How should I set the target measurement objective function (i.e. the PEST PHIMLIM control variable)?
Often this is a matter of trial and error. Ideally, it should be set at a level that is commensurate with the level of measurement noise. For example, if all of your observations have a weight that is the inverse of expected measurement noise, then the expected average value of each squared residual is about 1.0; the expected value of the objective function should then be about equal to the number of observations. The trouble is, most observation “noise” that we encounter when calibrating a model is in fact model structural noise. A model simply cannot replicate every nuance of system behaviour, and we are often quite happy to consider our model calibrated with a model-to-measurement misfit which is considerably greater than that which would be incurred by measurement noise alone. Furthermore we normally find out what this fit actually is (and hence the level of structural noise that actually exists) during the calibration process itself. In classical parameter estimation where we use
