Evaluating the Presence of Preferential Flow Using Remote Sensing, Inverse Modeling, and Machine Learning Algorithms

Contaminant fate and transport via Preferential Flow (PF) has been shown to increase Groundwater (GW) contamination due to the rapid flow of fluid and solute through continuous channels, bypassing the natural filtration within the active upper layer of the soil matrix. As the number of GW wells across the Continental United States (CONUS) report increased levels of contamination, improved techniques and technologies for tracking PF are becoming more essential. The lack of a complete understanding for PF behaviors stems from the complexity behind studying this phenomenon. To understand the mechanisms of PF, a significant number of resources such as time, money and labor lie behind in-situ and laboratory experimentation. Currently, these technologies are unable to provide complete data for PF across the entire CONUS effectively and efficiently. In this study, we aim to use remote sensing as a tool to characterize the effective PF above shallow water tables (~2 m from land surface) across CONUS and confirm our analyses through GW level predictions. Using a combination of an inverse modeling approach using the dual-porosity model in HYDRUS-1D, SMAP 9 km soil moisture, surface temperature, and vegetation opacity remote sensing data, static signatures of PF such as cation exchange capacity and soil organic carbon, and a multiple linear regression machine learning algorithm, we developed a new PF estimation framework. This algorithm resulted in successful GW level predictions for all the study sites across all four seasons in the year. Overall, with and values of 0.71 and 59 cm for Spring, 0.77 and 50 cm for Summer, 0.74 and 56 cm for Fall, and 0.81 and 52 cm for Winter; respectively. Additionally, the performance of GW simulations in HYDRUS-1D were improved by using the dual-porosity model replacing the originally implemented single-porosity model, reporting an increase in the value from nearly 0 to 0.85. Results from this study provide momentum towards the improved understanding of PF and the knowledge to develop efficient land management practices such as effective application of fertilizer which may otherwise bypass the root zone via PF and lead to increased GW contamination.