Bayesian calibration of a lumped model to estimate catchment nitrate fluxes from monthly monitoring data

Abstract

Nutrient exports from headwater catchments are commonly monitored by monthly water quality sampling at the catchment outlet. Comparison with high resolution stream flow data, however, highlights how poorly these monthly samples capture storm flow conditions, which may represent a significant component of annual load for many nutrients. We used a daily time step lumped-parameter model to estimate water and nutrient contributions arriving at the catchment outlet via near-surface, shallow seasonal groundwater, and deeper older groundwater flow paths in three mesoscale catchments in the Waikato region of New Zealand. Markov Chain Monte Carlo calibration of the model in each catchment, to monthly nitrate-nitrogen and daily stream flow data simultaneously, allowed estimation of annual average fluxes along the different flow paths, as well as the uncertainty in these estimates due to data accuracy and coverage. Results confirmed the relatively high uncertainty in near-surface flux predictions estimated from this type of data. Shallow and deeper groundwater contributions, on the other hand, were able to be estimated relatively accurately. Some suggestions for improved water quality monitoring strategies will be discussed.