Bayesian Nitrate Source Apportionment to Individual Groundwater Wells in the Central Valley by Use of Elemental and Isotopic Tracers
Groundwater quality is a concern in alluvial aquifers that underlie agricultural areas, such as in the San Joaquin Valley of California. Nitrate from fertilizers and animal waste can leach to groundwater and contaminate drinking water resources. Dairy manure and synthetic fertilizers are prevailing sources of nitrate in groundwater for the San Joaquin Valley with septic waste contributing as a major source in some areas. The rural population in the San Joaquin Valley relies almost exclusively on shallow domestic wells (less than 150 m deep), of which many have been affected by nitrate. Knowledge of the proportion of each of the three main nitrate sources (manure, synthetic fertilizer, and septic waste) contributing to individual well nitrate can aid future regulatory decisions. Mixing models quantify the proportional contributions of sources to a mixture by using the concentration of conservative tracers within each source as a source signature. Deterministic mixing models are common, but do not allow for variability in the tracer source concentration or overlap of tracer concentrations between sources. In contrast, Bayesian mixing models treat source contributions probabilistically, building statistical variation into the inferences for each well. The authors developed a Bayesian mixing model on a pilot network of 56 private domestic wells in the San Joaquin Valley for which nitrogen, oxygen, and boron isotopes as well as nitrate and iodine were measured. Nitrogen, oxygen, and boron isotopes as well as iodine can be used as tracers to differentiate between manure, fertilizers, septic waste, and natural sources of nitrate (which can contribute nitrate in concentrations up to 4 mg/L). In this work, the isotopic and elemental tracers were used to estimate the proportional contribution of manure, fertilizers, septic waste, and natural sources to overall groundwater nitrate concentration in individual wells. Prior distributions for the four tracers for each of the four sources were estimated based on end member measurements, literature, or as a part of our previous work. The Bayesian method produces estimates of the fractional source contributions to each well, which can be compared to surrounding landuse types. Estimated source contributions were broadly consistent with nearby landuse types in this sample.