Uncertainties on nitrate water quality indicators associated with infrequent sampling in Brittany, France

Excessive nutrient export in watersheds has led to the development of large monitoring programs and the implementation of Best Management Practices throughout the western countries. Documenting water quality improvement or degradation is not an easy endeavor. On a particular site, perceived improvement or degradation may result from climate variability from year to year. High uncertainties may actually result from the actual sampling and monitoring design, and particularly on the frequencies at which water is sampled and analyzed. We conducted an analysis of uncertainties on nitrate fluxes and concentration indicators (average, median, 90th and 95th percentiles, maximum) associated with infrequent sampling in Brittany, France. We used a database of 50 watershed-year datasets for which high temporal resolution data (hourly, daily at worst) was available for flow and nitrate concentrations. For each dataset, we calculated yearly reference fluxes and indicators. We then numerically simulated sampling for a set frequency and calculation strategy and compared the result to the reference ones. The choice of the algorithm used to compute the nitrate fluxes largely determines the expected accuracy and imprecision of the strategy. All methods that do not use the continuous record of flow performed very poorly and we recommend not to use them. The flow weighted average concentration ratio method was showed to best perform across the 50 datasets in Brittany. Ranges of errors were found to be correlated to a flow duration indicator, which lead to the drawing of reference guideline curves. These may serve as a template to determine, for a given watershed and its hydrological reactivity, expected errors for a given sampling frequency or to better design sampling frequency for a given uncertainty level. Annual values computed using monthly sampling showed that for a moderately ‘flashy’ watershed nitrate flux estimates would be between -12% and +11% of the ‘true’ load, average and 95th percentile concentrations would be within ±6% and between -7% and +2% of the reference value, respectively.