Year: 2005
Type: conference-paper
Abstract: Parameter calibration and prediction uncertainty of a model are intimately related. Although much literature is devoted to calibration and uncertainty analysis in hydrologic modelling, a clear definition of when a model is calibrated is generally lacking. In recent years, inverse modelling has become a popular alternative to direct parameter measurement. However, in contrast to obtaining model outputs (model variables) from model inputs (model parameters), which is unique, obtaining model parameters from model variables is non-unique by nature. Inversely obtained hydrologic parameters, therefore, are always uncertain (non-unique) because of parameter correlations, errors associated with the measurements, and Level of details and simplifications in the models, among other factors. Quantification of the uncertainty in a calibrated model is vital for a meaningful application of the model. Optimization of a goal (objective) function in a problem with many parameters is often difficult because of the complexities in the structure of the goal function. One complexity is the large number of local minima associated with any given goal function. To draw an analogy, the space of the goal function, g, could be likened to a block of "Swiss cheese" (if simplified to two parameters only) (Fig. 1) with many holes. Each hole represents a local minimum, with the size of the hole in any direction representing the range of uncertainty.
Author: Karim C. Abbaspour
Language: English
Subfield: Water Science and Technology
Field: Environmental Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W3036665271
