with scalable performance. Our CEP framework requires minimal coding from the user and can be easily extended to other similar environmental monitoring applications. Highlights • Wide use of smart sensors leads to explosive growth of data quantity and types. • We present an intelligent monitoring system enabled by complex event processing (CEP). • Machine learning algorithms are used to perform CEP. • Our system is built on an open-source stack of high-performance scalable components. • We demonstrate the system for a use case in geological carbon storage monitoring. [ABSTRACT FROM AUTHOR]

Abstract In a geological carbon storage project, leakage should be monitored to ensure safe long-term storage of injected CO 2. Leakage can be detected early and cost-effectively by monitoring subsurface pressure. The uncertainty in geological models also needs to be sufficiently reduced to detect leakage based on pressure monitoring data. This study presents numerical results of field pulse testing experiments that are designed to detect leakage based on pressure monitoring data for periodical CO 2 injection at a CO 2 enhanced oil recovery field in Mississippi, USA. In the pulse test, sinusoidal pressure patterns are captured in transitional pressure data because CO 2 injection and shut-in are repeated. The patterns are parameterized and history-matched efficiently in the frequency domain. Sensitivity analyses of pulse test parameters such as injection period and rate show that the frequency domain is more advantageous than the time domain for estimating leakage probability and well connectivity. We also conduct multi-objective history matching of pulse testing parameters in the frequency domain for reducing the uncertainty in geological models. This history matching reveals a clearer trade-off relationship between the matching qualities than conventional global-objective history matching, thereby being advantageous to yielding converged and diversified geological models for uncertainty quantification. Highlights • We assess the potential of pulse testing as a cost-effective monitoring tool to detect CO 2 leakage with application to Cranfield, USA. • Agreement between observations and simulations was improved by conditioning BHP and pulse testing data to the Cranfield model. • Attenuation in Fourier transform parameters is utilized as an indicator of CO 2 leakage as well as inter-well connectivity. • Integrating a flow-geomechanics simulator with a history-matching tool provides trade-off solutions in terms of matching qualities at wells. [ABSTRACT FROM AUTHOR]

Quantification of the predictive uncertainty of subsurface models has long been investigated. The traditional workflow is to calibrate prior models to match observed data, and then use the posterior models to simulate future system performance. Not only are these procedures computationally expensive, but they also have issues in maintaining geological model constraints during the calibration step. Data space inversion (DSI) was introduced recently to predict future system performance without the iterative history matching or model calibration step. In general, DSI approaches seek to establish a statistical relationship between the observed and forecast variables, as well as to quantify the predictive uncertainty of the forecast variables, by using an ensemble of uncalibrated prior models. Existing DSI approaches all require a number of complex transformation and mapping operations, which may deter their widespread use. In this study, we introduce a new and simpler DSI approach, the learning-based, data-driven forecast approach (LDFA), by combining dimension reduction and machine learning techniques to quickly provide accurate forecast results and reliably quantify corresponding uncertainty in the results. Our LDFA framework is demonstrated using two supervised learning algorithms, artificial neural network (ANN) and support vector regression (SVR), on two representative examples from reservoir engineering and geological carbon storage. Results suggest that our approach provides accurate forecast results (e.g., future oil production rate or cumulative injected CO 2 ) and reasonable predictive uncertainty intervals. Our framework is generic and may be applied to other surface and subsurface problems. [ABSTRACT FROM AUTHOR]

Carbon capture and storage (CCS) is being evaluated globally as a geoengineering measure for significantly reducing greenhouse emission. However, long-term liability associated with potential leakage from these geologic repositories is perceived as a main barrier of entry to site operators. Risk quantification and impact assessment help CCS operators to screen candidate sites for suitability of CO 2 storage. Leakage risks are highly site dependent, and a quantitative understanding and categorization of these risks can only be made possible through broad participation and deliberation of stakeholders, with the use of site-specific, process-based models as the decision basis. Online decision making, however, requires that scenarios be run in real time. In this work, a Python based, Leakage Assessment and Cost Estimation (PyLACE) web application was developed for quantifying financial risks associated with potential leakage from geologic carbon sequestration sites. PyLACE aims to assist a collaborative, analytic-deliberative decision making processes by automating metamodel creation, knowledge sharing, and online collaboration. In PyLACE, metamodeling, which is a process of developing faster-to-run surrogates of process-level models, is enabled using a special stochastic response surface method and the Gaussian process regression. Both methods allow consideration of model parameter uncertainties and the use of that information to generate confidence intervals on model outputs. Training of the metamodels is delegated to a high performance computing cluster and is orchestrated by a set of asynchronous job scheduling tools for job submission and result retrieval. As a case study, workflow and main features of PyLACE are demonstrated using a multilayer, carbon storage model. [ABSTRACT FROM AUTHOR]

Assessing reliability of global models is critical because of increasing reliance on these models to address past and projected future climate and human stresses on global water resources. Here, we evaluate model reliability based on a comprehensive comparison of decadal trends (2002-2014) in land water storage from seven global models (WGHM, PCR-GLOBWB, GLDAS NOAH,MOSAIC, VIC, CLM and CLSM) to trends from three Gravity Recovery and Climate Experiment (GRACE) satellite solutions in 186 river basins (∼60% of global land area). Medians of modeled basin water storage trends greatly underestimate GRACE-derived large decreasing (⩽-0.5 km3/y) and increasing (⩾0.5 km3/y) trends. Decreasing trends from GRACE are mostly related to human use (irrigation) and climate variations, whereas increasing trends reflect climate variations. For example, in the Amazon, GRACE estimates a large increasing trend of ∼43 km3/y, whereas most models estimate decreasing trends (-71 to 11 km3/y). Land water storage trends, summed over all basins, are positive for GRACE (∼71-82 km3/y) but negative for models (-450 to -12 km3/y), contributing opposing trends to global mean sea level change. Impacts of climate forcing on decadal land water storage trends exceed those of modeled human intervention by about a factor of 2. The model-GRACE comparison highlights potential areas of future model development, particularly simulated water storage. The inability of models to capture large decadal water storage trends based on GRACE indicates that model projections of climate and humaninduced water storage changes may be underestimated. [ABSTRACT FROM AUTHOR]

Understanding of the spatial and temporal dynamics of extreme precipitation not only improves prediction skills, but also helps to prioritize hazard mitigation efforts. This study seeks to enhance the understanding of spatiotemporal covariation patterns embedded in precipitation (P) and soil moisture (SM) by using an event-based, complex-network-theoretic approach. Events concurrences are quantified using a nonparametric event synchronization measure, and spatial patterns of hydroclimate variables are analyzed by using several network measures and a community detection algorithm. SM–P coupling is examined using a directional event coincidence analysis measure that takes the order of event occurrences into account. The complex network approach is demonstrated for Texas, US, a region possessing a rich set of hydroclimate features and is frequented by catastrophic flooding. Gridded daily observed P data and simulated SM data are used to create complex networks of P and SM extremes. The uncovered high degree centrality regions and community structures are qualitatively in agreement with the overall existing knowledge of hydroclimate extremes in the study region. Our analyses provide new visual insights on the propagation, connectivity, and synchronicity of P extremes, as well as the SM–P coupling, in this flood-prone region, and can be readily used as a basis for event-driven predictive analytics for other regions. [ABSTRACT FROM AUTHOR]

Global assessment of the spatiotemporal variability in terrestrial total water storage anomalies (TWSA) in response to hydrologic extremes is critical for water resources management. Using TWSA derived from the gravity recovery and climate experiment (GRACE) satellites, this study systematically assessed the skill of the TWSA-climatology (TC) approach and breakpoint (BP) detection method for identifying large-scale hydrologic extremes. The TC approach calculates standardized anomalies by using the mean and standard deviation of the GRACE TWSA corresponding to each month. In the BP detection method, the empirical mode decomposition (EMD) is first applied to identify the mean return period of TWSA extremes, and then a statistical procedure is used to identify the actual occurrence times of abrupt changes (i.e., BPs) in TWSA. Both detection methods were demonstrated on basin-averaged TWSA time series for the world's 35 largest river basins. A nonlinear event coincidence analysis measure was applied to cross-examine abrupt changes detected by these methods with those detected by the Standardized Precipitation Index (SPI). Results show that our EMD-assisted BP procedure is a promising tool for identifying hydrologic extremes using GRACE TWSA data. Abrupt changes detected by the BP method coincide well with those of the SPI anomalies and with documented hydrologic extreme events. Event timings obtained by the TC method were ambiguous for a number of river basins studied, probably because the GRACE data length is too short to derive long-term climatology at this time. The BP approach demonstrates a robust wet-dry anomaly detection capability, which will be important for applications with the upcoming GRACE Follow-On mission. [ABSTRACT FROM AUTHOR]

Leakage through abandoned wells represents one of the greatest risks to secure storage of CO 2 in geologic formations. In this work, we evaluate the feasibility of using temperature as an indicator for detecting leakage in and around wellbores. Numerical simulation tools are used to simulate temperature propagation both along and around the wellbore when leakage occurs. Typical overall heat transfer coefficient of formation rock is applied to quantify heat dissipation from the fluids leaked in well. We have tested the cases of CO 2 phase only and the mixture of CO 2 and brine. Thermodynamic phase change of CO 2 throughout the wellbore depth is accounted by an equation-of-state. Both vertical and slant (off-shore) wells are examined. Our results indicate that strong temperature variation is observed within 100 m from borehole and then virtual hot layers circulating the fluid column within 3 m radius are formed. However, differences from the ambient temperatures are not significant. The hot CO 2 column loses heat in 70 m of well depth from reservoir compared to 50 m height in the case of CO 2 /brine mixture. [ABSTRACT FROM AUTHOR]

Geologic repositories are extensively used for disposing byproducts in mineral and energy industries. The safety and reliability of these repositories are a primary concern to environmental regulators and the public. Time-lapse oscillatory pumping test (OPT) has been introduced recently as a pressure-based technique for detecting potential leakage in geologic repositories. By routinely conducting OPT at a number of pulsing frequencies, an operator may identify the potential repository anomalies in the frequency domain, alleviating the ambiguity caused by reservoir noise and improving the signal-to-noise ratio. Building on previous theoretical and field studies, this work performed a series of laboratory experiments to validate the concept of time-lapse OPT using a custom made, stainless steel tank under relatively high pressures. The experimental configuration simulates a miniature geologic storage repository consisting of three layers (i.e., injection zone, caprock, and above-zone aquifer). Results show that leakage in the injection zone led to deviations in the power spectrum of observed pressure data, and the amplitude of which also increases with decreasing pulsing frequencies. The experimental results are further analyzed by developing a 3D flow model, using which the model parameters are estimated through frequency domain inversion. [ABSTRACT FROM AUTHOR]

Management of shale gas wastewater treatment, disposal, and reuse has become a significant environmental challenge, driven by an ongoing boom in development of U.S. shale gas reservoirs. Systems-analysis based decision support is helpful for effective management of wastewater, and provision of cost-effective decision alternatives from a whole-system perspective. Uncertainties are inherent in many modeling parameters, affecting the generated decisions. In order to effectively deal with the recourse issue in decision making, in this work a two-stage stochastic fracturing wastewater management model, named TSWM, is developed to provide decision support for wastewater management planning in shale plays. Using the TSWM model, probabilistic and nonprobabilistic uncertainties are effectively handled. The TSWM model provides flexibility in generating shale gas wastewater management strategies, in which the first-stage decision predefined by decision makers before uncertainties are unfolded is corrected in the second stage to achieve the whole-system's optimality. Application of the TSWM model to a comprehensive synthetic example demonstrates its practical applicability and feasibility. Optimal results are generated for allowable wastewater quantities, excess wastewater, and capacity expansions of hazardous wastewater treatment plants to achieve the minimized total system cost. The obtained interval solutions encompass both optimistic and conservative decisions. Trade-offs between economic and environmental objectives are made depending on decision makers' knowledge and judgment, as well as site-specific information. The proposed model is helpful in forming informed decisions for wastewater management associated with shale gas development. [ABSTRACT FROM AUTHOR]

This work presents an optimization framework for evaluating different wastewater treatment/disposal options for water management during hydraulic fracturing (HF) operations. This framework takes into account both cost-effectiveness and system uncertainty. HF has enabled rapid development of shale gas resources. However, wastewater management has been one of the most contentious and widely publicized issues in shale gas production. The flowback and produced water (known as FP water) generated by HF may pose a serious risk to the surrounding environment and public health because this wastewater usually contains many toxic chemicals and high levels of total dissolved solids (TDS). Various treatment/disposal options are available for FP water management, such as underground injection, hazardous wastewater treatment plants, and/or reuse. In order to cost-effectively plan FP water management practices, including allocating FP water to different options and planning treatment facility capacity expansion, an optimization model named UO-FPW is developed in this study. The UO-FPW model can handle the uncertain information expressed in the form of fuzzy membership functions and probability density functions in the modeling parameters. The UO-FPW model is applied to a representative hypothetical case study to demonstrate its applicability in practice. The modeling results reflect the tradeoffs between economic objective (i.e., minimizing total-system cost) and system reliability (i.e., risk of violating fuzzy and/or random constraints, and meeting FP water treatment/disposal requirements). Using the developed optimization model, decision makers can make and adjust appropriate FP water management strategies through refining the values of feasibility degrees for fuzzy constraints and the probability levels for random constraints if the solutions are not satisfactory. The optimization model can be easily integrated into decision support systems for shale oil/gas lifecycle management. [ABSTRACT FROM AUTHOR]

Dissolution trapping of supercritical CO 2 into formation brine has been investigated as a potential mechanism for reducing buoyancy force in carbon storage formations. This study attempts to quantify how much CO 2 can be stored through dissolution trapping assuming the free-phase CO 2 will be dissolved continuously on the top of perturbed brine phase. Most former investigations focused on physical explanations of density-driven free convection instability. Our aim is to compute the amount of CO 2 (by mass) captured by dissolution trapping until the model reservoir reaches steady state. The numerical experimentation is done using dimensionless mass and momentum conservation laws. The major problem parameter here is the Rayleigh number, for which we carry out an extensive survey to find out its low and high ends based on field and observed data from the literature and in-house database. Because density difference is the main driving force, we also investigate the effects of impurities retained in CO 2 stream on density contrast. We study both homogeneous and heterogeneous reservoirs. Also, different boundary conditions (Neumann, Dirichlet, and periodic) are compared to understand their effects. The simulations are run until nearly complete saturation (∼99%) is reached. For a test case ( T = 40 °C, P = 50 bar) of geologic and thermophysical conditions, we have found that on average 0.33–15 g CO 2 will dissolve per year until a heterogeneous unit reservoir volume of 1 m 3 reaches complete saturations. For the case of homogeneous reservoir this amount is 0.28–6 g. [ABSTRACT FROM AUTHOR]

Highlights: [•] Gaussian Process Regression (GPR) is applied to monthly streamflow forecasting. [•] The efficacy of GPR is demonstrated for over 400 stations in the continental U.S. [•] Budyko framework is used to examine dependence of prediction skill on aridity index. [Copyright &y& Elsevier]

Abstract: We present an efficient methodology for assessing leakage detectability at geologic carbon sequestration sites under parameter uncertainty. Uncertainty quantification (UQ) and risk assessment are integral and, in many countries, mandatory components of geologic carbon sequestration projects. A primary goal of risk assessment is to evaluate leakage potential from anthropogenic and natural features, which constitute one of the greatest threats to the integrity of carbon sequestration repositories. The backbone of our detectability assessment framework is the probability collocation method (PCM), an efficient, nonintrusive, uncertainty-quantification technique that can enable large-scale stochastic simulations that are based on results from only a small number of forward-model runs. The metric for detectability is expressed through an extended signal-to-noise ratio (SNR), which incorporates epistemic uncertainty associated with both reservoir and aquifer parameters. The spatially heterogeneous aquifer hydraulic conductivity is parameterized using Karhunen–Loève (KL) expansion. Our methodology is demonstrated numerically for generating probability maps of pressure anomalies and for calculating SNRs. Results indicate that the likelihood of detecting anomalies depends on the level of uncertainty and location of monitoring wells. A monitoring well located close to leaky locations may not always yield the strongest signal of leakage when the level of uncertainty is high. Therefore, our results highlight the need for closed-loop site characterization, monitoring network design, and leakage source detection. [Copyright &y& Elsevier]

Abstract: Leakage from abandoned wells and geologic faults represents one of the greatest risks to the integrity of geologic CO2 sequestration sites. The ability to detect leakage in a timely manner is, therefore, crucial for mitigating the potential adverse impacts of leakage to the public and environment. We present an inversion approach for recovering both leakage locations and rates by using observed pressure anomaly data. The approach is based on formulation of a linear system of equations using the unit-step response method, which is applicable to both analytical and numerical models. Because the resulting system is often ill conditioned, we investigate the efficacy of regularization methods for stabilizing the solutions. Further, when prior information is insufficient to restrict the number of search locations, a global optimization algorithm is used to solve the challenging problem of joint location and leakage history inversion. The performance of several linear inversion solvers is compared while considering effects such as measurement error and spatial heterogeneity. The results are promising and suggest that our pressure-anomaly-based leakage detection algorithm can be used to identify leaky wells in practice. It can be deployed as an integrated component of CO2 risk management frameworks. [Copyright &y& Elsevier]

Summary: Regional groundwater models are increasingly used for short- and long-term water resources planning, in anticipation of greater climate variability and population growth. However, many of these models are subject to structural and parametric uncertainties because of the lack of field measurements. In recent years, the Gravity Recovery and Climate Experiment (GRACE) satellite mission has shown great potential for tracking total water storage changes over large regions. The pattern of groundwater storage changes inferred from GRACE may be incorporated as an additional regularization mechanism for calibrating regional groundwater models. Motivated by the demonstrated success of GRACE for monitoring groundwater storage changes, this study explores the combined use of in situ water level measurements and GRACE-derived groundwater storage changes for calibrating regional groundwater models. The resulting optimization problem is solved using an evolutionary optimization algorithm. We demonstrate the proposed calibration strategy for the hydraulically connected Edwards–Trinity Plateau and Pecos Valley aquifers (total area 115,000km2) in west Texas. Monthly GRACE data from 2002 to 2007 were used to recalibrate a regional groundwater model developed for the area. Our results indicate that (i) calibration using in situ data alone may yield multiple plausible solutions, a phenomenon well known to hydrologists; and (ii) GRACE data helped further constrain model parameters over the study period and, thus, may be continuously assimilated, among other sources of data, for enhancing existing regional groundwater models. [ABSTRACT FROM AUTHOR]

Abstract: Groundwater models are critical decision support tools for water resources management and environmental remediation. However, limitations in site characterization data and conceptual models can adversely affect the reliability of groundwater models. Therefore, there is a strong need for continuous model uncertainty reduction. Ensemble filters have recently emerged as promising high-dimensional data assimilation techniques. Two general categories of ensemble filters exist in the literature: perturbation-based and deterministic. Deterministic ensemble filters have been extensively studied for their better performance and robustness in assimilating oceanographic and atmospheric data. In hydrogeology, while a number of previous studies demonstrated the usefulness of the perturbation-based ensemble Kalman filter (EnKF) for joint parameter and state estimation, there have been few systematic studies investigating the performance of deterministic ensemble filters. This paper presents a comparative study of four commonly used deterministic ensemble filters for sequentially estimating the hydraulic conductivity parameter in low- and moderately high-dimensional groundwater models. The performance of the filters is assessed on the basis of twin experiments in which the true hydraulic conductivity field is assumed known. The test results indicate that the deterministic ensemble Kalman filter (DEnKF) is the most robust filter and achieves the best performance at relatively small ensemble sizes. Deterministic ensemble filters often make use of covariance inflation and localization to stabilize filter performance. Sensitivity studies demonstrate the effects of covariance inflation, localization, observation density, and conditioning on filter performance. [Copyright &y& Elsevier]

Model-based contaminant source identification plays an important role in effective site remediation. In this article, a contaminant source identification toolbox (CONSID) is introduced as a framework for solving contaminant source identification problems. It is known that the presence of various types of model uncertainties can severely undermine the performance of many existing source estimators. The current version of CONSID consists of two robust estimators for recovering source release histories under model uncertainty; one was developed in the deterministic framework and the other in the stochastic framework. To use the robust estimators provided in CONSID, the user is required to have only modest prior knowledge about the model uncertainty and be able to estimate the bound of model deviations resulting from the uncertainty. The toolbox is designed so that other source estimators can be added easily. A step-by-step guidance for using CONSID is described and an example is provided. [ABSTRACT FROM AUTHOR]

Abstract: Contamination source identification is a crucial step in environmental remediation. The exact contaminant source locations and release histories are often unknown due to lack of records and therefore must be identified through inversion. Coupled source location and release history identification is a complex nonlinear optimization problem. Existing strategies for contaminant source identification have important practical limitations. In many studies, analytical solutions for point sources are used; the problem is often formulated and solved via nonlinear optimization; and model uncertainty is seldom considered. In practice, model uncertainty can be significant because of the uncertainty in model structure and parameters, and the error in numerical solutions. An inaccurate model can lead to erroneous inversion of contaminant sources. In this work, a constrained robust least squares (CRLS) estimator is combined with a branch-and-bound global optimization solver for iteratively identifying source release histories and source locations. CRLS is used for source release history recovery and the global optimization solver is used for location search. CRLS is a robust estimator that was developed to incorporate directly a modeler''s prior knowledge of model uncertainty and measurement error. The robustness of CRLS is essential for systems that are ill-conditioned. Because of this decoupling, the total solution time can be reduced significantly. Our numerical experiments show that the combination of CRLS with the global optimization solver achieved better performance than the combination of a non-robust estimator, i.e., the nonnegative least squares (NNLS) method, with the same solver. [Copyright &y& Elsevier]

Summary: Objective:: The radical treatment of locally advanced non-small cell lung cancer (LA-NSCLC) currently involves combined modality therapy (CMT) with the use of chemotherapy in addition to radiation therapy and/or surgery. Chemotherapy has been shown to improve survival, but does not alter brain relapse. We reviewed the outcomes of Stage IIIA and IIIB LA-NSCLC patients treated with CMT at our institution. We assessed the incidence of brain metastases and the management and outcome of these patients. Methods:: Using our radiation-planning database (RSTS), we identified 230 consecutive patients from the years 1999 and 2000 who received radical radiation therapy to the lung. Extracting data from the chart, we identified 83 patients who were treated radically with chemotherapy, radiation and possibly surgery. These patients form the basis of this study. Results:: At 2 years, the actuarial rates for any brain failure, first failure in the brain and sole failure in the brain were 34.2%, 24.6% and 11.0%, respectively. Age was the only factor among sex, histology, stage, weight loss and the timing of chemotherapy and radiation that predicted for an increased risk of first failure in the brain. Patients less than age 60 had a risk of 25.6% versus 11.4% for those greater than 60 (p =0.022). Among the patients who failed first in the brain, those who had aggressive management of their brain metastases with surgical resection in addition to whole brain radiotherapy had a median survival of 26.3 months compared with 3.3 months for those treated with palliative whole brain radiotherapy alone. Conclusion:: Brain metastases are common in patients with LA-NSCLC treated with CMT. These patients may benefit from either prophylactic cranial irradiation or early detection and aggressive treatment of brain metastases. [Copyright &y& Elsevier]

Model-based optimization plays a central role in energy system design and management. The complexity and high-dimensionality of many process-level models, especially those used for geosystem energy exploration and utilization, often lead to formidable computational costs when the dimension of decision space is also large. This work adopts elements of recently advanced deep learning techniques to solve a sequential decision-making problem in applied geosystem management. Specifically, a deep reinforcement learning framework was formed for optimal multiperiod planning, in which a deep Q-learning network (DQN) agent was trained to maximize rewards by learning from high-dimensional inputs and from exploitation of its past experiences. To expedite computation, deep multitask learning was used to approximate high-dimensional, multistate transition functions. Both DQN and deep multitask learning are pattern based. As a demonstration, the framework was applied to optimal carbon sequestration reservoir planning using two different types of management strategies: monitoring only and brine extraction. Both strategies are designed to mitigate potential risks due to pressure buildup. Results show that the DQN agent can identify the optimal policies to maximize the reward for given risk and cost constraints. Experiments also show that knowledge the agent gained from interacting with one environment is largely preserved when deploying the same agent in other similar environments. • Many energy management applications require sequential decision making. • Identification of the optimal policies for sequential decision making is challenging for high-dimensional problems. • A deep reinforcement learning framework is introduced for sequential decision making. • Deep reinforcement learning is demonstrated on optimal carbon sequestration reservoir management. • Results show deep reinforcement learning agent can learn to find the optimal policy. [ABSTRACT FROM AUTHOR]

Waterflooding, during which water is injected in the reservoir to increase pressure and therefore boost oil production, is extensively used as a secondary oil recovery technology. Tracking the extent and efficacy of waterflooding (i.e., fluid distributions) is a primary task of reservoir engineers and is traditionally achieved by running full reservoir models. In this work, we design and implement a proxy model using a conditional deep convolutional generative neural network (cDC-GAN), which can be used to quickly calculate the dynamic fluid distribution of a heterogeneous reservoir under waterflooding. Zero-sum game theory is the basis for the cDC-GAN, which includes a pair of generative discriminative models. The generative model tries to learn the relationship between input and output and makes the generated output as close as possible to the training data, while the discriminative model tries to distinguish the fake output and the real data used for training, such that the cDC-GAN learns the real data distribution at the end. In our cDC-GAN formulation, the reservoir properties (permeability distribution in this research) and forecast time information are treated as input, and water saturation is the desired output. The reservoir fluid production rate can be calculated based on the material balance principle. The most significant contribution of this work resides in training a cDC-GAN proxy model to accurately predict fluid saturation. A cDC-GAN has several advantages over the traditional full-model based workflow. First, the model parameters estimated from history matching help to improve reservoir characterization. Second, this proposed proxy model can predict the water and oil saturation distributions simultaneously, which can be used to calculate the water and oil flow rates. Third, this proposed proxy model can be used for waterflooding optimization and uncertainty analysis with far less computational effort than with the traditional method, which uses a reservoir simulator. • cDC-GAN proxy model can be used to predict the reservoir fluid distribution computational efficiently. • The cDC-GAN can learn the nonlinear mapping function between input space and output space. • Reservoir fluid production rate can be calculated based on the principle of material balance. [ABSTRACT FROM AUTHOR]

Characterization of the CO 2 plume evolution in deep aquifers under geological uncertainties is crucial for designing geological carbon storage (GCS) projects. In this study, we are concerned with uncertainty quantification (UQ) of CO 2 plume migration in uncertain permeability fields following non-Gaussian distributions. Specifically, we are concerned with the effect of intra-facies heterogeneity on plume migration in a multi-facies sedimentary setting. To alleviate the huge computational burden required for UQ, machine learning (ML) methods are broadly used to develop surrogate models. However, the quality of ML surrogate models is often not qualified. Here we present a probabilistic convolutional neural network (pCNN) method for fast surrogate modeling of the CO 2 plume evolution. The dense and residual connections are adopted in the pCNN's network architecture to enhance its performance in approximating the highly complex input–output mapping of the GCS model induced by non-Gaussian permeability heterogeneity. In addition, a Bayesian training strategy is utilized to train the pCNN surrogate model, enabling it to give predictions together with an estimate of the surrogate uncertainty. As a result, uncertainties associated with the pCNN surrogate can be quantified in addition to the model parameter uncertainty. The proposed method is demonstrated using a 3-D GCS model with an uncertain, non-Gaussian distributed, multi-facies permeability field. Results show that pCNN is able to provide a reliable and fast approximation of the highly complex GCS model. The pCNN-derived UQ estimates, obtained with negligible additional computational costs, show a good agreement with those derived from the computationally intensive GCS model. The generalization performance of pCNN is further investigated using "out-of-distribution" permeability samples. For such permeability fields, pCNN can still provide relatively accurate approximations, only showing slightly increased predictive errors and uncertainties relative to those having similar distributions to the training samples. • pCNN provides both reliable predictions and predictive uncertainties. • UQ of CO 2 plume migration in non-Gaussian permeability fields using pCNN. • pCNN obtains accurate UQ results at a relatively low computational cost. • pCNN gives satisfactory performance on "out-of-distribution" permeability samples. [ABSTRACT FROM AUTHOR]

A fundamental problem in geophysical modeling is related to the identification and approximation of causal structures among physical processes. However, resolving the bidirectional mappings between physical parameters and model state variables (i.e., solving the forward and inverse problems) is challenging, especially when parameter dimensionality is high. Deep learning has opened a new door toward knowledge representation and complex pattern identification. In particular, the recently introduced generative adversarial networks (GANs) hold strong promises in learning cross‐domain mappings for image translation. This study presents a state‐parameter identification GAN (SPID‐GAN) for simultaneously learning bidirectional mappings between a high‐dimensional parameter space and the corresponding model state space. SPID‐GAN is demonstrated using a series of representative problems from subsurface flow modeling. Results show that SPID‐GAN achieves satisfactory performance in identifying the bidirectional state‐parameter mappings, providing a new deep‐learning‐based, knowledge representation paradigm for a wide array of complex geophysical problems. Plain Language Summary: Development of physically based models requires two steps, mathematical abstraction (forward modeling) and parameter estimation (inverse modeling). A high‐fidelity model requires high‐quality parameter support. The need for identifying forward and reverse mappings (i.e., a function that associates element of one set to another) is thus ubiquitous in geophysical research. A significant challenge in geosciences is that geoparameters are spatially heterogeneous and high dimensional and yet can only be observed at limited locations. The conventional workflow, built on minimizing the model‐observation mismatch at measurement locations, does not offer an efficient way for estimating the spatial structure of high‐dimensional parameter fields. This work presents a deep‐learning‐based framework for identifying the state‐parameter bidirectional mappings using the recently introduced generative adversarial networks (GANs). GANs have been shown to be adept at associating images from one domain to another. Its potential for discovering mappings in physically based models has not been demonstrated so far. This work shows that GAN can achieve high performance in learning bidirectional parameter‐to‐state mappings in physically based models, thus providing a new way of thinking and doing things in geosciences. The implication for additional applications in subsurface modeling is significant. Key Points: The need for identifying forward and reverse mappings is ubiquitous in all geophysical research fieldsThis study shows that generative adversarial networks (GANs) can be used to learn the forward and inverse mappings at the same timeGAN provides a new way of thinking in combining physical‐based modeling with data‐driven modeling and has broad applications for geosciences [ABSTRACT FROM AUTHOR]

Highlights • A dynamic Budyko model was improved by linking climate and vegetation. • Catchments with larger portion of karst landform (POK) have lower vegetation available water. • Evapotranspiration in catchments with larger POK rely more on precipitation. Abstract Karst landscape, covers more than 10% of the global land surface and plays an important role in supporting ecosystems and human society, may be strongly influenced by climate change. Vegetation available water (VAW) is a key variable impacting the sensitivity of ecosystems to a changing environment. However, VAW in karst region is difficult to determine and remains uncertain. This study improved a dynamic Budyko-type water balance model, by introducing a nonlinear equation linking climate, vegetation and Budyko-type water balance. This model was calibrated using evolutionary algorithm based on monthly runoff. Comparison results in the Pearl, Yangtze and 12 karst catchments in south China, suggested high effectiveness of this improved model in simulating monthly runoff and evapotranspiration. Furthermore, in the 12 karst catchments, the max VAW was negatively correlated to the portion of karst landscape of the catchment (POK, r = −0.63, α = 0.027) and the elasticity of evapotranspiration to precipitation (r = −0.60, α = 0.04). These implied that karst catchments with higher POK had lower VAW, making the ecosystems rely more on precipitation. [ABSTRACT FROM AUTHOR]

ABSTRACT: Southwest (SW) China is a drought prone region. Knowledge of the risks and coupling characteristics of drought duration and severity is essential for hazards mitigation and water management. Based on probability distributions, copulas, Mann–Kendall test, overlapping moving windows approach and the self‐calibrated Palmer Drought Severity Index (PDSI) at 69 stations (1959–2013), this study focused on the trends and spatial variability (SV) of (joint) probability drought duration and severity in SW China, and further investigated the potential contributors by analysing the behaviours of precipitation (P), temperature (T), relative humidity (RH), wind speed (WS), atmospheric moisture flux (MF) and divergence (MFD). In particular, the SV was represented by the standard deviation, the drought duration and severity were characterized by the number of months with PDSI < −0.5 in a year [drought months (DM)] and the absolute mean PDSI in drought months (DS), respectively. Results showed that, the DM, DS, DM5, DS5 (5‐year return levels for DM and DS) increased by 0.058 month year−2, 0.012 month−1 year−1, 0.039 month year−2, and 0.009 month−1 year−1, respectively, and the joint return period T(DM > 4,DS > 2) decreased by −0.20 year year−1. Furthermore, due to the decreasing SV in P, T, RH, WS, MF, and MFD, SV of DM5, DS5 and T(DM > 4,DS > 2) significantly decreased in the past decades, by −0.003 month year−2, −0.006 month−1 year−1 and −0.03 year year−1, respectively. These results indicated that drought risks and severity increased significantly in the past decades over SW China. Most importantly, the decreasing SV implied that drought risks and severity might have become more homogeneous in this region. In other words, droughts became more widely spread throughout SW China in the past decades. Therefore, water agencies and governments should be more proactive in seeking measures to improve water resources management and drought mitigation in this region. [ABSTRACT FROM AUTHOR]

A two-parameter annual water balance model was developed for reconstructing annual terrestrial water storage change (Δ TWS ) and groundwater storage change (Δ GWS ). The model was integrated with the Gravity Recovery and Climate Experiment (GRACE) data and applied to the Punjab province in Pakistan for reconstructing Δ TWS and Δ GWS during 1980–2015 based on multiple input data sources. Model parameters were estimated through minimizing the root-mean-square error between the Budyko-modeled and GRACE-derived Δ TWS during 2003–2015. The correlation of ensemble means between Budyko-modeled and GRACE-derived Δ TWS is 0.68 with p -value <0.05. The Δ GWS was reconstructed by subtracting soil moisture storage change from the Budyko-modeled Δ TWS and was validated (i.e., r = 0.74, p -value < 0.05) against well observations during the pre-GRACE period (i.e., 1985–1994). The negative values of the cumulative sum of the reconstructed Δ GWS during 1980–2015 (i.e., −13.6 ± 9.7 cm) indicate that the aquifer in Punjab has experienced depletion. The estimated depletion rate is −0.3 ± 0.2 cm/year and it has a negative correlation (i.e., r = −0.70, p -value < 0.0001) with the total number of tube wells installed in Punjab. The integration of the developed Budyko model with GRACE data provides a new way for evaluating long-term groundwater depletion in large-scale irrigation regions with parsimonious models. [ABSTRACT FROM AUTHOR]

Prediction of drought plays an important role in drought preparedness and mitigation, especially because of large impacts of drought and increasing demand for water resources. An important aspect for improving drought prediction skills is the identification of drought predictability sources. In general, a drought originates from precipitation deficit and thus the antecedent meteorological drought may provide predictive information for other types of drought. In this study, a hydrological drought (represented by Standardized Runoff Index (SRI)) prediction method is proposed based on the meta-Gaussian model taking into account the persistence and its prior meteorological drought condition (represented by Standardized Precipitation Index (SPI)). Considering the inherent nature of standardized drought indices, the meta-Gaussian model arises as a suitable model for constructing the joint distribution of multiple drought indices. Accordingly, the conditional distribution of hydrological drought can be derived analytically, which enables the probabilistic prediction of hydrological drought in the target period and uncertainty quantifications. Based on monthly precipitation and surface runoff of climate divisions of Texas, U.S., 1-month and 2-month lead predictions of hydrological drought are illustrated and compared to the prediction from Ensemble Streamflow Prediction (ESP). Results, based on 10 climate divisions in Texas, show that the proposed meta-Gaussian model provides useful drought prediction information with performance depending on regions and seasons. [ABSTRACT FROM AUTHOR]

Summary Karst landform represents about 10% of the continental area and plays key roles in water supplies for almost a quarter of the global population. Knowledge of ecohydrological responses of karst landform to climate change is critical for both water resources management and ecological protection in these regions. This study investigated the effects of karst landform on the elasticity of actual evapotranspiration (derived by the Budyko equation), estimated the contribution of climate change and evaluated the implications, on the basis of 13 typical catchments that have different karst landform coverages in southwest China. Catchment properties, including the vegetation coverage, portion of karst landform (POK), drainage area, surface roughness, mean topographic wetness index, mean slope, and mean aspect, were selected to test the influencing factors for the elasticity of actual evapotranspiration. Results indicate that POK is the most influencing factor for the elasticity of actual evapotranspiration in this region. Moreover, the actual evapotranspiration in karst catchments is more sensitive to precipitation change and less sensitive to the potential evapotranspiration change than that in the non-karst catchments. On the other hand, the contribution of climate change to actual evapotranspiration was generally negative in this region. Furthermore, relatively large negative contributions mainly occurred in the karst-dominated catchments, suggesting that the karst catchments were exposed to higher degradation stress brought by the climate change than that in non-karst catchments. [ABSTRACT FROM AUTHOR]

Purpose: To assess the impact of prophylactic cranial irradiation (PCI) on self-reported cognitive functioning (SRCF), a functional scale on the European Organization for Research and Treatment of Cancer Core Quality of Life Questionnaire (EORTC QLQ-C30). Methods and Materials: Radiation Therapy Oncology Group (RTOG) protocol 0214 randomized patients with locally advanced non-small cell lung cancer to PCI or observation; RTOG 0212 randomized patients with limited-disease small cell lung cancer to high- or standard-dose PCI. In both trials, Hopkins Verbal Learning Test (HVLT)-Recall and -Delayed Recall and SRCF were assessed at baseline (after locoregional therapy but before PCI or observation) and at 6 and 12 months. Patients developing brain relapse before follow-up evaluation were excluded. Decline was defined using the reliable change index method and correlated with receipt of PCI versus observation using logistic regression modeling. Fisher's exact test correlated decline in SRCF with HVLT decline. Results: Of the eligible patients pooled from RTOG 0212 and RTOG 0214, 410 (93%) receiving PCI and 173 (96%) undergoing observation completed baseline HVLT or EORTC QLQ-C30 testing and were included in this analysis. Prophylactic cranial irradiation was associated with a higher risk of decline in SRCF at 6 months (odds ratio 3.60, 95% confidence interval 2.34-6.37, P<.0001) and 12 months (odds ratio 3.44, 95% confidence interval 1.84-6.44, P<.0001). Decline on HVLT-Recall at 6 and 12 months was also associated with PCI (P=.002 and P=.002, respectively) but was not closely correlated with decline in SRCF at the same time points (P=.05 and P=.86, respectively). Conclusions: In lung cancer patients who do not develop brain relapse, PCI is associated with decline in HVLT-tested and self-reported cognitive functioning. Decline in HVLT and decline in SRCF are not closely correlated, suggesting that they may represent distinct elements of the cognitive spectrum. [Copyright &y& Elsevier]

Fractional calculus-based differential equations were found by previous studies to be promising tools in simulating local-scale anomalous diffusion for pollutants transport in natural geological media (geomedia), but efficient models are still needed for simulating anomalous transport over a broad spectrum of scales. This study proposed a hierarchical framework of fractional advection-dispersion equations (FADEs) for modeling pollutants moving in the river corridor at a full spectrum of scales. Applications showed that the fixed-index FADE could model bed sediment and manganese transport in streams at the geomorphologic unit scale, whereas the variable-index FADE well fitted bedload snapshots at the reach scale with spatially varying indices. Further analyses revealed that the selection of the FADEs depended on the scale, type of the geomedium (i.e., riverbed, aquifer, or soil), and the type of available observation dataset (i.e., the tracer snapshot or breakthrough curve (BTC)). When the pollutant BTC was used, a single-index FADE with scale-dependent parameters could fit the data by upscaling anomalous transport without mapping the sub-grid, intermediate multi-index anomalous diffusion. Pollutant transport in geomedia, therefore, may exhibit complex anomalous scaling in space (and/or time), and the identification of the FADE's index for the reach-scale anomalous transport, which links the geomorphologic unit and watershed scales, is the core for reliable applications of fractional calculus in hydrology. [ABSTRACT FROM AUTHOR]