By Neil Manewell
AGE have been using automated calibration techniques for a number of years, generally choosing singular value decomposition (SVDA assist with PEST) to orthogonally project complex parameter interactions to improve the efficiency of the calibration process.
The PEST automation process relies on gradient based improvements to the objective function (a measure of the difference between observed and simulated calibration targets), which means calibration is progressed by trying to find the nearest optimum. Groundwater models are usually highly non-linear, with numerous local optimum, and a single global optimum. Depending on the starting conditions assigned to the parameter values, automated software can easily be ill informed, meaning the true global optima may not be defined.
AGE have developed a process whereby a ‘shuffled complex evolution’ method is applied to find the true ‘global optimum’, resulting in the best calibration statistics achievable. It also leads to thorough exploration of possible aquifer parameter fields. Enquiries can be directed to [email protected]