Author(s): R. Khatibi; D. Jackson; G. Pender; C. Whitlow; M. Werner; T. Harrison
Linked Author(s): Rahman Khatibi, Gareth Pender
Keywords: No Keywords
Abstract: Calibration has been an integral part of modelling since its emergence in the late 1950s as a design and decision-making tool. Later research works determined that calibration is an “inverse problem” and a distinction was made between calibration lacking a mathematical rigour and “identification” with a mathematical rigour. However, defensibility of traditional inverse problems is becoming an issue if inherent deterministic methods are not reformed. This paper critically reviews inverse problems to unravel some of the latent assumptions, which often remain unchallenged in deterministic modelling practices. These include (i) the selection of the modelling solution and (ii) a range of other assumptions in model building procedures. The paper discusses a tentative best practice in inverse problems composed of (i) a mechanism to challenge the above assumptions by an iterative procedure, (ii) making use of parameters quality conditions of identifiability, uniqueness and stability and (iii) treating their inherent uncertainties through Monte Carlo simulations and Bayesian approaches.
Year: 2003