Uncertainty in flow and water quality measurement data: A case study in the Daning River watershed in the Three Gorges Reservoir region, China

Abstract

A B S T R A C TMeasurement data are used to calibrate and validate the effect of models. In this study, four types of uncertainty categories covering flow measurement, sample collection, sample preservation/storage and laboratory analysis were quantified for flow, sediment and water quality measurement data under best-, typical-and worst-case scenarios. The root mean square error propagation method was used to calculate the overall uncertainty. A case study was conducted in the Daning River with the Soil and Water Assessment Tool. Based on the results, the probable error range values of flow were lower than those of sediment and nutrients, indicating greater uncertainty in the sediment and total phosphorus results. It is also indicated that the uncertainty was smaller when the model outputs had a normal distribution. This research provides scientific data of measurement uncertainty in China, which can assist modelers in evaluating model performance by quantifying the "quality" of the response data.

Improvement of model evaluation by incorporating prediction and measurement uncertainty

“…Some results showed that spatial interpolation techniques resulted in considerable uncertainty of rainfall spatial variability and transferred larger uncertainty to H/NPS modelling. In addition, Shen et al (2013)'s study has been carried out into the effect of GIS data on water quality modelling and the uncertainty related to the combination of the available GIS maps. All these kinds of prediction uncertainty relating to limited model structures, or model input data, could result in discrete variables.…”

“…To incorporate this type of uncertainty, MCA was implemented using the Monte Carlo technique, which has been used in many hydrological uncertainty studies (Sun et al, 2008;Zhang et al, 2016). The Monte Carlo technique is a type of random sampling method that considers combinations of different input components and determines a statistical distribution for the output data (Shen et al, 2013). A key step is sampling variables randomly for discrete data so that the measurement and prediction data can be expressed as certain distributions.…”

“…Thus, model performance might be influenced not only by the error range, but also by the choice of PDF. In this study, the prediction uncertainty was assumed as three certain distribution functions, but it is always difficult to ensure which PDFs should be L. Chen et al: Improvement of model evaluation by incorporating prediction and measurement uncertainty used (Shen et al, 2013). In this section, we further quantified the impacts of the different PDFs on model performance.…”

“…Similar variations in the evaluation results were observed for the MCA method. The NSE for the flow evaluation was 63.5 % for the ideal case (normal distribution) and was 40.9 % and 10.6 % for the typical case and the worst case, respectively (Harmel and Smith, 2007;Shen et al, 2013). For flow prediction, the evaluation Table 3.…”

“…The assignment of PDFs might be the most difficult and subjective task in the application of uncertainty analysis to hydrological models (Shen et al, 2013). Thus, model performance might be influenced not only by the error range, but also by the choice of PDF.…”