Basic Research Needs for Geosciences - Energetics Meetings and ...

PANEL REPORT: MODELING AND SIMULATION OF GEOLOGIC SYSTEMSModeling complex systems with diverse data setsCurrent practice in inverse modeling tends to decouple processes, to aggregate parameters acrossscales, and to include only a limited amount of the available, real data. It is anticipated thatadvances in characterization methods will produce increasingly larger data sets as the spatialextent and the spatial and temporal resolutions of measurements increase. In addition, with thedevelopment of new measurement devices, data types not currently commonplace may becomeroutinely available. The scientific challenge is to incorporate these larger data volumes anddisparate data types into inverse modeling while simultaneously satisfying diverse physical andchemical constraints. In particular, in the context of monitoring perturbed geological systems,models will have to handle streams of different data types measured over a variety of spatial andtemporal resolutions, including data that are only indirectly related to system state variables ofinterest and model parameters. The computational burden of such comprehensive inversemodeling requires fundamental advances in mathematical methods and numerical algorithms.Revolutionary advances in the modeling workflow are required to achieve automatic, real-timeincorporation of diverse data sets into models.Hubbard and Rubin (2005) described several obstacles that hinder the routine use of geophysicsfor quantitative estimation of subsurface properties, including limitations inherent in geophysicaldata inversion and associated artifacts, integration or joint inversion of datasets that sampledifferent properties, petrophysical models, scaling issues, and non-unique geophysical responsesto heterogeneities. In spite of these obstacles, recent synthetic and field studies have illustratedhow geophysical data improve the prediction of transport and contaminant remediation in naturalsystems through better model parameterization and validation (e.g., Scheibe and Chien 2003;Scheibe et al. 2006). For example, geophysical methods have been used to provide highresolutionestimates of subsurface properties; some examples of hydrogeophysical estimation areshown in Figure 19.Quantifying and reducing predictive uncertaintyUncertainty in model results arises primarily from lack of information, inconsistencies betweenthe available information, and the modeled quantities, and errors in the existing information.Quantifying the uncertainty requires a careful accounting of these contributions, numericalmethods that are accurate and robust, use of optimization methods for model calibration, and useof sensitivity methods to reveal important information deficiencies. Recent and anticipatedadvances in data collection are expected to increase the demands on this process. Despitesignificant advances in recent years, the fields of model calibration, sensitivity analysis, dataneeds assessment, long-term monitoring, and uncertainty evaluation require fundamentalimprovements to optimally use anticipated data sets while fully accounting for multiscalefeatures and nonlinearly coupled processes.At the more practical level of decision making in a regulatory context, additional benefits wouldaccrue from the development of these methods. In both CO 2 storage and nuclear waste disposal,operators are obligated to demonstrate that decisions based on models are appropriate. In today’sterminology, this generally means that models must be demonstrated to be “valid,” that is, shownto be consistent with data. The concept of model validity is fraught with philosophicalBasic Research Needs for Geosciences: Facilitating 21 st Century Energy Systems 57