Evaluating the information content of multiple groundwater age tracers in projecting nitrate vulnerability

Abstract:

Nitrate is one of the major sources of contamination of groundwater in the United States and around the world. In this study the applicability of multiple groundwater age tracers including ³H, ³He, ⁴He, ¹⁴C, ¹³C, and ⁸⁵Kr in projecting future trends of nitrate concentration in several long-screened, public drinking water wells in Turlock, California, where nitrate concentrations are increasing toward the regulatory limit, is studied. Several lumped parameter models (LPM)s were considered to represent the groundwater age distribution at each well, including binary mixtures between Inverse Gaussian(young) and Dirac(old), generalized inverse Gaussian, and Levy distributions . LPM model parameters and unknown physical parameters (crustal production rate of ⁴He, dissolved inorganic carbon contribution from rock dissolution) were estimated using a Bayesian inference, resulting in the posterior probability distribution of the parameters and therefore the uncertainty associated with each. The performance of each LPM in reproducing the data while accounting for the level of model complexity is evaluated using deviance information criteria (DIC) and Bayes Factors (BF). Historical nitrate concentration data are also evaluated as an additional tracer to refine the age distribution. We found that historical nitrate levels can reduce the uncertainty about the age distribution. LPMs with a distinct feature to represent the old fraction of groundwater (for example Inverse Gaussian-Dirac) are better at reproducing the tracer data but with the price of a larger number of parameters, which results in a larger uncertainty about the age distribution itself. Although the uncertainty regarding the shape of the age distribution remains relatively high, whether nitrate is included as a tracer or not, different models predict similar future trends in nitrate concentration.