USING SUB-SPACE METHODS TO DEFINE INFORMATION TRANSFER FROM CALIBRATION DATA TO MODEL BASED ESTIMATES

Sensitivities calculated by numerical models that link parameters to observations of system state in the course of the calibration exercise can be subjected to singular value decomposition. The observation eigencomponents of the weighted sensitivity matrix, termed “observation eigencomponents” herein, can provide assistance in data worth exploration. Firstly they expose important linkages between those components of the observation datasets which contain the most information, and the parameters these components inform. Secondly their use can overcome an issue that is often ignored in traditional data worth analysis, whereby two complimentary observations can contain significantly more information than the sum of two individual information contents.

Of particular interest is the fact that observation eigencomponents can be used to assess which components of the observation dataset collectively best inform one or a number of model predictions that are salient to management of an environmental system. The relevance of data suites comprising a particular observation eigencomponent to any prediction is exposed most fruitfully through calculating the projection of prediction sensitivities on each observation eigencomponent. This, in turn, allows data worth analysis to be targeted at reducing the uncertainty of predictions of interest by the maximum possible amount, especially in terms of the spatial and temporal distribution of measurements. Furthermore, because the methodology relies on sensitivities rather than actual observation or parameter values, the relative worth of different data acquisition strategies can be ascertained using a model before the data has actually been gathered.

We demonstrate, using the case of the Lockyer Valley water allocation scheme in South East Queensland, how the use of observation information content, forthcoming from this type of analysis, can provide a useful foundation for optimizing both existing monitoring networks and future data acquisition strategies. This is informing model based environmental management in an area where major water import infrastructure solutions are considered to allow for continued groundwater use while maintaining drought buffers and servicing environmental needs.