A quantitative approach of the impact of sampling frequency on uncertainties associated with the estimation of the annual nitrate load and average concentration in Brittany, France

Abstract

The mission of most water quality monitoring networks is to evaluate the long term water quality pattern of water bodies. These networks often use compiled indicators (annual average concentration, maximum concentration, annual exported flux, etc.) calculated from a limited number of measurements. Infrequent measurements induce uncertainties on the indicator values. These uncertainties must be known to better interpret observed patterns. The objective of this article is to evaluate, using scientific approaches, uncertainties on nitrate indicators induced by infrequent sampling, and by the algorithms used to compute fluxes. A 50 year-basin data set of hourly to daily flow and concentration data gathered from nine watersheds (5 to 252 km2 in size) in Brittany, France, was analyzed to evaluate the uncertainties. Original data was numerically sampled to simulate common sampling frequencies including bimonthly, monthly and bimestrial frequencies. Water quality indicators (annual concentration average, maximum, median, 90th and 95th percentile and flux) calculated from the simulated samples were compared to the reference indicators calculated from the high frequency original data. Several algorithms proposed in the literature to calculate annual fluxes were tested. Results show that for all indicators, uncertainties increase as sampling intervals increase. Results also show that all the algorithms that do not use continuous flow data to compute nitrate fluxes yield very large uncertainties, even for biweekly sampling intervals, prompting the recommendation not to use these types of algorithms. Simulation results of watersheds with average hydrological reactivity show that monthly sampling procedures (typical sampling frequency in many monitoring networks) yield uncertainties ranging from – 6 to +4 % for the annual average concentration indicator and uncertainties ranging from –12 to +11 % for the annual flux. The boundary values of the uncertainty ranges were found to be correlated to an indicator of hydrological reactivity of the watersheds. This flow duration indicator was the percentage of flow occurring in 2 % of the time corresponding to the highest flow. Using these correlations, guideline curves were drawn relating sampling frequencies, uncertainty levels and hydrological reactivity of watersheds. Using these curves, the general perception that average concentration and flux values obtained using the existing sampling scheme (monthly or bimestrial intervals) in the Brittany water quality monitoring networks are quite reliable is validated. However, uncertainties become very high for watersheds of high hydrological reactivity including artificially drained watersheds. The guideline curves offer for the first time a tool that can be used to optimize sampling schemes of surface water monitoring networks. This is particularly valuable in light of the tightening of existing monitoring scheme resulting from the implementation
of the European Water Framework Directive.