Metamodels to bridge the gap between modeling and decision support
Explore the source record for details and available documents.
Groundwater↗
Geology topics
Michael N. Fienen
Publications and source records attributed to Michael N. Fienen.
High-throughput computing vs. high-performance computing for groundwater applications
No abstract available.
Groundwater↗
Understanding the Day Cent model: Calibration, sensitivity, and identifiability through inverse modeling
The ability of biogeochemical ecosystem models to represent agro-ecosystems depends on their correct integration with field observations. We report simultaneous calibration of 67 DayCent model parameters using multiple observation types through inverse modeling using the PEST parameter estimation software. Parameter estimation reduced the total sum of weighted squared residuals by 56% and improved model fit to crop productivity, soil carbon, volumetric soil water content, soil temperature, N 2 O, and soil 3 NO − compared to the default simulation. Inverse modeling substantially reduced predictive model error relative to the default model for all model predictions, except for soil 3 NO − and 4 NH + . Post-processing analyses provided insights into parameter–observation relationships based on parameter correlations, sensitivity and identifiability. Inverse modeling tools are shown to be a powerful way to systematize and accelerate the process of biogeochemical model interrogation, improving our understanding of model function and the underlying ecosystem biogeochemical processes that they represent.
Environmental Modelling and Software↗
A statistical learning framework for groundwater nitrate models of the Central Valley, California, USA
We used a statistical learning framework to evaluate the ability of three machine-learning methods to predict nitrate concentration in shallow groundwater of the Central Valley, California: boosted regression trees (BRT), artificial neural networks (ANN), and Bayesian networks (BN). Machine learning methods can learn complex patterns in the data but because of overfitting may not generalize well to new data. The statistical learning framework involves cross-validation (CV) training and testing data and a separate hold-out data set for model evaluation, with the goal of optimizing predictive performance by controlling for model overfit. The order of prediction performance according to both CV testing R 2 and that for the hold-out data set was BRT > BN > ANN. For each method we identified two models based on CV testing results: that with maximum testing R 2 and a version with R 2 within one standard error of the maximum (the 1SE model). The former yielded CV training R 2 values of 0.94–1.0. Cross-validation testing R 2 values indicate predictive performance, and these were 0.22–0.39 for the maximum R 2 models and 0.19–0.36 for the 1SE models. Evaluation with hold-out data suggested that the 1SE BRT and ANN models predicted better for an independent data set compared with the maximum R 2 versions, which is relevant to extrapolation by mapping. Scatterplots of predicted vs. observed hold-out data obtained for final models helped identify prediction bias, which was fairly pronounced for ANN and BN. Lastly, the models were compared with multiple linear regression (MLR) and a previous random forest regression (RFR) model. Whereas BRT results were comparable to RFR, MLR had low hold-out R 2 (0.07) and explained less than half the variation in the training data. Spatial patterns of predictions by the final, 1SE BRT model agreed reasonably well with previously observed patterns of nitrate occurrence in groundwater of the Central Valley.
California↗
A cross-validation package driving Netica with python
Bayesian networks (BNs) are powerful tools for probabilistically simulating natural systems and emulating process models. Cross validation is a technique to avoid overfitting resulting from overly complex BNs. Overfitting reduces predictive skill. Cross-validation for BNs is known but rarely implemented due partly to a lack of software tools designed to work with available BN packages. CVNetica is open-source, written in Python, and extends the Netica software package to perform cross-validation and read, rebuild, and learn BNs from data. Insights gained from cross-validation and implications on prediction versus description are illustrated with: a data-driven oceanographic application; and a model-emulation application. These examples show that overfitting occurs when BNs become more complex than allowed by supporting data and overfitting incurs computational costs as well as causing a reduction in prediction skill. CVNetica evaluates overfitting using several complexity metrics (we used level of discretization) and its impact on performance metrics (we used skill).
Environmental Modelling and Software↗
Simulation of groundwater flow and interaction of groundwater and surface water on the Lac du Flambeau Reservation, Wisconsin
The Lac du Flambeau Band of Lake Superior Chippewa and Indian Health Service are interested in improving the understanding of groundwater flow and groundwater/surface-water interaction on the Lac du Flambeau Reservation (Reservation) in southwest Vilas County and southeast Iron County, Wisconsin, with particular interest in an understanding of the potential for contamination of groundwater supply wells and the fate of wastewater that is infiltrated from treatment lagoons on the Reservation. This report describes the construction, calibration, and application of a regional groundwater flow model used to simulate the shallow groundwater flow system of the Reservation and water-quality results for groundwater and surface-water samples collected near a system of waste-water-treatment lagoons. Groundwater flows through a permeable glacial aquifer that ranges in thickness from 60 to more than 200 feet (ft). Seepage and drainage lakes are common in the area and influence groundwater flow patterns on the Reservation. A two-dimensional, steady-state analytic element groundwater flow model was constructed using the program GFLOW. The model was calibrated by matching target water levels and stream base flows through the use of the parameter-estimation program, PEST. Simulated results illustrate that groundwater flow within most of the Reservation is toward the Bear River and the chain of lakes that feed the Bear River. Results of analyses of groundwater and surface-water samples collected downgradient from the wastewater infiltration lagoons show elevated levels of ammonia and dissolved phosphorus. In addition, wastewater indicator chemicals detected in three downgradient wells and a small downgradient stream indicate that infiltrated wastewater is moving southwest of the lagoons toward Moss Lake. Potential effects of extended wet and dry periods (within historical ranges) were evaluated by adjusting precipitation and groundwater recharge in the model and comparing the resulting simulated lake stage and water budgets to stages and water budgets from the calibrated model. Simulated lake water budgets and water level changes illustrate the importance of understanding the position of a lake within the hydrologic system (headwater or downstream), the type of lake (surface-water drainage or seepage lake), and the role of groundwater in dampening the effects of large-scale changes in weather patterns on lake levels. Areas contributing recharge to drinking-water supply wells on the Reservation were delineated using forward particle tracking from the water table to the well. Monte Carlo uncertainty analyses were used to produce maps showing the probability of groundwater capture for areas around each well nest. At the Main Pumphouse site near the Village of Lac du Flambeau, most of the area contributing recharge to the wells occurs downgradient from a large wetland between the wells and the wastewater infiltration lagoons. Nonetheless, a small potential for the wells to capture infiltrated wastewater is apparent when considering uncertainty in the model parameter values. At the West Pumphouse wells south of Flambeau Lake, most of the area contributing recharge is between the wells and Tippecanoe Lake. The extent of infiltrated wastewater from two infiltration lagoons was tracked using the groundwater flow model and Monte Carlo uncertainty analyses. Wastewater infiltrated from the lagoons flows predominantly south toward Moss Lake as it integrates with the regional groundwater flow system. The wastewater-plume-extent simulations support the area-contributing-recharge simulations, indicating that there is a possibility, albeit at low probability, that some wastewater could be captured by water-supply wells. Comparison of simulated water-table contours indicate that the lagoons may mound the water table approximately 4 ft, with diminishing levels of mounding outward from the lagoons. Four scenarios, representing potential alternatives for wastewater management, were simulated (at current discharge rates) to evaluate the potential extent of wastewater in the aquifer and discharge to surface-water bodies associated with each management scenario. Wastewater simulated to infiltrate through a hypothetical diffuser below a wetland south of the current lagoons appears to discharge to the overlying wetland and would likely discharge to Moss Lake as overland flow. Wastewater simulated to discharge to a small lake (Mindy Lake) between Moss and Fence Lakes appears to spread radically over a large area between the lakes. Wastewater simulated to discharge to lagoons south and northeast of the current lagoons also appears to spread radially, but the areas of the aquifer with the highest probability of encountering waste-water contamination would likely be between the lagoons and the nearest lake, where the wastewater would eventually discharge. Probability results for the wastewater-plume-extent scenarios are sensitive to the number of mathematical water particles used to represent infiltrating wastewater and the level of detail in the synthetic grid used for the probability analysis. Thus, probability results from wastewater-plume-extent simulations are qualitative only; however, it is expected that illustrations of relatively high or low probability will be useful as a general guide for decision making. Management problems requiring quantitative estimates of probability are best re-cast into problems evaluating the area that contributes recharge to the location of interest, which is not dependent upon the number of simulated particles or the resolution of a synthetic grid.
Wisconsin↗
Effects of sea-level rise on barrier island groundwater system dynamics: ecohydrological implications
We used a numerical model to investigate how a barrier island groundwater system responds to increases of up to 60 cm in sea level. We found that a sea-level rise of 20 cm leads to substantial changes in the depth of the water table and the extent and depth of saltwater intrusion, which are key determinants in the establishment, distribution and succession of vegetation assemblages and habitat suitability in barrier islands ecosystems. In our simulations, increases in water-table height in areas with a shallow depth to water (or thin vadose zone) resulted in extensive groundwater inundation of land surface and a thinning of the underlying freshwater lens. We demonstrated the interdependence of the groundwater response to island morphology by evaluating changes at three sites. This interdependence can have a profound effect on ecosystem composition in these fragile coastal landscapes under long-term changing climatic conditions.
Maryl;Virginia↗
Bridging groundwater models and decision support with a Bayesian network
Resource managers need to make decisions to plan for future environmental conditions, particularly sea level rise, in the face of substantial uncertainty. Many interacting processes factor in to the decisions they face. Advances in process models and the quantification of uncertainty have made models a valuable tool for this purpose. Long-simulation runtimes and, often, numerical instability make linking process models impractical in many cases. A method for emulating the important connections between model input and forecasts, while propagating uncertainty, has the potential to provide a bridge between complicated numerical process models and the efficiency and stability needed for decision making. We explore this using a Bayesian network (BN) to emulate a groundwater flow model. We expand on previous approaches to validating a BN by calculating forecasting skill using cross validation of a groundwater model of Assateague Island in Virginia and Maryland, USA. This BN emulation was shown to capture the important groundwater-flow characteristics and uncertainty of the groundwater system because of its connection to island morphology and sea level. Forecast power metrics associated with the validation of multiple alternative BN designs guided the selection of an optimal level of BN complexity. Assateague island is an ideal test case for exploring a forecasting tool based on current conditions because the unique hydrogeomorphological variability of the island includes a range of settings indicative of past, current, and future conditions. The resulting BN is a valuable tool for exploring the response of groundwater conditions to sea level rise in decision support.
Water Resources Research↗
Nitrous oxide emissions from cropland: a procedure for calibrating the DayCent biogeochemical model using inverse modelling
DayCent is a biogeochemical model of intermediate complexity widely used to simulate greenhouse gases (GHG), soil organic carbon and nutrients in crop, grassland, forest and savannah ecosystems. Although this model has been applied to a wide range of ecosystems, it is still typically parameterized through a traditional “trial and error” approach and has not been calibrated using statistical inverse modelling (i.e. algorithmic parameter estimation). The aim of this study is to establish and demonstrate a procedure for calibration of DayCent to improve estimation of GHG emissions. We coupled DayCent with the parameter estimation (PEST) software for inverse modelling. The PEST software can be used for calibration through regularized inversion as well as model sensitivity and uncertainty analysis. The DayCent model was analysed and calibrated using N2O flux data collected over 2 years at the Iowa State University Agronomy and Agricultural Engineering Research Farms, Boone, IA. Crop year 2003 data were used for model calibration and 2004 data were used for validation. The optimization of DayCent model parameters using PEST significantly reduced model residuals relative to the default DayCent parameter values. Parameter estimation improved the model performance by reducing the sum of weighted squared residual difference between measured and modelled outputs by up to 67 %. For the calibration period, simulation with the default model parameter values underestimated mean daily N2O flux by 98 %. After parameter estimation, the model underestimated the mean daily fluxes by 35 %. During the validation period, the calibrated model reduced sum of weighted squared residuals by 20 % relative to the default simulation. Sensitivity analysis performed provides important insights into the model structure providing guidance for model improvement.
Water, Air, & Soil Pollution↗
Partial least squares for efficient models of fecal indicator bacteria on Great Lakes beaches
At public beaches, it is now common to mitigate the impact of water-borne pathogens by posting a swimmer's advisory when the concentration of fecal indicator bacteria (FIB) exceeds an action threshold. Since culturing the bacteria delays public notification when dangerous conditions exist, regression models are sometimes used to predict the FIB concentration based on readily-available environmental measurements. It is hard to know which environmental parameters are relevant to predicting FIB concentration, and the parameters are usually correlated, which can hurt the predictive power of a regression model. Here the method of partial least squares (PLS) is introduced to automate the regression modeling process. Model selection is reduced to the process of setting a tuning parameter to control the decision threshold that separates predicted exceedances of the standard from predicted non-exceedances. The method is validated by application to four Great Lakes beaches during the summer of 2010. Performance of the PLS models compares favorably to that of the existing state-of-the-art regression models at these four sites.
Journal of Environmental Management↗
Development of a numerical model to simulate groundwater flow in the shallow aquifer system of Assateague Island, Maryland and Virginia
A three-dimensional groundwater-flow model was developed for Assateague Island in eastern Maryland and Virginia to simulate both groundwater flow and solute (salt) transport to evaluate the groundwater system response to sea-level rise. The model was constructed using geologic and spatial information to represent the island geometry, boundaries, and physical properties and was calibrated using an inverse modeling parameter-estimation technique. An initial transient solute-transport simulation was used to establish the freshwater-saltwater boundary for a final calibrated steady-state model of groundwater flow. This model was developed as part of an ongoing investigation by the U.S. Geological Survey Climate and Land Use Change Research and Development Program to improve capabilities for predicting potential climate-change effects and provide the necessary tools for adaptation and mitigation of potentially adverse impacts.
Open-File Report↗
Simulation of the shallow groundwater-flow system in the Forest County Potawatomi Community, Forest County, Wisconsin
The shallow groundwater system in the Forest County Potawatomi Comminity, Forest County, Wisconsin, was simulated by expanding and recalibrating a previously calibrated regional model. The existing model was updated using newly collected water-level measurements, inclusion of surface-water features beyond the previous near-field boundary, and refinements to surface-water features. The updated model then was used to calculate the area contributing recharge for seven existing and three proposed pumping locations on lands of the Forest County Potawatomi Community. The existing wells were the subject of a 2004 source-water evaluation in which areas contributing recharge were calculated using the fixed-radius method. The motivation for the present (2012) project was to improve the level of detail of areas contributing recharge for the existing wells and to provide similar analysis for the proposed wells. Delineated 5- and 10-year areas contributing recharge for existing and proposed wells extend from the areas of pumping to delineate the area at the surface contributing recharge to the wells. Steady-state pumping was simulated for two scenarios: a base-pumping scenario using pumping rates that reflect what the Community currently (2012) pumps (or plans to in the case of proposed wells), and a high-pumping scenario in which the rate was set to the maximum expected from wells installed in this area, according to the Forest County Potawatomi Community Natural Resources Department. In general, the 10-year areas contributing recharge did not intersect surface-water bodies. The 5- and 10-year areas contributing recharge simulated at the maximum pumping rate at Bug Lake Road may intersect Bug Lake. At the casino near the Town of Carter, Wisconsin, the 10-year areas contributing recharge intersect infiltration ponds. At the Devils Lake and Lois Crow Drive wells, areas contributing recharge are near cultural features, including residences.
Wisconsin↗
Approaches in highly parameterized inversion: bgaPEST, a Bayesian geostatistical approach implementation with PEST: documentation and instructions
The application bgaPEST is a highly parameterized inversion software package implementing the Bayesian Geostatistical Approach in a framework compatible with the parameter estimation suite PEST. Highly parameterized inversion refers to cases in which parameters are distributed in space or time and are correlated with one another. The Bayesian aspect of bgaPEST is related to Bayesian probability theory in which prior information about parameters is formally revised on the basis of the calibration dataset used for the inversion. Conceptually, this approach formalizes the conditionality of estimated parameters on the specific data and model available. The geostatistical component of the method refers to the way in which prior information about the parameters is used. A geostatistical autocorrelation function is used to enforce structure on the parameters to avoid overfitting and unrealistic results. Bayesian Geostatistical Approach is designed to provide the smoothest solution that is consistent with the data. Optionally, users can specify a level of fit or estimate a balance between fit and model complexity informed by the data. Groundwater and surface-water applications are used as examples in this text, but the possible uses of bgaPEST extend to any distributed parameter applications.
Techniques and Methods↗
Crowdsourcing to Acquire Hydrologic Data and Engage Citizen Scientists: CrowdHydrology
Spatially and temporally distributed measurements of processes, such as baseflow at the watershed scale, come at substantial equipment and personnel cost. Research presented here focuses on building a crowdsourced database of inexpensive distributed stream stage measurements. Signs on staff gauges encourage citizen scientists to voluntarily send hydrologic measurements (e.g., stream stage) via text message to a server that stores and displays the data on the web. Based on the crowdsourced stream stage, we evaluate the accuracy of citizen scientist measurements and measurement approach. The results show that crowdsourced data collection is a supplemental method for collecting hydrologic data and a promising method of public engagement.
Ground Water↗
Social.Water - A crowdsourcing tool for environmental data acquisition
Remote telemetry has a long history of use for collection of environmental measurements. With the rise of mobile phones and SMS text-messaging capacity, many members of the general pubic carry communications equipment in their pockets at all times. Enabling the general public to provide environmental data through text messages has the potential both to provide additional data to scientific projects and also to raise awareness of the projects through participation. Hydrologic measurements – some of which can be made without training, involve a single measurement, and are often made in rural areas – are well-suited to text-message conveyance. Many other environmental measurements are similarly well-suited for this technology. Social.Water is a software package, written in Python, that collects, parses, and categorizes text messages sent to a dedicated phone number, updates a simple database, and posts both graphical results and the database on the Web. Social.Water was designed as the backend to the Crowdhydrology project and is written in an object-oriented design that makes customization and modification straightforward.
Computers & Geosciences↗
Regression modeling of particle size distributions in urban stormwater: Advancements through improved sample collection methods
A new sample collection system was developed to improve the representation of sediment entrained in urban storm water by integrating water quality samples from the entire water column. The depth-integrated sampler arm (DISA) was able to mitigate sediment stratification bias in storm water, thereby improving the characterization of suspended-sediment concentration and particle size distribution at three independent study locations. Use of the DISA decreased variability, which improved statistical regression to predict particle size distribution using surrogate environmental parameters, such as precipitation depth and intensity. The performance of this statistical modeling technique was compared to results using traditional fixed-point sampling methods and was found to perform better. When environmental parameters can be used to predict particle size distributions, environmental managers have more options when characterizing concentrations, loads, and particle size distributions in urban runoff.
Wisconsin↗
MODFLOW-style parameters in underdetermined parameter estimation
In this article, we discuss the use of MODFLOW-Style parameters in the numerical codes MODFLOW_2005 and MODFLOW_2005-Adjoint for the definition of variables in the Layer Property Flow package. Parameters are a useful tool to represent aquifer properties in both codes and are the only option available in the adjoint version. Moreover, for overdetermined parameter estimation problems, the parameter approach for model input can make data input easier. We found that if each estimable parameter is defined by one parameter , the codes require a large computational effort and substantial gains in efficiency are achieved by removing logical comparison of character strings that represent the names and types of the parameters. An alternative formulation already available in the current implementation of the code can also alleviate the efficiency degradation due to character comparisons in the special case of distributed parameters defined through multiplication matrices. The authors also hope that lessons learned in analyzing the performance of the MODFLOW family codes will be enlightening to developers of other Fortran implementations of numerical codes.
Groundwater↗