Improvement of model evaluation by incorporating prediction and measurement uncertainty

Abstract

Numerous studies have been conducted to assess uncertainty in hydrological and non-point source pollution predictions, but few studies have considered both prediction and measurement uncertainty in the model evaluation process. In this study, the cumulative distribution function approach (CDFA) and the Monte Carlo approach (MCA) were developed as two new approaches for model evaluation within an uncertainty condition. For the CDFA, a new distance between the cumulative distribution functions of the predicted data and the measured data was established in the model evaluation process, whereas the MCA was proposed to address conditions with dispersed data points. These new approaches were then applied in combination with the Soil and Water Assessment Tool in the Three Gorges Region, China. Based on the results, these two new approaches provided more accurate goodness-of-fit indicators for model evaluation compared to traditional methods. The model performance worsened when the error range became larger, and the choice of probability density functions (PDFs) affected model performance, especially for non-point source (NPS) predictions. The case study showed that if the measured error is small and if the distribution can be specified, the CDFA and MCA could be extended to other model evaluations within an uncertainty framework and even be used to calibrate and validate hydrological and NPS pollution (H/NPS) models.