Reproducibility of hydraulic tomography estimates and their predictions: A two-year case study in Taiwan

Abstract Over the past decades, a new aquifer test technology (sequential pumping tests or hydraulic tomography, HT) has been developed and successfully applied to many field sites to delineate the spatial distributions of hydraulic properties (e.g., transmissivity ( T ) and storage coefficient ( S )). Yet, the reproducibility of its estimated T and S fields and the predictive capabilities of the estimates for different flow scenarios at different time periods remain unexplored. That is to say, if the estimated fields based on sequential pumping tests conducted during different years are the same since the geologic formation and processes may have undergone changes. In order to answer this important question, this study first compares the drawdown-time behaviors from the sequential pumping tests (SPTs) conducted in 2010 with those conducted in 2012 at a field site and then finds they are similar but different in detail. It then uses these data to estimate the T and S fields and checks the reproducibility of the estimates. The estimated heterogeneity patterns are found to be generally reproducible in spite of uncertainties. In addition, the estimates from each year are capable of predicting the observed drawdowns, induced by independent pumping tests during the corresponding year (i.e., self-validation). Moreover, the estimated fields are cross-validated. That is, this study uses the estimates obtained from the 2010 pumping tests to predict the observed drawdowns of the independent pumping tests conducted in 2012. Likewise, it uses the estimates from 2012 pumping tests to forecast the drawdowns of the independent pumping tests of 2010. The results of both self-validation and cross-validation indicate that the estimated T and S fields based on the test in one year can be used to predict bulk flow behavior in the other year. Differences in detailed behaviors may be attributed to changes in the processes, omitted in the depth-averaged flow model.