Improving SNMR data sensitivity to infiltrating water in the presence of large bodies of surface water

Friday, 19 December 2014
Sam Falzone1, Kristina Keating1, Elliot D Grunewald2 and David O Walsh2, (1)Rutgers University Newark, Newark, NJ, United States, (2)Vista Clara Inc., Mukilteo, WA, United States
Surface nuclear magnetic resonance (SNMR) is a geophysical method used to image water content with depth. Recently SNMR has been used to monitor infiltration events in the vadose zone; however, this application can be complicated by the presence of large signals associated with the ponded surface water. In this study, we develop algorithms to reduce this surface water signal for improved sensitivity to the infiltrated groundwater. Using synthetic models, we examine the accuracy of these algorithms. We then assess our approach using a field dataset collected from a five-week SNMR survey conducted during an infiltration event at the South Aura Valley Storage and Recovery Project (SAVSARP) site in Tucson, AZ.

Three different algorithms were developed to remove the surface water from the SNMR data: (1) late time mono-exponential subtraction, in which signal from late in the measurement is used to model surface water signal; (2) model subtraction, in which the Earth’s magnetic field subsurface conductive structure, and water layer thickness are used to model the surface water signal; and (3) late time inversion correction, in which model parameters in the relaxation time distributions corresponding to slower relaxation times are zeroed. We used two readily available SNMR inversion codes to verify the three approaches: the GMR Inversion software and the MRS Matlab toolkit. Synthetic models were recovered using both inversion codes by applying the late time mono-exponential subtraction and the model subtraction algorithms, while the late time inversion correction algorithm produced poorly resolved relaxation time distribution models. The corrected dataset from the start of the SAVSARP survey contained features in the relaxation time distribution and water content versus depth models that were consistent with observed features present in other datasets from the survey. We conclude that either the late time mono-exponential subtraction or the model subtraction algorithm are sufficient for removing the surface water signal and improving data sensitivity to infiltrated water.