The delineation of hydraulic conductivity (K) heterogeneity is essential to support effective remediation of environmental sites and sustainability management of groundwater basins. However, K characterization by conventional methods is a difficult and expensive task at sites with complex hydrogeology. Inadequate K characterization has resulted in poor remediation performance at many legacy environmental sites and excessive potentiometric head decline at many over-drafted wellfields. Recently, a technique based on sequential pumping tests and stochastic hydraulic tomography (HT) inversion using successive linearization estimator (SLE) has been demonstrated to be effective for delineating K heterogeneity The HT technique was initially developed to treat the K-distribution in a groundwater model as a correlated random field and to invert the model stochastically using the hydraulic head response data from individual aquifer pumping/injection tests. Although many environmental sites with pump-and-treat systems and water suppliers with production wellfields have collected an abundant amount of operational and monitoring data, sequential pumping tests by temporarily shutting down individual wells might not have been performed. Performing such tests might not be possible due to operational constraints. Even if they have been performed, the duration of operational changes might have been too short. The monitoring data during a temporary operational change always contain the complex signature of previous operations prior to the change. Although all historical data can be theoretically utilized for HT inversion, the computational effort needed is practically infeasible.
We will present a computationally efficient approach based on a combination of Transfer Function Noise (TFN) analysis and Hydraulic Tomography (HT) analysis to estimate the spatial K-distribution and the associated uncertainty using abundant extraction/injection operational and monitoring data. We will use the operation and monitoring data collected at an environmental site in Arizona as an example. The TFN technique can be applied to estimate the hydraulic head response to constant-rate pumping/injection at each relevant well through convolution integration. Such step-response functions are equivalent to the data collected from individual aquifer pumping/injection test without carrying any long-duration pumping/injection operations signals and can be utilized directly in HT analysis. We will present an efficient approach to simplify the HT computations by incorporating principal component analysis (PCA) to reduce the numbers of parameters and the number of calibration targets. This method allows for the removal of both parameter and calibration target dependencies without losing significant information. In addition, we will present a Markov Chain Monte Carlo (MCMC) approach for HT inversion. This approach can be easily adopted to use any forward simulation models.