Current Projects Read or Print all Projects

Reconciling Models with Data (System Identification). It is in our nature to build ever more complex, higher-order models: VHOMs, in other words. There are those for whom such a program of model-building constitutes the program of reducing the scientific uncertainties of today. There are also those who, by their attachment to the pragmatism of seeking the robustly preferred decision, above all else and, in particular, above all the uncertainty of environmental modeling and management, might tempt us to give up on eliminating ambiguity in the interpretation of the past. Indeed, with an eye on a future in which VHOMs will be available for reconciliation with high volume, high quality (HVHQ) data, such as those we can now acquire with the Environmental Process Control Laboratory (EPCL), the tasks of system identification look daunting. There is a paradox. For any given system, we can have a VHOM that is not capable of being unambiguously reconciled with the observed record of the past; yet we can have unambiguously identified low order models (LOMs) incapable of supporting a satisfactory theoretical interpretation of the data. Should we indeed give up our quest: to reconcile a theoretically insightful model with the observed data? Our current research is developing an approach to system identification in which the gap between the two maximally different experiences of the system's behavior, {Data} and {Theory}, is to be bridged by a chain of transcriptions, from Data-based Mechanistic (DBM) models to Theory-based Mechanistic (TBM) models. At some point in the chain, in the parametric (not output) spaces of the two types of model, a transcribed interpretation of the {Data} can be compared with the transcribed interpretation of {Theory}. What is more, this comparison may be made all the more effective through the use of novel algorithms of recursive estimation, in particular, those currently being developed around the Recursive Prediction Error (RPE) and Fixed Interval Smoothing (FIS) algorithms. And these, in turn, can have very practical outcomes, for the detection of faults in the instrumentation of networks for monitoring water quality in real time, specifically in the Lagoon of Venice (through the Chair's collaboration with the University of Venice).