Uncertainty analysis at large scales: limitations and subjectivity of current practices - a water quality case study

Abstract

Uncertainty analysis for large-scale model studies is a challenging activity that requires a different approach to uncertainty analysis at a smaller scale. However, in river basin studies, the practice of uncertainty analysis at a large scale is mostly derived from practice at a small scale. The limitations and inherent subjectivity of some current practices and assumptions are identified, based on the results of a quantitative uncertainty analysis exploring the effects of input data and parameter uncertainty on surface water nutrient concentration. We show that: (i) although the results from small- scale sensitivity analysis are often applied at larger scales, this is not always valid; (ii) the current restriction of the uncertainty assessment to uncertainty types with a strong evidence base gives structurally conservative estimates; (iii) uncertainty due to bias is usually not assessed, but it may easily outweigh the effects of variability; (iv) the uncertainty bandwidth may increase for higher aggregation levels, although the opposite is the standard assumption.

Advances in the study of uncertainty quantification of large-scale hydrological modeling system

“…Uncertainty quantification for a large-scale model is a challenging activity that requires a different approach to uncertainty analysis at a relatively small scale (Bijlsma et al, 2007). And Bijlsma et al (2007) stated that the some limitations and inherent subjectivity existed, and it was not always valid when the practice of uncertainty analysis approach at a large scale was derived from a small scale.…”

“…And Bijlsma et al (2007) stated that the some limitations and inherent subjectivity existed, and it was not always valid when the practice of uncertainty analysis approach at a large scale was derived from a small scale. Also, the current restriction of uncertainty assessment was mostly based the given structurally conservative estimates, and the unknown bias was usually not assessed though it may easily outweigh the effects of variability.…”