Upscaling Submarine Groundwater Discharge Using Local Long-Term High-Resolution Radon Measurements and Deep Learning

Abstract:

The significance of submarine groundwater discharge (SGD) to local coastal water and chemical budgets has been documented worldwide. Because SGD is spatiotemporally variable and measurements tend to be scale dependent, upscaling local data to regional or global extents can be challenging. For instance, ²²²Rn (radon), a naturally occurring radionuclide, is frequently used as geochemical tracer for SGD but only provides information on local scales. In this study, we explore one possible way to take advantage of coastal radon data to project SGD on a regional scale. We use a long-term hourly resolution SGD dataset collected by an autonomous gamma spectrometer that was deployed for four years off the Kona coast of Hawaiʻi. Data collected include coastal radon activities, salinity, and water temperature, which are coupled with metrological, groundwater level, and oceanic datasets. These data were used to train a deep neural network that accurately predicted test data with a normalized mean absolute error of less than 2x10^-2. Using this model, we explore a possible path for upscaling of SGD to regional and global scales with evidence from multiple locations within the Hawaiian Islands. This research combines field-based measurements with big-data approaches and demonstrates the efficacy of these methods for SGD modeling beyond local scales.