Radar interferometry provides for the acquisition of basin wide high precision vertical deformation data that reveal the spatially complex and structurally dependent nature of land subsidence in heavily pumped sedimentary basins. In Las Vegas Valley, for example, InSAR interferograms portray compartmentalized subsidence bowls bounded by basin-fill faults. Earth fissures are known to occur adjacent to many such faults where differential subsidence is observed from interferograms. Recovering water levels associated with reduced groundwater pumping during summer months and a rigorous ASR program during winter months have resulted in seasonal patterns of subsidence and rebound that are reflected in the InSAR time-series data. A new modeling strategy is proposed whereby these seasonal deformation patterns are coupled with observed water-level data to quantify the storage characteristics of the aquifer and confining units at a pixel resolution of the interferogram, or about 40m. Inverse models that use these observations are hindered by the fact that parameter zone distributions for storage and hydraulic conductivity are typically user defined and often formulated in an ad-hoc fashion. The objective of this proposed research is to develop an adjoint-based parameter estimation model that systematically produces the optimal storage and conductivity zone distributions that leads to a superior conceptual model and yields not only the best parameter distribution, but provides the details necessary to reflect the intricacies of the fault-bounded storage bowls observed in Las Vegas Valley.
The broader impacts of this proposed research are significant because they address the socioeconomic and hydrogeologic problems facing Las Vegas Valley using a multi-disciplinary approach. The economic well being of Las Vegas Valley is dependent on long-term growth and management of developable land. This research will provide a high-resolution groundwater and subsidence model that can be used as a water-management tool well into the future. The adjoint-based parameter estimation model (APE) developed in this research automates zones used in inverse modeling. This new model provides a distinct advantage over techniques requiring analysis of numerous alternative models. This new modeling package will be made available to all modeling practitioners. An Outreach Coordinator will use the VT Museum of Geosciences to provide groundwater and water resources education to audiences that include pK-12 students, teachers, undergraduates and the general public. A physical groundwater model will be used during planned workshops to instruct teachers and demonstrate safe water practices. A poster and worksheet will be developed for high school classrooms and for the Museum to highlight Virginia Tech groundwater research in an effort to attract more future groundwater scientists.
We developed an adjoint parameter estimation (APE) algorithm to automate the creation of parameter zonations, which has historically been done ad hoc based on best "guesses". The new algorithm uses a multilevel approach to evaluate whether the initial zone should be either increased or decreased in size based on minimization of an objective function based on observations of both hydraulic head and land subsidence. We evaluated our algorithm against a number of other methods such as pilot points, and markov chain monte carlo. Our algorithm is shown to be more computationally efficient and accurate, leading to a parameter zonation that leads to far more accurate simulated heads and subsidence patterns. The way this works is an initial parameter zonation is provided. Modflow-2005 is run to obtain simulated heads and subsidence, UCODE_2005 is run to obtain an objective function and sensitivities. Then our APE algorithm is run to refine the zonations. This process is repeated until the objective function is minimized (global minimum as opposed to a local minimum). A second outcome is that we've developed a new basin-wide groundwater and subsidence model for Las Vegas Valley, which has a much finer grid than has been previously used. We have also incorporated basin-fill faults, which we know influences subsidence patterns. We are currently completing a final water-management model for the entire basin, based on our new parameter zonations from our algorithm. The final model represents the most detailed and robust model that has ever been developed for Las Vegas Valley. In terms of broader impacts, the new APE algorithm can be applied to any model in which parameter estimation is used. No longer will the user have to manually guess or assume what the parameter zonations look like based on best estimates of hydrogeological conditions. Rather, the algorithm will automatically refine the zones by minimizing the residuals of observations of hydraulic head and land subsidence measurements. Thus, this outcome is a powerful new tool that can greatly aid modelers.
