Successive bootstrapping deep learning approach and airborne EM-borehole data fusion to understand salt water in the Mississippi River Valley Alluvial Aquifer

Abstract

Increasing demands from agriculture and urbanization have decreased groundwater level and increased salinity worldwide. Better aquifer characterization and soil salinity mapping are important for proactive groundwater management. Airborne electromagnetic (AEM) is a powerful tool for aquifer characterization and salinity delineation. However, AEM needs to be interpreted with caution before being used for groundwater quality analysis. This study introduces a framework that utilizes the AEM data for both lithologic modeling and salinity delineation. A resistivity-to-lithology (R2L) model is developed to interpret AEM resistivity to lithology based a depth-dependent multi-resistivity thresholds. Then, a cokriging method is used to integrate AEM data from two different EM systems to predict resistivity at the aquifer. Finally, a resistivity-to-chloride concentration (R2C) model utilizes the resistivity model to estimate chloride concentrations at sand facies. A deep learning artificial neural network (DL-ANN) model is introduced with a successive bootstrapping approach to estimate total dissolved solids first and then use it together with resistivity data to estimate chloride concentration. The methodology was applied to delineating salinity plumes in the Mississippi River Valley alluvial aquifer (MRVA). This study found that the salinity distribution in MRVA is highly correlated with the Jurassic salt basin, salt domes, faulting, seismicity, and river water quality. The result indicates salinity upconing due to excessive pumping.