Modelling changes in river water quality, and by extension developing river management strategies, has historically been reliant on empirical data collected at relatively low temporal resolutions. With access to data collected at higher temporal resolutions, this study investigated how these new dataset types could be employed to assess the precision and accuracy of two phosphorus (P) load apportionment models (LAMs) developed on lower resolution empirical data. Predictions were made of point and diffuse sources of P across ten different sampling scenarios. Sampling resolution ranged from hourly to monthly through the use of 2000 newly created datasets from high frequency P and discharge data collected from a eutrophic river draining a 9.48 km2 catchment. Outputs from the two LAMs were found to differ significantly in the P load apportionment (51.4% versus 4.6% from point sources) with reducing precision and increasing bias as sampling frequency decreased. Residual analysis identified a large deviation from observed data at high flows. This deviation affected the apportionment of P from diffuse sources in particular. The study demonstrated the potential problems in developing empirical models such as LAMs based on temporally relatively poorly-resolved data (the level of resolution that is available for the majority of catchments). When these models are applied ad hoc and outside an expert modelling framework using extant datasets of lower resolution, interpretations of their outputs could potentially reduce the effectiveness of management decisions aimed at improving water quality.
Keywords: Agriculture; High frequency data; Modelling; Phosphorus; Pollution; Water quality.